]>
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 | ||
ca0cab65 VB |
117 | #ifdef CONFIG_SLUB_DEBUG |
118 | #ifdef CONFIG_SLUB_DEBUG_ON | |
119 | DEFINE_STATIC_KEY_TRUE(slub_debug_enabled); | |
120 | #else | |
121 | DEFINE_STATIC_KEY_FALSE(slub_debug_enabled); | |
122 | #endif | |
123 | #endif | |
124 | ||
59052e89 VB |
125 | static inline bool kmem_cache_debug(struct kmem_cache *s) |
126 | { | |
127 | return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS); | |
af537b0a | 128 | } |
5577bd8a | 129 | |
117d54df | 130 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be | 131 | { |
59052e89 | 132 | if (kmem_cache_debug_flags(s, SLAB_RED_ZONE)) |
d86bd1be JK |
133 | p += s->red_left_pad; |
134 | ||
135 | return p; | |
136 | } | |
137 | ||
345c905d JK |
138 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
139 | { | |
140 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
141 | return !kmem_cache_debug(s); | |
142 | #else | |
143 | return false; | |
144 | #endif | |
145 | } | |
146 | ||
81819f0f CL |
147 | /* |
148 | * Issues still to be resolved: | |
149 | * | |
81819f0f CL |
150 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
151 | * | |
81819f0f CL |
152 | * - Variable sizing of the per node arrays |
153 | */ | |
154 | ||
155 | /* Enable to test recovery from slab corruption on boot */ | |
156 | #undef SLUB_RESILIENCY_TEST | |
157 | ||
b789ef51 CL |
158 | /* Enable to log cmpxchg failures */ |
159 | #undef SLUB_DEBUG_CMPXCHG | |
160 | ||
2086d26a CL |
161 | /* |
162 | * Mininum number of partial slabs. These will be left on the partial | |
163 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
164 | */ | |
76be8950 | 165 | #define MIN_PARTIAL 5 |
e95eed57 | 166 | |
2086d26a CL |
167 | /* |
168 | * Maximum number of desirable partial slabs. | |
169 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 170 | * sort the partial list by the number of objects in use. |
2086d26a CL |
171 | */ |
172 | #define MAX_PARTIAL 10 | |
173 | ||
becfda68 | 174 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 175 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 176 | |
149daaf3 LA |
177 | /* |
178 | * These debug flags cannot use CMPXCHG because there might be consistency | |
179 | * issues when checking or reading debug information | |
180 | */ | |
181 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
182 | SLAB_TRACE) | |
183 | ||
184 | ||
fa5ec8a1 | 185 | /* |
3de47213 DR |
186 | * Debugging flags that require metadata to be stored in the slab. These get |
187 | * disabled when slub_debug=O is used and a cache's min order increases with | |
188 | * metadata. | |
fa5ec8a1 | 189 | */ |
3de47213 | 190 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 191 | |
210b5c06 CG |
192 | #define OO_SHIFT 16 |
193 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 194 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 195 | |
81819f0f | 196 | /* Internal SLUB flags */ |
d50112ed | 197 | /* Poison object */ |
4fd0b46e | 198 | #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U) |
d50112ed | 199 | /* Use cmpxchg_double */ |
4fd0b46e | 200 | #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U) |
81819f0f | 201 | |
02cbc874 CL |
202 | /* |
203 | * Tracking user of a slab. | |
204 | */ | |
d6543e39 | 205 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 206 | struct track { |
ce71e27c | 207 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
208 | #ifdef CONFIG_STACKTRACE |
209 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
210 | #endif | |
02cbc874 CL |
211 | int cpu; /* Was running on cpu */ |
212 | int pid; /* Pid context */ | |
213 | unsigned long when; /* When did the operation occur */ | |
214 | }; | |
215 | ||
216 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
217 | ||
ab4d5ed5 | 218 | #ifdef CONFIG_SYSFS |
81819f0f CL |
219 | static int sysfs_slab_add(struct kmem_cache *); |
