]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
81819f0f CL |
2 | /* |
3 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
4 | * objects in per cpu and per node lists. | |
5 | * | |
881db7fb CL |
6 | * The allocator synchronizes using per slab locks or atomic operatios |
7 | * and only uses a centralized lock to manage a pool of partial slabs. | |
81819f0f | 8 | * |
cde53535 | 9 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 10 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
11 | */ |
12 | ||
13 | #include <linux/mm.h> | |
1eb5ac64 | 14 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
15 | #include <linux/module.h> |
16 | #include <linux/bit_spinlock.h> | |
17 | #include <linux/interrupt.h> | |
18 | #include <linux/bitops.h> | |
19 | #include <linux/slab.h> | |
97d06609 | 20 | #include "slab.h" |
7b3c3a50 | 21 | #include <linux/proc_fs.h> |
81819f0f | 22 | #include <linux/seq_file.h> |
a79316c6 | 23 | #include <linux/kasan.h> |
81819f0f CL |
24 | #include <linux/cpu.h> |
25 | #include <linux/cpuset.h> | |
26 | #include <linux/mempolicy.h> | |
27 | #include <linux/ctype.h> | |
3ac7fe5a | 28 | #include <linux/debugobjects.h> |
81819f0f | 29 | #include <linux/kallsyms.h> |
b9049e23 | 30 | #include <linux/memory.h> |
f8bd2258 | 31 | #include <linux/math64.h> |
773ff60e | 32 | #include <linux/fault-inject.h> |
bfa71457 | 33 | #include <linux/stacktrace.h> |
4de900b4 | 34 | #include <linux/prefetch.h> |
2633d7a0 | 35 | #include <linux/memcontrol.h> |
2482ddec | 36 | #include <linux/random.h> |
81819f0f | 37 | |
4a92379b RK |
38 | #include <trace/events/kmem.h> |
39 | ||
072bb0aa MG |
40 | #include "internal.h" |
41 | ||
81819f0f CL |
42 | /* |
43 | * Lock order: | |
18004c5d | 44 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
45 | * 2. node->list_lock |
46 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 47 | * |
18004c5d | 48 | * slab_mutex |
881db7fb | 49 | * |
18004c5d | 50 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
51 | * and to synchronize major metadata changes to slab cache structures. |
52 | * | |
53 | * The slab_lock is only used for debugging and on arches that do not | |
b7ccc7f8 | 54 | * have the ability to do a cmpxchg_double. It only protects: |
881db7fb | 55 | * A. page->freelist -> List of object free in a page |
b7ccc7f8 MW |
56 | * B. page->inuse -> Number of objects in use |
57 | * C. page->objects -> Number of objects in page | |
58 | * D. page->frozen -> frozen state | |
881db7fb CL |
59 | * |
60 | * If a slab is frozen then it is exempt from list management. It is not | |
632b2ef0 LX |
61 | * on any list except per cpu partial list. The processor that froze the |
62 | * slab is the one who can perform list operations on the page. Other | |
63 | * processors may put objects onto the freelist but the processor that | |
64 | * froze the slab is the only one that can retrieve the objects from the | |
65 | * page's freelist. | |
81819f0f CL |
66 | * |
67 | * The list_lock protects the partial and full list on each node and | |
68 | * the partial slab counter. If taken then no new slabs may be added or | |
69 | * removed from the lists nor make the number of partial slabs be modified. | |
70 | * (Note that the total number of slabs is an atomic value that may be | |
71 | * modified without taking the list lock). | |
72 | * | |
73 | * The list_lock is a centralized lock and thus we avoid taking it as | |
74 | * much as possible. As long as SLUB does not have to handle partial | |
75 | * slabs, operations can continue without any centralized lock. F.e. | |
76 | * allocating a long series of objects that fill up slabs does not require | |
77 | * the list lock. | |
81819f0f CL |
78 | * Interrupts are disabled during allocation and deallocation in order to |
79 | * make the slab allocator safe to use in the context of an irq. In addition | |
80 | * interrupts are disabled to ensure that the processor does not change | |
81 | * while handling per_cpu slabs, due to kernel preemption. | |
82 | * | |
83 | * SLUB assigns one slab for allocation to each processor. | |
84 | * Allocations only occur from these slabs called cpu slabs. | |
85 | * | |
672bba3a CL |
86 | * Slabs with free elements are kept on a partial list and during regular |
87 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 88 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
89 | * We track full slabs for debugging purposes though because otherwise we |
90 | * cannot scan all objects. | |
81819f0f CL |
91 | * |
92 | * Slabs are freed when they become empty. Teardown and setup is | |
93 | * minimal so we rely on the page allocators per cpu caches for | |
94 | * fast frees and allocs. | |
95 | * | |
aed68148 | 96 | * page->frozen The slab is frozen and exempt from list processing. |
4b6f0750 CL |
97 | * This means that the slab is dedicated to a purpose |
98 | * such as satisfying allocations for a specific | |
99 | * processor. Objects may be freed in the slab while | |
100 | * it is frozen but slab_free will then skip the usual | |
101 | * list operations. It is up to the processor holding | |
102 | * the slab to integrate the slab into the slab lists | |
103 | * when the slab is no longer needed. | |
104 | * | |
105 | * One use of this flag is to mark slabs that are | |
106 | * used for allocations. Then such a slab becomes a cpu | |
107 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 108 | * freelist that allows lockless access to |
894b8788 CL |
109 | * free objects in addition to the regular freelist |
110 | * that requires the slab lock. | |
81819f0f | 111 | * |
aed68148 | 112 | * SLAB_DEBUG_FLAGS Slab requires special handling due to debug |
81819f0f | 113 | * options set. This moves slab handling out of |
894b8788 | 114 | * the fast path and disables lockless freelists. |
81819f0f CL |
115 | */ |
116 | ||
af537b0a CL |
117 | static inline int kmem_cache_debug(struct kmem_cache *s) |
118 | { | |
5577bd8a | 119 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 120 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 121 | #else |
af537b0a | 122 | return 0; |
5577bd8a | 123 | #endif |
af537b0a | 124 | } |
5577bd8a | 125 | |
117d54df | 126 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be JK |
127 | { |
128 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) | |
129 | p += s->red_left_pad; | |
130 | ||
131 | return p; | |
132 | } | |
133 | ||
345c905d JK |
134 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
135 | { | |
136 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
137 | return !kmem_cache_debug(s); | |
138 | #else | |
139 | return false; | |
140 | #endif | |
141 | } | |
142 | ||
81819f0f CL |
143 | /* |
144 | * Issues still to be resolved: | |
145 | * | |
81819f0f CL |
146 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
147 | * | |
81819f0f CL |
148 | * - Variable sizing of the per node arrays |
149 | */ | |
150 | ||
151 | /* Enable to test recovery from slab corruption on boot */ | |
152 | #undef SLUB_RESILIENCY_TEST | |
153 | ||
b789ef51 CL |
154 | /* Enable to log cmpxchg failures */ |
155 | #undef SLUB_DEBUG_CMPXCHG | |
156 | ||
2086d26a CL |
157 | /* |
158 | * Mininum number of partial slabs. These will be left on the partial | |
159 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
160 | */ | |
76be8950 | 161 | #define MIN_PARTIAL 5 |
e95eed57 | 162 | |
2086d26a CL |
163 | /* |
164 | * Maximum number of desirable partial slabs. | |
165 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 166 | * sort the partial list by the number of objects in use. |
2086d26a CL |
167 | */ |
168 | #define MAX_PARTIAL 10 | |
169 | ||
becfda68 | 170 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 171 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 172 | |
149daaf3 LA |
173 | /* |
174 | * These debug flags cannot use CMPXCHG because there might be consistency | |
175 | * issues when checking or reading debug information | |
176 | */ | |
177 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
178 | SLAB_TRACE) | |
179 | ||
180 | ||
fa5ec8a1 | 181 | /* |
3de47213 DR |
182 | * Debugging flags that require metadata to be stored in the slab. These get |
183 | * disabled when slub_debug=O is used and a cache's min order increases with | |
184 | * metadata. | |
fa5ec8a1 | 185 | */ |
3de47213 | 186 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 187 | |
210b5c06 CG |
188 | #define OO_SHIFT 16 |
189 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 190 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 191 | |
81819f0f | 192 | /* Internal SLUB flags */ |
d50112ed | 193 | /* Poison object */ |
4fd0b46e | 194 | #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U) |
d50112ed | 195 | /* Use cmpxchg_double */ |
4fd0b46e | 196 | #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U) |
81819f0f | 197 | |
02cbc874 CL |
198 | /* |
199 | * Tracking user of a slab. | |
200 | */ | |
d6543e39 | 201 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 202 | struct track { |
ce71e27c | 203 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
204 | #ifdef CONFIG_STACKTRACE |
205 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
206 | #endif | |
02cbc874 CL |
207 | int cpu; /* Was running on cpu */ |
208 | int pid; /* Pid context */ | |
209 | unsigned long when; /* When did the operation occur */ | |
210 | }; | |
211 | ||
212 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
213 | ||
ab4d5ed5 | 214 | #ifdef CONFIG_SYSFS |
81819f0f CL |
215 | static int sysfs_slab_add(struct kmem_cache *); |
216 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 217 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
bf5eb3de | 218 | static void sysfs_slab_remove(struct kmem_cache *s); |
81819f0f | 219 | #else |
0c710013 CL |
220 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
221 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
222 | { return 0; } | |
107dab5c | 223 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
bf5eb3de | 224 | static inline void sysfs_slab_remove(struct kmem_cache *s) { } |
81819f0f CL |
225 | #endif |
226 | ||
4fdccdfb | 227 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
228 | { |
229 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
230 | /* |
231 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
232 | * avoid this_cpu_add()'s irq-disable overhead. | |
233 | */ | |
234 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
235 | #endif |
236 | } | |
237 | ||
81819f0f CL |
238 | /******************************************************************** |
239 | * Core slab cache functions | |
240 | *******************************************************************/ | |
241 | ||
2482ddec KC |
242 | /* |
243 | * Returns freelist pointer (ptr). With hardening, this is obfuscated | |
244 | * with an XOR of the address where the pointer is held and a per-cache | |
245 | * random number. | |
246 | */ | |
247 | static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr, | |
248 | unsigned long ptr_addr) | |
249 | { | |
250 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
d36a63a9 AK |
251 | /* |
252 | * When CONFIG_KASAN_SW_TAGS is enabled, ptr_addr might be tagged. | |
253 | * Normally, this doesn't cause any issues, as both set_freepointer() | |
254 | * and get_freepointer() are called with a pointer with the same tag. | |
255 | * However, there are some issues with CONFIG_SLUB_DEBUG code. For | |
256 | * example, when __free_slub() iterates over objects in a cache, it | |
257 | * passes untagged pointers to check_object(). check_object() in turns | |
258 | * calls get_freepointer() with an untagged pointer, which causes the | |
259 | * freepointer to be restored incorrectly. | |
260 | */ | |
261 | return (void *)((unsigned long)ptr ^ s->random ^ | |
1ad53d9f | 262 | swab((unsigned long)kasan_reset_tag((void *)ptr_addr))); |
2482ddec KC |
263 | #else |
264 | return ptr; | |
265 | #endif | |
266 | } | |
267 | ||
268 | /* Returns the freelist pointer recorded at location ptr_addr. */ | |
269 | static inline void *freelist_dereference(const struct kmem_cache *s, | |
270 | void *ptr_addr) | |
271 | { | |
272 | return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr), | |
273 | (unsigned long)ptr_addr); | |
274 | } | |
275 | ||
7656c72b CL |
276 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
277 | { | |
2482ddec | 278 | return freelist_dereference(s, object + s->offset); |
7656c72b CL |
279 | } |
280 | ||
0ad9500e ED |
281 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
282 | { | |
0882ff91 | 283 | prefetch(object + s->offset); |
0ad9500e ED |
284 | } |
285 | ||
1393d9a1 CL |
286 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
287 | { | |
2482ddec | 288 | unsigned long freepointer_addr; |
1393d9a1 CL |
289 | void *p; |
290 | ||
8e57f8ac | 291 | if (!debug_pagealloc_enabled_static()) |
922d566c JK |
292 | return get_freepointer(s, object); |
293 | ||
2482ddec KC |
294 | freepointer_addr = (unsigned long)object + s->offset; |
295 | probe_kernel_read(&p, (void **)freepointer_addr, sizeof(p)); | |
296 | return freelist_ptr(s, p, freepointer_addr); | |
1393d9a1 CL |
297 | } |
298 | ||
7656c72b CL |
299 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
300 | { | |
2482ddec KC |
301 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
302 | ||
ce6fa91b AP |
303 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
304 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
305 | #endif | |
306 | ||
2482ddec | 307 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); |
7656c72b CL |
308 | } |
309 | ||
310 | /* Loop over all objects in a slab */ | |
224a88be | 311 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
312 | for (__p = fixup_red_left(__s, __addr); \ |
313 | __p < (__addr) + (__objects) * (__s)->size; \ | |
314 | __p += (__s)->size) | |
7656c72b | 315 | |
7656c72b | 316 | /* Determine object index from a given position */ |
284b50dd | 317 | static inline unsigned int slab_index(void *p, struct kmem_cache *s, void *addr) |
7656c72b | 318 | { |
6373dca1 | 319 | return (kasan_reset_tag(p) - addr) / s->size; |
7656c72b CL |
320 | } |
321 | ||
9736d2a9 | 322 | static inline unsigned int order_objects(unsigned int order, unsigned int size) |
ab9a0f19 | 323 | { |
9736d2a9 | 324 | return ((unsigned int)PAGE_SIZE << order) / size; |
ab9a0f19 LJ |
325 | } |
326 | ||
19af27af | 327 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, |
9736d2a9 | 328 | unsigned int size) |
834f3d11 CL |
329 | { |
330 | struct kmem_cache_order_objects x = { | |
9736d2a9 | 331 | (order << OO_SHIFT) + order_objects(order, size) |
834f3d11 CL |
332 | }; |
333 | ||
334 | return x; | |
335 | } | |
336 | ||
19af27af | 337 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) |
834f3d11 | 338 | { |
210b5c06 | 339 | return x.x >> OO_SHIFT; |
834f3d11 CL |
340 | } |
341 | ||
19af27af | 342 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) |
834f3d11 | 343 | { |
210b5c06 | 344 | return x.x & OO_MASK; |
834f3d11 CL |
345 | } |
346 | ||
881db7fb CL |
347 | /* |
348 | * Per slab locking using the pagelock | |
349 | */ | |
350 | static __always_inline void slab_lock(struct page *page) | |
351 | { | |
48c935ad | 352 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
353 | bit_spin_lock(PG_locked, &page->flags); |
354 | } | |
355 | ||
356 | static __always_inline void slab_unlock(struct page *page) | |
357 | { | |
48c935ad | 358 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
359 | __bit_spin_unlock(PG_locked, &page->flags); |
360 | } | |
361 | ||
1d07171c CL |
362 | /* Interrupts must be disabled (for the fallback code to work right) */ |
363 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
364 | void *freelist_old, unsigned long counters_old, | |
365 | void *freelist_new, unsigned long counters_new, | |
366 | const char *n) | |
367 | { | |
368 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
369 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
370 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 371 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 372 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
373 | freelist_old, counters_old, |
374 | freelist_new, counters_new)) | |
6f6528a1 | 375 | return true; |
1d07171c CL |
376 | } else |
377 | #endif | |
378 | { | |
379 | slab_lock(page); | |
d0e0ac97 CG |
380 | if (page->freelist == freelist_old && |
381 | page->counters == counters_old) { | |
1d07171c | 382 | page->freelist = freelist_new; |
7d27a04b | 383 | page->counters = counters_new; |
1d07171c | 384 | slab_unlock(page); |
6f6528a1 | 385 | return true; |
1d07171c CL |
386 | } |
387 | slab_unlock(page); | |
388 | } | |
389 | ||
390 | cpu_relax(); | |
391 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
392 | ||
393 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 394 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
395 | #endif |
396 | ||
6f6528a1 | 397 | return false; |
1d07171c CL |
398 | } |
399 | ||
b789ef51 CL |
400 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
401 | void *freelist_old, unsigned long counters_old, | |
402 | void *freelist_new, unsigned long counters_new, | |
403 | const char *n) | |
404 | { | |
2565409f HC |
405 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
406 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 407 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 408 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
409 | freelist_old, counters_old, |
410 | freelist_new, counters_new)) | |
6f6528a1 | 411 | return true; |
b789ef51 CL |
412 | } else |
413 | #endif | |
414 | { | |
1d07171c CL |
415 | unsigned long flags; |
416 | ||
417 | local_irq_save(flags); | |
881db7fb | 418 | slab_lock(page); |
d0e0ac97 CG |
419 | if (page->freelist == freelist_old && |
420 | page->counters == counters_old) { | |
b789ef51 | 421 | page->freelist = freelist_new; |
7d27a04b | 422 | page->counters = counters_new; |
881db7fb | 423 | slab_unlock(page); |
1d07171c | 424 | local_irq_restore(flags); |
6f6528a1 | 425 | return true; |
b789ef51 | 426 | } |
881db7fb | 427 | slab_unlock(page); |
1d07171c | 428 | local_irq_restore(flags); |
b789ef51 CL |
429 | } |
430 | ||
431 | cpu_relax(); | |
432 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
433 | ||
434 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 435 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
436 | #endif |
437 | ||
6f6528a1 | 438 | return false; |
b789ef51 CL |
439 | } |
440 | ||
41ecc55b | 441 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 YZ |
442 | static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)]; |
443 | static DEFINE_SPINLOCK(object_map_lock); | |
444 | ||
5f80b13a CL |
445 | /* |
446 | * Determine a map of object in use on a page. | |
447 | * | |
881db7fb | 448 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
449 | * not vanish from under us. |
450 | */ | |
90e9f6a6 | 451 | static unsigned long *get_map(struct kmem_cache *s, struct page *page) |
31364c2e | 452 | __acquires(&object_map_lock) |
5f80b13a CL |
453 | { |
454 | void *p; | |
455 | void *addr = page_address(page); | |
456 | ||
90e9f6a6 YZ |
457 | VM_BUG_ON(!irqs_disabled()); |
458 | ||
459 | spin_lock(&object_map_lock); | |
460 | ||
461 | bitmap_zero(object_map, page->objects); | |
462 | ||
5f80b13a | 463 | for (p = page->freelist; p; p = get_freepointer(s, p)) |
90e9f6a6 YZ |
464 | set_bit(slab_index(p, s, addr), object_map); |
465 | ||
466 | return object_map; | |
467 | } | |
468 | ||
81aba9e0 | 469 | static void put_map(unsigned long *map) __releases(&object_map_lock) |
90e9f6a6 YZ |
470 | { |
471 | VM_BUG_ON(map != object_map); | |
472 | lockdep_assert_held(&object_map_lock); | |
473 | ||
474 | spin_unlock(&object_map_lock); | |
5f80b13a CL |
475 | } |
476 | ||
870b1fbb | 477 | static inline unsigned int size_from_object(struct kmem_cache *s) |
d86bd1be JK |
478 | { |
479 | if (s->flags & SLAB_RED_ZONE) | |
480 | return s->size - s->red_left_pad; | |
481 | ||
482 | return s->size; | |
483 | } | |
484 | ||
485 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
486 | { | |
487 | if (s->flags & SLAB_RED_ZONE) | |
488 | p -= s->red_left_pad; | |
489 | ||
490 | return p; | |
491 | } | |
492 | ||
41ecc55b CL |
493 | /* |
494 | * Debug settings: | |
495 | */ | |
89d3c87e | 496 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 497 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 498 | #else |
d50112ed | 499 | static slab_flags_t slub_debug; |
f0630fff | 500 | #endif |
41ecc55b CL |
501 | |
502 | static char *slub_debug_slabs; | |
fa5ec8a1 | 503 | static int disable_higher_order_debug; |
41ecc55b | 504 | |
a79316c6 AR |
505 | /* |
506 | * slub is about to manipulate internal object metadata. This memory lies | |
507 | * outside the range of the allocated object, so accessing it would normally | |
508 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
509 | * to tell kasan that these accesses are OK. | |
510 | */ | |
511 | static inline void metadata_access_enable(void) | |
512 | { | |
513 | kasan_disable_current(); | |
514 | } | |
515 | ||
516 | static inline void metadata_access_disable(void) | |
517 | { | |
518 | kasan_enable_current(); | |
519 | } | |
520 | ||
81819f0f CL |
521 | /* |
522 | * Object debugging | |
523 | */ | |
d86bd1be JK |
524 | |
525 | /* Verify that a pointer has an address that is valid within a slab page */ | |
526 | static inline int check_valid_pointer(struct kmem_cache *s, | |
527 | struct page *page, void *object) | |
528 | { | |
529 | void *base; | |
530 | ||
531 | if (!object) | |
532 | return 1; | |
533 | ||
534 | base = page_address(page); | |
338cfaad | 535 | object = kasan_reset_tag(object); |
d86bd1be JK |
536 | object = restore_red_left(s, object); |
537 | if (object < base || object >= base + page->objects * s->size || | |
538 | (object - base) % s->size) { | |
539 | return 0; | |
540 | } | |
541 | ||
542 | return 1; | |
543 | } | |
544 | ||
aa2efd5e DT |
545 | static void print_section(char *level, char *text, u8 *addr, |
546 | unsigned int length) | |
81819f0f | 547 | { |
a79316c6 | 548 | metadata_access_enable(); |
aa2efd5e | 549 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
ffc79d28 | 550 | length, 1); |
a79316c6 | 551 | metadata_access_disable(); |
81819f0f CL |
552 | } |
553 | ||
81819f0f CL |
554 | static struct track *get_track(struct kmem_cache *s, void *object, |
555 | enum track_item alloc) | |
556 | { | |
557 | struct track *p; | |
558 | ||
559 | if (s->offset) | |
560 | p = object + s->offset + sizeof(void *); | |
561 | else | |
562 | p = object + s->inuse; | |
563 | ||
564 | return p + alloc; | |
565 | } | |
566 | ||
567 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 568 | enum track_item alloc, unsigned long addr) |
81819f0f | 569 | { |
1a00df4a | 570 | struct track *p = get_track(s, object, alloc); |
81819f0f | 571 | |
81819f0f | 572 | if (addr) { |
d6543e39 | 573 | #ifdef CONFIG_STACKTRACE |
79716799 | 574 | unsigned int nr_entries; |
d6543e39 | 575 | |
a79316c6 | 576 | metadata_access_enable(); |
79716799 | 577 | nr_entries = stack_trace_save(p->addrs, TRACK_ADDRS_COUNT, 3); |
a79316c6 | 578 | metadata_access_disable(); |
d6543e39 | 579 | |
79716799 TG |
580 | if (nr_entries < TRACK_ADDRS_COUNT) |
581 | p->addrs[nr_entries] = 0; | |
d6543e39 | 582 | #endif |
81819f0f CL |
583 | p->addr = addr; |
584 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 585 | p->pid = current->pid; |
81819f0f | 586 | p->when = jiffies; |
b8ca7ff7 | 587 | } else { |
81819f0f | 588 | memset(p, 0, sizeof(struct track)); |
b8ca7ff7 | 589 | } |
81819f0f CL |
590 | } |
591 | ||
81819f0f CL |
592 | static void init_tracking(struct kmem_cache *s, void *object) |
593 | { | |
24922684 CL |
594 | if (!(s->flags & SLAB_STORE_USER)) |
595 | return; | |
596 | ||
ce71e27c EGM |
597 | set_track(s, object, TRACK_FREE, 0UL); |
598 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
599 | } |
600 | ||
86609d33 | 601 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f CL |
602 | { |
603 | if (!t->addr) | |
604 | return; | |
605 | ||
f9f58285 | 606 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 607 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
d6543e39 BG |
608 | #ifdef CONFIG_STACKTRACE |
609 | { | |
610 | int i; | |
611 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
612 | if (t->addrs[i]) | |
f9f58285 | 613 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
614 | else |
615 | break; | |
616 | } | |
617 | #endif | |
24922684 CL |
618 | } |
619 | ||
620 | static void print_tracking(struct kmem_cache *s, void *object) | |
621 | { | |
86609d33 | 622 | unsigned long pr_time = jiffies; |
24922684 CL |
623 | if (!(s->flags & SLAB_STORE_USER)) |
624 | return; | |
625 | ||
86609d33 CP |
626 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
627 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
628 | } |
629 | ||
630 | static void print_page_info(struct page *page) | |
631 | { | |
f9f58285 | 632 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 633 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
634 | |
635 | } | |
636 | ||
637 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
638 | { | |
ecc42fbe | 639 | struct va_format vaf; |
24922684 | 640 | va_list args; |
24922684 CL |
641 | |
642 | va_start(args, fmt); | |
ecc42fbe FF |
643 | vaf.fmt = fmt; |
644 | vaf.va = &args; | |
f9f58285 | 645 | pr_err("=============================================================================\n"); |
ecc42fbe | 646 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 647 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 648 | |
373d4d09 | 649 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 650 | va_end(args); |
81819f0f CL |
651 | } |
652 | ||
24922684 CL |
653 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
654 | { | |
ecc42fbe | 655 | struct va_format vaf; |
24922684 | 656 | va_list args; |
24922684 CL |
657 | |
658 | va_start(args, fmt); | |
ecc42fbe FF |
659 | vaf.fmt = fmt; |
660 | vaf.va = &args; | |
661 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 662 | va_end(args); |
24922684 CL |
663 | } |
664 | ||
665 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
666 | { |
667 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 668 | u8 *addr = page_address(page); |
24922684 CL |
669 | |
670 | print_tracking(s, p); | |
671 | ||
672 | print_page_info(page); | |
673 | ||
f9f58285 FF |
674 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
675 | p, p - addr, get_freepointer(s, p)); | |
24922684 | 676 | |
d86bd1be | 677 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e DT |
678 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
679 | s->red_left_pad); | |
d86bd1be | 680 | else if (p > addr + 16) |
aa2efd5e | 681 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 682 | |
aa2efd5e | 683 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 684 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 685 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e | 686 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 687 | s->inuse - s->object_size); |
81819f0f | 688 | |
81819f0f CL |
689 | if (s->offset) |
690 | off = s->offset + sizeof(void *); | |
691 | else | |
692 | off = s->inuse; | |
693 | ||
24922684 | 694 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 695 | off += 2 * sizeof(struct track); |
81819f0f | 696 | |
80a9201a AP |
697 | off += kasan_metadata_size(s); |
698 | ||
d86bd1be | 699 | if (off != size_from_object(s)) |
81819f0f | 700 | /* Beginning of the filler is the free pointer */ |
aa2efd5e DT |
701 | print_section(KERN_ERR, "Padding ", p + off, |
702 | size_from_object(s) - off); | |
24922684 CL |
703 | |
704 | dump_stack(); | |
81819f0f CL |
705 | } |
706 | ||
75c66def | 707 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
708 | u8 *object, char *reason) |
709 | { | |
3dc50637 | 710 | slab_bug(s, "%s", reason); |
24922684 | 711 | print_trailer(s, page, object); |
81819f0f CL |
712 | } |
713 | ||
a38965bf | 714 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page, |
d0e0ac97 | 715 | const char *fmt, ...) |
81819f0f CL |
716 | { |
717 | va_list args; | |
718 | char buf[100]; | |
719 | ||
24922684 CL |
720 | va_start(args, fmt); |
721 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 722 | va_end(args); |
3dc50637 | 723 | slab_bug(s, "%s", buf); |
24922684 | 724 | print_page_info(page); |
81819f0f CL |
725 | dump_stack(); |
726 | } | |
727 | ||
f7cb1933 | 728 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
729 | { |
730 | u8 *p = object; | |
731 | ||
d86bd1be JK |
732 | if (s->flags & SLAB_RED_ZONE) |
733 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
734 | ||
81819f0f | 735 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
736 | memset(p, POISON_FREE, s->object_size - 1); |
737 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
738 | } |
739 | ||
740 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 741 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
742 | } |
743 | ||
24922684 CL |
744 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
745 | void *from, void *to) | |
746 | { | |
747 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
748 | memset(from, data, to - from); | |
749 | } | |
750 | ||
751 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
752 | u8 *object, char *what, | |
06428780 | 753 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
754 | { |
755 | u8 *fault; | |
756 | u8 *end; | |
e1b70dd1 | 757 | u8 *addr = page_address(page); |
24922684 | 758 | |
a79316c6 | 759 | metadata_access_enable(); |
79824820 | 760 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 761 | metadata_access_disable(); |
24922684 CL |
762 | if (!fault) |
763 | return 1; | |
764 | ||
765 | end = start + bytes; | |
766 | while (end > fault && end[-1] == value) | |
767 | end--; | |
768 | ||
769 | slab_bug(s, "%s overwritten", what); | |
e1b70dd1 MC |
770 | pr_err("INFO: 0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n", |
771 | fault, end - 1, fault - addr, | |
772 | fault[0], value); | |
24922684 CL |
773 | print_trailer(s, page, object); |
774 | ||
775 | restore_bytes(s, what, value, fault, end); | |
776 | return 0; | |
81819f0f CL |
777 | } |
778 | ||
81819f0f CL |
779 | /* |
780 | * Object layout: | |
781 | * | |
782 | * object address | |
783 | * Bytes of the object to be managed. | |
784 | * If the freepointer may overlay the object then the free | |
785 | * pointer is the first word of the object. | |
672bba3a | 786 | * |
81819f0f CL |
787 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
788 | * 0xa5 (POISON_END) | |
789 | * | |
3b0efdfa | 790 | * object + s->object_size |
81819f0f | 791 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 792 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 793 | * object_size == inuse. |
672bba3a | 794 | * |
81819f0f CL |
795 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
796 | * 0xcc (RED_ACTIVE) for objects in use. | |
797 | * | |
798 | * object + s->inuse | |
672bba3a CL |
799 | * Meta data starts here. |
800 | * | |
81819f0f CL |
801 | * A. Free pointer (if we cannot overwrite object on free) |
802 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 803 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 804 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
805 | * before the word boundary. |
806 | * | |
807 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
808 | * |
809 | * object + s->size | |
672bba3a | 810 | * Nothing is used beyond s->size. |
81819f0f | 811 | * |
3b0efdfa | 812 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 813 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
814 | * may be used with merged slabcaches. |
815 | */ | |
816 | ||
81819f0f CL |
817 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
818 | { | |
819 | unsigned long off = s->inuse; /* The end of info */ | |
820 | ||
821 | if (s->offset) | |
822 | /* Freepointer is placed after the object. */ | |
823 | off += sizeof(void *); | |
824 | ||
825 | if (s->flags & SLAB_STORE_USER) | |
826 | /* We also have user information there */ | |
827 | off += 2 * sizeof(struct track); | |
828 | ||
80a9201a AP |
829 | off += kasan_metadata_size(s); |
830 | ||
d86bd1be | 831 | if (size_from_object(s) == off) |
81819f0f CL |
832 | return 1; |
833 | ||
24922684 | 834 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 835 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
836 | } |
837 | ||
39b26464 | 838 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
839 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
840 | { | |
24922684 CL |
841 | u8 *start; |
842 | u8 *fault; | |
843 | u8 *end; | |
5d682681 | 844 | u8 *pad; |
24922684 CL |
845 | int length; |
846 | int remainder; | |
81819f0f CL |
847 | |
848 | if (!(s->flags & SLAB_POISON)) | |
849 | return 1; | |
850 | ||
a973e9dd | 851 | start = page_address(page); |
a50b854e | 852 | length = page_size(page); |
39b26464 CL |
853 | end = start + length; |
854 | remainder = length % s->size; | |
81819f0f CL |
855 | if (!remainder) |
856 | return 1; | |
857 | ||
5d682681 | 858 | pad = end - remainder; |
a79316c6 | 859 | metadata_access_enable(); |
5d682681 | 860 | fault = memchr_inv(pad, POISON_INUSE, remainder); |
a79316c6 | 861 | metadata_access_disable(); |
24922684 CL |
862 | if (!fault) |
863 | return 1; | |
864 | while (end > fault && end[-1] == POISON_INUSE) | |
865 | end--; | |
866 | ||
e1b70dd1 MC |
867 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p @offset=%tu", |
868 | fault, end - 1, fault - start); | |
5d682681 | 869 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 870 | |
5d682681 | 871 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
24922684 | 872 | return 0; |
81819f0f CL |
873 | } |
874 | ||
875 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 876 | void *object, u8 val) |
81819f0f CL |
877 | { |
878 | u8 *p = object; | |
3b0efdfa | 879 | u8 *endobject = object + s->object_size; |
81819f0f CL |
880 | |
881 | if (s->flags & SLAB_RED_ZONE) { | |
d86bd1be JK |
882 | if (!check_bytes_and_report(s, page, object, "Redzone", |
883 | object - s->red_left_pad, val, s->red_left_pad)) | |
884 | return 0; | |
885 | ||
24922684 | 886 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 887 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 888 | return 0; |
81819f0f | 889 | } else { |
3b0efdfa | 890 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 891 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
892 | endobject, POISON_INUSE, |
893 | s->inuse - s->object_size); | |
3adbefee | 894 | } |
81819f0f CL |
895 | } |
896 | ||
897 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 898 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 899 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 900 | POISON_FREE, s->object_size - 1) || |
24922684 | 901 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 902 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 903 | return 0; |
81819f0f CL |
904 | /* |
905 | * check_pad_bytes cleans up on its own. | |
906 | */ | |
907 | check_pad_bytes(s, page, p); | |
908 | } | |
909 | ||
f7cb1933 | 910 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
911 | /* |
912 | * Object and freepointer overlap. Cannot check | |
913 | * freepointer while object is allocated. | |
914 | */ | |
915 | return 1; | |
916 | ||
917 | /* Check free pointer validity */ | |
918 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
919 | object_err(s, page, p, "Freepointer corrupt"); | |
920 | /* | |
9f6c708e | 921 | * No choice but to zap it and thus lose the remainder |
81819f0f | 922 | * of the free objects in this slab. May cause |
672bba3a | 923 | * another error because the object count is now wrong. |
81819f0f | 924 | */ |
a973e9dd | 925 | set_freepointer(s, p, NULL); |
81819f0f CL |
926 | return 0; |
927 | } | |
928 | return 1; | |
929 | } | |
930 | ||
931 | static int check_slab(struct kmem_cache *s, struct page *page) | |
932 | { | |
39b26464 CL |
933 | int maxobj; |
934 | ||
81819f0f CL |
935 | VM_BUG_ON(!irqs_disabled()); |
936 | ||
937 | if (!PageSlab(page)) { | |
24922684 | 938 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
939 | return 0; |
940 | } | |
39b26464 | 941 | |
9736d2a9 | 942 | maxobj = order_objects(compound_order(page), s->size); |
39b26464 CL |
943 | if (page->objects > maxobj) { |
944 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 945 | page->objects, maxobj); |
39b26464 CL |
946 | return 0; |
947 | } | |
948 | if (page->inuse > page->objects) { | |
24922684 | 949 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 950 | page->inuse, page->objects); |
81819f0f CL |
951 | return 0; |
952 | } | |
953 | /* Slab_pad_check fixes things up after itself */ | |
954 | slab_pad_check(s, page); | |
955 | return 1; | |
956 | } | |
957 | ||
958 | /* | |
672bba3a CL |
959 | * Determine if a certain object on a page is on the freelist. Must hold the |
960 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
961 | */ |
962 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
963 | { | |
964 | int nr = 0; | |
881db7fb | 965 | void *fp; |
81819f0f | 966 | void *object = NULL; |
f6edde9c | 967 | int max_objects; |
81819f0f | 968 | |
881db7fb | 969 | fp = page->freelist; |
39b26464 | 970 | while (fp && nr <= page->objects) { |
81819f0f CL |
971 | if (fp == search) |
972 | return 1; | |
973 | if (!check_valid_pointer(s, page, fp)) { | |
974 | if (object) { | |
975 | object_err(s, page, object, | |
976 | "Freechain corrupt"); | |
a973e9dd | 977 | set_freepointer(s, object, NULL); |
81819f0f | 978 | } else { |
24922684 | 979 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 980 | page->freelist = NULL; |
39b26464 | 981 | page->inuse = page->objects; |
24922684 | 982 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
983 | return 0; |
984 | } | |
985 | break; | |
986 | } | |
987 | object = fp; | |
988 | fp = get_freepointer(s, object); | |
989 | nr++; | |
990 | } | |
991 | ||
9736d2a9 | 992 | max_objects = order_objects(compound_order(page), s->size); |
210b5c06 CG |
993 | if (max_objects > MAX_OBJS_PER_PAGE) |
994 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
995 | |
996 | if (page->objects != max_objects) { | |
756a025f JP |
997 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
998 | page->objects, max_objects); | |
224a88be CL |
999 | page->objects = max_objects; |
1000 | slab_fix(s, "Number of objects adjusted."); | |
1001 | } | |
39b26464 | 1002 | if (page->inuse != page->objects - nr) { |
756a025f JP |
1003 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
1004 | page->inuse, page->objects - nr); | |
39b26464 | 1005 | page->inuse = page->objects - nr; |
24922684 | 1006 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
1007 | } |
1008 | return search == NULL; | |
1009 | } | |
1010 | ||
0121c619 CL |
1011 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
1012 | int alloc) | |
3ec09742 CL |
1013 | { |
1014 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 1015 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
1016 | s->name, |
1017 | alloc ? "alloc" : "free", | |
1018 | object, page->inuse, | |
1019 | page->freelist); | |
1020 | ||
1021 | if (!alloc) | |
aa2efd5e | 1022 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1023 | s->object_size); |
3ec09742 CL |
1024 | |
1025 | dump_stack(); | |
1026 | } | |
1027 | } | |
1028 | ||
643b1138 | 1029 | /* |
672bba3a | 1030 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1031 | */ |
5cc6eee8 CL |
1032 | static void add_full(struct kmem_cache *s, |
1033 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1034 | { |
5cc6eee8 CL |
1035 | if (!(s->flags & SLAB_STORE_USER)) |
1036 | return; | |
1037 | ||
255d0884 | 1038 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1039 | list_add(&page->slab_list, &n->full); |
643b1138 CL |
1040 | } |
1041 | ||
c65c1877 | 1042 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1043 | { |
643b1138 CL |
1044 | if (!(s->flags & SLAB_STORE_USER)) |
1045 | return; | |
1046 | ||
255d0884 | 1047 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1048 | list_del(&page->slab_list); |
643b1138 CL |
1049 | } |
1050 | ||
0f389ec6 CL |
1051 | /* Tracking of the number of slabs for debugging purposes */ |
1052 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1053 | { | |
1054 | struct kmem_cache_node *n = get_node(s, node); | |
1055 | ||
1056 | return atomic_long_read(&n->nr_slabs); | |
1057 | } | |
1058 | ||
26c02cf0 AB |
1059 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1060 | { | |
1061 | return atomic_long_read(&n->nr_slabs); | |
1062 | } | |
1063 | ||
205ab99d | 1064 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1065 | { |
1066 | struct kmem_cache_node *n = get_node(s, node); | |
1067 | ||
1068 | /* | |
1069 | * May be called early in order to allocate a slab for the | |
1070 | * kmem_cache_node structure. Solve the chicken-egg | |
1071 | * dilemma by deferring the increment of the count during | |
1072 | * bootstrap (see early_kmem_cache_node_alloc). | |
1073 | */ | |
338b2642 | 1074 | if (likely(n)) { |
0f389ec6 | 1075 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1076 | atomic_long_add(objects, &n->total_objects); |
1077 | } | |
0f389ec6 | 1078 | } |
205ab99d | 1079 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1080 | { |
1081 | struct kmem_cache_node *n = get_node(s, node); | |
1082 | ||
1083 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1084 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1085 | } |
1086 | ||
1087 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1088 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1089 | void *object) | |
1090 | { | |
1091 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1092 | return; | |
1093 | ||
f7cb1933 | 1094 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1095 | init_tracking(s, object); |
1096 | } | |
1097 | ||
a50b854e MWO |
1098 | static |
1099 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) | |
a7101224 AK |
1100 | { |
1101 | if (!(s->flags & SLAB_POISON)) | |
1102 | return; | |
1103 | ||
1104 | metadata_access_enable(); | |
a50b854e | 1105 | memset(addr, POISON_INUSE, page_size(page)); |
a7101224 AK |
1106 | metadata_access_disable(); |
1107 | } | |
1108 | ||
becfda68 | 1109 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
278d7756 | 1110 | struct page *page, void *object) |
81819f0f CL |
1111 | { |
1112 | if (!check_slab(s, page)) | |
becfda68 | 1113 | return 0; |
81819f0f | 1114 | |
81819f0f CL |
1115 | if (!check_valid_pointer(s, page, object)) { |
1116 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1117 | return 0; |
81819f0f CL |
1118 | } |
1119 | ||
f7cb1933 | 1120 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1121 | return 0; |
1122 | ||
1123 | return 1; | |
1124 | } | |
1125 | ||
1126 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1127 | struct page *page, | |
1128 | void *object, unsigned long addr) | |
1129 | { | |
1130 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
278d7756 | 1131 | if (!alloc_consistency_checks(s, page, object)) |
becfda68 LA |
1132 | goto bad; |
1133 | } | |
81819f0f | 1134 | |
3ec09742 CL |
1135 | /* Success perform special debug activities for allocs */ |
1136 | if (s->flags & SLAB_STORE_USER) | |
1137 | set_track(s, object, TRACK_ALLOC, addr); | |
1138 | trace(s, page, object, 1); | |
f7cb1933 | 1139 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1140 | return 1; |
3ec09742 | 1141 | |
81819f0f CL |
1142 | bad: |
1143 | if (PageSlab(page)) { | |
1144 | /* | |
1145 | * If this is a slab page then lets do the best we can | |
1146 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1147 | * as used avoids touching the remaining objects. |
81819f0f | 1148 | */ |
24922684 | 1149 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1150 | page->inuse = page->objects; |
a973e9dd | 1151 | page->freelist = NULL; |
81819f0f CL |
1152 | } |
1153 | return 0; | |
1154 | } | |
1155 | ||
becfda68 LA |
1156 | static inline int free_consistency_checks(struct kmem_cache *s, |
1157 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1158 | { |
81819f0f | 1159 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1160 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1161 | return 0; |
81819f0f CL |
1162 | } |
1163 | ||
1164 | if (on_freelist(s, page, object)) { | |
24922684 | 1165 | object_err(s, page, object, "Object already free"); |
becfda68 | 1166 | return 0; |
81819f0f CL |
1167 | } |
1168 | ||
f7cb1933 | 1169 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1170 | return 0; |
81819f0f | 1171 | |
1b4f59e3 | 1172 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1173 | if (!PageSlab(page)) { |
756a025f JP |
1174 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1175 | object); | |
1b4f59e3 | 1176 | } else if (!page->slab_cache) { |
f9f58285 FF |
1177 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1178 | object); | |
70d71228 | 1179 | dump_stack(); |
06428780 | 1180 | } else |
24922684 CL |
1181 | object_err(s, page, object, |
1182 | "page slab pointer corrupt."); | |
becfda68 LA |
1183 | return 0; |
1184 | } | |
1185 | return 1; | |
1186 | } | |
1187 | ||
1188 | /* Supports checking bulk free of a constructed freelist */ | |
1189 | static noinline int free_debug_processing( | |
1190 | struct kmem_cache *s, struct page *page, | |
1191 | void *head, void *tail, int bulk_cnt, | |
1192 | unsigned long addr) | |
1193 | { | |
1194 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1195 | void *object = head; | |
1196 | int cnt = 0; | |
1197 | unsigned long uninitialized_var(flags); | |
1198 | int ret = 0; | |
1199 | ||
1200 | spin_lock_irqsave(&n->list_lock, flags); | |
1201 | slab_lock(page); | |
1202 | ||
1203 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1204 | if (!check_slab(s, page)) | |
1205 | goto out; | |
1206 | } | |
1207 | ||
1208 | next_object: | |
1209 | cnt++; | |
1210 | ||
1211 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1212 | if (!free_consistency_checks(s, page, object, addr)) | |
1213 | goto out; | |
81819f0f | 1214 | } |
3ec09742 | 1215 | |
3ec09742 CL |
1216 | if (s->flags & SLAB_STORE_USER) |
1217 | set_track(s, object, TRACK_FREE, addr); | |
1218 | trace(s, page, object, 0); | |
81084651 | 1219 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1220 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1221 | |
1222 | /* Reached end of constructed freelist yet? */ | |
1223 | if (object != tail) { | |
1224 | object = get_freepointer(s, object); | |
1225 | goto next_object; | |
1226 | } | |
804aa132 LA |
1227 | ret = 1; |
1228 | ||
5c2e4bbb | 1229 | out: |
81084651 JDB |
1230 | if (cnt != bulk_cnt) |
1231 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1232 | bulk_cnt, cnt); | |
1233 | ||
881db7fb | 1234 | slab_unlock(page); |
282acb43 | 1235 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1236 | if (!ret) |
1237 | slab_fix(s, "Object at 0x%p not freed", object); | |
1238 | return ret; | |
81819f0f CL |
1239 | } |
1240 | ||
41ecc55b CL |
1241 | static int __init setup_slub_debug(char *str) |
1242 | { | |
f0630fff CL |
1243 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1244 | if (*str++ != '=' || !*str) | |
1245 | /* | |
1246 | * No options specified. Switch on full debugging. | |
1247 | */ | |
1248 | goto out; | |
1249 | ||
1250 | if (*str == ',') | |
1251 | /* | |
1252 | * No options but restriction on slabs. This means full | |
1253 | * debugging for slabs matching a pattern. | |
1254 | */ | |
1255 | goto check_slabs; | |
1256 | ||
1257 | slub_debug = 0; | |
1258 | if (*str == '-') | |
1259 | /* | |
1260 | * Switch off all debugging measures. | |
1261 | */ | |
1262 | goto out; | |
1263 | ||
1264 | /* | |
1265 | * Determine which debug features should be switched on | |
1266 | */ | |
06428780 | 1267 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1268 | switch (tolower(*str)) { |
1269 | case 'f': | |
becfda68 | 1270 | slub_debug |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1271 | break; |
1272 | case 'z': | |
1273 | slub_debug |= SLAB_RED_ZONE; | |
1274 | break; | |
1275 | case 'p': | |
1276 | slub_debug |= SLAB_POISON; | |
1277 | break; | |
1278 | case 'u': | |
1279 | slub_debug |= SLAB_STORE_USER; | |
1280 | break; | |
1281 | case 't': | |
1282 | slub_debug |= SLAB_TRACE; | |
1283 | break; | |
4c13dd3b DM |
1284 | case 'a': |
1285 | slub_debug |= SLAB_FAILSLAB; | |
1286 | break; | |
08303a73 CA |
1287 | case 'o': |
1288 | /* | |
1289 | * Avoid enabling debugging on caches if its minimum | |
1290 | * order would increase as a result. | |
1291 | */ | |
1292 | disable_higher_order_debug = 1; | |
1293 | break; | |
f0630fff | 1294 | default: |
f9f58285 FF |
1295 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1296 | *str); | |
f0630fff | 1297 | } |
41ecc55b CL |
1298 | } |
1299 | ||
f0630fff | 1300 | check_slabs: |
41ecc55b CL |
1301 | if (*str == ',') |
1302 | slub_debug_slabs = str + 1; | |
f0630fff | 1303 | out: |
6471384a AP |
1304 | if ((static_branch_unlikely(&init_on_alloc) || |
1305 | static_branch_unlikely(&init_on_free)) && | |
1306 | (slub_debug & SLAB_POISON)) | |
1307 | pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n"); | |
41ecc55b CL |
1308 | return 1; |
1309 | } | |
1310 | ||
1311 | __setup("slub_debug", setup_slub_debug); | |
1312 | ||
c5fd3ca0 AT |
1313 | /* |
1314 | * kmem_cache_flags - apply debugging options to the cache | |
1315 | * @object_size: the size of an object without meta data | |
1316 | * @flags: flags to set | |
1317 | * @name: name of the cache | |
1318 | * @ctor: constructor function | |
1319 | * | |
1320 | * Debug option(s) are applied to @flags. In addition to the debug | |
1321 | * option(s), if a slab name (or multiple) is specified i.e. | |
1322 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1323 | * then only the select slabs will receive the debug option(s). | |
1324 | */ | |
0293d1fd | 1325 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1326 | slab_flags_t flags, const char *name, |
51cc5068 | 1327 | void (*ctor)(void *)) |
41ecc55b | 1328 | { |
c5fd3ca0 AT |
1329 | char *iter; |
1330 | size_t len; | |
1331 | ||
1332 | /* If slub_debug = 0, it folds into the if conditional. */ | |
1333 | if (!slub_debug_slabs) | |
1334 | return flags | slub_debug; | |
1335 | ||
1336 | len = strlen(name); | |
1337 | iter = slub_debug_slabs; | |
1338 | while (*iter) { | |
1339 | char *end, *glob; | |
1340 | size_t cmplen; | |
1341 | ||
9cf3a8d8 | 1342 | end = strchrnul(iter, ','); |
c5fd3ca0 AT |
1343 | |
1344 | glob = strnchr(iter, end - iter, '*'); | |
1345 | if (glob) | |
1346 | cmplen = glob - iter; | |
1347 | else | |
1348 | cmplen = max_t(size_t, len, (end - iter)); | |
1349 | ||
1350 | if (!strncmp(name, iter, cmplen)) { | |
1351 | flags |= slub_debug; | |
1352 | break; | |
1353 | } | |
1354 | ||
1355 | if (!*end) | |
1356 | break; | |
1357 | iter = end + 1; | |
1358 | } | |
ba0268a8 CL |
1359 | |
1360 | return flags; | |
41ecc55b | 1361 | } |
b4a64718 | 1362 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1363 | static inline void setup_object_debug(struct kmem_cache *s, |
1364 | struct page *page, void *object) {} | |
a50b854e MWO |
1365 | static inline |
1366 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {} | |
41ecc55b | 1367 | |
3ec09742 | 1368 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1369 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1370 | |
282acb43 | 1371 | static inline int free_debug_processing( |
81084651 JDB |
1372 | struct kmem_cache *s, struct page *page, |
1373 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1374 | unsigned long addr) { return 0; } |
41ecc55b | 1375 | |
41ecc55b CL |
1376 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1377 | { return 1; } | |
1378 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1379 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1380 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1381 | struct page *page) {} | |
c65c1877 PZ |
1382 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1383 | struct page *page) {} | |
0293d1fd | 1384 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1385 | slab_flags_t flags, const char *name, |
51cc5068 | 1386 | void (*ctor)(void *)) |
ba0268a8 CL |
1387 | { |
1388 | return flags; | |
1389 | } | |
41ecc55b | 1390 | #define slub_debug 0 |
0f389ec6 | 1391 | |
fdaa45e9 IM |
1392 | #define disable_higher_order_debug 0 |
1393 | ||
0f389ec6 CL |
1394 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1395 | { return 0; } | |
26c02cf0 AB |
1396 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1397 | { return 0; } | |
205ab99d CL |
1398 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1399 | int objects) {} | |
1400 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1401 | int objects) {} | |
7d550c56 | 1402 | |
02e72cc6 AR |
1403 | #endif /* CONFIG_SLUB_DEBUG */ |
1404 | ||
1405 | /* | |
1406 | * Hooks for other subsystems that check memory allocations. In a typical | |
1407 | * production configuration these hooks all should produce no code at all. | |
1408 | */ | |
0116523c | 1409 | static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
d56791b3 | 1410 | { |
53128245 | 1411 | ptr = kasan_kmalloc_large(ptr, size, flags); |
a2f77575 | 1412 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ |
d56791b3 | 1413 | kmemleak_alloc(ptr, size, 1, flags); |
53128245 | 1414 | return ptr; |
d56791b3 RB |
1415 | } |
1416 | ||
ee3ce779 | 1417 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1418 | { |
1419 | kmemleak_free(x); | |
ee3ce779 | 1420 | kasan_kfree_large(x, _RET_IP_); |
d56791b3 RB |
1421 | } |
1422 | ||
c3895391 | 1423 | static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x) |
d56791b3 RB |
1424 | { |
1425 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1426 | |
02e72cc6 AR |
1427 | /* |
1428 | * Trouble is that we may no longer disable interrupts in the fast path | |
1429 | * So in order to make the debug calls that expect irqs to be | |
1430 | * disabled we need to disable interrupts temporarily. | |
1431 | */ | |
4675ff05 | 1432 | #ifdef CONFIG_LOCKDEP |
02e72cc6 AR |
1433 | { |
1434 | unsigned long flags; | |
1435 | ||
1436 | local_irq_save(flags); | |
02e72cc6 AR |
1437 | debug_check_no_locks_freed(x, s->object_size); |
1438 | local_irq_restore(flags); | |
1439 | } | |
1440 | #endif | |
1441 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1442 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1443 | |
c3895391 AK |
1444 | /* KASAN might put x into memory quarantine, delaying its reuse */ |
1445 | return kasan_slab_free(s, x, _RET_IP_); | |
02e72cc6 | 1446 | } |
205ab99d | 1447 | |
c3895391 AK |
1448 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
1449 | void **head, void **tail) | |
81084651 | 1450 | { |
6471384a AP |
1451 | |
1452 | void *object; | |
1453 | void *next = *head; | |
1454 | void *old_tail = *tail ? *tail : *head; | |
1455 | int rsize; | |
1456 | ||
aea4df4c LA |
1457 | /* Head and tail of the reconstructed freelist */ |