220 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
81819f0f | 221 | #else |
0c710013 CL |
222 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
223 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
224 | { return 0; } | |
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 | 251 | /* |
aa1ef4d7 | 252 | * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged. |
d36a63a9 AK |
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 | { | |
aa1ef4d7 | 278 | object = kasan_reset_tag(object); |
2482ddec | 279 | return freelist_dereference(s, object + s->offset); |
7656c72b CL |
280 | } |
281 | ||
0ad9500e ED |
282 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
283 | { | |
0882ff91 | 284 | prefetch(object + s->offset); |
0ad9500e ED |
285 | } |
286 | ||
1393d9a1 CL |
287 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
288 | { | |
2482ddec | 289 | unsigned long freepointer_addr; |
1393d9a1 CL |
290 | void *p; |
291 | ||
8e57f8ac | 292 | if (!debug_pagealloc_enabled_static()) |
922d566c JK |
293 | return get_freepointer(s, object); |
294 | ||
2482ddec | 295 | freepointer_addr = (unsigned long)object + s->offset; |
fe557319 | 296 | copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p)); |
2482ddec | 297 | return freelist_ptr(s, p, freepointer_addr); |
1393d9a1 CL |
298 | } |
299 | ||
7656c72b CL |
300 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
301 | { | |
2482ddec KC |
302 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
303 | ||
ce6fa91b AP |
304 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
305 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
306 | #endif | |
307 | ||
aa1ef4d7 | 308 | freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr); |
2482ddec | 309 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); |
7656c72b CL |
310 | } |
311 | ||
312 | /* Loop over all objects in a slab */ | |
224a88be | 313 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
314 | for (__p = fixup_red_left(__s, __addr); \ |
315 | __p < (__addr) + (__objects) * (__s)->size; \ | |
316 | __p += (__s)->size) | |
7656c72b | 317 | |
9736d2a9 | 318 | static inline unsigned int order_objects(unsigned int order, unsigned int size) |
ab9a0f19 | 319 | { |
9736d2a9 | 320 | return ((unsigned int)PAGE_SIZE << order) / size; |
ab9a0f19 LJ |
321 | } |
322 | ||
19af27af | 323 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, |
9736d2a9 | 324 | unsigned int size) |
834f3d11 CL |
325 | { |
326 | struct kmem_cache_order_objects x = { | |
9736d2a9 | 327 | (order << OO_SHIFT) + order_objects(order, size) |
834f3d11 CL |
328 | }; |
329 | ||
330 | return x; | |
331 | } | |
332 | ||
19af27af | 333 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) |
834f3d11 | 334 | { |
210b5c06 | 335 | return x.x >> OO_SHIFT; |
834f3d11 CL |
336 | } |
337 | ||
19af27af | 338 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) |
834f3d11 | 339 | { |
210b5c06 | 340 | return x.x & OO_MASK; |
834f3d11 CL |
341 | } |
342 | ||
881db7fb CL |
343 | /* |
344 | * Per slab locking using the pagelock | |
345 | */ | |
346 | static __always_inline void slab_lock(struct page *page) | |
347 | { | |
48c935ad | 348 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
349 | bit_spin_lock(PG_locked, &page->flags); |
350 | } | |
351 | ||
352 | static __always_inline void slab_unlock(struct page *page) | |
353 | { | |
48c935ad | 354 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
355 | __bit_spin_unlock(PG_locked, &page->flags); |
356 | } | |
357 | ||
1d07171c CL |
358 | /* Interrupts must be disabled (for the fallback code to work right) */ |
359 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
360 | void *freelist_old, unsigned long counters_old, | |
361 | void *freelist_new, unsigned long counters_new, | |
362 | const char *n) | |
363 | { | |
364 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
365 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