1458 | *head = NULL; | |
1459 | *tail = NULL; | |
1b7e816f | 1460 | |
aea4df4c LA |
1461 | do { |
1462 | object = next; | |
1463 | next = get_freepointer(s, object); | |
1464 | ||
1465 | if (slab_want_init_on_free(s)) { | |
6471384a AP |
1466 | /* |
1467 | * Clear the object and the metadata, but don't touch | |
1468 | * the redzone. | |
1469 | */ | |
1470 | memset(object, 0, s->object_size); | |
1471 | rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad | |
1472 | : 0; | |
1473 | memset((char *)object + s->inuse, 0, | |
1474 | s->size - s->inuse - rsize); | |
81084651 | 1475 | |
aea4df4c | 1476 | } |
c3895391 AK |
1477 | /* If object's reuse doesn't have to be delayed */ |
1478 | if (!slab_free_hook(s, object)) { | |
1479 | /* Move object to the new freelist */ | |
1480 | set_freepointer(s, object, *head); | |
1481 | *head = object; | |
1482 | if (!*tail) | |
1483 | *tail = object; | |
1484 | } | |
1485 | } while (object != old_tail); | |
1486 | ||
1487 | if (*head == *tail) | |
1488 | *tail = NULL; | |
1489 | ||
1490 | return *head != NULL; | |
81084651 JDB |
1491 | } |
1492 | ||
4d176711 | 1493 | static void *setup_object(struct kmem_cache *s, struct page *page, |
588f8ba9 TG |
1494 | void *object) |
1495 | { | |
1496 | setup_object_debug(s, page, object); | |
4d176711 | 1497 | object = kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1498 | if (unlikely(s->ctor)) { |
1499 | kasan_unpoison_object_data(s, object); | |
1500 | s->ctor(object); | |
1501 | kasan_poison_object_data(s, object); | |
1502 | } | |
4d176711 | 1503 | return object; |
588f8ba9 TG |
1504 | } |
1505 | ||
81819f0f CL |
1506 | /* |
1507 | * Slab allocation and freeing | |
1508 | */ | |
5dfb4175 VD |
1509 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1510 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1511 | { |
5dfb4175 | 1512 | struct page *page; |
19af27af | 1513 | unsigned int order = oo_order(oo); |
65c3376a | 1514 | |
2154a336 | 1515 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1516 | page = alloc_pages(flags, order); |
65c3376a | 1517 | else |
96db800f | 1518 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1519 | |
6cea1d56 | 1520 | if (page && charge_slab_page(page, flags, order, s)) { |
f3ccb2c4 VD |
1521 | __free_pages(page, order); |
1522 | page = NULL; | |
1523 | } | |
5dfb4175 VD |
1524 | |
1525 | return page; | |
65c3376a CL |
1526 | } |
1527 | ||
210e7a43 TG |
1528 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1529 | /* Pre-initialize the random sequence cache */ | |
1530 | static int init_cache_random_seq(struct kmem_cache *s) | |
1531 | { | |
19af27af | 1532 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 1533 | int err; |
210e7a43 | 1534 | |
a810007a SR |
1535 | /* Bailout if already initialised */ |
1536 | if (s->random_seq) | |
1537 | return 0; | |
1538 | ||
210e7a43 TG |
1539 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1540 | if (err) { | |
1541 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1542 | s->name); | |
1543 | return err; | |
1544 | } | |
1545 | ||
1546 | /* Transform to an offset on the set of pages */ | |
1547 | if (s->random_seq) { | |
19af27af AD |
1548 | unsigned int i; |
1549 | ||
210e7a43 TG |
1550 | for (i = 0; i < count; i++) |
1551 | s->random_seq[i] *= s->size; | |
1552 | } | |
1553 | return 0; | |
1554 | } | |
1555 | ||
1556 | /* Initialize each random sequence freelist per cache */ | |
1557 | static void __init init_freelist_randomization(void) | |
1558 | { | |
1559 | struct kmem_cache *s; | |
1560 | ||
1561 | mutex_lock(&slab_mutex); | |
1562 | ||
1563 | list_for_each_entry(s, &slab_caches, list) | |
1564 | init_cache_random_seq(s); | |
1565 | ||
1566 | mutex_unlock(&slab_mutex); | |
1567 | } | |
1568 | ||
1569 | /* Get the next entry on the pre-computed freelist randomized */ | |
1570 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1571 | unsigned long *pos, void *start, | |
1572 | unsigned long page_limit, | |
1573 | unsigned long freelist_count) | |
1574 | { | |
1575 | unsigned int idx; | |
1576 | ||
1577 | /* | |
1578 | * If the target page allocation failed, the number of objects on the | |
1579 | * page might be smaller than the usual size defined by the cache. | |
1580 | */ | |
1581 | do { | |
1582 | idx = s->random_seq[*pos]; | |
1583 | *pos += 1; | |
1584 | if (*pos >= freelist_count) | |
1585 | *pos = 0; | |
1586 | } while (unlikely(idx >= page_limit)); | |
1587 | ||
1588 | return (char *)start + idx; | |
1589 | } | |
1590 | ||
1591 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1592 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1593 | { | |
1594 | void *start; | |
1595 | void *cur; | |
1596 | void *next; | |
1597 | unsigned long idx, pos, page_limit, freelist_count; | |
1598 | ||
1599 | if (page->objects < 2 || !s->random_seq) | |
1600 | return false; | |
1601 | ||
1602 | freelist_count = oo_objects(s->oo); | |
1603 | pos = get_random_int() % freelist_count; | |
1604 | ||
1605 | page_limit = page->objects * s->size; | |
1606 | start = fixup_red_left(s, page_address(page)); | |
1607 | ||
1608 | /* First entry is used as the base of the freelist */ | |
1609 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1610 | freelist_count); | |
4d176711 | 1611 | cur = setup_object(s, page, cur); |
210e7a43 TG |
1612 | page->freelist = cur; |
1613 | ||
1614 | for (idx = 1; idx < page->objects; idx++) { | |
210e7a43 TG |
1615 | next = next_freelist_entry(s, page, &pos, start, page_limit, |
1616 | freelist_count); | |
4d176711 | 1617 | next = setup_object(s, page, next); |
210e7a43 TG |
1618 | set_freepointer(s, cur, next); |
1619 | cur = next; | |
1620 | } | |
210e7a43 TG |
1621 | set_freepointer(s, cur, NULL); |
1622 | ||
1623 | return true; | |
1624 | } | |
1625 | #else | |
1626 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1627 | { | |
1628 | return 0; | |
1629 | } | |
1630 | static inline void init_freelist_randomization(void) { } | |
1631 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1632 | { | |
1633 | return false; | |
1634 | } | |
1635 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1636 | ||
81819f0f CL |
1637 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1638 | { | |
06428780 | 1639 | struct page *page; |
834f3d11 | 1640 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1641 | gfp_t alloc_gfp; |
4d176711 | 1642 | void *start, *p, *next; |
a50b854e | 1643 | int idx; |
210e7a43 | 1644 | bool shuffle; |
81819f0f | 1645 | |
7e0528da CL |
1646 | flags &= gfp_allowed_mask; |
1647 | ||
d0164adc | 1648 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1649 | local_irq_enable(); |
1650 | ||
b7a49f0d | 1651 | flags |= s->allocflags; |
e12ba74d | 1652 | |
ba52270d PE |
1653 | /* |
1654 | * Let the initial higher-order allocation fail under memory pressure | |
1655 | * so we fall-back to the minimum order allocation. | |
1656 | */ | |
1657 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1658 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1659 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1660 | |
5dfb4175 | 1661 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1662 | if (unlikely(!page)) { |
1663 | oo = s->min; | |
80c3a998 | 1664 | alloc_gfp = flags; |
65c3376a CL |
1665 | /* |
1666 | * Allocation may have failed due to fragmentation. | |
1667 | * Try a lower order alloc if possible | |
1668 | */ | |
5dfb4175 | 1669 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1670 | if (unlikely(!page)) |
1671 | goto out; | |
1672 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1673 | } |
5a896d9e | 1674 | |
834f3d11 | 1675 | page->objects = oo_objects(oo); |
81819f0f | 1676 | |
1b4f59e3 | 1677 | page->slab_cache = s; |
c03f94cc | 1678 | __SetPageSlab(page); |
2f064f34 | 1679 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1680 | SetPageSlabPfmemalloc(page); |
81819f0f | 1681 | |
a7101224 | 1682 | kasan_poison_slab(page); |
81819f0f | 1683 | |
a7101224 | 1684 | start = page_address(page); |
81819f0f | 1685 | |
a50b854e | 1686 | setup_page_debug(s, page, start); |
0316bec2 | 1687 | |
210e7a43 TG |
1688 | shuffle = shuffle_freelist(s, page); |
1689 | ||
1690 | if (!shuffle) { | |
4d176711 AK |
1691 | start = fixup_red_left(s, start); |
1692 | start = setup_object(s, page, start); | |
1693 | page->freelist = start; | |
18e50661 AK |
1694 | for (idx = 0, p = start; idx < page->objects - 1; idx++) { |
1695 | next = p + s->size; | |
1696 | next = setup_object(s, page, next); | |
1697 | set_freepointer(s, p, next); | |
1698 | p = next; | |
1699 | } | |
1700 | set_freepointer(s, p, NULL); | |
81819f0f | 1701 | } |
81819f0f | 1702 | |
e6e82ea1 | 1703 | page->inuse = page->objects; |
8cb0a506 | 1704 | page->frozen = 1; |
588f8ba9 | 1705 | |
81819f0f | 1706 | out: |
d0164adc | 1707 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1708 | local_irq_disable(); |
1709 | if (!page) | |
1710 | return NULL; | |
1711 | ||
588f8ba9 TG |
1712 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1713 | ||
81819f0f CL |
1714 | return page; |
1715 | } | |
1716 | ||
588f8ba9 TG |
1717 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1718 | { | |
1719 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
bacdcb34 | 1720 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
1721 | flags &= ~GFP_SLAB_BUG_MASK; |
1722 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
1723 | invalid_mask, &invalid_mask, flags, &flags); | |
65b9de75 | 1724 | dump_stack(); |
588f8ba9 TG |
1725 | } |
1726 | ||
1727 | return allocate_slab(s, | |
1728 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1729 | } | |
1730 | ||
81819f0f CL |
1731 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1732 | { | |
834f3d11 CL |
1733 | int order = compound_order(page); |
1734 | int pages = 1 << order; | |
81819f0f | 1735 | |
becfda68 | 1736 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { |
81819f0f CL |
1737 | void *p; |
1738 | ||
1739 | slab_pad_check(s, page); | |
224a88be CL |
1740 | for_each_object(p, s, page_address(page), |
1741 | page->objects) | |
f7cb1933 | 1742 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1743 | } |
1744 | ||
072bb0aa | 1745 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1746 | __ClearPageSlab(page); |
1f458cbf | 1747 | |
d4fc5069 | 1748 | page->mapping = NULL; |
1eb5ac64 NP |
1749 | if (current->reclaim_state) |
1750 | current->reclaim_state->reclaimed_slab += pages; | |
6cea1d56 | 1751 | uncharge_slab_page(page, order, s); |
27ee57c9 | 1752 | __free_pages(page, order); |
81819f0f CL |
1753 | } |
1754 | ||
1755 | static void rcu_free_slab(struct rcu_head *h) | |
1756 | { | |
bf68c214 | 1757 | struct page *page = container_of(h, struct page, rcu_head); |
da9a638c | 1758 | |
1b4f59e3 | 1759 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1760 | } |
1761 | ||
1762 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1763 | { | |
5f0d5a3a | 1764 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bf68c214 | 1765 | call_rcu(&page->rcu_head, rcu_free_slab); |
81819f0f CL |
1766 | } else |
1767 | __free_slab(s, page); | |
1768 | } | |
1769 | ||
1770 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1771 | { | |
205ab99d | 1772 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1773 | free_slab(s, page); |
1774 | } | |
1775 | ||
1776 | /* | |
5cc6eee8 | 1777 | * Management of partially allocated slabs. |
81819f0f | 1778 | */ |
1e4dd946 SR |
1779 | static inline void |
1780 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1781 | { |
e95eed57 | 1782 | n->nr_partial++; |
136333d1 | 1783 | if (tail == DEACTIVATE_TO_TAIL) |
916ac052 | 1784 | list_add_tail(&page->slab_list, &n->partial); |
7c2e132c | 1785 | else |
916ac052 | 1786 | list_add(&page->slab_list, &n->partial); |
81819f0f CL |
1787 | } |
1788 | ||
1e4dd946 SR |
1789 | static inline void add_partial(struct kmem_cache_node *n, |
1790 | struct page *page, int tail) | |
62e346a8 | 1791 | { |
c65c1877 | 1792 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1793 | __add_partial(n, page, tail); |
1794 | } | |
c65c1877 | 1795 | |
1e4dd946 SR |
1796 | static inline void remove_partial(struct kmem_cache_node *n, |
1797 | struct page *page) | |
1798 | { | |
1799 | lockdep_assert_held(&n->list_lock); | |
916ac052 | 1800 | list_del(&page->slab_list); |
52b4b950 | 1801 | n->nr_partial--; |
1e4dd946 SR |
1802 | } |
1803 | ||
81819f0f | 1804 | /* |
7ced3719 CL |
1805 | * Remove slab from the partial list, freeze it and |
1806 | * return the pointer to the freelist. | |
81819f0f | 1807 | * |
497b66f2 | 1808 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1809 | */ |
497b66f2 | 1810 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1811 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1812 | int mode, int *objects) |
81819f0f | 1813 | { |
2cfb7455 CL |
1814 | void *freelist; |
1815 | unsigned long counters; | |
1816 | struct page new; | |
1817 | ||
c65c1877 PZ |
1818 | lockdep_assert_held(&n->list_lock); |
1819 | ||
2cfb7455 CL |
1820 | /* |
1821 | * Zap the freelist and set the frozen bit. | |
1822 | * The old freelist is the list of objects for the | |
1823 | * per cpu allocation list. | |
1824 | */ | |
7ced3719 CL |
1825 | freelist = page->freelist; |
1826 | counters = page->counters; | |
1827 | new.counters = counters; | |
633b0764 | 1828 | *objects = new.objects - new.inuse; |
23910c50 | 1829 | if (mode) { |
7ced3719 | 1830 | new.inuse = page->objects; |
23910c50 PE |
1831 | new.freelist = NULL; |
1832 | } else { | |
1833 | new.freelist = freelist; | |
1834 | } | |
2cfb7455 | 1835 | |
a0132ac0 | 1836 | VM_BUG_ON(new.frozen); |
7ced3719 | 1837 | new.frozen = 1; |
2cfb7455 | 1838 | |
7ced3719 | 1839 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1840 | freelist, counters, |
02d7633f | 1841 | new.freelist, new.counters, |
7ced3719 | 1842 | "acquire_slab")) |
7ced3719 | 1843 | return NULL; |
2cfb7455 CL |
1844 | |
1845 | remove_partial(n, page); | |
7ced3719 | 1846 | WARN_ON(!freelist); |
49e22585 | 1847 | return freelist; |
81819f0f CL |
1848 | } |
1849 | ||
633b0764 | 1850 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1851 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1852 | |
81819f0f | 1853 | /* |
672bba3a | 1854 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1855 | */ |
8ba00bb6 JK |
1856 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1857 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1858 | { |
49e22585 CL |
1859 | struct page *page, *page2; |
1860 | void *object = NULL; | |
e5d9998f | 1861 | unsigned int available = 0; |
633b0764 | 1862 | int objects; |
81819f0f CL |
1863 | |
1864 | /* | |
1865 | * Racy check. If we mistakenly see no partial slabs then we | |
1866 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1867 | * partial slab and there is none available then get_partials() |
1868 | * will return NULL. | |
81819f0f CL |
1869 | */ |
1870 | if (!n || !n->nr_partial) | |
1871 | return NULL; | |
1872 | ||
1873 | spin_lock(&n->list_lock); | |
916ac052 | 1874 | list_for_each_entry_safe(page, page2, &n->partial, slab_list) { |
8ba00bb6 | 1875 | void *t; |
49e22585 | 1876 | |
8ba00bb6 JK |
1877 | if (!pfmemalloc_match(page, flags)) |
1878 | continue; | |
1879 | ||
633b0764 | 1880 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1881 | if (!t) |
1882 | break; | |
1883 | ||
633b0764 | 1884 | available += objects; |
12d79634 | 1885 | if (!object) { |
49e22585 | 1886 | c->page = page; |
49e22585 | 1887 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1888 | object = t; |
49e22585 | 1889 | } else { |
633b0764 | 1890 | put_cpu_partial(s, page, 0); |
8028dcea | 1891 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1892 | } |
345c905d | 1893 | if (!kmem_cache_has_cpu_partial(s) |
e6d0e1dc | 1894 | || available > slub_cpu_partial(s) / 2) |
49e22585 CL |
1895 | break; |
1896 | ||
497b66f2 | 1897 | } |
81819f0f | 1898 | spin_unlock(&n->list_lock); |
497b66f2 | 1899 | return object; |
81819f0f CL |
1900 | } |
1901 | ||
1902 | /* | |
672bba3a | 1903 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1904 | */ |
de3ec035 | 1905 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1906 | struct kmem_cache_cpu *c) |
81819f0f CL |
1907 | { |
1908 | #ifdef CONFIG_NUMA | |
1909 | struct zonelist *zonelist; | |
dd1a239f | 1910 | struct zoneref *z; |
54a6eb5c MG |
1911 | struct zone *zone; |
1912 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1913 | void *object; |
cc9a6c87 | 1914 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1915 | |
1916 | /* | |
672bba3a CL |
1917 | * The defrag ratio allows a configuration of the tradeoffs between |
1918 | * inter node defragmentation and node local allocations. A lower | |
1919 | * defrag_ratio increases the tendency to do local allocations | |
1920 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1921 | * |
672bba3a CL |
1922 | * If the defrag_ratio is set to 0 then kmalloc() always |
1923 | * returns node local objects. If the ratio is higher then kmalloc() | |
1924 | * may return off node objects because partial slabs are obtained | |
1925 | * from other nodes and filled up. | |
81819f0f | 1926 | * |
43efd3ea LP |
1927 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
1928 | * (which makes defrag_ratio = 1000) then every (well almost) | |
1929 | * allocation will first attempt to defrag slab caches on other nodes. | |
1930 | * This means scanning over all nodes to look for partial slabs which | |
1931 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 1932 | * with available objects. |
81819f0f | 1933 | */ |
9824601e CL |
1934 | if (!s->remote_node_defrag_ratio || |
1935 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1936 | return NULL; |
1937 | ||
cc9a6c87 | 1938 | do { |
d26914d1 | 1939 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1940 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1941 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1942 | struct kmem_cache_node *n; | |
1943 | ||
1944 | n = get_node(s, zone_to_nid(zone)); | |
1945 | ||
dee2f8aa | 1946 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1947 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1948 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1949 | if (object) { |
1950 | /* | |
d26914d1 MG |
1951 | * Don't check read_mems_allowed_retry() |
1952 | * here - if mems_allowed was updated in | |
1953 | * parallel, that was a harmless race | |
1954 | * between allocation and the cpuset | |
1955 | * update | |
cc9a6c87 | 1956 | */ |
cc9a6c87 MG |
1957 | return object; |
1958 | } | |
c0ff7453 | 1959 | } |
81819f0f | 1960 | } |
d26914d1 | 1961 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
6dfd1b65 | 1962 | #endif /* CONFIG_NUMA */ |
81819f0f CL |
1963 | return NULL; |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * Get a partial page, lock it and return it. | |
1968 | */ | |
497b66f2 | 1969 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1970 | struct kmem_cache_cpu *c) |
81819f0f | 1971 | { |
497b66f2 | 1972 | void *object; |
a561ce00 JK |
1973 | int searchnode = node; |
1974 | ||
1975 | if (node == NUMA_NO_NODE) | |
1976 | searchnode = numa_mem_id(); | |
81819f0f | 1977 | |
8ba00bb6 | 1978 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1979 | if (object || node != NUMA_NO_NODE) |
1980 | return object; | |
81819f0f | 1981 | |
acd19fd1 | 1982 | return get_any_partial(s, flags, c); |
81819f0f CL |
1983 | } |
1984 | ||
923717cb | 1985 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba CL |
1986 | /* |
1987 | * Calculate the next globally unique transaction for disambiguiation | |
1988 | * during cmpxchg. The transactions start with the cpu number and are then | |
1989 | * incremented by CONFIG_NR_CPUS. | |
1990 | */ | |
1991 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1992 | #else | |
1993 | /* | |
1994 | * No preemption supported therefore also no need to check for | |
1995 | * different cpus. | |
1996 | */ | |
1997 | #define TID_STEP 1 | |
1998 | #endif | |
1999 | ||
2000 | static inline unsigned long next_tid(unsigned long tid) | |
2001 | { | |
2002 | return tid + TID_STEP; | |
2003 | } | |
2004 | ||
9d5f0be0 | 2005 | #ifdef SLUB_DEBUG_CMPXCHG |
8a5ec0ba CL |
2006 | static inline unsigned int tid_to_cpu(unsigned long tid) |
2007 | { | |
2008 | return tid % TID_STEP; | |
2009 | } | |
2010 | ||
2011 | static inline unsigned long tid_to_event(unsigned long tid) | |
2012 | { | |
2013 | return tid / TID_STEP; | |
2014 | } | |
9d5f0be0 | 2015 | #endif |
8a5ec0ba CL |
2016 | |
2017 | static inline unsigned int init_tid(int cpu) | |
2018 | { | |
2019 | return cpu; | |
2020 | } | |
2021 | ||
2022 | static inline void note_cmpxchg_failure(const char *n, | |
2023 | const struct kmem_cache *s, unsigned long tid) | |
2024 | { | |
2025 | #ifdef SLUB_DEBUG_CMPXCHG | |
2026 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2027 | ||
f9f58285 | 2028 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba | 2029 | |
923717cb | 2030 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2031 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) |
f9f58285 | 2032 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2033 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2034 | else | |
2035 | #endif | |
2036 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2037 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2038 | tid_to_event(tid), tid_to_event(actual_tid)); |
2039 | else | |
f9f58285 | 2040 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2041 | actual_tid, tid, next_tid(tid)); |
2042 | #endif | |
4fdccdfb | 2043 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2044 | } |
2045 | ||
788e1aad | 2046 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2047 | { |
8a5ec0ba CL |
2048 | int cpu; |
2049 | ||
2050 | for_each_possible_cpu(cpu) | |
2051 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 2052 | } |
2cfb7455 | 2053 | |
81819f0f CL |
2054 | /* |
2055 | * Remove the cpu slab | |
2056 | */ | |
d0e0ac97 | 2057 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
d4ff6d35 | 2058 | void *freelist, struct kmem_cache_cpu *c) |
81819f0f | 2059 | { |
2cfb7455 | 2060 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
2061 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
2062 | int lock = 0; | |
2063 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 2064 | void *nextfree; |
136333d1 | 2065 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2066 | struct page new; |
2067 | struct page old; | |
2068 | ||
2069 | if (page->freelist) { | |
84e554e6 | 2070 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2071 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2072 | } |
2073 | ||
894b8788 | 2074 | /* |
2cfb7455 CL |
2075 | * Stage one: Free all available per cpu objects back |
2076 | * to the page freelist while it is still frozen. Leave the | |
2077 | * last one. | |
2078 | * | |
2079 | * There is no need to take the list->lock because the page | |
2080 | * is still frozen. | |
2081 | */ | |
2082 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2083 | void *prior; | |
2084 | unsigned long counters; | |
2085 | ||
2086 | do { | |
2087 | prior = page->freelist; | |
2088 | counters = page->counters; | |
2089 | set_freepointer(s, freelist, prior); | |
2090 | new.counters = counters; | |
2091 | new.inuse--; | |
a0132ac0 | 2092 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 2093 | |
1d07171c | 2094 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2095 | prior, counters, |
2096 | freelist, new.counters, | |
2097 | "drain percpu freelist")); | |
2098 | ||
2099 | freelist = nextfree; | |
2100 | } | |
2101 | ||
894b8788 | 2102 | /* |
2cfb7455 CL |
2103 | * Stage two: Ensure that the page is unfrozen while the |
2104 | * list presence reflects the actual number of objects | |
2105 | * during unfreeze. | |
2106 | * | |
2107 | * We setup the list membership and then perform a cmpxchg | |
2108 | * with the count. If there is a mismatch then the page | |
2109 | * is not unfrozen but the page is on the wrong list. | |
2110 | * | |
2111 | * Then we restart the process which may have to remove | |
2112 | * the page from the list that we just put it on again | |
2113 | * because the number of objects in the slab may have | |
2114 | * changed. | |
894b8788 | 2115 | */ |
2cfb7455 | 2116 | redo: |
894b8788 | 2117 | |
2cfb7455 CL |
2118 | old.freelist = page->freelist; |
2119 | old.counters = page->counters; | |
a0132ac0 | 2120 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2121 | |
2cfb7455 CL |
2122 | /* Determine target state of the slab */ |
2123 | new.counters = old.counters; | |
2124 | if (freelist) { | |
2125 | new.inuse--; | |
2126 | set_freepointer(s, freelist, old.freelist); | |
2127 | new.freelist = freelist; | |
2128 | } else | |
2129 | new.freelist = old.freelist; | |
2130 | ||
2131 | new.frozen = 0; | |
2132 | ||
8a5b20ae | 2133 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2134 | m = M_FREE; |
2135 | else if (new.freelist) { | |
2136 | m = M_PARTIAL; | |
2137 | if (!lock) { | |
2138 | lock = 1; | |
2139 | /* | |
8bb4e7a2 | 2140 | * Taking the spinlock removes the possibility |
2cfb7455 CL |
2141 | * that acquire_slab() will see a slab page that |
2142 | * is frozen | |
2143 | */ | |
2144 | spin_lock(&n->list_lock); | |
2145 | } | |
2146 | } else { | |
2147 | m = M_FULL; | |
2148 | if (kmem_cache_debug(s) && !lock) { | |
2149 | lock = 1; | |
2150 | /* | |
2151 | * This also ensures that the scanning of full | |
2152 | * slabs from diagnostic functions will not see | |
2153 | * any frozen slabs. | |
2154 | */ | |
2155 | spin_lock(&n->list_lock); | |
2156 | } | |
2157 | } | |
2158 | ||
2159 | if (l != m) { | |
2cfb7455 | 2160 | if (l == M_PARTIAL) |
2cfb7455 | 2161 | remove_partial(n, page); |
2cfb7455 | 2162 | else if (l == M_FULL) |
c65c1877 | 2163 | remove_full(s, n, page); |
2cfb7455 | 2164 | |
88349a28 | 2165 | if (m == M_PARTIAL) |
2cfb7455 | 2166 | add_partial(n, page, tail); |
88349a28 | 2167 | else if (m == M_FULL) |
2cfb7455 | 2168 | add_full(s, n, page); |
2cfb7455 CL |
2169 | } |
2170 | ||
2171 | l = m; | |
1d07171c | 2172 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2173 | old.freelist, old.counters, |
2174 | new.freelist, new.counters, | |
2175 | "unfreezing slab")) | |
2176 | goto redo; | |
2177 | ||
2cfb7455 CL |
2178 | if (lock) |
2179 | spin_unlock(&n->list_lock); | |
2180 | ||
88349a28 WY |
2181 | if (m == M_PARTIAL) |
2182 | stat(s, tail); | |
2183 | else if (m == M_FULL) | |
2184 | stat(s, DEACTIVATE_FULL); | |
2185 | else if (m == M_FREE) { | |
2cfb7455 CL |
2186 | stat(s, DEACTIVATE_EMPTY); |
2187 | discard_slab(s, page); | |
2188 | stat(s, FREE_SLAB); | |
894b8788 | 2189 | } |
d4ff6d35 WY |
2190 | |
2191 | c->page = NULL; | |
2192 | c->freelist = NULL; | |
81819f0f CL |
2193 | } |
2194 | ||
d24ac77f JK |
2195 | /* |
2196 | * Unfreeze all the cpu partial slabs. | |
2197 | * | |
59a09917 CL |
2198 | * This function must be called with interrupts disabled |
2199 | * for the cpu using c (or some other guarantee must be there | |
2200 | * to guarantee no concurrent accesses). | |
d24ac77f | 2201 | */ |
59a09917 CL |
2202 | static void unfreeze_partials(struct kmem_cache *s, |
2203 | struct kmem_cache_cpu *c) | |
49e22585 | 2204 | { |
345c905d | 2205 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2206 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2207 | struct page *page, *discard_page = NULL; |
49e22585 | 2208 | |
4c7ba22e | 2209 | while ((page = slub_percpu_partial(c))) { |
49e22585 CL |
2210 | struct page new; |
2211 | struct page old; | |
2212 | ||
4c7ba22e | 2213 | slub_set_percpu_partial(c, page); |
43d77867 JK |
2214 | |
2215 | n2 = get_node(s, page_to_nid(page)); | |
2216 | if (n != n2) { | |
2217 | if (n) | |
2218 | spin_unlock(&n->list_lock); | |
2219 | ||
2220 | n = n2; | |
2221 | spin_lock(&n->list_lock); | |
2222 | } | |
49e22585 CL |
2223 | |
2224 | do { | |
2225 | ||
2226 | old.freelist = page->freelist; | |
2227 | old.counters = page->counters; | |
a0132ac0 | 2228 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2229 | |
2230 | new.counters = old.counters; | |
2231 | new.freelist = old.freelist; | |
2232 | ||
2233 | new.frozen = 0; | |
2234 | ||
d24ac77f | 2235 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2236 | old.freelist, old.counters, |
2237 | new.freelist, new.counters, | |
2238 | "unfreezing slab")); | |
2239 | ||
8a5b20ae | 2240 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2241 | page->next = discard_page; |
2242 | discard_page = page; | |
43d77867 JK |
2243 | } else { |
2244 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2245 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2246 | } |
2247 | } | |
2248 | ||
2249 | if (n) | |
2250 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2251 | |
2252 | while (discard_page) { | |
2253 | page = discard_page; | |
2254 | discard_page = discard_page->next; | |
2255 | ||
2256 | stat(s, DEACTIVATE_EMPTY); | |
2257 | discard_slab(s, page); | |
2258 | stat(s, FREE_SLAB); | |
2259 | } | |
6dfd1b65 | 2260 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2261 | } |
2262 | ||
2263 | /* | |
9234bae9 WY |
2264 | * Put a page that was just frozen (in __slab_free|get_partial_node) into a |
2265 | * partial page slot if available. | |
49e22585 CL |
2266 | * |
2267 | * If we did not find a slot then simply move all the partials to the | |
2268 | * per node partial list. | |
2269 | */ | |
633b0764 | 2270 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2271 | { |
345c905d | 2272 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2273 | struct page *oldpage; |
2274 | int pages; | |
2275 | int pobjects; | |
2276 | ||
d6e0b7fa | 2277 | preempt_disable(); |
49e22585 CL |
2278 | do { |
2279 | pages = 0; | |
2280 | pobjects = 0; | |
2281 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2282 | ||
2283 | if (oldpage) { | |
2284 | pobjects = oldpage->pobjects; | |
2285 | pages = oldpage->pages; | |
bbd4e305 | 2286 | if (drain && pobjects > slub_cpu_partial(s)) { |
49e22585 CL |
2287 | unsigned long flags; |
2288 | /* | |
2289 | * partial array is full. Move the existing | |
2290 | * set to the per node partial list. | |
2291 | */ | |
2292 | local_irq_save(flags); | |
59a09917 | 2293 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2294 | local_irq_restore(flags); |
e24fc410 | 2295 | oldpage = NULL; |
49e22585 CL |
2296 | pobjects = 0; |
2297 | pages = 0; | |
8028dcea | 2298 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2299 | } |
2300 | } | |
2301 | ||
2302 | pages++; | |
2303 | pobjects += page->objects - page->inuse; | |
2304 | ||
2305 | page->pages = pages; | |
2306 | page->pobjects = pobjects; | |
2307 | page->next = oldpage; | |
2308 | ||
d0e0ac97 CG |
2309 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2310 | != oldpage); | |
bbd4e305 | 2311 | if (unlikely(!slub_cpu_partial(s))) { |
d6e0b7fa VD |
2312 | unsigned long flags; |
2313 | ||
2314 | local_irq_save(flags); | |
2315 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2316 | local_irq_restore(flags); | |
2317 | } | |
2318 | preempt_enable(); | |
6dfd1b65 | 2319 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2320 | } |
2321 | ||
dfb4f096 | 2322 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2323 | { |
84e554e6 | 2324 | stat(s, CPUSLAB_FLUSH); |
d4ff6d35 | 2325 | deactivate_slab(s, c->page, c->freelist, c); |
c17dda40 CL |
2326 | |
2327 | c->tid = next_tid(c->tid); | |
81819f0f CL |
2328 | } |
2329 | ||
2330 | /* | |
2331 | * Flush cpu slab. | |
6446faa2 | 2332 | * |
81819f0f CL |
2333 | * Called from IPI handler with interrupts disabled. |
2334 | */ | |
0c710013 | 2335 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2336 | { |
9dfc6e68 | 2337 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2338 | |
1265ef2d WY |
2339 | if (c->page) |
2340 | flush_slab(s, c); | |
49e22585 | 2341 | |
1265ef2d | 2342 | unfreeze_partials(s, c); |
81819f0f CL |
2343 | } |
2344 | ||
2345 | static void flush_cpu_slab(void *d) | |
2346 | { | |
2347 | struct kmem_cache *s = d; | |
81819f0f | 2348 | |
dfb4f096 | 2349 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2350 | } |
2351 | ||
a8364d55 GBY |
2352 | static bool has_cpu_slab(int cpu, void *info) |
2353 | { | |
2354 | struct kmem_cache *s = info; | |
2355 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2356 | ||
a93cf07b | 2357 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2358 | } |
2359 | ||
81819f0f CL |
2360 | static void flush_all(struct kmem_cache *s) |
2361 | { | |
cb923159 | 2362 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1); |
81819f0f CL |
2363 | } |
2364 | ||
a96a87bf SAS |
2365 | /* |
2366 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2367 | * necessary. | |
2368 | */ | |
2369 | static int slub_cpu_dead(unsigned int cpu) | |
2370 | { | |
2371 | struct kmem_cache *s; | |
2372 | unsigned long flags; | |
2373 | ||
2374 | mutex_lock(&slab_mutex); | |
2375 | list_for_each_entry(s, &slab_caches, list) { | |
2376 | local_irq_save(flags); | |
2377 | __flush_cpu_slab(s, cpu); | |
2378 | local_irq_restore(flags); | |
2379 | } | |
2380 | mutex_unlock(&slab_mutex); | |
2381 | return 0; | |
2382 | } | |
2383 | ||
dfb4f096 CL |
2384 | /* |
2385 | * Check if the objects in a per cpu structure fit numa | |
2386 | * locality expectations. | |
2387 | */ | |
57d437d2 | 2388 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2389 | { |
2390 | #ifdef CONFIG_NUMA | |
6159d0f5 | 2391 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2392 | return 0; |
2393 | #endif | |
2394 | return 1; | |
2395 | } | |
2396 | ||
9a02d699 | 2397 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2398 | static int count_free(struct page *page) |
2399 | { | |
2400 | return page->objects - page->inuse; | |
2401 | } | |
2402 | ||
9a02d699 DR |
2403 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2404 | { | |
2405 | return atomic_long_read(&n->total_objects); | |
2406 | } | |
2407 | #endif /* CONFIG_SLUB_DEBUG */ | |
2408 | ||
2409 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2410 | static unsigned long count_partial(struct kmem_cache_node *n, |
2411 | int (*get_count)(struct page *)) | |
2412 | { | |
2413 | unsigned long flags; | |
2414 | unsigned long x = 0; | |
2415 | struct page *page; | |
2416 | ||
2417 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 2418 | list_for_each_entry(page, &n->partial, slab_list) |
781b2ba6 PE |
2419 | x += get_count(page); |
2420 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2421 | return x; | |
2422 | } | |
9a02d699 | 2423 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2424 | |
781b2ba6 PE |
2425 | static noinline void |
2426 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2427 | { | |
9a02d699 DR |
2428 | #ifdef CONFIG_SLUB_DEBUG |
2429 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2430 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2431 | int node; |
fa45dc25 | 2432 | struct kmem_cache_node *n; |
781b2ba6 | 2433 | |
9a02d699 DR |
2434 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2435 | return; | |
2436 | ||
5b3810e5 VB |
2437 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2438 | nid, gfpflags, &gfpflags); | |
19af27af | 2439 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2440 | s->name, s->object_size, s->size, oo_order(s->oo), |
2441 | oo_order(s->min)); | |
781b2ba6 | 2442 | |
3b0efdfa | 2443 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2444 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2445 | s->name); | |
fa5ec8a1 | 2446 | |
fa45dc25 | 2447 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2448 | unsigned long nr_slabs; |
2449 | unsigned long nr_objs; | |
2450 | unsigned long nr_free; | |
2451 | ||
26c02cf0 AB |
2452 | nr_free = count_partial(n, count_free); |
2453 | nr_slabs = node_nr_slabs(n); | |
2454 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2455 | |
f9f58285 | 2456 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2457 | node, nr_slabs, nr_objs, nr_free); |
2458 | } | |
9a02d699 | 2459 | #endif |
781b2ba6 PE |
2460 | } |
2461 | ||
497b66f2 CL |
2462 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2463 | int node, struct kmem_cache_cpu **pc) | |
2464 | { | |
6faa6833 | 2465 | void *freelist; |
188fd063 CL |
2466 | struct kmem_cache_cpu *c = *pc; |
2467 | struct page *page; | |
497b66f2 | 2468 | |
128227e7 MW |
2469 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
2470 | ||
188fd063 | 2471 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2472 | |
188fd063 CL |
2473 | if (freelist) |
2474 | return freelist; | |
2475 | ||
2476 | page = new_slab(s, flags, node); | |
497b66f2 | 2477 | if (page) { |
7c8e0181 | 2478 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2479 | if (c->page) |
2480 | flush_slab(s, c); | |
2481 | ||
2482 | /* | |
2483 | * No other reference to the page yet so we can | |
2484 | * muck around with it freely without cmpxchg | |
2485 | */ | |
6faa6833 | 2486 | freelist = page->freelist; |
497b66f2 CL |
2487 | page->freelist = NULL; |
2488 | ||
2489 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2490 | c->page = page; |
2491 | *pc = c; | |
edde82b6 | 2492 | } |
497b66f2 | 2493 | |
6faa6833 | 2494 | return freelist; |
497b66f2 CL |
2495 | } |
2496 | ||
072bb0aa MG |
2497 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2498 | { | |
2499 | if (unlikely(PageSlabPfmemalloc(page))) | |
2500 | return gfp_pfmemalloc_allowed(gfpflags); | |
2501 | ||
2502 | return true; | |
2503 | } | |
2504 | ||
213eeb9f | 2505 | /* |
d0e0ac97 CG |
2506 | * Check the page->freelist of a page and either transfer the freelist to the |
2507 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2508 | * |
2509 | * The page is still frozen if the return value is not NULL. | |
2510 | * | |
2511 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2512 | * |
2513 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2514 | */ |
2515 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2516 | { | |
2517 | struct page new; | |
2518 | unsigned long counters; | |
2519 | void *freelist; | |
2520 | ||
2521 | do { | |
2522 | freelist = page->freelist; | |
2523 | counters = page->counters; | |
6faa6833 | 2524 | |
213eeb9f | 2525 | new.counters = counters; |
a0132ac0 | 2526 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2527 | |
2528 | new.inuse = page->objects; | |
2529 | new.frozen = freelist != NULL; | |
2530 | ||
d24ac77f | 2531 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2532 | freelist, counters, |
2533 | NULL, new.counters, | |
2534 | "get_freelist")); | |
2535 | ||
2536 | return freelist; | |
2537 | } | |
2538 | ||
81819f0f | 2539 | /* |
894b8788 CL |
2540 | * Slow path. The lockless freelist is empty or we need to perform |
2541 | * debugging duties. | |
2542 | * | |
894b8788 CL |
2543 | * Processing is still very fast if new objects have been freed to the |
2544 | * regular freelist. In that case we simply take over the regular freelist | |
2545 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2546 | * |
894b8788 CL |
2547 | * If that is not working then we fall back to the partial lists. We take the |
2548 | * first element of the freelist as the object to allocate now and move the | |
2549 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2550 | * |
894b8788 | 2551 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2552 | * we need to allocate a new slab. This is the slowest path since it involves |
2553 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2554 | * |
2555 | * Version of __slab_alloc to use when we know that interrupts are | |
2556 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2557 | */ |
a380a3c7 | 2558 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2559 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2560 | { |
6faa6833 | 2561 | void *freelist; |
f6e7def7 | 2562 | struct page *page; |
81819f0f | 2563 | |
f6e7def7 | 2564 | page = c->page; |
0715e6c5 VB |
2565 | if (!page) { |
2566 | /* | |
2567 | * if the node is not online or has no normal memory, just | |
2568 | * ignore the node constraint | |
2569 | */ | |
2570 | if (unlikely(node != NUMA_NO_NODE && | |
2571 | !node_state(node, N_NORMAL_MEMORY))) | |
2572 | node = NUMA_NO_NODE; | |
81819f0f | 2573 | goto new_slab; |
0715e6c5 | 2574 | } |
49e22585 | 2575 | redo: |
6faa6833 | 2576 | |
57d437d2 | 2577 | if (unlikely(!node_match(page, node))) { |
0715e6c5 VB |
2578 | /* |
2579 | * same as above but node_match() being false already | |
2580 | * implies node != NUMA_NO_NODE | |
2581 | */ | |
2582 | if (!node_state(node, N_NORMAL_MEMORY)) { | |
2583 | node = NUMA_NO_NODE; | |
2584 | goto redo; | |
2585 | } else { | |
a561ce00 | 2586 | stat(s, ALLOC_NODE_MISMATCH); |
d4ff6d35 | 2587 | deactivate_slab(s, page, c->freelist, c); |
a561ce00 JK |
2588 | goto new_slab; |
2589 | } | |
fc59c053 | 2590 | } |
6446faa2 | 2591 | |
072bb0aa MG |
2592 | /* |
2593 | * By rights, we should be searching for a slab page that was | |
2594 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2595 | * information when the page leaves the per-cpu allocator | |
2596 | */ | |
2597 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
d4ff6d35 | 2598 | deactivate_slab(s, page, c->freelist, c); |
072bb0aa MG |
2599 | goto new_slab; |
2600 | } | |
2601 | ||
73736e03 | 2602 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2603 | freelist = c->freelist; |
2604 | if (freelist) | |
73736e03 | 2605 | goto load_freelist; |
03e404af | 2606 | |
f6e7def7 | 2607 | freelist = get_freelist(s, page); |
6446faa2 | 2608 | |
6faa6833 | 2609 | if (!freelist) { |
03e404af CL |
2610 | c->page = NULL; |
2611 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2612 | goto new_slab; |
03e404af | 2613 | } |
6446faa2 | 2614 | |
84e554e6 | 2615 | stat(s, ALLOC_REFILL); |
6446faa2 | 2616 | |
894b8788 | 2617 | load_freelist: |
507effea CL |
2618 | /* |
2619 | * freelist is pointing to the list of objects to be used. | |
2620 | * page is pointing to the page from which the objects are obtained. | |
2621 | * That page must be frozen for per cpu allocations to work. | |
2622 | */ | |
a0132ac0 | 2623 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2624 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2625 | c->tid = next_tid(c->tid); |
6faa6833 | 2626 | return freelist; |
81819f0f | 2627 | |
81819f0f | 2628 | new_slab: |
2cfb7455 | 2629 | |
a93cf07b WY |
2630 | if (slub_percpu_partial(c)) { |
2631 | page = c->page = slub_percpu_partial(c); | |
2632 | slub_set_percpu_partial(c, page); | |
49e22585 | 2633 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2634 | goto redo; |
81819f0f CL |
2635 | } |
2636 | ||
188fd063 | 2637 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2638 | |
f4697436 | 2639 | if (unlikely(!freelist)) { |
9a02d699 | 2640 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2641 | return NULL; |
81819f0f | 2642 | } |
2cfb7455 | 2643 | |
f6e7def7 | 2644 | page = c->page; |
5091b74a | 2645 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2646 | goto load_freelist; |
2cfb7455 | 2647 | |
497b66f2 | 2648 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2649 | if (kmem_cache_debug(s) && |
2650 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2651 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2652 | |
d4ff6d35 | 2653 | deactivate_slab(s, page, get_freepointer(s, freelist), c); |
6faa6833 | 2654 | return freelist; |
894b8788 CL |
2655 | } |
2656 | ||
a380a3c7 CL |
2657 | /* |
2658 | * Another one that disabled interrupt and compensates for possible | |
2659 | * cpu changes by refetching the per cpu area pointer. | |
2660 | */ | |
2661 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2662 | unsigned long addr, struct kmem_cache_cpu *c) | |
2663 | { | |
2664 | void *p; | |
2665 | unsigned long flags; | |
2666 | ||
2667 | local_irq_save(flags); | |
923717cb | 2668 | #ifdef CONFIG_PREEMPTION |
a380a3c7 CL |
2669 | /* |
2670 | * We may have been preempted and rescheduled on a different | |
2671 | * cpu before disabling interrupts. Need to reload cpu area | |
2672 | * pointer. | |
2673 | */ | |
2674 | c = this_cpu_ptr(s->cpu_slab); | |
2675 | #endif | |
2676 | ||
2677 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2678 | local_irq_restore(flags); | |
2679 | return p; | |
2680 | } | |
2681 | ||
0f181f9f AP |
2682 | /* |
2683 | * If the object has been wiped upon free, make sure it's fully initialized by | |
2684 | * zeroing out freelist pointer. | |
2685 | */ | |
2686 | static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s, | |
2687 | void *obj) | |
2688 | { | |
2689 | if (unlikely(slab_want_init_on_free(s)) && obj) | |
2690 | memset((void *)((char *)obj + s->offset), 0, sizeof(void *)); | |
2691 | } | |
2692 | ||
894b8788 CL |
2693 | /* |
2694 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2695 | * have the fastpath folded into their functions. So no function call | |
2696 | * overhead for requests that can be satisfied on the fastpath. | |
2697 | * | |
2698 | * The fastpath works by first checking if the lockless freelist can be used. | |
2699 | * If not then __slab_alloc is called for slow processing. | |
2700 | * | |
2701 | * Otherwise we can simply pick the next object from the lockless free list. | |
2702 | */ | |
2b847c3c | 2703 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2704 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2705 | { |
03ec0ed5 | 2706 | void *object; |
dfb4f096 | 2707 | struct kmem_cache_cpu *c; |
57d437d2 | 2708 | struct page *page; |
8a5ec0ba | 2709 | unsigned long tid; |
1f84260c | 2710 | |
8135be5a VD |
2711 | s = slab_pre_alloc_hook(s, gfpflags); |
2712 | if (!s) | |
773ff60e | 2713 | return NULL; |
8a5ec0ba | 2714 | redo: |
8a5ec0ba CL |
2715 | /* |
2716 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2717 | * enabled. We may switch back and forth between cpus while | |
2718 | * reading from one cpu area. That does not matter as long | |
2719 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2720 | * |
9aabf810 | 2721 | * We should guarantee that tid and kmem_cache are retrieved on |
923717cb | 2722 | * the same cpu. It could be different if CONFIG_PREEMPTION so we need |
9aabf810 | 2723 | * to check if it is matched or not. |
8a5ec0ba | 2724 | */ |
9aabf810 JK |
2725 | do { |
2726 | tid = this_cpu_read(s->cpu_slab->tid); | |
2727 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 2728 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 2729 | unlikely(tid != READ_ONCE(c->tid))); |
9aabf810 JK |
2730 | |
2731 | /* | |
2732 | * Irqless object alloc/free algorithm used here depends on sequence | |
2733 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2734 | * on c to guarantee that object and page associated with previous tid | |
2735 | * won't be used with current tid. If we fetch tid first, object and | |
2736 | * page could be one associated with next tid and our alloc/free | |
2737 | * request will be failed. In this case, we will retry. So, no problem. | |
2738 | */ | |
2739 | barrier(); | |
8a5ec0ba | 2740 | |
8a5ec0ba CL |
2741 | /* |
2742 | * The transaction ids are globally unique per cpu and per operation on | |
2743 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2744 | * occurs on the right processor and that there was no operation on the | |
2745 | * linked list in between. | |
2746 | */ | |
8a5ec0ba | 2747 | |
9dfc6e68 | 2748 | object = c->freelist; |
57d437d2 | 2749 | page = c->page; |
8eae1492 | 2750 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2751 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2752 | stat(s, ALLOC_SLOWPATH); |
2753 | } else { | |
0ad9500e ED |
2754 | void *next_object = get_freepointer_safe(s, object); |
2755 | ||
8a5ec0ba | 2756 | /* |
25985edc | 2757 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2758 | * operation and if we are on the right processor. |
2759 | * | |
d0e0ac97 CG |
2760 | * The cmpxchg does the following atomically (without lock |
2761 | * semantics!) | |
8a5ec0ba CL |
2762 | * 1. Relocate first pointer to the current per cpu area. |
2763 | * 2. Verify that tid and freelist have not been changed | |
2764 | * 3. If they were not changed replace tid and freelist | |
2765 | * | |
d0e0ac97 CG |
2766 | * Since this is without lock semantics the protection is only |
2767 | * against code executing on this cpu *not* from access by | |
2768 | * other cpus. | |
8a5ec0ba | 2769 | */ |
933393f5 | 2770 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2771 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2772 | object, tid, | |
0ad9500e | 2773 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2774 | |
2775 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2776 | goto redo; | |
2777 | } | |
0ad9500e | 2778 | prefetch_freepointer(s, next_object); |
84e554e6 | 2779 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2780 | } |
0f181f9f AP |
2781 | |
2782 | maybe_wipe_obj_freeptr(s, object); | |
8a5ec0ba | 2783 | |
6471384a | 2784 | if (unlikely(slab_want_init_on_alloc(gfpflags, s)) && object) |
3b0efdfa | 2785 | memset(object, 0, s->object_size); |
d07dbea4 | 2786 | |
03ec0ed5 | 2787 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2788 | |
894b8788 | 2789 | return object; |
81819f0f CL |
2790 | } |
2791 | ||
2b847c3c EG |
2792 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2793 | gfp_t gfpflags, unsigned long addr) | |
2794 | { | |
2795 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2796 | } | |
2797 | ||
81819f0f CL |
2798 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2799 | { | |
2b847c3c | 2800 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2801 | |
d0e0ac97 CG |
2802 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2803 | s->size, gfpflags); | |
5b882be4 EGM |
2804 | |
2805 | return ret; | |
81819f0f CL |
2806 | } |
2807 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2808 | ||
0f24f128 | 2809 | #ifdef CONFIG_TRACING |
4a92379b RK |
2810 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2811 | { | |
2b847c3c | 2812 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2813 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0116523c | 2814 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2815 | return ret; |
2816 | } | |
2817 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2818 | #endif |
2819 | ||
81819f0f CL |
2820 | #ifdef CONFIG_NUMA |
2821 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2822 | { | |
2b847c3c | 2823 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2824 | |
ca2b84cb | 2825 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2826 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2827 | |
2828 | return ret; | |
81819f0f CL |
2829 | } |
2830 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2831 | |
0f24f128 | 2832 | #ifdef CONFIG_TRACING |
4a92379b | 2833 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2834 | gfp_t gfpflags, |
4a92379b | 2835 | int node, size_t size) |
5b882be4 | 2836 | { |
2b847c3c | 2837 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2838 | |
2839 | trace_kmalloc_node(_RET_IP_, ret, | |
2840 | size, s->size, gfpflags, node); | |
0316bec2 | 2841 | |
0116523c | 2842 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2843 | return ret; |
5b882be4 | 2844 | } |
4a92379b | 2845 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2846 | #endif |
6dfd1b65 | 2847 | #endif /* CONFIG_NUMA */ |
5b882be4 | 2848 | |
81819f0f | 2849 | /* |
94e4d712 | 2850 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2851 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2852 | * |
894b8788 CL |
2853 | * So we still attempt to reduce cache line usage. Just take the slab |
2854 | * lock and free the item. If there is no additional partial page | |
2855 | * handling required then we can return immediately. | |
81819f0f | 2856 | */ |
894b8788 | 2857 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2858 | void *head, void *tail, int cnt, |
2859 | unsigned long addr) | |
2860 | ||
81819f0f CL |
2861 | { |
2862 | void *prior; | |
2cfb7455 | 2863 | int was_frozen; |
2cfb7455 CL |
2864 | struct page new; |
2865 | unsigned long counters; | |
2866 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2867 | unsigned long uninitialized_var(flags); |
81819f0f | 2868 | |
8a5ec0ba | 2869 | stat(s, FREE_SLOWPATH); |
81819f0f | 2870 | |
19c7ff9e | 2871 | if (kmem_cache_debug(s) && |
282acb43 | 2872 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2873 | return; |
6446faa2 | 2874 | |
2cfb7455 | 2875 | do { |
837d678d JK |
2876 | if (unlikely(n)) { |
2877 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2878 | n = NULL; | |
2879 | } | |
2cfb7455 CL |
2880 | prior = page->freelist; |
2881 | counters = page->counters; | |
81084651 | 2882 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2883 | new.counters = counters; |
2884 | was_frozen = new.frozen; | |
81084651 | 2885 | new.inuse -= cnt; |
837d678d | 2886 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2887 | |
c65c1877 | 2888 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2889 | |
2890 | /* | |
d0e0ac97 CG |
2891 | * Slab was on no list before and will be |
2892 | * partially empty | |
2893 | * We can defer the list move and instead | |
2894 | * freeze it. | |
49e22585 CL |
2895 | */ |
2896 | new.frozen = 1; | |
2897 | ||
c65c1877 | 2898 | } else { /* Needs to be taken off a list */ |
49e22585 | 2899 | |
b455def2 | 2900 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2901 | /* |
2902 | * Speculatively acquire the list_lock. | |
2903 | * If the cmpxchg does not succeed then we may | |
2904 | * drop the list_lock without any processing. | |
2905 | * | |
2906 | * Otherwise the list_lock will synchronize with | |
2907 | * other processors updating the list of slabs. | |
2908 | */ | |
2909 | spin_lock_irqsave(&n->list_lock, flags); | |
2910 | ||
2911 | } | |
2cfb7455 | 2912 | } |
81819f0f | 2913 | |
2cfb7455 CL |
2914 | } while (!cmpxchg_double_slab(s, page, |
2915 | prior, counters, | |
81084651 | 2916 | head, new.counters, |
2cfb7455 | 2917 | "__slab_free")); |
81819f0f | 2918 | |
2cfb7455 | 2919 | if (likely(!n)) { |
49e22585 CL |
2920 | |
2921 | /* | |
2922 | * If we just froze the page then put it onto the | |
2923 | * per cpu partial list. | |
2924 | */ | |
8028dcea | 2925 | if (new.frozen && !was_frozen) { |
49e22585 | 2926 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2927 | stat(s, CPU_PARTIAL_FREE); |
2928 | } | |
49e22585 | 2929 | /* |
2cfb7455 CL |
2930 | * The list lock was not taken therefore no list |
2931 | * activity can be necessary. | |
2932 | */ | |
b455def2 L |
2933 | if (was_frozen) |
2934 | stat(s, FREE_FROZEN); | |
2935 | return; | |
2936 | } | |
81819f0f | 2937 | |
8a5b20ae | 2938 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2939 | goto slab_empty; |
2940 | ||
81819f0f | 2941 | /* |
837d678d JK |
2942 | * Objects left in the slab. If it was not on the partial list before |
2943 | * then add it. | |
81819f0f | 2944 | */ |
345c905d | 2945 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
a4d3f891 | 2946 | remove_full(s, n, page); |