366 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 367 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 368 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
369 | freelist_old, counters_old, |
370 | freelist_new, counters_new)) | |
6f6528a1 | 371 | return true; |
1d07171c CL |
372 | } else |
373 | #endif | |
374 | { | |
375 | slab_lock(page); | |
d0e0ac97 CG |
376 | if (page->freelist == freelist_old && |
377 | page->counters == counters_old) { | |
1d07171c | 378 | page->freelist = freelist_new; |
7d27a04b | 379 | page->counters = counters_new; |
1d07171c | 380 | slab_unlock(page); |
6f6528a1 | 381 | return true; |
1d07171c CL |
382 | } |
383 | slab_unlock(page); | |
384 | } | |
385 | ||
386 | cpu_relax(); | |
387 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
388 | ||
389 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 390 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
391 | #endif |
392 | ||
6f6528a1 | 393 | return false; |
1d07171c CL |
394 | } |
395 | ||
b789ef51 CL |
396 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
397 | void *freelist_old, unsigned long counters_old, | |
398 | void *freelist_new, unsigned long counters_new, | |
399 | const char *n) | |
400 | { | |
2565409f HC |
401 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
402 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 403 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 404 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
405 | freelist_old, counters_old, |
406 | freelist_new, counters_new)) | |
6f6528a1 | 407 | return true; |
b789ef51 CL |
408 | } else |
409 | #endif | |
410 | { | |
1d07171c CL |
411 | unsigned long flags; |
412 | ||
413 | local_irq_save(flags); | |
881db7fb | 414 | slab_lock(page); |
d0e0ac97 CG |
415 | if (page->freelist == freelist_old && |
416 | page->counters == counters_old) { | |
b789ef51 | 417 | page->freelist = freelist_new; |
7d27a04b | 418 | page->counters = counters_new; |
881db7fb | 419 | slab_unlock(page); |
1d07171c | 420 | local_irq_restore(flags); |
6f6528a1 | 421 | return true; |
b789ef51 | 422 | } |
881db7fb | 423 | slab_unlock(page); |
1d07171c | 424 | local_irq_restore(flags); |
b789ef51 CL |
425 | } |
426 | ||
427 | cpu_relax(); | |
428 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
429 | ||
430 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 431 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
432 | #endif |
433 | ||
6f6528a1 | 434 | return false; |
b789ef51 CL |
435 | } |
436 | ||
41ecc55b | 437 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 YZ |
438 | static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)]; |
439 | static DEFINE_SPINLOCK(object_map_lock); | |
440 | ||
5f80b13a CL |
441 | /* |
442 | * Determine a map of object in use on a page. | |
443 | * | |
881db7fb | 444 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
445 | * not vanish from under us. |
446 | */ | |
90e9f6a6 | 447 | static unsigned long *get_map(struct kmem_cache *s, struct page *page) |
31364c2e | 448 | __acquires(&object_map_lock) |
5f80b13a CL |
449 | { |
450 | void *p; | |
451 | void *addr = page_address(page); | |
452 | ||
90e9f6a6 YZ |
453 | VM_BUG_ON(!irqs_disabled()); |
454 | ||
455 | spin_lock(&object_map_lock); | |
456 | ||
457 | bitmap_zero(object_map, page->objects); | |
458 | ||
5f80b13a | 459 | for (p = page->freelist; p; p = get_freepointer(s, p)) |
4138fdfc | 460 | set_bit(__obj_to_index(s, addr, p), object_map); |
90e9f6a6 YZ |
461 | |
462 | return object_map; | |
463 | } | |
464 | ||
81aba9e0 | 465 | static void put_map(unsigned long *map) __releases(&object_map_lock) |
90e9f6a6 YZ |
466 | { |
467 | VM_BUG_ON(map != object_map); | |
90e9f6a6 | 468 | spin_unlock(&object_map_lock); |
5f80b13a CL |
469 | } |
470 | ||
870b1fbb | 471 | static inline unsigned int size_from_object(struct kmem_cache *s) |
d86bd1be JK |
472 | { |