837d678d JK |
2947 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2948 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2949 | } |
80f08c19 | 2950 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2951 | return; |
2952 | ||
2953 | slab_empty: | |
a973e9dd | 2954 | if (prior) { |
81819f0f | 2955 | /* |
6fbabb20 | 2956 | * Slab on the partial list. |
81819f0f | 2957 | */ |
5cc6eee8 | 2958 | remove_partial(n, page); |
84e554e6 | 2959 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2960 | } else { |
6fbabb20 | 2961 | /* Slab must be on the full list */ |
c65c1877 PZ |
2962 | remove_full(s, n, page); |
2963 | } | |
2cfb7455 | 2964 | |
80f08c19 | 2965 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2966 | stat(s, FREE_SLAB); |
81819f0f | 2967 | discard_slab(s, page); |
81819f0f CL |
2968 | } |
2969 | ||
894b8788 CL |
2970 | /* |
2971 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2972 | * can perform fastpath freeing without additional function calls. | |
2973 | * | |
2974 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2975 | * of this processor. This typically the case if we have just allocated | |
2976 | * the item before. | |
2977 | * | |
2978 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2979 | * with all sorts of special processing. | |
81084651 JDB |
2980 | * |
2981 | * Bulk free of a freelist with several objects (all pointing to the | |
2982 | * same page) possible by specifying head and tail ptr, plus objects | |
2983 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2984 | */ |
80a9201a AP |
2985 | static __always_inline void do_slab_free(struct kmem_cache *s, |
2986 | struct page *page, void *head, void *tail, | |
2987 | int cnt, unsigned long addr) | |
894b8788 | 2988 | { |
81084651 | 2989 | void *tail_obj = tail ? : head; |
dfb4f096 | 2990 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2991 | unsigned long tid; |
8a5ec0ba CL |
2992 | redo: |
2993 | /* | |
2994 | * Determine the currently cpus per cpu slab. | |
2995 | * The cpu may change afterward. However that does not matter since | |
2996 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2997 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2998 | */ |
9aabf810 JK |
2999 | do { |
3000 | tid = this_cpu_read(s->cpu_slab->tid); | |
3001 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 3002 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 3003 | unlikely(tid != READ_ONCE(c->tid))); |
c016b0bd | 3004 | |
9aabf810 JK |
3005 | /* Same with comment on barrier() in slab_alloc_node() */ |
3006 | barrier(); | |
c016b0bd | 3007 | |
442b06bc | 3008 | if (likely(page == c->page)) { |
5076190d LT |
3009 | void **freelist = READ_ONCE(c->freelist); |
3010 | ||
3011 | set_freepointer(s, tail_obj, freelist); | |
8a5ec0ba | 3012 | |
933393f5 | 3013 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba | 3014 | s->cpu_slab->freelist, s->cpu_slab->tid, |
5076190d | 3015 | freelist, tid, |
81084651 | 3016 | head, next_tid(tid)))) { |
8a5ec0ba CL |
3017 | |
3018 | note_cmpxchg_failure("slab_free", s, tid); | |
3019 | goto redo; | |
3020 | } | |
84e554e6 | 3021 | stat(s, FREE_FASTPATH); |
894b8788 | 3022 | } else |
81084651 | 3023 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 3024 | |
894b8788 CL |
3025 | } |
3026 | ||
80a9201a AP |
3027 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
3028 | void *head, void *tail, int cnt, | |
3029 | unsigned long addr) | |
3030 | { | |
80a9201a | 3031 | /* |
c3895391 AK |
3032 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
3033 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 3034 | */ |
c3895391 AK |
3035 | if (slab_free_freelist_hook(s, &head, &tail)) |
3036 | do_slab_free(s, page, head, tail, cnt, addr); | |
80a9201a AP |
3037 | } |
3038 | ||
2bd926b4 | 3039 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
3040 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
3041 | { | |
3042 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
3043 | } | |
3044 | #endif | |
3045 | ||
81819f0f CL |
3046 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3047 | { | |
b9ce5ef4 GC |
3048 | s = cache_from_obj(s, x); |
3049 | if (!s) | |
79576102 | 3050 | return; |
81084651 | 3051 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 3052 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
3053 | } |
3054 | EXPORT_SYMBOL(kmem_cache_free); | |
3055 | ||
d0ecd894 | 3056 | struct detached_freelist { |
fbd02630 | 3057 | struct page *page; |
d0ecd894 JDB |
3058 | void *tail; |
3059 | void *freelist; | |
3060 | int cnt; | |
376bf125 | 3061 | struct kmem_cache *s; |
d0ecd894 | 3062 | }; |
fbd02630 | 3063 | |
d0ecd894 JDB |
3064 | /* |
3065 | * This function progressively scans the array with free objects (with | |
3066 | * a limited look ahead) and extract objects belonging to the same | |
3067 | * page. It builds a detached freelist directly within the given | |
3068 | * page/objects. This can happen without any need for | |
3069 | * synchronization, because the objects are owned by running process. | |
3070 | * The freelist is build up as a single linked list in the objects. | |
3071 | * The idea is, that this detached freelist can then be bulk | |
3072 | * transferred to the real freelist(s), but only requiring a single | |
3073 | * synchronization primitive. Look ahead in the array is limited due | |
3074 | * to performance reasons. | |
3075 | */ | |
376bf125 JDB |
3076 | static inline |
3077 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3078 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3079 | { |
3080 | size_t first_skipped_index = 0; | |
3081 | int lookahead = 3; | |
3082 | void *object; | |
ca257195 | 3083 | struct page *page; |
fbd02630 | 3084 | |
d0ecd894 JDB |
3085 | /* Always re-init detached_freelist */ |
3086 | df->page = NULL; | |
fbd02630 | 3087 | |
d0ecd894 JDB |
3088 | do { |
3089 | object = p[--size]; | |
ca257195 | 3090 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3091 | } while (!object && size); |
3eed034d | 3092 | |
d0ecd894 JDB |
3093 | if (!object) |
3094 | return 0; | |
fbd02630 | 3095 | |
ca257195 JDB |
3096 | page = virt_to_head_page(object); |
3097 | if (!s) { | |
3098 | /* Handle kalloc'ed objects */ | |
3099 | if (unlikely(!PageSlab(page))) { | |
3100 | BUG_ON(!PageCompound(page)); | |
3101 | kfree_hook(object); | |
4949148a | 3102 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3103 | p[size] = NULL; /* mark object processed */ |
3104 | return size; | |
3105 | } | |
3106 | /* Derive kmem_cache from object */ | |
3107 | df->s = page->slab_cache; | |
3108 | } else { | |
3109 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3110 | } | |
376bf125 | 3111 | |
d0ecd894 | 3112 | /* Start new detached freelist */ |
ca257195 | 3113 | df->page = page; |
376bf125 | 3114 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3115 | df->tail = object; |
3116 | df->freelist = object; | |
3117 | p[size] = NULL; /* mark object processed */ | |
3118 | df->cnt = 1; | |
3119 | ||
3120 | while (size) { | |
3121 | object = p[--size]; | |
3122 | if (!object) | |
3123 | continue; /* Skip processed objects */ | |
3124 | ||
3125 | /* df->page is always set at this point */ | |
3126 | if (df->page == virt_to_head_page(object)) { | |
3127 | /* Opportunity build freelist */ | |
376bf125 | 3128 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3129 | df->freelist = object; |
3130 | df->cnt++; | |
3131 | p[size] = NULL; /* mark object processed */ | |
3132 | ||
3133 | continue; | |
fbd02630 | 3134 | } |
d0ecd894 JDB |
3135 | |
3136 | /* Limit look ahead search */ | |
3137 | if (!--lookahead) | |
3138 | break; | |
3139 | ||
3140 | if (!first_skipped_index) | |
3141 | first_skipped_index = size + 1; | |
fbd02630 | 3142 | } |
d0ecd894 JDB |
3143 | |
3144 | return first_skipped_index; | |
3145 | } | |
3146 | ||
d0ecd894 | 3147 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3148 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3149 | { |
3150 | if (WARN_ON(!size)) | |
3151 | return; | |
3152 | ||
3153 | do { | |
3154 | struct detached_freelist df; | |
3155 | ||
3156 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3157 | if (!df.page) |
d0ecd894 JDB |
3158 | continue; |
3159 | ||
376bf125 | 3160 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 3161 | } while (likely(size)); |
484748f0 CL |
3162 | } |
3163 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3164 | ||
994eb764 | 3165 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3166 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3167 | void **p) | |
484748f0 | 3168 | { |
994eb764 JDB |
3169 | struct kmem_cache_cpu *c; |
3170 | int i; | |
3171 | ||
03ec0ed5 JDB |
3172 | /* memcg and kmem_cache debug support */ |
3173 | s = slab_pre_alloc_hook(s, flags); | |
3174 | if (unlikely(!s)) | |
3175 | return false; | |
994eb764 JDB |
3176 | /* |
3177 | * Drain objects in the per cpu slab, while disabling local | |
3178 | * IRQs, which protects against PREEMPT and interrupts | |
3179 | * handlers invoking normal fastpath. | |
3180 | */ | |
3181 | local_irq_disable(); | |
3182 | c = this_cpu_ptr(s->cpu_slab); | |
3183 | ||
3184 | for (i = 0; i < size; i++) { | |
3185 | void *object = c->freelist; | |
3186 | ||
ebe909e0 | 3187 | if (unlikely(!object)) { |
fd4d9c7d JH |
3188 | /* |
3189 | * We may have removed an object from c->freelist using | |
3190 | * the fastpath in the previous iteration; in that case, | |
3191 | * c->tid has not been bumped yet. | |
3192 | * Since ___slab_alloc() may reenable interrupts while | |
3193 | * allocating memory, we should bump c->tid now. | |
3194 | */ | |
3195 | c->tid = next_tid(c->tid); | |
3196 | ||
ebe909e0 JDB |
3197 | /* |
3198 | * Invoking slow path likely have side-effect | |
3199 | * of re-populating per CPU c->freelist | |
3200 | */ | |
87098373 | 3201 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3202 | _RET_IP_, c); |
87098373 CL |
3203 | if (unlikely(!p[i])) |
3204 | goto error; | |
3205 | ||
ebe909e0 | 3206 | c = this_cpu_ptr(s->cpu_slab); |
0f181f9f AP |
3207 | maybe_wipe_obj_freeptr(s, p[i]); |
3208 | ||
ebe909e0 JDB |
3209 | continue; /* goto for-loop */ |
3210 | } | |
994eb764 JDB |
3211 | c->freelist = get_freepointer(s, object); |
3212 | p[i] = object; | |
0f181f9f | 3213 | maybe_wipe_obj_freeptr(s, p[i]); |
994eb764 JDB |
3214 | } |
3215 | c->tid = next_tid(c->tid); | |
3216 | local_irq_enable(); | |
3217 | ||
3218 | /* Clear memory outside IRQ disabled fastpath loop */ | |
6471384a | 3219 | if (unlikely(slab_want_init_on_alloc(flags, s))) { |
994eb764 JDB |
3220 | int j; |
3221 | ||
3222 | for (j = 0; j < i; j++) | |
3223 | memset(p[j], 0, s->object_size); | |
3224 | } | |
3225 | ||
03ec0ed5 JDB |
3226 | /* memcg and kmem_cache debug support */ |
3227 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 3228 | return i; |
87098373 | 3229 | error: |
87098373 | 3230 | local_irq_enable(); |
03ec0ed5 JDB |
3231 | slab_post_alloc_hook(s, flags, i, p); |
3232 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 3233 | return 0; |
484748f0 CL |
3234 | } |
3235 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3236 | ||
3237 | ||
81819f0f | 3238 | /* |
672bba3a CL |
3239 | * Object placement in a slab is made very easy because we always start at |
3240 | * offset 0. If we tune the size of the object to the alignment then we can | |
3241 | * get the required alignment by putting one properly sized object after | |
3242 | * another. | |
81819f0f CL |
3243 | * |
3244 | * Notice that the allocation order determines the sizes of the per cpu | |
3245 | * caches. Each processor has always one slab available for allocations. | |
3246 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3247 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3248 | * locking overhead. |
81819f0f CL |
3249 | */ |
3250 | ||
3251 | /* | |
3252 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3253 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3254 | * and increases the number of allocations possible without having to | |
3255 | * take the list_lock. | |
3256 | */ | |
19af27af AD |
3257 | static unsigned int slub_min_order; |
3258 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3259 | static unsigned int slub_min_objects; | |
81819f0f | 3260 | |
81819f0f CL |
3261 | /* |
3262 | * Calculate the order of allocation given an slab object size. | |
3263 | * | |
672bba3a CL |
3264 | * The order of allocation has significant impact on performance and other |
3265 | * system components. Generally order 0 allocations should be preferred since | |
3266 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3267 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3268 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3269 | * would be wasted. |
3270 | * | |
3271 | * In order to reach satisfactory performance we must ensure that a minimum | |
3272 | * number of objects is in one slab. Otherwise we may generate too much | |
3273 | * activity on the partial lists which requires taking the list_lock. This is | |
3274 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3275 | * |
672bba3a CL |
3276 | * slub_max_order specifies the order where we begin to stop considering the |
3277 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3278 | * we try to keep the page order as low as possible. So we accept more waste | |
3279 | * of space in favor of a small page order. | |
81819f0f | 3280 | * |
672bba3a CL |
3281 | * Higher order allocations also allow the placement of more objects in a |
3282 | * slab and thereby reduce object handling overhead. If the user has | |
3283 | * requested a higher mininum order then we start with that one instead of | |
3284 | * the smallest order which will fit the object. | |
81819f0f | 3285 | */ |
19af27af AD |
3286 | static inline unsigned int slab_order(unsigned int size, |
3287 | unsigned int min_objects, unsigned int max_order, | |
9736d2a9 | 3288 | unsigned int fract_leftover) |
81819f0f | 3289 | { |
19af27af AD |
3290 | unsigned int min_order = slub_min_order; |
3291 | unsigned int order; | |
81819f0f | 3292 | |
9736d2a9 | 3293 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3294 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3295 | |
9736d2a9 | 3296 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3297 | order <= max_order; order++) { |
81819f0f | 3298 | |
19af27af AD |
3299 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3300 | unsigned int rem; | |
81819f0f | 3301 | |
9736d2a9 | 3302 | rem = slab_size % size; |
81819f0f | 3303 | |
5e6d444e | 3304 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3305 | break; |
81819f0f | 3306 | } |
672bba3a | 3307 | |
81819f0f CL |
3308 | return order; |
3309 | } | |
3310 | ||
9736d2a9 | 3311 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3312 | { |
19af27af AD |
3313 | unsigned int order; |
3314 | unsigned int min_objects; | |
3315 | unsigned int max_objects; | |
5e6d444e CL |
3316 | |
3317 | /* | |
3318 | * Attempt to find best configuration for a slab. This | |
3319 | * works by first attempting to generate a layout with | |
3320 | * the best configuration and backing off gradually. | |
3321 | * | |
422ff4d7 | 3322 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3323 | * we reduce the minimum objects required in a slab. |
3324 | */ | |
3325 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3326 | if (!min_objects) |
3327 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
9736d2a9 | 3328 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3329 | min_objects = min(min_objects, max_objects); |
3330 | ||
5e6d444e | 3331 | while (min_objects > 1) { |
19af27af AD |
3332 | unsigned int fraction; |
3333 | ||
c124f5b5 | 3334 | fraction = 16; |
5e6d444e CL |
3335 | while (fraction >= 4) { |
3336 | order = slab_order(size, min_objects, | |
9736d2a9 | 3337 | slub_max_order, fraction); |
5e6d444e CL |
3338 | if (order <= slub_max_order) |
3339 | return order; | |
3340 | fraction /= 2; | |
3341 | } | |
5086c389 | 3342 | min_objects--; |
5e6d444e CL |
3343 | } |
3344 | ||
3345 | /* | |
3346 | * We were unable to place multiple objects in a slab. Now | |
3347 | * lets see if we can place a single object there. | |
3348 | */ | |
9736d2a9 | 3349 | order = slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3350 | if (order <= slub_max_order) |
3351 | return order; | |
3352 | ||
3353 | /* | |
3354 | * Doh this slab cannot be placed using slub_max_order. | |
3355 | */ | |
9736d2a9 | 3356 | order = slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3357 | if (order < MAX_ORDER) |
5e6d444e CL |
3358 | return order; |
3359 | return -ENOSYS; | |
3360 | } | |
3361 | ||
5595cffc | 3362 | static void |
4053497d | 3363 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3364 | { |
3365 | n->nr_partial = 0; | |
81819f0f CL |
3366 | spin_lock_init(&n->list_lock); |
3367 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3368 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3369 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3370 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3371 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3372 | #endif |
81819f0f CL |
3373 | } |
3374 | ||
55136592 | 3375 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3376 | { |
6c182dc0 | 3377 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3378 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3379 | |
8a5ec0ba | 3380 | /* |
d4d84fef CM |
3381 | * Must align to double word boundary for the double cmpxchg |
3382 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3383 | */ |
d4d84fef CM |
3384 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3385 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3386 | |
3387 | if (!s->cpu_slab) | |
3388 | return 0; | |
3389 | ||
3390 | init_kmem_cache_cpus(s); | |
4c93c355 | 3391 | |
8a5ec0ba | 3392 | return 1; |
4c93c355 | 3393 | } |
4c93c355 | 3394 | |
51df1142 CL |
3395 | static struct kmem_cache *kmem_cache_node; |
3396 | ||
81819f0f CL |
3397 | /* |
3398 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3399 | * slab on the node for this slabcache. There are no concurrent accesses | |
3400 | * possible. | |
3401 | * | |
721ae22a ZYW |
3402 | * Note that this function only works on the kmem_cache_node |
3403 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3404 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3405 | */ |
55136592 | 3406 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3407 | { |
3408 | struct page *page; | |
3409 | struct kmem_cache_node *n; | |
3410 | ||
51df1142 | 3411 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3412 | |
51df1142 | 3413 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3414 | |
3415 | BUG_ON(!page); | |
a2f92ee7 | 3416 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3417 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3418 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3419 | } |
3420 | ||
81819f0f CL |
3421 | n = page->freelist; |
3422 | BUG_ON(!n); | |
8ab1372f | 3423 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3424 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3425 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3426 | #endif |
12b22386 | 3427 | n = kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
505f5dcb | 3428 | GFP_KERNEL); |
12b22386 AK |
3429 | page->freelist = get_freepointer(kmem_cache_node, n); |
3430 | page->inuse = 1; | |
3431 | page->frozen = 0; | |
3432 | kmem_cache_node->node[node] = n; | |
4053497d | 3433 | init_kmem_cache_node(n); |
51df1142 | 3434 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3435 | |
67b6c900 | 3436 | /* |
1e4dd946 SR |
3437 | * No locks need to be taken here as it has just been |
3438 | * initialized and there is no concurrent access. | |
67b6c900 | 3439 | */ |
1e4dd946 | 3440 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3441 | } |
3442 | ||
3443 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3444 | { | |
3445 | int node; | |
fa45dc25 | 3446 | struct kmem_cache_node *n; |
81819f0f | 3447 | |
fa45dc25 | 3448 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3449 | s->node[node] = NULL; |
ea37df54 | 3450 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3451 | } |
3452 | } | |
3453 | ||
52b4b950 DS |
3454 | void __kmem_cache_release(struct kmem_cache *s) |
3455 | { | |
210e7a43 | 3456 | cache_random_seq_destroy(s); |
52b4b950 DS |
3457 | free_percpu(s->cpu_slab); |
3458 | free_kmem_cache_nodes(s); | |
3459 | } | |
3460 | ||
55136592 | 3461 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3462 | { |
3463 | int node; | |
81819f0f | 3464 | |
f64dc58c | 3465 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3466 | struct kmem_cache_node *n; |
3467 | ||
73367bd8 | 3468 | if (slab_state == DOWN) { |
55136592 | 3469 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3470 | continue; |
3471 | } | |
51df1142 | 3472 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3473 | GFP_KERNEL, node); |
81819f0f | 3474 | |
73367bd8 AD |
3475 | if (!n) { |
3476 | free_kmem_cache_nodes(s); | |
3477 | return 0; | |
81819f0f | 3478 | } |
73367bd8 | 3479 | |
4053497d | 3480 | init_kmem_cache_node(n); |
ea37df54 | 3481 | s->node[node] = n; |
81819f0f CL |
3482 | } |
3483 | return 1; | |
3484 | } | |
81819f0f | 3485 | |
c0bdb232 | 3486 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3487 | { |
3488 | if (min < MIN_PARTIAL) | |
3489 | min = MIN_PARTIAL; | |
3490 | else if (min > MAX_PARTIAL) | |
3491 | min = MAX_PARTIAL; | |
3492 | s->min_partial = min; | |
3493 | } | |
3494 | ||
e6d0e1dc WY |
3495 | static void set_cpu_partial(struct kmem_cache *s) |
3496 | { | |
3497 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3498 | /* | |
3499 | * cpu_partial determined the maximum number of objects kept in the | |
3500 | * per cpu partial lists of a processor. | |
3501 | * | |
3502 | * Per cpu partial lists mainly contain slabs that just have one | |
3503 | * object freed. If they are used for allocation then they can be | |
3504 | * filled up again with minimal effort. The slab will never hit the | |
3505 | * per node partial lists and therefore no locking will be required. | |
3506 | * | |
3507 | * This setting also determines | |
3508 | * | |
3509 | * A) The number of objects from per cpu partial slabs dumped to the | |
3510 | * per node list when we reach the limit. | |
3511 | * B) The number of objects in cpu partial slabs to extract from the | |
3512 | * per node list when we run out of per cpu objects. We only fetch | |
3513 | * 50% to keep some capacity around for frees. | |
3514 | */ | |
3515 | if (!kmem_cache_has_cpu_partial(s)) | |
bbd4e305 | 3516 | slub_set_cpu_partial(s, 0); |
e6d0e1dc | 3517 | else if (s->size >= PAGE_SIZE) |
bbd4e305 | 3518 | slub_set_cpu_partial(s, 2); |
e6d0e1dc | 3519 | else if (s->size >= 1024) |
bbd4e305 | 3520 | slub_set_cpu_partial(s, 6); |
e6d0e1dc | 3521 | else if (s->size >= 256) |
bbd4e305 | 3522 | slub_set_cpu_partial(s, 13); |
e6d0e1dc | 3523 | else |
bbd4e305 | 3524 | slub_set_cpu_partial(s, 30); |
e6d0e1dc WY |
3525 | #endif |
3526 | } | |
3527 | ||
81819f0f CL |
3528 | /* |
3529 | * calculate_sizes() determines the order and the distribution of data within | |
3530 | * a slab object. | |
3531 | */ | |
06b285dc | 3532 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 3533 | { |
d50112ed | 3534 | slab_flags_t flags = s->flags; |
be4a7988 | 3535 | unsigned int size = s->object_size; |
19af27af | 3536 | unsigned int order; |
81819f0f | 3537 | |
d8b42bf5 CL |
3538 | /* |
3539 | * Round up object size to the next word boundary. We can only | |
3540 | * place the free pointer at word boundaries and this determines | |
3541 | * the possible location of the free pointer. | |
3542 | */ | |
3543 | size = ALIGN(size, sizeof(void *)); | |
3544 | ||
3545 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3546 | /* |
3547 | * Determine if we can poison the object itself. If the user of | |
3548 | * the slab may touch the object after free or before allocation | |
3549 | * then we should never poison the object itself. | |
3550 | */ | |
5f0d5a3a | 3551 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3552 | !s->ctor) |
81819f0f CL |
3553 | s->flags |= __OBJECT_POISON; |
3554 | else | |
3555 | s->flags &= ~__OBJECT_POISON; | |
3556 | ||
81819f0f CL |
3557 | |
3558 | /* | |
672bba3a | 3559 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3560 | * end of the object and the free pointer. If not then add an |
672bba3a | 3561 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3562 | */ |
3b0efdfa | 3563 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3564 | size += sizeof(void *); |
41ecc55b | 3565 | #endif |
81819f0f CL |
3566 | |
3567 | /* | |
672bba3a CL |
3568 | * With that we have determined the number of bytes in actual use |
3569 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3570 | */ |
3571 | s->inuse = size; | |
3572 | ||
5f0d5a3a | 3573 | if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
c59def9f | 3574 | s->ctor)) { |
81819f0f CL |
3575 | /* |
3576 | * Relocate free pointer after the object if it is not | |
3577 | * permitted to overwrite the first word of the object on | |
3578 | * kmem_cache_free. | |
3579 | * | |
3580 | * This is the case if we do RCU, have a constructor or | |
3581 | * destructor or are poisoning the objects. | |
3582 | */ | |
3583 | s->offset = size; | |
3584 | size += sizeof(void *); | |
3202fa62 KC |
3585 | } else if (size > sizeof(void *)) { |
3586 | /* | |
3587 | * Store freelist pointer near middle of object to keep | |
3588 | * it away from the edges of the object to avoid small | |
3589 | * sized over/underflows from neighboring allocations. | |
3590 | */ | |
3591 | s->offset = ALIGN(size / 2, sizeof(void *)); | |
81819f0f CL |
3592 | } |
3593 | ||
c12b3c62 | 3594 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3595 | if (flags & SLAB_STORE_USER) |
3596 | /* | |
3597 | * Need to store information about allocs and frees after | |
3598 | * the object. | |
3599 | */ | |
3600 | size += 2 * sizeof(struct track); | |
80a9201a | 3601 | #endif |
81819f0f | 3602 | |
80a9201a AP |
3603 | kasan_cache_create(s, &size, &s->flags); |
3604 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3605 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3606 | /* |
3607 | * Add some empty padding so that we can catch | |
3608 | * overwrites from earlier objects rather than let | |
3609 | * tracking information or the free pointer be | |
0211a9c8 | 3610 | * corrupted if a user writes before the start |
81819f0f CL |
3611 | * of the object. |
3612 | */ | |
3613 | size += sizeof(void *); | |
d86bd1be JK |
3614 | |
3615 | s->red_left_pad = sizeof(void *); | |
3616 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3617 | size += s->red_left_pad; | |
3618 | } | |
41ecc55b | 3619 | #endif |
672bba3a | 3620 | |
81819f0f CL |
3621 | /* |
3622 | * SLUB stores one object immediately after another beginning from | |
3623 | * offset 0. In order to align the objects we have to simply size | |
3624 | * each object to conform to the alignment. | |
3625 | */ | |
45906855 | 3626 | size = ALIGN(size, s->align); |
81819f0f | 3627 | s->size = size; |
06b285dc CL |
3628 | if (forced_order >= 0) |
3629 | order = forced_order; | |
3630 | else | |
9736d2a9 | 3631 | order = calculate_order(size); |
81819f0f | 3632 | |
19af27af | 3633 | if ((int)order < 0) |
81819f0f CL |
3634 | return 0; |
3635 | ||
b7a49f0d | 3636 | s->allocflags = 0; |
834f3d11 | 3637 | if (order) |
b7a49f0d CL |
3638 | s->allocflags |= __GFP_COMP; |
3639 | ||
3640 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3641 | s->allocflags |= GFP_DMA; |
b7a49f0d | 3642 | |
6d6ea1e9 NB |
3643 | if (s->flags & SLAB_CACHE_DMA32) |
3644 | s->allocflags |= GFP_DMA32; | |
3645 | ||
b7a49f0d CL |
3646 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
3647 | s->allocflags |= __GFP_RECLAIMABLE; | |
3648 | ||
81819f0f CL |
3649 | /* |
3650 | * Determine the number of objects per slab | |
3651 | */ | |
9736d2a9 MW |
3652 | s->oo = oo_make(order, size); |
3653 | s->min = oo_make(get_order(size), size); | |
205ab99d CL |
3654 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3655 | s->max = s->oo; | |
81819f0f | 3656 | |
834f3d11 | 3657 | return !!oo_objects(s->oo); |
81819f0f CL |
3658 | } |
3659 | ||
d50112ed | 3660 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 3661 | { |
8a13a4cc | 3662 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
2482ddec KC |
3663 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
3664 | s->random = get_random_long(); | |
3665 | #endif | |
81819f0f | 3666 | |
06b285dc | 3667 | if (!calculate_sizes(s, -1)) |
81819f0f | 3668 | goto error; |
3de47213 DR |
3669 | if (disable_higher_order_debug) { |
3670 | /* | |
3671 | * Disable debugging flags that store metadata if the min slab | |
3672 | * order increased. | |
3673 | */ | |
3b0efdfa | 3674 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3675 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3676 | s->offset = 0; | |
3677 | if (!calculate_sizes(s, -1)) | |
3678 | goto error; | |
3679 | } | |
3680 | } | |
81819f0f | 3681 | |
2565409f HC |
3682 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3683 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3684 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3685 | /* Enable fast mode */ |
3686 | s->flags |= __CMPXCHG_DOUBLE; | |
3687 | #endif | |
3688 | ||
3b89d7d8 DR |
3689 | /* |
3690 | * The larger the object size is, the more pages we want on the partial | |
3691 | * list to avoid pounding the page allocator excessively. | |
3692 | */ | |
49e22585 CL |
3693 | set_min_partial(s, ilog2(s->size) / 2); |
3694 | ||
e6d0e1dc | 3695 | set_cpu_partial(s); |
49e22585 | 3696 | |
81819f0f | 3697 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3698 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3699 | #endif |
210e7a43 TG |
3700 | |
3701 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3702 | if (slab_state >= UP) { | |
3703 | if (init_cache_random_seq(s)) | |
3704 | goto error; | |
3705 | } | |
3706 | ||
55136592 | 3707 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3708 | goto error; |
81819f0f | 3709 | |
55136592 | 3710 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3711 | return 0; |
ff12059e | 3712 | |
4c93c355 | 3713 | free_kmem_cache_nodes(s); |
81819f0f | 3714 | error: |
278b1bb1 | 3715 | return -EINVAL; |
81819f0f | 3716 | } |
81819f0f | 3717 | |
33b12c38 CL |
3718 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3719 | const char *text) | |
3720 | { | |
3721 | #ifdef CONFIG_SLUB_DEBUG | |
3722 | void *addr = page_address(page); | |
3723 | void *p; | |
90e9f6a6 YZ |
3724 | unsigned long *map; |
3725 | ||
945cf2b6 | 3726 | slab_err(s, page, text, s->name); |
33b12c38 | 3727 | slab_lock(page); |
33b12c38 | 3728 | |
90e9f6a6 | 3729 | map = get_map(s, page); |
33b12c38 CL |
3730 | for_each_object(p, s, addr, page->objects) { |
3731 | ||
3732 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3733 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3734 | print_tracking(s, p); |
3735 | } | |
3736 | } | |
90e9f6a6 YZ |
3737 | put_map(map); |
3738 | ||
33b12c38 CL |
3739 | slab_unlock(page); |
3740 | #endif | |
3741 | } | |
3742 | ||
81819f0f | 3743 | /* |
599870b1 | 3744 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3745 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3746 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3747 | */ |
599870b1 | 3748 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3749 | { |
60398923 | 3750 | LIST_HEAD(discard); |
81819f0f CL |
3751 | struct page *page, *h; |
3752 | ||
52b4b950 DS |
3753 | BUG_ON(irqs_disabled()); |
3754 | spin_lock_irq(&n->list_lock); | |
916ac052 | 3755 | list_for_each_entry_safe(page, h, &n->partial, slab_list) { |
81819f0f | 3756 | if (!page->inuse) { |
52b4b950 | 3757 | remove_partial(n, page); |
916ac052 | 3758 | list_add(&page->slab_list, &discard); |
33b12c38 CL |
3759 | } else { |
3760 | list_slab_objects(s, page, | |
52b4b950 | 3761 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3762 | } |
33b12c38 | 3763 | } |
52b4b950 | 3764 | spin_unlock_irq(&n->list_lock); |
60398923 | 3765 | |
916ac052 | 3766 | list_for_each_entry_safe(page, h, &discard, slab_list) |
60398923 | 3767 | discard_slab(s, page); |
81819f0f CL |
3768 | } |
3769 | ||
f9e13c0a SB |
3770 | bool __kmem_cache_empty(struct kmem_cache *s) |
3771 | { | |
3772 | int node; | |
3773 | struct kmem_cache_node *n; | |
3774 | ||
3775 | for_each_kmem_cache_node(s, node, n) | |
3776 | if (n->nr_partial || slabs_node(s, node)) | |
3777 | return false; | |
3778 | return true; | |
3779 | } | |
3780 | ||
81819f0f | 3781 | /* |
672bba3a | 3782 | * Release all resources used by a slab cache. |
81819f0f | 3783 | */ |
52b4b950 | 3784 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3785 | { |
3786 | int node; | |
fa45dc25 | 3787 | struct kmem_cache_node *n; |
81819f0f CL |
3788 | |
3789 | flush_all(s); | |
81819f0f | 3790 | /* Attempt to free all objects */ |
fa45dc25 | 3791 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3792 | free_partial(s, n); |
3793 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3794 | return 1; |
3795 | } | |
bf5eb3de | 3796 | sysfs_slab_remove(s); |
81819f0f CL |
3797 | return 0; |
3798 | } | |
3799 | ||
81819f0f CL |
3800 | /******************************************************************** |
3801 | * Kmalloc subsystem | |
3802 | *******************************************************************/ | |
3803 | ||
81819f0f CL |
3804 | static int __init setup_slub_min_order(char *str) |
3805 | { | |
19af27af | 3806 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
3807 | |
3808 | return 1; | |
3809 | } | |
3810 | ||
3811 | __setup("slub_min_order=", setup_slub_min_order); | |
3812 | ||
3813 | static int __init setup_slub_max_order(char *str) | |
3814 | { | |
19af27af AD |
3815 | get_option(&str, (int *)&slub_max_order); |
3816 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
3817 | |
3818 | return 1; | |
3819 | } | |
3820 | ||
3821 | __setup("slub_max_order=", setup_slub_max_order); | |
3822 | ||
3823 | static int __init setup_slub_min_objects(char *str) | |
3824 | { | |
19af27af | 3825 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
3826 | |
3827 | return 1; | |
3828 | } | |
3829 | ||
3830 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3831 | ||
81819f0f CL |
3832 | void *__kmalloc(size_t size, gfp_t flags) |
3833 | { | |
aadb4bc4 | 3834 | struct kmem_cache *s; |
5b882be4 | 3835 | void *ret; |
81819f0f | 3836 | |
95a05b42 | 3837 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3838 | return kmalloc_large(size, flags); |
aadb4bc4 | 3839 | |
2c59dd65 | 3840 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3841 | |
3842 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3843 | return s; |
3844 | ||
2b847c3c | 3845 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3846 | |
ca2b84cb | 3847 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3848 | |
0116523c | 3849 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3850 | |
5b882be4 | 3851 | return ret; |
81819f0f CL |
3852 | } |
3853 | EXPORT_SYMBOL(__kmalloc); | |
3854 | ||
5d1f57e4 | 3855 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3856 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3857 | { | |
b1eeab67 | 3858 | struct page *page; |
e4f7c0b4 | 3859 | void *ptr = NULL; |
6a486c0a | 3860 | unsigned int order = get_order(size); |
f619cfe1 | 3861 | |
75f296d9 | 3862 | flags |= __GFP_COMP; |
6a486c0a VB |
3863 | page = alloc_pages_node(node, flags, order); |
3864 | if (page) { | |
e4f7c0b4 | 3865 | ptr = page_address(page); |
6a486c0a VB |
3866 | mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE, |
3867 | 1 << order); | |
3868 | } | |
e4f7c0b4 | 3869 | |
0116523c | 3870 | return kmalloc_large_node_hook(ptr, size, flags); |
f619cfe1 CL |
3871 | } |
3872 | ||
81819f0f CL |
3873 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3874 | { | |
aadb4bc4 | 3875 | struct kmem_cache *s; |
5b882be4 | 3876 | void *ret; |
81819f0f | 3877 | |
95a05b42 | 3878 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3879 | ret = kmalloc_large_node(size, flags, node); |
3880 | ||
ca2b84cb EGM |
3881 | trace_kmalloc_node(_RET_IP_, ret, |
3882 | size, PAGE_SIZE << get_order(size), | |
3883 | flags, node); | |
5b882be4 EGM |
3884 | |
3885 | return ret; | |
3886 | } | |
aadb4bc4 | 3887 | |
2c59dd65 | 3888 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3889 | |
3890 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3891 | return s; |
3892 | ||
2b847c3c | 3893 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3894 | |
ca2b84cb | 3895 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3896 | |
0116523c | 3897 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3898 | |
5b882be4 | 3899 | return ret; |
81819f0f CL |
3900 | } |
3901 | EXPORT_SYMBOL(__kmalloc_node); | |
6dfd1b65 | 3902 | #endif /* CONFIG_NUMA */ |
81819f0f | 3903 | |
ed18adc1 KC |
3904 | #ifdef CONFIG_HARDENED_USERCOPY |
3905 | /* | |
afcc90f8 KC |
3906 | * Rejects incorrectly sized objects and objects that are to be copied |
3907 | * to/from userspace but do not fall entirely within the containing slab | |
3908 | * cache's usercopy region. | |
ed18adc1 KC |
3909 | * |
3910 | * Returns NULL if check passes, otherwise const char * to name of cache | |
3911 | * to indicate an error. | |
3912 | */ | |
f4e6e289 KC |
3913 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
3914 | bool to_user) | |
ed18adc1 KC |
3915 | { |
3916 | struct kmem_cache *s; | |
44065b2e | 3917 | unsigned int offset; |
ed18adc1 KC |
3918 | size_t object_size; |
3919 | ||
96fedce2 AK |
3920 | ptr = kasan_reset_tag(ptr); |
3921 | ||
ed18adc1 KC |
3922 | /* Find object and usable object size. */ |
3923 | s = page->slab_cache; | |
ed18adc1 KC |
3924 | |
3925 | /* Reject impossible pointers. */ | |
3926 | if (ptr < page_address(page)) | |
f4e6e289 KC |
3927 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
3928 | to_user, 0, n); | |
ed18adc1 KC |
3929 | |
3930 | /* Find offset within object. */ | |
3931 | offset = (ptr - page_address(page)) % s->size; | |
3932 | ||
3933 | /* Adjust for redzone and reject if within the redzone. */ | |
3934 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) { | |
3935 | if (offset < s->red_left_pad) | |
f4e6e289 KC |
3936 | usercopy_abort("SLUB object in left red zone", |
3937 | s->name, to_user, offset, n); | |
ed18adc1 KC |
3938 | offset -= s->red_left_pad; |
3939 | } | |
3940 | ||
afcc90f8 KC |
3941 | /* Allow address range falling entirely within usercopy region. */ |
3942 | if (offset >= s->useroffset && | |
3943 | offset - s->useroffset <= s->usersize && | |
3944 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 3945 | return; |
ed18adc1 | 3946 | |
afcc90f8 KC |
3947 | /* |
3948 | * If the copy is still within the allocated object, produce | |
3949 | * a warning instead of rejecting the copy. This is intended | |
3950 | * to be a temporary method to find any missing usercopy | |
3951 | * whitelists. | |
3952 | */ | |
3953 | object_size = slab_ksize(s); | |
2d891fbc KC |
3954 | if (usercopy_fallback && |
3955 | offset <= object_size && n <= object_size - offset) { | |
afcc90f8 KC |
3956 | usercopy_warn("SLUB object", s->name, to_user, offset, n); |
3957 | return; | |
3958 | } | |
ed18adc1 | 3959 | |
f4e6e289 | 3960 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
3961 | } |
3962 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
3963 | ||
10d1f8cb | 3964 | size_t __ksize(const void *object) |
81819f0f | 3965 | { |
272c1d21 | 3966 | struct page *page; |
81819f0f | 3967 | |
ef8b4520 | 3968 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3969 | return 0; |
3970 | ||
294a80a8 | 3971 | page = virt_to_head_page(object); |
294a80a8 | 3972 | |
76994412 PE |
3973 | if (unlikely(!PageSlab(page))) { |
3974 | WARN_ON(!PageCompound(page)); | |
a50b854e | 3975 | return page_size(page); |
76994412 | 3976 | } |
81819f0f | 3977 | |
1b4f59e3 | 3978 | return slab_ksize(page->slab_cache); |
81819f0f | 3979 | } |
10d1f8cb | 3980 | EXPORT_SYMBOL(__ksize); |
81819f0f CL |
3981 | |
3982 | void kfree(const void *x) | |
3983 | { | |
81819f0f | 3984 | struct page *page; |
5bb983b0 | 3985 | void *object = (void *)x; |
81819f0f | 3986 | |
2121db74 PE |
3987 | trace_kfree(_RET_IP_, x); |
3988 | ||
2408c550 | 3989 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3990 | return; |
3991 | ||
b49af68f | 3992 | page = virt_to_head_page(x); |
aadb4bc4 | 3993 | if (unlikely(!PageSlab(page))) { |
6a486c0a VB |
3994 | unsigned int order = compound_order(page); |
3995 | ||
0937502a | 3996 | BUG_ON(!PageCompound(page)); |
47adccce | 3997 | kfree_hook(object); |
6a486c0a VB |
3998 | mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE, |
3999 | -(1 << order)); | |
4000 | __free_pages(page, order); | |
aadb4bc4 CL |
4001 | return; |
4002 | } | |
81084651 | 4003 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
4004 | } |
4005 | EXPORT_SYMBOL(kfree); | |
4006 | ||
832f37f5 VD |
4007 | #define SHRINK_PROMOTE_MAX 32 |
4008 | ||
2086d26a | 4009 | /* |
832f37f5 VD |
4010 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
4011 | * up most to the head of the partial lists. New allocations will then | |
4012 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
4013 | * |
4014 | * The slabs with the least items are placed last. This results in them | |
4015 | * being allocated from last increasing the chance that the last objects | |
4016 | * are freed in them. | |
2086d26a | 4017 | */ |
c9fc5864 | 4018 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
4019 | { |
4020 | int node; | |
4021 | int i; | |
4022 | struct kmem_cache_node *n; | |
4023 | struct page *page; | |
4024 | struct page *t; | |
832f37f5 VD |
4025 | struct list_head discard; |
4026 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 4027 | unsigned long flags; |
ce3712d7 | 4028 | int ret = 0; |
2086d26a | 4029 | |
2086d26a | 4030 | flush_all(s); |
fa45dc25 | 4031 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
4032 | INIT_LIST_HEAD(&discard); |
4033 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
4034 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
4035 | |
4036 | spin_lock_irqsave(&n->list_lock, flags); | |
4037 | ||
4038 | /* | |
832f37f5 | 4039 | * Build lists of slabs to discard or promote. |
2086d26a | 4040 | * |
672bba3a CL |
4041 | * Note that concurrent frees may occur while we hold the |
4042 | * list_lock. page->inuse here is the upper limit. | |
2086d26a | 4043 | */ |
916ac052 | 4044 | list_for_each_entry_safe(page, t, &n->partial, slab_list) { |
832f37f5 VD |
4045 | int free = page->objects - page->inuse; |
4046 | ||
4047 | /* Do not reread page->inuse */ | |
4048 | barrier(); | |
4049 | ||
4050 | /* We do not keep full slabs on the list */ | |
4051 | BUG_ON(free <= 0); | |
4052 | ||
4053 | if (free == page->objects) { | |
916ac052 | 4054 | list_move(&page->slab_list, &discard); |
69cb8e6b | 4055 | n->nr_partial--; |
832f37f5 | 4056 | } else if (free <= SHRINK_PROMOTE_MAX) |
916ac052 | 4057 | list_move(&page->slab_list, promote + free - 1); |
2086d26a CL |
4058 | } |
4059 | ||
2086d26a | 4060 | /* |
832f37f5 VD |
4061 | * Promote the slabs filled up most to the head of the |
4062 | * partial list. | |
2086d26a | 4063 | */ |
832f37f5 VD |
4064 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
4065 | list_splice(promote + i, &n->partial); | |
2086d26a | 4066 | |
2086d26a | 4067 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
4068 | |
4069 | /* Release empty slabs */ | |
916ac052 | 4070 | list_for_each_entry_safe(page, t, &discard, slab_list) |
69cb8e6b | 4071 | discard_slab(s, page); |
ce3712d7 VD |
4072 | |
4073 | if (slabs_node(s, node)) | |
4074 | ret = 1; | |
2086d26a CL |
4075 | } |
4076 | ||
ce3712d7 | 4077 | return ret; |
2086d26a | 4078 | } |
2086d26a | 4079 | |
c9fc5864 | 4080 | #ifdef CONFIG_MEMCG |
43486694 | 4081 | void __kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s) |
01fb58bc | 4082 | { |
50862ce7 TH |
4083 | /* |
4084 | * Called with all the locks held after a sched RCU grace period. | |
4085 | * Even if @s becomes empty after shrinking, we can't know that @s | |
4086 | * doesn't have allocations already in-flight and thus can't | |
4087 | * destroy @s until the associated memcg is released. | |
4088 | * | |
4089 | * However, let's remove the sysfs files for empty caches here. | |
4090 | * Each cache has a lot of interface files which aren't | |
4091 | * particularly useful for empty draining caches; otherwise, we can | |
4092 | * easily end up with millions of unnecessary sysfs files on | |
4093 | * systems which have a lot of memory and transient cgroups. | |
4094 | */ | |
4095 | if (!__kmem_cache_shrink(s)) | |
4096 | sysfs_slab_remove(s); | |
01fb58bc TH |
4097 | } |
4098 | ||
c9fc5864 TH |
4099 | void __kmemcg_cache_deactivate(struct kmem_cache *s) |
4100 | { | |
4101 | /* | |
4102 | * Disable empty slabs caching. Used to avoid pinning offline | |
4103 | * memory cgroups by kmem pages that can be freed. | |
4104 | */ | |
e6d0e1dc | 4105 | slub_set_cpu_partial(s, 0); |
c9fc5864 | 4106 | s->min_partial = 0; |
c9fc5864 | 4107 | } |
6dfd1b65 | 4108 | #endif /* CONFIG_MEMCG */ |
c9fc5864 | 4109 | |
b9049e23 YG |
4110 | static int slab_mem_going_offline_callback(void *arg) |
4111 | { | |
4112 | struct kmem_cache *s; | |
4113 | ||
18004c5d | 4114 | mutex_lock(&slab_mutex); |
b9049e23 | 4115 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4116 | __kmem_cache_shrink(s); |
18004c5d | 4117 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4118 | |
4119 | return 0; | |
4120 | } | |
4121 | ||
4122 | static void slab_mem_offline_callback(void *arg) | |
4123 | { | |
4124 | struct kmem_cache_node *n; | |
4125 | struct kmem_cache *s; | |
4126 | struct memory_notify *marg = arg; | |
4127 | int offline_node; | |
4128 | ||
b9d5ab25 | 4129 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4130 | |
4131 | /* | |
4132 | * If the node still has available memory. we need kmem_cache_node | |
4133 | * for it yet. | |
4134 | */ | |
4135 | if (offline_node < 0) | |
4136 | return; | |
4137 | ||
18004c5d | 4138 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4139 | list_for_each_entry(s, &slab_caches, list) { |
4140 | n = get_node(s, offline_node); | |
4141 | if (n) { | |
4142 | /* | |
4143 | * if n->nr_slabs > 0, slabs still exist on the node | |
4144 | * that is going down. We were unable to free them, | |
c9404c9c | 4145 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
4146 | * callback. So, we must fail. |
4147 | */ | |
0f389ec6 | 4148 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
4149 | |
4150 | s->node[offline_node] = NULL; | |
8de66a0c | 4151 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
4152 | } |
4153 | } | |
18004c5d | 4154 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4155 | } |
4156 | ||
4157 | static int slab_mem_going_online_callback(void *arg) | |
4158 | { | |
4159 | struct kmem_cache_node *n; | |
4160 | struct kmem_cache *s; | |
4161 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4162 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4163 | int ret = 0; |
4164 | ||
4165 | /* | |
4166 | * If the node's memory is already available, then kmem_cache_node is | |
4167 | * already created. Nothing to do. | |
4168 | */ | |
4169 | if (nid < 0) | |
4170 | return 0; | |
4171 | ||
4172 | /* | |
0121c619 | 4173 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4174 | * allocate a kmem_cache_node structure in order to bring the node |
4175 | * online. | |
4176 | */ | |
18004c5d | 4177 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4178 | list_for_each_entry(s, &slab_caches, list) { |
4179 | /* | |
4180 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4181 | * since memory is not yet available from the node that | |
4182 | * is brought up. | |
4183 | */ | |
8de66a0c | 4184 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4185 | if (!n) { |
4186 | ret = -ENOMEM; | |
4187 | goto out; | |
4188 | } | |
4053497d | 4189 | init_kmem_cache_node(n); |
b9049e23 YG |
4190 | s->node[nid] = n; |
4191 | } | |
4192 | out: | |
18004c5d | 4193 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4194 | return ret; |
4195 | } | |
4196 | ||
4197 | static int slab_memory_callback(struct notifier_block *self, | |
4198 | unsigned long action, void *arg) | |
4199 | { | |
4200 | int ret = 0; | |
4201 | ||
4202 | switch (action) { | |
4203 | case MEM_GOING_ONLINE: | |
4204 | ret = slab_mem_going_online_callback(arg); | |
4205 | break; | |
4206 | case MEM_GOING_OFFLINE: | |
4207 | ret = slab_mem_going_offline_callback(arg); | |
4208 | break; | |
4209 | case MEM_OFFLINE: | |
4210 | case MEM_CANCEL_ONLINE: | |
4211 | slab_mem_offline_callback(arg); | |
4212 | break; | |
4213 | case MEM_ONLINE: | |
4214 | case MEM_CANCEL_OFFLINE: | |
4215 | break; | |
4216 | } | |
dc19f9db KH |
4217 | if (ret) |
4218 | ret = notifier_from_errno(ret); | |
4219 | else | |
4220 | ret = NOTIFY_OK; | |
b9049e23 YG |
4221 | return ret; |
4222 | } | |
4223 | ||
3ac38faa AM |
4224 | static struct notifier_block slab_memory_callback_nb = { |
4225 | .notifier_call = slab_memory_callback, | |
4226 | .priority = SLAB_CALLBACK_PRI, | |
4227 | }; | |
b9049e23 | 4228 | |
81819f0f CL |
4229 | /******************************************************************** |
4230 | * Basic setup of slabs | |
4231 | *******************************************************************/ | |
4232 | ||
51df1142 CL |
4233 | /* |
4234 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4235 | * the page allocator. Allocate them properly then fix up the pointers |
4236 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4237 | */ |
4238 | ||
dffb4d60 | 4239 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4240 | { |
4241 | int node; | |
dffb4d60 | 4242 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4243 | struct kmem_cache_node *n; |
51df1142 | 4244 | |
dffb4d60 | 4245 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4246 | |
7d557b3c GC |
4247 | /* |
4248 | * This runs very early, and only the boot processor is supposed to be | |
4249 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4250 | * IPIs around. | |
4251 | */ | |
4252 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4253 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4254 | struct page *p; |
4255 | ||
916ac052 | 4256 | list_for_each_entry(p, &n->partial, slab_list) |
fa45dc25 | 4257 | p->slab_cache = s; |
51df1142 | 4258 | |
607bf324 | 4259 | #ifdef CONFIG_SLUB_DEBUG |
916ac052 | 4260 | list_for_each_entry(p, &n->full, slab_list) |
fa45dc25 | 4261 | p->slab_cache = s; |
51df1142 | 4262 | #endif |
51df1142 | 4263 | } |
f7ce3190 | 4264 | slab_init_memcg_params(s); |
dffb4d60 | 4265 | list_add(&s->list, &slab_caches); |
c03914b7 | 4266 | memcg_link_cache(s, NULL); |
dffb4d60 | 4267 | return s; |
51df1142 CL |
4268 | } |
4269 | ||
81819f0f CL |
4270 | void __init kmem_cache_init(void) |
4271 | { | |
dffb4d60 CL |
4272 | static __initdata struct kmem_cache boot_kmem_cache, |
4273 | boot_kmem_cache_node; | |
51df1142 | 4274 | |
fc8d8620 SG |
4275 | if (debug_guardpage_minorder()) |
4276 | slub_max_order = 0; | |
4277 | ||
dffb4d60 CL |
4278 | kmem_cache_node = &boot_kmem_cache_node; |
4279 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4280 | |
dffb4d60 | 4281 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4282 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4283 | |
3ac38faa | 4284 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4285 | |
4286 | /* Able to allocate the per node structures */ | |
4287 | slab_state = PARTIAL; | |
4288 | ||
dffb4d60 CL |
4289 | create_boot_cache(kmem_cache, "kmem_cache", |
4290 | offsetof(struct kmem_cache, node) + | |
4291 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4292 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4293 | |
dffb4d60 | 4294 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4295 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4296 | |
4297 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4298 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4299 | create_kmalloc_caches(0); |
81819f0f | 4300 | |
210e7a43 TG |
4301 | /* Setup random freelists for each cache */ |
4302 | init_freelist_randomization(); | |
4303 | ||
a96a87bf SAS |
4304 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4305 | slub_cpu_dead); | |
81819f0f | 4306 | |
b9726c26 | 4307 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", |
f97d5f63 | 4308 | cache_line_size(), |
81819f0f CL |
4309 | slub_min_order, slub_max_order, slub_min_objects, |
4310 | nr_cpu_ids, nr_node_ids); | |
4311 | } | |
4312 | ||
7e85ee0c PE |
4313 | void __init kmem_cache_init_late(void) |
4314 | { | |
7e85ee0c PE |
4315 | } |
4316 | ||
2633d7a0 | 4317 | struct kmem_cache * |
f4957d5b | 4318 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4319 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4320 | { |
426589f5 | 4321 | struct kmem_cache *s, *c; |
81819f0f | 4322 | |
a44cb944 | 4323 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4324 | if (s) { |
4325 | s->refcount++; | |
84d0ddd6 | 4326 | |
81819f0f CL |
4327 | /* |
4328 | * Adjust the object sizes so that we clear | |
4329 | * the complete object on kzalloc. | |
4330 | */ | |
1b473f29 | 4331 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4332 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4333 | |
426589f5 | 4334 | for_each_memcg_cache(c, s) { |
84d0ddd6 | 4335 | c->object_size = s->object_size; |
52ee6d74 | 4336 | c->inuse = max(c->inuse, ALIGN(size, sizeof(void *))); |
84d0ddd6 VD |
4337 | } |
4338 | ||
7b8f3b66 | 4339 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4340 | s->refcount--; |
cbb79694 | 4341 | s = NULL; |
7b8f3b66 | 4342 | } |
a0e1d1be | 4343 | } |
6446faa2 | 4344 | |
cbb79694 CL |
4345 | return s; |
4346 | } | |
84c1cf62 | 4347 | |
d50112ed | 4348 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4349 | { |
aac3a166 PE |
4350 | int err; |
4351 | ||
4352 | err = kmem_cache_open(s, flags); | |
4353 | if (err) | |
4354 | return err; | |
20cea968 | 4355 | |
45530c44 CL |
4356 | /* Mutex is not taken during early boot */ |
4357 | if (slab_state <= UP) | |
4358 | return 0; | |
4359 | ||
107dab5c | 4360 | memcg_propagate_slab_attrs(s); |
aac3a166 | 4361 | err = sysfs_slab_add(s); |
aac3a166 | 4362 | if (err) |
52b4b950 | 4363 | __kmem_cache_release(s); |
20cea968 | 4364 | |
aac3a166 | 4365 | return err; |
81819f0f | 4366 | } |
81819f0f | 4367 | |
ce71e27c | 4368 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4369 | { |
aadb4bc4 | 4370 | struct kmem_cache *s; |
94b528d0 | 4371 | void *ret; |
aadb4bc4 | 4372 | |
95a05b42 | 4373 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4374 | return kmalloc_large(size, gfpflags); |
4375 | ||
2c59dd65 | 4376 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4377 | |