473 | if (s->flags & SLAB_RED_ZONE) | |
474 | return s->size - s->red_left_pad; | |
475 | ||
476 | return s->size; | |
477 | } | |
478 | ||
479 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
480 | { | |
481 | if (s->flags & SLAB_RED_ZONE) | |
482 | p -= s->red_left_pad; | |
483 | ||
484 | return p; | |
485 | } | |
486 | ||
41ecc55b CL |
487 | /* |
488 | * Debug settings: | |
489 | */ | |
89d3c87e | 490 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 491 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 492 | #else |
d50112ed | 493 | static slab_flags_t slub_debug; |
f0630fff | 494 | #endif |
41ecc55b | 495 | |
e17f1dfb | 496 | static char *slub_debug_string; |
fa5ec8a1 | 497 | static int disable_higher_order_debug; |
41ecc55b | 498 | |
a79316c6 AR |
499 | /* |
500 | * slub is about to manipulate internal object metadata. This memory lies | |
501 | * outside the range of the allocated object, so accessing it would normally | |
502 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
503 | * to tell kasan that these accesses are OK. | |
504 | */ | |
505 | static inline void metadata_access_enable(void) | |
506 | { | |
507 | kasan_disable_current(); | |
508 | } | |
509 | ||
510 | static inline void metadata_access_disable(void) | |
511 | { | |
512 | kasan_enable_current(); | |
513 | } | |
514 | ||
81819f0f CL |
515 | /* |
516 | * Object debugging | |
517 | */ | |
d86bd1be JK |
518 | |
519 | /* Verify that a pointer has an address that is valid within a slab page */ | |
520 | static inline int check_valid_pointer(struct kmem_cache *s, | |
521 | struct page *page, void *object) | |
522 | { | |
523 | void *base; | |
524 | ||
525 | if (!object) | |
526 | return 1; | |
527 | ||
528 | base = page_address(page); | |
338cfaad | 529 | object = kasan_reset_tag(object); |
d86bd1be JK |
530 | object = restore_red_left(s, object); |
531 | if (object < base || object >= base + page->objects * s->size || | |
532 | (object - base) % s->size) { | |
533 | return 0; | |
534 | } | |
535 | ||
536 | return 1; | |
537 | } | |
538 | ||
aa2efd5e DT |
539 | static void print_section(char *level, char *text, u8 *addr, |
540 | unsigned int length) | |
81819f0f | 541 | { |
a79316c6 | 542 | metadata_access_enable(); |
aa1ef4d7 AK |
543 | print_hex_dump(level, kasan_reset_tag(text), DUMP_PREFIX_ADDRESS, |
544 | 16, 1, addr, length, 1); | |
a79316c6 | 545 | metadata_access_disable(); |
81819f0f CL |
546 | } |
547 | ||
cbfc35a4 WL |
548 | /* |
549 | * See comment in calculate_sizes(). | |
550 | */ | |
551 | static inline bool freeptr_outside_object(struct kmem_cache *s) | |
552 | { | |
553 | return s->offset >= s->inuse; | |
554 | } | |
555 | ||
556 | /* | |
557 | * Return offset of the end of info block which is inuse + free pointer if | |
558 | * not overlapping with object. | |
559 | */ | |
560 | static inline unsigned int get_info_end(struct kmem_cache *s) | |
561 | { | |
562 | if (freeptr_outside_object(s)) | |
563 | return s->inuse + sizeof(void *); | |
564 | else | |
565 | return s->inuse; | |
566 | } | |
567 | ||
81819f0f CL |
568 | static struct track *get_track(struct kmem_cache *s, void *object, |
569 | enum track_item alloc) | |
570 | { | |
571 | struct track *p; | |
572 | ||
cbfc35a4 | 573 | p = object + get_info_end(s); |
81819f0f | 574 | |
aa1ef4d7 | 575 | return kasan_reset_tag(p + alloc); |
81819f0f CL |
576 | } |
577 | ||
578 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 579 | enum track_item alloc, unsigned long addr) |
81819f0f | 580 | { |
1a00df4a | 581 | struct track *p = get_track(s, object, alloc); |
81819f0f | 582 | |
81819f0f | 583 | if (addr) { |
d6543e39 | 584 | #ifdef CONFIG_STACKTRACE |
79716799 | 585 | unsigned int nr_entries; |
d6543e39 | 586 | |
a79316c6 | 587 | metadata_access_enable(); |
aa1ef4d7 AK |
588 | nr_entries = stack_trace_save(kasan_reset_tag(p->addrs), |
589 | TRACK_ADDRS_COUNT, 3); | |
a79316c6 | 590 | metadata_access_disable(); |