2408c550 | 4378 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4379 | return s; |
81819f0f | 4380 | |
2b847c3c | 4381 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4382 | |
25985edc | 4383 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4384 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4385 | |
4386 | return ret; | |
81819f0f CL |
4387 | } |
4388 | ||
5d1f57e4 | 4389 | #ifdef CONFIG_NUMA |
81819f0f | 4390 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4391 | int node, unsigned long caller) |
81819f0f | 4392 | { |
aadb4bc4 | 4393 | struct kmem_cache *s; |
94b528d0 | 4394 | void *ret; |
aadb4bc4 | 4395 | |
95a05b42 | 4396 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4397 | ret = kmalloc_large_node(size, gfpflags, node); |
4398 | ||
4399 | trace_kmalloc_node(caller, ret, | |
4400 | size, PAGE_SIZE << get_order(size), | |
4401 | gfpflags, node); | |
4402 | ||
4403 | return ret; | |
4404 | } | |
eada35ef | 4405 | |
2c59dd65 | 4406 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4407 | |
2408c550 | 4408 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4409 | return s; |
81819f0f | 4410 | |
2b847c3c | 4411 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4412 | |
25985edc | 4413 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4414 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4415 | |
4416 | return ret; | |
81819f0f | 4417 | } |
5d1f57e4 | 4418 | #endif |
81819f0f | 4419 | |
ab4d5ed5 | 4420 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4421 | static int count_inuse(struct page *page) |
4422 | { | |
4423 | return page->inuse; | |
4424 | } | |
4425 | ||
4426 | static int count_total(struct page *page) | |
4427 | { | |
4428 | return page->objects; | |
4429 | } | |
ab4d5ed5 | 4430 | #endif |
205ab99d | 4431 | |
ab4d5ed5 | 4432 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 | 4433 | static void validate_slab(struct kmem_cache *s, struct page *page) |
53e15af0 CL |
4434 | { |
4435 | void *p; | |
a973e9dd | 4436 | void *addr = page_address(page); |
90e9f6a6 YZ |
4437 | unsigned long *map; |
4438 | ||
4439 | slab_lock(page); | |
53e15af0 | 4440 | |
dd98afd4 | 4441 | if (!check_slab(s, page) || !on_freelist(s, page, NULL)) |
90e9f6a6 | 4442 | goto unlock; |
53e15af0 CL |
4443 | |
4444 | /* Now we know that a valid freelist exists */ | |
90e9f6a6 | 4445 | map = get_map(s, page); |
5f80b13a | 4446 | for_each_object(p, s, addr, page->objects) { |
dd98afd4 YZ |
4447 | u8 val = test_bit(slab_index(p, s, addr), map) ? |
4448 | SLUB_RED_INACTIVE : SLUB_RED_ACTIVE; | |
53e15af0 | 4449 | |
dd98afd4 YZ |
4450 | if (!check_object(s, page, p, val)) |
4451 | break; | |
4452 | } | |
90e9f6a6 YZ |
4453 | put_map(map); |
4454 | unlock: | |
881db7fb | 4455 | slab_unlock(page); |
53e15af0 CL |
4456 | } |
4457 | ||
434e245d | 4458 | static int validate_slab_node(struct kmem_cache *s, |
90e9f6a6 | 4459 | struct kmem_cache_node *n) |
53e15af0 CL |
4460 | { |
4461 | unsigned long count = 0; | |
4462 | struct page *page; | |
4463 | unsigned long flags; | |
4464 | ||
4465 | spin_lock_irqsave(&n->list_lock, flags); | |
4466 | ||
916ac052 | 4467 | list_for_each_entry(page, &n->partial, slab_list) { |
90e9f6a6 | 4468 | validate_slab(s, page); |
53e15af0 CL |
4469 | count++; |
4470 | } | |
4471 | if (count != n->nr_partial) | |
f9f58285 FF |
4472 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4473 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4474 | |
4475 | if (!(s->flags & SLAB_STORE_USER)) | |
4476 | goto out; | |
4477 | ||
916ac052 | 4478 | list_for_each_entry(page, &n->full, slab_list) { |
90e9f6a6 | 4479 | validate_slab(s, page); |
53e15af0 CL |
4480 | count++; |
4481 | } | |
4482 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4483 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4484 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4485 | |
4486 | out: | |
4487 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4488 | return count; | |
4489 | } | |
4490 | ||
434e245d | 4491 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4492 | { |
4493 | int node; | |
4494 | unsigned long count = 0; | |
fa45dc25 | 4495 | struct kmem_cache_node *n; |
53e15af0 CL |
4496 | |
4497 | flush_all(s); | |
fa45dc25 | 4498 | for_each_kmem_cache_node(s, node, n) |
90e9f6a6 YZ |
4499 | count += validate_slab_node(s, n); |
4500 | ||
53e15af0 CL |
4501 | return count; |
4502 | } | |
88a420e4 | 4503 | /* |
672bba3a | 4504 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4505 | * and freed. |
4506 | */ | |
4507 | ||
4508 | struct location { | |
4509 | unsigned long count; | |
ce71e27c | 4510 | unsigned long addr; |
45edfa58 CL |
4511 | long long sum_time; |
4512 | long min_time; | |
4513 | long max_time; | |
4514 | long min_pid; | |
4515 | long max_pid; | |
174596a0 | 4516 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4517 | nodemask_t nodes; |
88a420e4 CL |
4518 | }; |
4519 | ||
4520 | struct loc_track { | |
4521 | unsigned long max; | |
4522 | unsigned long count; | |
4523 | struct location *loc; | |
4524 | }; | |
4525 | ||
4526 | static void free_loc_track(struct loc_track *t) | |
4527 | { | |
4528 | if (t->max) | |
4529 | free_pages((unsigned long)t->loc, | |
4530 | get_order(sizeof(struct location) * t->max)); | |
4531 | } | |
4532 | ||
68dff6a9 | 4533 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4534 | { |
4535 | struct location *l; | |
4536 | int order; | |
4537 | ||
88a420e4 CL |
4538 | order = get_order(sizeof(struct location) * max); |
4539 | ||
68dff6a9 | 4540 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4541 | if (!l) |
4542 | return 0; | |
4543 | ||
4544 | if (t->count) { | |
4545 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4546 | free_loc_track(t); | |
4547 | } | |
4548 | t->max = max; | |
4549 | t->loc = l; | |
4550 | return 1; | |
4551 | } | |
4552 | ||
4553 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4554 | const struct track *track) |
88a420e4 CL |
4555 | { |
4556 | long start, end, pos; | |
4557 | struct location *l; | |
ce71e27c | 4558 | unsigned long caddr; |
45edfa58 | 4559 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4560 | |
4561 | start = -1; | |
4562 | end = t->count; | |
4563 | ||
4564 | for ( ; ; ) { | |
4565 | pos = start + (end - start + 1) / 2; | |
4566 | ||
4567 | /* | |
4568 | * There is nothing at "end". If we end up there | |
4569 | * we need to add something to before end. | |
4570 | */ | |
4571 | if (pos == end) | |
4572 | break; | |
4573 | ||
4574 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4575 | if (track->addr == caddr) { |
4576 | ||
4577 | l = &t->loc[pos]; | |
4578 | l->count++; | |
4579 | if (track->when) { | |
4580 | l->sum_time += age; | |
4581 | if (age < l->min_time) | |
4582 | l->min_time = age; | |
4583 | if (age > l->max_time) | |
4584 | l->max_time = age; | |
4585 | ||
4586 | if (track->pid < l->min_pid) | |
4587 | l->min_pid = track->pid; | |
4588 | if (track->pid > l->max_pid) | |
4589 | l->max_pid = track->pid; | |
4590 | ||
174596a0 RR |
4591 | cpumask_set_cpu(track->cpu, |
4592 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4593 | } |
4594 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4595 | return 1; |
4596 | } | |
4597 | ||
45edfa58 | 4598 | if (track->addr < caddr) |
88a420e4 CL |
4599 | end = pos; |
4600 | else | |
4601 | start = pos; | |
4602 | } | |
4603 | ||
4604 | /* | |
672bba3a | 4605 | * Not found. Insert new tracking element. |
88a420e4 | 4606 | */ |
68dff6a9 | 4607 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4608 | return 0; |
4609 | ||
4610 | l = t->loc + pos; | |
4611 | if (pos < t->count) | |
4612 | memmove(l + 1, l, | |
4613 | (t->count - pos) * sizeof(struct location)); | |
4614 | t->count++; | |
4615 | l->count = 1; | |
45edfa58 CL |
4616 | l->addr = track->addr; |
4617 | l->sum_time = age; | |
4618 | l->min_time = age; | |
4619 | l->max_time = age; | |
4620 | l->min_pid = track->pid; | |
4621 | l->max_pid = track->pid; | |
174596a0 RR |
4622 | cpumask_clear(to_cpumask(l->cpus)); |
4623 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4624 | nodes_clear(l->nodes); |
4625 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4626 | return 1; |
4627 | } | |
4628 | ||
4629 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
90e9f6a6 | 4630 | struct page *page, enum track_item alloc) |
88a420e4 | 4631 | { |
a973e9dd | 4632 | void *addr = page_address(page); |
88a420e4 | 4633 | void *p; |
90e9f6a6 | 4634 | unsigned long *map; |
88a420e4 | 4635 | |
90e9f6a6 | 4636 | map = get_map(s, page); |
224a88be | 4637 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4638 | if (!test_bit(slab_index(p, s, addr), map)) |
4639 | add_location(t, s, get_track(s, p, alloc)); | |
90e9f6a6 | 4640 | put_map(map); |
88a420e4 CL |
4641 | } |
4642 | ||
4643 | static int list_locations(struct kmem_cache *s, char *buf, | |
4644 | enum track_item alloc) | |
4645 | { | |
e374d483 | 4646 | int len = 0; |
88a420e4 | 4647 | unsigned long i; |
68dff6a9 | 4648 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4649 | int node; |
fa45dc25 | 4650 | struct kmem_cache_node *n; |
88a420e4 | 4651 | |
90e9f6a6 YZ |
4652 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4653 | GFP_KERNEL)) { | |
68dff6a9 | 4654 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4655 | } |
88a420e4 CL |
4656 | /* Push back cpu slabs */ |
4657 | flush_all(s); | |
4658 | ||
fa45dc25 | 4659 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4660 | unsigned long flags; |
4661 | struct page *page; | |
4662 | ||
9e86943b | 4663 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4664 | continue; |
4665 | ||
4666 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 4667 | list_for_each_entry(page, &n->partial, slab_list) |
90e9f6a6 | 4668 | process_slab(&t, s, page, alloc); |
916ac052 | 4669 | list_for_each_entry(page, &n->full, slab_list) |
90e9f6a6 | 4670 | process_slab(&t, s, page, alloc); |
88a420e4 CL |
4671 | spin_unlock_irqrestore(&n->list_lock, flags); |
4672 | } | |
4673 | ||
4674 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4675 | struct location *l = &t.loc[i]; |
88a420e4 | 4676 | |
9c246247 | 4677 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4678 | break; |
e374d483 | 4679 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4680 | |
4681 | if (l->addr) | |
62c70bce | 4682 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4683 | else |
e374d483 | 4684 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4685 | |
4686 | if (l->sum_time != l->min_time) { | |
e374d483 | 4687 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4688 | l->min_time, |
4689 | (long)div_u64(l->sum_time, l->count), | |
4690 | l->max_time); | |
45edfa58 | 4691 | } else |
e374d483 | 4692 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4693 | l->min_time); |
4694 | ||
4695 | if (l->min_pid != l->max_pid) | |
e374d483 | 4696 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4697 | l->min_pid, l->max_pid); |
4698 | else | |
e374d483 | 4699 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4700 | l->min_pid); |
4701 | ||
174596a0 RR |
4702 | if (num_online_cpus() > 1 && |
4703 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4704 | len < PAGE_SIZE - 60) |
4705 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4706 | " cpus=%*pbl", | |
4707 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4708 | |
62bc62a8 | 4709 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4710 | len < PAGE_SIZE - 60) |
4711 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4712 | " nodes=%*pbl", | |
4713 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4714 | |
e374d483 | 4715 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4716 | } |
4717 | ||
4718 | free_loc_track(&t); | |
4719 | if (!t.count) | |
e374d483 HH |
4720 | len += sprintf(buf, "No data\n"); |
4721 | return len; | |
88a420e4 | 4722 | } |
6dfd1b65 | 4723 | #endif /* CONFIG_SLUB_DEBUG */ |
88a420e4 | 4724 | |
a5a84755 | 4725 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4726 | static void __init resiliency_test(void) |
a5a84755 CL |
4727 | { |
4728 | u8 *p; | |
cc252eae | 4729 | int type = KMALLOC_NORMAL; |
a5a84755 | 4730 | |
95a05b42 | 4731 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4732 | |
f9f58285 FF |
4733 | pr_err("SLUB resiliency testing\n"); |
4734 | pr_err("-----------------------\n"); | |
4735 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4736 | |
4737 | p = kzalloc(16, GFP_KERNEL); | |
4738 | p[16] = 0x12; | |
f9f58285 FF |
4739 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4740 | p + 16); | |
a5a84755 | 4741 | |
cc252eae | 4742 | validate_slab_cache(kmalloc_caches[type][4]); |
a5a84755 CL |
4743 | |
4744 | /* Hmmm... The next two are dangerous */ | |
4745 | p = kzalloc(32, GFP_KERNEL); | |
4746 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4747 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4748 | p); | |
4749 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 | 4750 | |
cc252eae | 4751 | validate_slab_cache(kmalloc_caches[type][5]); |
a5a84755 CL |
4752 | p = kzalloc(64, GFP_KERNEL); |
4753 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4754 | *p = 0x56; | |
f9f58285 FF |
4755 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4756 | p); | |
4757 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
cc252eae | 4758 | validate_slab_cache(kmalloc_caches[type][6]); |
a5a84755 | 4759 | |
f9f58285 | 4760 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4761 | p = kzalloc(128, GFP_KERNEL); |
4762 | kfree(p); | |
4763 | *p = 0x78; | |
f9f58285 | 4764 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
cc252eae | 4765 | validate_slab_cache(kmalloc_caches[type][7]); |
a5a84755 CL |
4766 | |
4767 | p = kzalloc(256, GFP_KERNEL); | |
4768 | kfree(p); | |
4769 | p[50] = 0x9a; | |
f9f58285 | 4770 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
cc252eae | 4771 | validate_slab_cache(kmalloc_caches[type][8]); |
a5a84755 CL |
4772 | |
4773 | p = kzalloc(512, GFP_KERNEL); | |
4774 | kfree(p); | |
4775 | p[512] = 0xab; | |
f9f58285 | 4776 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
cc252eae | 4777 | validate_slab_cache(kmalloc_caches[type][9]); |
a5a84755 CL |
4778 | } |
4779 | #else | |
4780 | #ifdef CONFIG_SYSFS | |
4781 | static void resiliency_test(void) {}; | |
4782 | #endif | |
6dfd1b65 | 4783 | #endif /* SLUB_RESILIENCY_TEST */ |
a5a84755 | 4784 | |
ab4d5ed5 | 4785 | #ifdef CONFIG_SYSFS |
81819f0f | 4786 | enum slab_stat_type { |
205ab99d CL |
4787 | SL_ALL, /* All slabs */ |
4788 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4789 | SL_CPU, /* Only slabs used for cpu caches */ | |
4790 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4791 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4792 | }; |
4793 | ||
205ab99d | 4794 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4795 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4796 | #define SO_CPU (1 << SL_CPU) | |
4797 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4798 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4799 | |
1663f26d TH |
4800 | #ifdef CONFIG_MEMCG |
4801 | static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON); | |
4802 | ||
4803 | static int __init setup_slub_memcg_sysfs(char *str) | |
4804 | { | |
4805 | int v; | |
4806 | ||
4807 | if (get_option(&str, &v) > 0) | |
4808 | memcg_sysfs_enabled = v; | |
4809 | ||
4810 | return 1; | |
4811 | } | |
4812 | ||
4813 | __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs); | |
4814 | #endif | |
4815 | ||
62e5c4b4 CG |
4816 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4817 | char *buf, unsigned long flags) | |
81819f0f CL |
4818 | { |
4819 | unsigned long total = 0; | |
81819f0f CL |
4820 | int node; |
4821 | int x; | |
4822 | unsigned long *nodes; | |
81819f0f | 4823 | |
6396bb22 | 4824 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
4825 | if (!nodes) |
4826 | return -ENOMEM; | |
81819f0f | 4827 | |
205ab99d CL |
4828 | if (flags & SO_CPU) { |
4829 | int cpu; | |
81819f0f | 4830 | |
205ab99d | 4831 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4832 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4833 | cpu); | |
ec3ab083 | 4834 | int node; |
49e22585 | 4835 | struct page *page; |
dfb4f096 | 4836 | |
4db0c3c2 | 4837 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4838 | if (!page) |
4839 | continue; | |
205ab99d | 4840 | |
ec3ab083 CL |
4841 | node = page_to_nid(page); |
4842 | if (flags & SO_TOTAL) | |
4843 | x = page->objects; | |
4844 | else if (flags & SO_OBJECTS) | |
4845 | x = page->inuse; | |
4846 | else | |
4847 | x = 1; | |
49e22585 | 4848 | |
ec3ab083 CL |
4849 | total += x; |
4850 | nodes[node] += x; | |
4851 | ||
a93cf07b | 4852 | page = slub_percpu_partial_read_once(c); |
49e22585 | 4853 | if (page) { |
8afb1474 LZ |
4854 | node = page_to_nid(page); |
4855 | if (flags & SO_TOTAL) | |
4856 | WARN_ON_ONCE(1); | |
4857 | else if (flags & SO_OBJECTS) | |
4858 | WARN_ON_ONCE(1); | |
4859 | else | |
4860 | x = page->pages; | |
bc6697d8 ED |
4861 | total += x; |
4862 | nodes[node] += x; | |
49e22585 | 4863 | } |
81819f0f CL |
4864 | } |
4865 | } | |
4866 | ||
e4f8e513 QC |
4867 | /* |
4868 | * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex" | |
4869 | * already held which will conflict with an existing lock order: | |
4870 | * | |
4871 | * mem_hotplug_lock->slab_mutex->kernfs_mutex | |
4872 | * | |
4873 | * We don't really need mem_hotplug_lock (to hold off | |
4874 | * slab_mem_going_offline_callback) here because slab's memory hot | |
4875 | * unplug code doesn't destroy the kmem_cache->node[] data. | |
4876 | */ | |
4877 | ||
ab4d5ed5 | 4878 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4879 | if (flags & SO_ALL) { |
fa45dc25 CL |
4880 | struct kmem_cache_node *n; |
4881 | ||
4882 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4883 | |
d0e0ac97 CG |
4884 | if (flags & SO_TOTAL) |
4885 | x = atomic_long_read(&n->total_objects); | |
4886 | else if (flags & SO_OBJECTS) | |
4887 | x = atomic_long_read(&n->total_objects) - | |
4888 | count_partial(n, count_free); | |
81819f0f | 4889 | else |
205ab99d | 4890 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4891 | total += x; |
4892 | nodes[node] += x; | |
4893 | } | |
4894 | ||
ab4d5ed5 CL |
4895 | } else |
4896 | #endif | |
4897 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4898 | struct kmem_cache_node *n; |
81819f0f | 4899 | |
fa45dc25 | 4900 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4901 | if (flags & SO_TOTAL) |
4902 | x = count_partial(n, count_total); | |
4903 | else if (flags & SO_OBJECTS) | |
4904 | x = count_partial(n, count_inuse); | |
81819f0f | 4905 | else |
205ab99d | 4906 | x = n->nr_partial; |
81819f0f CL |
4907 | total += x; |
4908 | nodes[node] += x; | |
4909 | } | |
4910 | } | |
81819f0f CL |
4911 | x = sprintf(buf, "%lu", total); |
4912 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4913 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4914 | if (nodes[node]) |
4915 | x += sprintf(buf + x, " N%d=%lu", | |
4916 | node, nodes[node]); | |
4917 | #endif | |
4918 | kfree(nodes); | |
4919 | return x + sprintf(buf + x, "\n"); | |
4920 | } | |
4921 | ||
ab4d5ed5 | 4922 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4923 | static int any_slab_objects(struct kmem_cache *s) |
4924 | { | |
4925 | int node; | |
fa45dc25 | 4926 | struct kmem_cache_node *n; |
81819f0f | 4927 | |
fa45dc25 | 4928 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4929 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4930 | return 1; |
fa45dc25 | 4931 | |
81819f0f CL |
4932 | return 0; |
4933 | } | |
ab4d5ed5 | 4934 | #endif |
81819f0f CL |
4935 | |
4936 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4937 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4938 | |
4939 | struct slab_attribute { | |
4940 | struct attribute attr; | |
4941 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4942 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4943 | }; | |
4944 | ||
4945 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4946 | static struct slab_attribute _name##_attr = \ |
4947 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4948 | |
4949 | #define SLAB_ATTR(_name) \ | |
4950 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4951 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4952 | |
81819f0f CL |
4953 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4954 | { | |
44065b2e | 4955 | return sprintf(buf, "%u\n", s->size); |
81819f0f CL |
4956 | } |
4957 | SLAB_ATTR_RO(slab_size); | |
4958 | ||
4959 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4960 | { | |
3a3791ec | 4961 | return sprintf(buf, "%u\n", s->align); |
81819f0f CL |
4962 | } |
4963 | SLAB_ATTR_RO(align); | |
4964 | ||
4965 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4966 | { | |
1b473f29 | 4967 | return sprintf(buf, "%u\n", s->object_size); |
81819f0f CL |
4968 | } |
4969 | SLAB_ATTR_RO(object_size); | |
4970 | ||
4971 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4972 | { | |
19af27af | 4973 | return sprintf(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
4974 | } |
4975 | SLAB_ATTR_RO(objs_per_slab); | |
4976 | ||
06b285dc CL |
4977 | static ssize_t order_store(struct kmem_cache *s, |
4978 | const char *buf, size_t length) | |
4979 | { | |
19af27af | 4980 | unsigned int order; |
0121c619 CL |
4981 | int err; |
4982 | ||
19af27af | 4983 | err = kstrtouint(buf, 10, &order); |
0121c619 CL |
4984 | if (err) |
4985 | return err; | |
06b285dc CL |
4986 | |
4987 | if (order > slub_max_order || order < slub_min_order) | |
4988 | return -EINVAL; | |
4989 | ||
4990 | calculate_sizes(s, order); | |
4991 | return length; | |
4992 | } | |
4993 | ||
81819f0f CL |
4994 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4995 | { | |
19af27af | 4996 | return sprintf(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 4997 | } |
06b285dc | 4998 | SLAB_ATTR(order); |
81819f0f | 4999 | |
73d342b1 DR |
5000 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
5001 | { | |
5002 | return sprintf(buf, "%lu\n", s->min_partial); | |
5003 | } | |
5004 | ||
5005 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
5006 | size_t length) | |
5007 | { | |
5008 | unsigned long min; | |
5009 | int err; | |
5010 | ||
3dbb95f7 | 5011 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
5012 | if (err) |
5013 | return err; | |
5014 | ||
c0bdb232 | 5015 | set_min_partial(s, min); |
73d342b1 DR |
5016 | return length; |
5017 | } | |
5018 | SLAB_ATTR(min_partial); | |
5019 | ||
49e22585 CL |
5020 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
5021 | { | |
e6d0e1dc | 5022 | return sprintf(buf, "%u\n", slub_cpu_partial(s)); |
49e22585 CL |
5023 | } |
5024 | ||
5025 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
5026 | size_t length) | |
5027 | { | |
e5d9998f | 5028 | unsigned int objects; |
49e22585 CL |
5029 | int err; |
5030 | ||
e5d9998f | 5031 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
5032 | if (err) |
5033 | return err; | |
345c905d | 5034 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 5035 | return -EINVAL; |
49e22585 | 5036 | |
e6d0e1dc | 5037 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
5038 | flush_all(s); |
5039 | return length; | |
5040 | } | |
5041 | SLAB_ATTR(cpu_partial); | |
5042 | ||
81819f0f CL |
5043 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
5044 | { | |
62c70bce JP |
5045 | if (!s->ctor) |
5046 | return 0; | |
5047 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
5048 | } |
5049 | SLAB_ATTR_RO(ctor); | |
5050 | ||
81819f0f CL |
5051 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
5052 | { | |
4307c14f | 5053 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
5054 | } |
5055 | SLAB_ATTR_RO(aliases); | |
5056 | ||
81819f0f CL |
5057 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5058 | { | |
d9acf4b7 | 5059 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5060 | } |
5061 | SLAB_ATTR_RO(partial); | |
5062 | ||
5063 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5064 | { | |
d9acf4b7 | 5065 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5066 | } |
5067 | SLAB_ATTR_RO(cpu_slabs); | |
5068 | ||
5069 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5070 | { | |
205ab99d | 5071 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5072 | } |
5073 | SLAB_ATTR_RO(objects); | |
5074 | ||
205ab99d CL |
5075 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5076 | { | |
5077 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5078 | } | |
5079 | SLAB_ATTR_RO(objects_partial); | |
5080 | ||
49e22585 CL |
5081 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5082 | { | |
5083 | int objects = 0; | |
5084 | int pages = 0; | |
5085 | int cpu; | |
5086 | int len; | |
5087 | ||
5088 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5089 | struct page *page; |
5090 | ||
5091 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5092 | |
5093 | if (page) { | |
5094 | pages += page->pages; | |
5095 | objects += page->pobjects; | |
5096 | } | |
5097 | } | |
5098 | ||
5099 | len = sprintf(buf, "%d(%d)", objects, pages); | |
5100 | ||
5101 | #ifdef CONFIG_SMP | |
5102 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5103 | struct page *page; |
5104 | ||
5105 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5106 | |
5107 | if (page && len < PAGE_SIZE - 20) | |
5108 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
5109 | page->pobjects, page->pages); | |
5110 | } | |
5111 | #endif | |
5112 | return len + sprintf(buf + len, "\n"); | |
5113 | } | |
5114 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5115 | ||
a5a84755 CL |
5116 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5117 | { | |
5118 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
5119 | } | |
5120 | ||
5121 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
5122 | const char *buf, size_t length) | |
5123 | { | |
5124 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
5125 | if (buf[0] == '1') | |
5126 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
5127 | return length; | |
5128 | } | |
5129 | SLAB_ATTR(reclaim_account); | |
5130 | ||
5131 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5132 | { | |
5133 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
5134 | } | |
5135 | SLAB_ATTR_RO(hwcache_align); | |
5136 | ||
5137 | #ifdef CONFIG_ZONE_DMA | |
5138 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5139 | { | |
5140 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
5141 | } | |
5142 | SLAB_ATTR_RO(cache_dma); | |
5143 | #endif | |
5144 | ||
8eb8284b DW |
5145 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5146 | { | |
7bbdb81e | 5147 | return sprintf(buf, "%u\n", s->usersize); |
8eb8284b DW |
5148 | } |
5149 | SLAB_ATTR_RO(usersize); | |
5150 | ||
a5a84755 CL |
5151 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5152 | { | |
5f0d5a3a | 5153 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5154 | } |
5155 | SLAB_ATTR_RO(destroy_by_rcu); | |
5156 | ||
ab4d5ed5 | 5157 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5158 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5159 | { | |
5160 | return show_slab_objects(s, buf, SO_ALL); | |
5161 | } | |
5162 | SLAB_ATTR_RO(slabs); | |
5163 | ||
205ab99d CL |
5164 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5165 | { | |
5166 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5167 | } | |
5168 | SLAB_ATTR_RO(total_objects); | |
5169 | ||
81819f0f CL |
5170 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5171 | { | |
becfda68 | 5172 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f CL |
5173 | } |
5174 | ||
5175 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
5176 | const char *buf, size_t length) | |
5177 | { | |
becfda68 | 5178 | s->flags &= ~SLAB_CONSISTENCY_CHECKS; |
b789ef51 CL |
5179 | if (buf[0] == '1') { |
5180 | s->flags &= ~__CMPXCHG_DOUBLE; | |
becfda68 | 5181 | s->flags |= SLAB_CONSISTENCY_CHECKS; |
b789ef51 | 5182 | } |
81819f0f CL |
5183 | return length; |
5184 | } | |
5185 | SLAB_ATTR(sanity_checks); | |
5186 | ||
5187 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5188 | { | |
5189 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
5190 | } | |
5191 | ||
5192 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
5193 | size_t length) | |
5194 | { | |
c9e16131 CL |
5195 | /* |
5196 | * Tracing a merged cache is going to give confusing results | |
5197 | * as well as cause other issues like converting a mergeable | |
5198 | * cache into an umergeable one. | |
5199 | */ | |
5200 | if (s->refcount > 1) | |
5201 | return -EINVAL; | |
5202 | ||
81819f0f | 5203 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
5204 | if (buf[0] == '1') { |
5205 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5206 | s->flags |= SLAB_TRACE; |
b789ef51 | 5207 | } |
81819f0f CL |
5208 | return length; |
5209 | } | |
5210 | SLAB_ATTR(trace); | |
5211 | ||
81819f0f CL |
5212 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5213 | { | |
5214 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
5215 | } | |
5216 | ||
5217 | static ssize_t red_zone_store(struct kmem_cache *s, | |
5218 | const char *buf, size_t length) | |
5219 | { | |
5220 | if (any_slab_objects(s)) | |
5221 | return -EBUSY; | |
5222 | ||
5223 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 | 5224 | if (buf[0] == '1') { |
81819f0f | 5225 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 5226 | } |
06b285dc | 5227 | calculate_sizes(s, -1); |
81819f0f CL |
5228 | return length; |
5229 | } | |
5230 | SLAB_ATTR(red_zone); | |
5231 | ||
5232 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5233 | { | |
5234 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
5235 | } | |
5236 | ||
5237 | static ssize_t poison_store(struct kmem_cache *s, | |
5238 | const char *buf, size_t length) | |
5239 | { | |
5240 | if (any_slab_objects(s)) | |
5241 | return -EBUSY; | |
5242 | ||
5243 | s->flags &= ~SLAB_POISON; | |
b789ef51 | 5244 | if (buf[0] == '1') { |
81819f0f | 5245 | s->flags |= SLAB_POISON; |
b789ef51 | 5246 | } |
06b285dc | 5247 | calculate_sizes(s, -1); |
81819f0f CL |
5248 | return length; |
5249 | } | |
5250 | SLAB_ATTR(poison); | |
5251 | ||
5252 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5253 | { | |
5254 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
5255 | } | |
5256 | ||
5257 | static ssize_t store_user_store(struct kmem_cache *s, | |
5258 | const char *buf, size_t length) | |
5259 | { | |
5260 | if (any_slab_objects(s)) | |
5261 | return -EBUSY; | |
5262 | ||
5263 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
5264 | if (buf[0] == '1') { |
5265 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5266 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 5267 | } |
06b285dc | 5268 | calculate_sizes(s, -1); |
81819f0f CL |
5269 | return length; |
5270 | } | |
5271 | SLAB_ATTR(store_user); | |
5272 | ||
53e15af0 CL |
5273 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5274 | { | |
5275 | return 0; | |
5276 | } | |
5277 | ||
5278 | static ssize_t validate_store(struct kmem_cache *s, | |
5279 | const char *buf, size_t length) | |
5280 | { | |
434e245d CL |
5281 | int ret = -EINVAL; |
5282 | ||
5283 | if (buf[0] == '1') { | |
5284 | ret = validate_slab_cache(s); | |
5285 | if (ret >= 0) | |
5286 | ret = length; | |
5287 | } | |
5288 | return ret; | |
53e15af0 CL |
5289 | } |
5290 | SLAB_ATTR(validate); | |
a5a84755 CL |
5291 | |
5292 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5293 | { | |
5294 | if (!(s->flags & SLAB_STORE_USER)) | |
5295 | return -ENOSYS; | |
5296 | return list_locations(s, buf, TRACK_ALLOC); | |
5297 | } | |
5298 | SLAB_ATTR_RO(alloc_calls); | |
5299 | ||
5300 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5301 | { | |
5302 | if (!(s->flags & SLAB_STORE_USER)) | |
5303 | return -ENOSYS; | |
5304 | return list_locations(s, buf, TRACK_FREE); | |
5305 | } | |
5306 | SLAB_ATTR_RO(free_calls); | |
5307 | #endif /* CONFIG_SLUB_DEBUG */ | |
5308 | ||
5309 | #ifdef CONFIG_FAILSLAB | |
5310 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5311 | { | |
5312 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
5313 | } | |
5314 | ||
5315 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
5316 | size_t length) | |
5317 | { | |
c9e16131 CL |
5318 | if (s->refcount > 1) |
5319 | return -EINVAL; | |
5320 | ||
a5a84755 CL |
5321 | s->flags &= ~SLAB_FAILSLAB; |
5322 | if (buf[0] == '1') | |
5323 | s->flags |= SLAB_FAILSLAB; | |
5324 | return length; | |
5325 | } | |
5326 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 5327 | #endif |
53e15af0 | 5328 | |
2086d26a CL |
5329 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5330 | { | |
5331 | return 0; | |
5332 | } | |
5333 | ||
5334 | static ssize_t shrink_store(struct kmem_cache *s, | |
5335 | const char *buf, size_t length) | |
5336 | { | |
832f37f5 | 5337 | if (buf[0] == '1') |
04f768a3 | 5338 | kmem_cache_shrink_all(s); |
832f37f5 | 5339 | else |
2086d26a CL |
5340 | return -EINVAL; |
5341 | return length; | |
5342 | } | |
5343 | SLAB_ATTR(shrink); | |
5344 | ||
81819f0f | 5345 | #ifdef CONFIG_NUMA |
9824601e | 5346 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5347 | { |
eb7235eb | 5348 | return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5349 | } |
5350 | ||
9824601e | 5351 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5352 | const char *buf, size_t length) |
5353 | { | |
eb7235eb | 5354 | unsigned int ratio; |
0121c619 CL |
5355 | int err; |
5356 | ||
eb7235eb | 5357 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5358 | if (err) |
5359 | return err; | |
eb7235eb AD |
5360 | if (ratio > 100) |
5361 | return -ERANGE; | |
0121c619 | 5362 | |
eb7235eb | 5363 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5364 | |
81819f0f CL |
5365 | return length; |
5366 | } | |
9824601e | 5367 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5368 | #endif |
5369 | ||
8ff12cfc | 5370 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5371 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5372 | { | |
5373 | unsigned long sum = 0; | |
5374 | int cpu; | |
5375 | int len; | |
6da2ec56 | 5376 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5377 | |
5378 | if (!data) | |
5379 | return -ENOMEM; | |
5380 | ||
5381 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5382 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5383 | |
5384 | data[cpu] = x; | |
5385 | sum += x; | |
5386 | } | |
5387 | ||
5388 | len = sprintf(buf, "%lu", sum); | |
5389 | ||
50ef37b9 | 5390 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5391 | for_each_online_cpu(cpu) { |
5392 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5393 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5394 | } |
50ef37b9 | 5395 | #endif |
8ff12cfc CL |
5396 | kfree(data); |
5397 | return len + sprintf(buf + len, "\n"); | |
5398 | } | |
5399 | ||
78eb00cc DR |
5400 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5401 | { | |
5402 | int cpu; | |
5403 | ||
5404 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5405 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5406 | } |
5407 | ||
8ff12cfc CL |
5408 | #define STAT_ATTR(si, text) \ |
5409 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5410 | { \ | |
5411 | return show_stat(s, buf, si); \ | |
5412 | } \ | |
78eb00cc DR |
5413 | static ssize_t text##_store(struct kmem_cache *s, \ |
5414 | const char *buf, size_t length) \ | |
5415 | { \ | |
5416 | if (buf[0] != '0') \ | |
5417 | return -EINVAL; \ | |
5418 | clear_stat(s, si); \ | |
5419 | return length; \ | |
5420 | } \ | |
5421 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5422 | |
5423 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5424 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5425 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5426 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5427 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5428 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5429 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5430 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5431 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5432 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5433 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5434 | STAT_ATTR(FREE_SLAB, free_slab); |
5435 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5436 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5437 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5438 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5439 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5440 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5441 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5442 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5443 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5444 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5445 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5446 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5447 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5448 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
6dfd1b65 | 5449 | #endif /* CONFIG_SLUB_STATS */ |
8ff12cfc | 5450 | |
06428780 | 5451 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5452 | &slab_size_attr.attr, |
5453 | &object_size_attr.attr, | |
5454 | &objs_per_slab_attr.attr, | |
5455 | &order_attr.attr, | |
73d342b1 | 5456 | &min_partial_attr.attr, |
49e22585 | 5457 | &cpu_partial_attr.attr, |
81819f0f | 5458 | &objects_attr.attr, |
205ab99d | 5459 | &objects_partial_attr.attr, |
81819f0f CL |
5460 | &partial_attr.attr, |
5461 | &cpu_slabs_attr.attr, | |
5462 | &ctor_attr.attr, | |
81819f0f CL |
5463 | &aliases_attr.attr, |
5464 | &align_attr.attr, | |
81819f0f CL |
5465 | &hwcache_align_attr.attr, |
5466 | &reclaim_account_attr.attr, | |
5467 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5468 | &shrink_attr.attr, |
49e22585 | 5469 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5470 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5471 | &total_objects_attr.attr, |
5472 | &slabs_attr.attr, | |
5473 | &sanity_checks_attr.attr, | |
5474 | &trace_attr.attr, | |
81819f0f CL |
5475 | &red_zone_attr.attr, |
5476 | &poison_attr.attr, | |
5477 | &store_user_attr.attr, | |
53e15af0 | 5478 | &validate_attr.attr, |
88a420e4 CL |
5479 | &alloc_calls_attr.attr, |
5480 | &free_calls_attr.attr, | |
ab4d5ed5 | 5481 | #endif |
81819f0f CL |
5482 | #ifdef CONFIG_ZONE_DMA |
5483 | &cache_dma_attr.attr, | |
5484 | #endif | |
5485 | #ifdef CONFIG_NUMA | |
9824601e | 5486 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5487 | #endif |
5488 | #ifdef CONFIG_SLUB_STATS | |
5489 | &alloc_fastpath_attr.attr, | |
5490 | &alloc_slowpath_attr.attr, | |
5491 | &free_fastpath_attr.attr, | |
5492 | &free_slowpath_attr.attr, | |
5493 | &free_frozen_attr.attr, | |
5494 | &free_add_partial_attr.attr, | |
5495 | &free_remove_partial_attr.attr, | |
5496 | &alloc_from_partial_attr.attr, | |
5497 | &alloc_slab_attr.attr, | |
5498 | &alloc_refill_attr.attr, | |
e36a2652 | 5499 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5500 | &free_slab_attr.attr, |
5501 | &cpuslab_flush_attr.attr, | |
5502 | &deactivate_full_attr.attr, | |
5503 | &deactivate_empty_attr.attr, | |
5504 | &deactivate_to_head_attr.attr, | |
5505 | &deactivate_to_tail_attr.attr, | |
5506 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5507 | &deactivate_bypass_attr.attr, |
65c3376a | 5508 | &order_fallback_attr.attr, |
b789ef51 CL |
5509 | &cmpxchg_double_fail_attr.attr, |
5510 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5511 | &cpu_partial_alloc_attr.attr, |
5512 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5513 | &cpu_partial_node_attr.attr, |
5514 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5515 | #endif |
4c13dd3b DM |
5516 | #ifdef CONFIG_FAILSLAB |
5517 | &failslab_attr.attr, | |
5518 | #endif | |
8eb8284b | 5519 | &usersize_attr.attr, |
4c13dd3b | 5520 | |
81819f0f CL |
5521 | NULL |
5522 | }; | |
5523 | ||
1fdaaa23 | 5524 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5525 | .attrs = slab_attrs, |
5526 | }; | |
5527 | ||
5528 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5529 | struct attribute *attr, | |
5530 | char *buf) | |
5531 | { | |
5532 | struct slab_attribute *attribute; | |
5533 | struct kmem_cache *s; | |
5534 | int err; | |
5535 | ||
5536 | attribute = to_slab_attr(attr); | |
5537 | s = to_slab(kobj); | |
5538 | ||
5539 | if (!attribute->show) | |
5540 | return -EIO; | |
5541 | ||
5542 | err = attribute->show(s, buf); | |
5543 | ||
5544 | return err; | |
5545 | } | |
5546 | ||
5547 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5548 | struct attribute *attr, | |
5549 | const char *buf, size_t len) | |
5550 | { | |
5551 | struct slab_attribute *attribute; | |
5552 | struct kmem_cache *s; | |
5553 | int err; | |
5554 | ||
5555 | attribute = to_slab_attr(attr); | |
5556 | s = to_slab(kobj); | |
5557 | ||
5558 | if (!attribute->store) | |
5559 | return -EIO; | |
5560 | ||
5561 | err = attribute->store(s, buf, len); | |
127424c8 | 5562 | #ifdef CONFIG_MEMCG |
107dab5c | 5563 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5564 | struct kmem_cache *c; |
81819f0f | 5565 | |
107dab5c GC |
5566 | mutex_lock(&slab_mutex); |
5567 | if (s->max_attr_size < len) | |
5568 | s->max_attr_size = len; | |
5569 | ||
ebe945c2 GC |
5570 | /* |
5571 | * This is a best effort propagation, so this function's return | |
5572 | * value will be determined by the parent cache only. This is | |
5573 | * basically because not all attributes will have a well | |
5574 | * defined semantics for rollbacks - most of the actions will | |
5575 | * have permanent effects. | |
5576 | * | |
5577 | * Returning the error value of any of the children that fail | |
5578 | * is not 100 % defined, in the sense that users seeing the | |
5579 | * error code won't be able to know anything about the state of | |
5580 | * the cache. | |
5581 | * | |
5582 | * Only returning the error code for the parent cache at least | |
5583 | * has well defined semantics. The cache being written to | |
5584 | * directly either failed or succeeded, in which case we loop | |
5585 | * through the descendants with best-effort propagation. | |
5586 | */ | |
426589f5 VD |
5587 | for_each_memcg_cache(c, s) |
5588 | attribute->store(c, buf, len); | |
107dab5c GC |
5589 | mutex_unlock(&slab_mutex); |
5590 | } | |
5591 | #endif | |
81819f0f CL |
5592 | return err; |
5593 | } | |
5594 | ||
107dab5c GC |
5595 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5596 | { | |
127424c8 | 5597 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5598 | int i; |
5599 | char *buffer = NULL; | |
93030d83 | 5600 | struct kmem_cache *root_cache; |
107dab5c | 5601 | |
93030d83 | 5602 | if (is_root_cache(s)) |
107dab5c GC |
5603 | return; |
5604 | ||
f7ce3190 | 5605 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5606 | |
107dab5c GC |
5607 | /* |
5608 | * This mean this cache had no attribute written. Therefore, no point | |
5609 | * in copying default values around | |
5610 | */ | |
93030d83 | 5611 | if (!root_cache->max_attr_size) |
107dab5c GC |
5612 | return; |
5613 | ||
5614 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5615 | char mbuf[64]; | |
5616 | char *buf; | |
5617 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
478fe303 | 5618 | ssize_t len; |
107dab5c GC |
5619 | |
5620 | if (!attr || !attr->store || !attr->show) | |
5621 | continue; | |
5622 | ||
5623 | /* | |
5624 | * It is really bad that we have to allocate here, so we will | |
5625 | * do it only as a fallback. If we actually allocate, though, | |
5626 | * we can just use the allocated buffer until the end. | |
5627 | * | |
5628 | * Most of the slub attributes will tend to be very small in | |
5629 | * size, but sysfs allows buffers up to a page, so they can | |
5630 | * theoretically happen. | |
5631 | */ | |
5632 | if (buffer) | |
5633 | buf = buffer; | |
93030d83 | 5634 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5635 | buf = mbuf; |
5636 | else { | |
5637 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5638 | if (WARN_ON(!buffer)) | |
5639 | continue; | |
5640 | buf = buffer; | |
5641 | } | |
5642 | ||
478fe303 TG |
5643 | len = attr->show(root_cache, buf); |
5644 | if (len > 0) | |
5645 | attr->store(s, buf, len); | |
107dab5c GC |
5646 | } |
5647 | ||
5648 | if (buffer) | |
5649 | free_page((unsigned long)buffer); | |
6dfd1b65 | 5650 | #endif /* CONFIG_MEMCG */ |
107dab5c GC |
5651 | } |
5652 | ||
41a21285 CL |
5653 | static void kmem_cache_release(struct kobject *k) |
5654 | { | |
5655 | slab_kmem_cache_release(to_slab(k)); | |
5656 | } | |
5657 | ||
52cf25d0 | 5658 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5659 | .show = slab_attr_show, |
5660 | .store = slab_attr_store, | |
5661 | }; | |
5662 | ||
5663 | static struct kobj_type slab_ktype = { | |
5664 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5665 | .release = kmem_cache_release, |
81819f0f CL |
5666 | }; |
5667 | ||
5668 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5669 | { | |
5670 | struct kobj_type *ktype = get_ktype(kobj); | |
5671 | ||
5672 | if (ktype == &slab_ktype) | |
5673 | return 1; | |
5674 | return 0; | |
5675 | } | |
5676 | ||
9cd43611 | 5677 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5678 | .filter = uevent_filter, |
5679 | }; | |
5680 | ||
27c3a314 | 5681 | static struct kset *slab_kset; |
81819f0f | 5682 | |
9a41707b VD |
5683 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5684 | { | |
127424c8 | 5685 | #ifdef CONFIG_MEMCG |
9a41707b | 5686 | if (!is_root_cache(s)) |
f7ce3190 | 5687 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5688 | #endif |
5689 | return slab_kset; | |
5690 | } | |
5691 | ||
81819f0f CL |
5692 | #define ID_STR_LENGTH 64 |
5693 | ||
5694 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5695 | * |
5696 | * Format :[flags-]size | |
81819f0f CL |
5697 | */ |
5698 | static char *create_unique_id(struct kmem_cache *s) | |
5699 | { | |
5700 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5701 | char *p = name; | |
5702 | ||
5703 | BUG_ON(!name); | |
5704 | ||
5705 | *p++ = ':'; | |
5706 | /* | |
5707 | * First flags affecting slabcache operations. We will only | |
5708 | * get here for aliasable slabs so we do not need to support | |
5709 | * too many flags. The flags here must cover all flags that | |
5710 | * are matched during merging to guarantee that the id is | |
5711 | * unique. | |
5712 | */ | |
5713 | if (s->flags & SLAB_CACHE_DMA) | |
5714 | *p++ = 'd'; | |
6d6ea1e9 NB |
5715 | if (s->flags & SLAB_CACHE_DMA32) |
5716 | *p++ = 'D'; | |
81819f0f CL |
5717 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
5718 | *p++ = 'a'; | |
becfda68 | 5719 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5720 | *p++ = 'F'; |
230e9fc2 VD |
5721 | if (s->flags & SLAB_ACCOUNT) |
5722 | *p++ = 'A'; | |
81819f0f CL |
5723 | if (p != name + 1) |
5724 | *p++ = '-'; | |
44065b2e | 5725 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5726 | |
81819f0f CL |
5727 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5728 | return name; | |
5729 | } | |
5730 | ||
3b7b3140 TH |
5731 | static void sysfs_slab_remove_workfn(struct work_struct *work) |
5732 | { | |
5733 | struct kmem_cache *s = | |
5734 | container_of(work, struct kmem_cache, kobj_remove_work); | |
5735 | ||
5736 | if (!s->kobj.state_in_sysfs) | |
5737 | /* | |
5738 | * For a memcg cache, this may be called during | |
5739 | * deactivation and again on shutdown. Remove only once. | |
5740 | * A cache is never shut down before deactivation is | |
5741 | * complete, so no need to worry about synchronization. | |
5742 | */ | |
f6ba4880 | 5743 | goto out; |
3b7b3140 TH |
5744 | |
5745 | #ifdef CONFIG_MEMCG | |
5746 | kset_unregister(s->memcg_kset); | |
5747 | #endif | |
5748 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
f6ba4880 | 5749 | out: |
3b7b3140 TH |
5750 | kobject_put(&s->kobj); |
5751 | } | |
5752 | ||
81819f0f CL |
5753 | static int sysfs_slab_add(struct kmem_cache *s) |
5754 | { | |
5755 | int err; | |
5756 | const char *name; | |
1663f26d | 5757 | struct kset *kset = cache_kset(s); |
45530c44 | 5758 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5759 | |
3b7b3140 TH |
5760 | INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn); |
5761 | ||
1663f26d TH |
5762 | if (!kset) { |
5763 | kobject_init(&s->kobj, &slab_ktype); | |
5764 | return 0; | |
5765 | } | |
5766 | ||
11066386 MC |
5767 | if (!unmergeable && disable_higher_order_debug && |
5768 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5769 | unmergeable = 1; | |
5770 | ||
81819f0f CL |
5771 | if (unmergeable) { |
5772 | /* | |
5773 | * Slabcache can never be merged so we can use the name proper. | |
5774 | * This is typically the case for debug situations. In that | |
5775 | * case we can catch duplicate names easily. | |
5776 | */ | |
27c3a314 | 5777 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5778 | name = s->name; |
5779 | } else { | |
5780 | /* | |
5781 | * Create a unique name for the slab as a target | |
5782 | * for the symlinks. | |
5783 | */ | |
5784 | name = create_unique_id(s); | |
5785 | } | |
5786 | ||
1663f26d | 5787 | s->kobj.kset = kset; |
26e4f205 | 5788 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5789 | if (err) |
80da026a | 5790 | goto out; |
81819f0f CL |
5791 | |
5792 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5793 | if (err) |
5794 | goto out_del_kobj; | |
9a41707b | 5795 | |
127424c8 | 5796 | #ifdef CONFIG_MEMCG |
1663f26d | 5797 | if (is_root_cache(s) && memcg_sysfs_enabled) { |
9a41707b VD |
5798 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); |
5799 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5800 | err = -ENOMEM; |
5801 | goto out_del_kobj; | |
9a41707b VD |
5802 | } |
5803 | } | |
5804 | #endif | |
5805 | ||
81819f0f CL |
5806 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5807 | if (!unmergeable) { | |
5808 | /* Setup first alias */ | |
5809 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5810 | } |
54b6a731 DJ |
5811 | out: |
5812 | if (!unmergeable) | |
5813 | kfree(name); | |
5814 | return err; | |
5815 | out_del_kobj: | |
5816 | kobject_del(&s->kobj); | |
54b6a731 | 5817 | goto out; |
81819f0f CL |
5818 | } |
5819 | ||
bf5eb3de | 5820 | static void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5821 | { |
97d06609 | 5822 | if (slab_state < FULL) |
2bce6485 CL |
5823 | /* |
5824 | * Sysfs has not been setup yet so no need to remove the | |
5825 | * cache from sysfs. | |
5826 | */ | |
5827 | return; | |
5828 | ||
3b7b3140 TH |
5829 | kobject_get(&s->kobj); |
5830 | schedule_work(&s->kobj_remove_work); | |
bf5eb3de TH |
5831 | } |
5832 | ||
d50d82fa MP |
5833 | void sysfs_slab_unlink(struct kmem_cache *s) |
5834 | { | |
5835 | if (slab_state >= FULL) | |
5836 | kobject_del(&s->kobj); | |
5837 | } | |
5838 | ||
bf5eb3de TH |
5839 | void sysfs_slab_release(struct kmem_cache *s) |
5840 | { | |
5841 | if (slab_state >= FULL) | |
5842 | kobject_put(&s->kobj); | |
81819f0f CL |
5843 | } |
5844 | ||
5845 | /* | |
5846 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5847 | * available lest we lose that information. |
81819f0f CL |
5848 | */ |
5849 | struct saved_alias { | |
5850 | struct kmem_cache *s; | |
5851 | const char *name; | |
5852 | struct saved_alias *next; | |
5853 | }; | |
5854 | ||
5af328a5 | 5855 | static struct saved_alias *alias_list; |
81819f0f CL |
5856 | |
5857 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5858 | { | |
5859 | struct saved_alias *al; | |
5860 | ||
97d06609 | 5861 | if (slab_state == FULL) { |
81819f0f CL |
5862 | /* |
5863 | * If we have a leftover link then remove it. | |
5864 | */ | |
27c3a314 GKH |
5865 | sysfs_remove_link(&slab_kset->kobj, name); |
5866 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5867 | } |
5868 | ||
5869 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5870 | if (!al) | |
5871 | return -ENOMEM; | |
5872 | ||
5873 | al->s = s; | |
5874 | al->name = name; | |
5875 | al->next = alias_list; | |
5876 | alias_list = al; | |
5877 | return 0; | |
5878 | } | |
5879 | ||
5880 | static int __init slab_sysfs_init(void) | |
5881 | { | |
5b95a4ac | 5882 | struct kmem_cache *s; |
81819f0f CL |
5883 | int err; |
5884 | ||
18004c5d | 5885 | mutex_lock(&slab_mutex); |
2bce6485 | 5886 | |
0ff21e46 | 5887 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5888 | if (!slab_kset) { |
18004c5d | 5889 | mutex_unlock(&slab_mutex); |
f9f58285 | 5890 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5891 | return -ENOSYS; |
5892 | } | |
5893 | ||
97d06609 | 5894 | slab_state = FULL; |
26a7bd03 | 5895 | |
5b95a4ac | 5896 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5897 | err = sysfs_slab_add(s); |
5d540fb7 | 5898 | if (err) |
f9f58285 FF |
5899 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5900 | s->name); | |
26a7bd03 | 5901 | } |
81819f0f CL |
5902 | |
5903 | while (alias_list) { | |
5904 | struct saved_alias *al = alias_list; | |
5905 | ||
5906 | alias_list = alias_list->next; | |
5907 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5908 | if (err) |
f9f58285 FF |
5909 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5910 | al->name); | |
81819f0f CL |
5911 | kfree(al); |
5912 | } | |
5913 | ||
18004c5d | 5914 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5915 | resiliency_test(); |
5916 | return 0; | |
5917 | } | |
5918 | ||
5919 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5920 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5921 | |
5922 | /* | |
5923 | * The /proc/slabinfo ABI | |
5924 | */ | |
5b365771 | 5925 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 5926 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5927 | { |
57ed3eda | 5928 | unsigned long nr_slabs = 0; |
205ab99d CL |
5929 | unsigned long nr_objs = 0; |
5930 | unsigned long nr_free = 0; | |
57ed3eda | 5931 | int node; |
fa45dc25 | 5932 | struct kmem_cache_node *n; |
57ed3eda | 5933 | |
fa45dc25 | 5934 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5935 | nr_slabs += node_nr_slabs(n); |
5936 | nr_objs += node_nr_objs(n); | |
205ab99d | 5937 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5938 | } |
5939 | ||
0d7561c6 GC |
5940 | sinfo->active_objs = nr_objs - nr_free; |
5941 | sinfo->num_objs = nr_objs; | |
5942 | sinfo->active_slabs = nr_slabs; | |
5943 | sinfo->num_slabs = nr_slabs; | |
5944 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5945 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5946 | } |
5947 | ||
0d7561c6 | 5948 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5949 | { |
7b3c3a50 AD |
5950 | } |
5951 | ||
b7454ad3 GC |
5952 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5953 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5954 | { |
b7454ad3 | 5955 | return -EIO; |
7b3c3a50 | 5956 | } |
5b365771 | 5957 | #endif /* CONFIG_SLUB_DEBUG */ |