d6543e39 | 591 | |
79716799 TG |
592 | if (nr_entries < TRACK_ADDRS_COUNT) |
593 | p->addrs[nr_entries] = 0; | |
d6543e39 | 594 | #endif |
81819f0f CL |
595 | p->addr = addr; |
596 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 597 | p->pid = current->pid; |
81819f0f | 598 | p->when = jiffies; |
b8ca7ff7 | 599 | } else { |
81819f0f | 600 | memset(p, 0, sizeof(struct track)); |
b8ca7ff7 | 601 | } |
81819f0f CL |
602 | } |
603 | ||
81819f0f CL |
604 | static void init_tracking(struct kmem_cache *s, void *object) |
605 | { | |
24922684 CL |
606 | if (!(s->flags & SLAB_STORE_USER)) |
607 | return; | |
608 | ||
ce71e27c EGM |
609 | set_track(s, object, TRACK_FREE, 0UL); |
610 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
611 | } |
612 | ||
86609d33 | 613 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f CL |
614 | { |
615 | if (!t->addr) | |
616 | return; | |
617 | ||
f9f58285 | 618 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 619 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
d6543e39 BG |
620 | #ifdef CONFIG_STACKTRACE |
621 | { | |
622 | int i; | |
623 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
624 | if (t->addrs[i]) | |
f9f58285 | 625 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
626 | else |
627 | break; | |
628 | } | |
629 | #endif | |
24922684 CL |
630 | } |
631 | ||
e42f174e | 632 | void print_tracking(struct kmem_cache *s, void *object) |
24922684 | 633 | { |
86609d33 | 634 | unsigned long pr_time = jiffies; |
24922684 CL |
635 | if (!(s->flags & SLAB_STORE_USER)) |
636 | return; | |
637 | ||
86609d33 CP |
638 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
639 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
640 | } |
641 | ||
642 | static void print_page_info(struct page *page) | |
643 | { | |
f9f58285 | 644 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 645 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
646 | |
647 | } | |
648 | ||
649 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
650 | { | |
ecc42fbe | 651 | struct va_format vaf; |
24922684 | 652 | va_list args; |
24922684 CL |
653 | |
654 | va_start(args, fmt); | |
ecc42fbe FF |
655 | vaf.fmt = fmt; |
656 | vaf.va = &args; | |
f9f58285 | 657 | pr_err("=============================================================================\n"); |
ecc42fbe | 658 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 659 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 660 | |
373d4d09 | 661 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 662 | va_end(args); |
81819f0f CL |
663 | } |
664 | ||
24922684 CL |
665 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
666 | { | |
ecc42fbe | 667 | struct va_format vaf; |
24922684 | 668 | va_list args; |
24922684 CL |
669 | |
670 | va_start(args, fmt); | |
ecc42fbe FF |
671 | vaf.fmt = fmt; |
672 | vaf.va = &args; | |
673 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 674 | va_end(args); |
24922684 CL |
675 | } |
676 | ||
52f23478 | 677 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 678 | void **freelist, void *nextfree) |
52f23478 DZ |
679 | { |
680 | if ((s->flags & SLAB_CONSISTENCY_CHECKS) && | |
dc07a728 ER |
681 | !check_valid_pointer(s, page, nextfree) && freelist) { |
682 | object_err(s, page, *freelist, "Freechain corrupt"); | |
683 | *freelist = NULL; | |
52f23478 DZ |
684 | slab_fix(s, "Isolate corrupted freechain"); |
685 | return true; | |
686 | } | |
687 | ||
688 | return false; | |
689 | } | |
690 | ||
24922684 | 691 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) |
81819f0f CL |
692 | { |
693 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 694 | u8 *addr = page_address(page); |
24922684 CL |
695 | |
696 | print_tracking(s, p); | |
697 | ||
698 | print_page_info(page); | |
699 | ||
f9f58285 FF |
700 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
701 | p, p - addr, get_freepointer(s, p)); | |
24922684 | 702 | |
d86bd1be | 703 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e DT |
704 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
705 | s->red_left_pad); | |
d86bd1be | 706 | else if (p > addr + 16) |
aa2efd5e | 707 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 708 | |
aa2efd5e | 709 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 710 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 711 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e | 712 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 713 | s->inuse - s->object_size); |
81819f0f | 714 | |
cbfc35a4 | 715 | off = get_info_end(s); |
81819f0f | 716 | |
24922684 | 717 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 718 | off += 2 * sizeof(struct track); |
81819f0f | 719 | |
80a9201a AP |
720 | off += kasan_metadata_size(s); |
721 | ||
d86bd1be | 722 | if (off != size_from_object(s)) |
81819f0f | 723 | /* Beginning of the filler is the free pointer */ |
aa2efd5e DT |
724 | print_section(KERN_ERR, "Padding ", p + off, |
725 | size_from_object(s) - off); | |
24922684 CL |
726 | |
727 | dump_stack(); | |
81819f0f CL |
728 | } |
729 | ||
75c66def | 730 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
731 | u8 *object, char *reason) |
732 | { | |
3dc50637 | 733 | slab_bug(s, "%s", reason); |
24922684 | 734 | print_trailer(s, page, object); |
81819f0f CL |
735 | } |
736 | ||
a38965bf | 737 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page, |
d0e0ac97 | 738 | const char *fmt, ...) |
81819f0f CL |
739 | { |
740 | va_list args; | |
741 | char buf[100]; | |
742 | ||
24922684 CL |
743 | va_start(args, fmt); |
744 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 745 | va_end(args); |
3dc50637 | 746 | slab_bug(s, "%s", buf); |
24922684 | 747 | print_page_info(page); |
81819f0f CL |
748 | dump_stack(); |
749 | } | |
750 | ||
f7cb1933 | 751 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f | 752 | { |
aa1ef4d7 | 753 | u8 *p = kasan_reset_tag(object); |
81819f0f | 754 | |
d86bd1be JK |
755 | if (s->flags & SLAB_RED_ZONE) |
756 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
757 | ||
81819f0f | 758 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
759 | memset(p, POISON_FREE, s->object_size - 1); |
760 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
761 | } |
762 | ||
763 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 764 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
765 | } |
766 | ||
24922684 CL |
767 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
768 | void *from, void *to) | |
769 | { | |
770 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
771 | memset(from, data, to - from); | |
772 | } | |
773 | ||
774 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
775 | u8 *object, char *what, | |
06428780 | 776 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
777 | { |
778 | u8 *fault; | |
779 | u8 *end; | |
e1b70dd1 | 780 | u8 *addr = page_address(page); |
24922684 | 781 | |
a79316c6 | 782 | metadata_access_enable(); |
aa1ef4d7 | 783 | fault = memchr_inv(kasan_reset_tag(start), value, bytes); |
a79316c6 | 784 | metadata_access_disable(); |
24922684 CL |
785 | if (!fault) |
786 | return 1; | |
787 | ||
788 | end = start + bytes; | |
789 | while (end > fault && end[-1] == value) | |
790 | end--; | |
791 | ||
792 | slab_bug(s, "%s overwritten", what); | |
e1b70dd1 MC |
793 | pr_err("INFO: 0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n", |
794 | fault, end - 1, fault - addr, | |
795 | fault[0], value); | |
24922684 CL |
796 | print_trailer(s, page, object); |
797 | ||
798 | restore_bytes(s, what, value, fault, end); | |
799 | return 0; | |
81819f0f CL |
800 | } |
801 | ||
81819f0f CL |
802 | /* |
803 | * Object layout: | |
804 | * | |
805 | * object address | |
806 | * Bytes of the object to be managed. | |
807 | * If the freepointer may overlay the object then the free | |
cbfc35a4 | 808 | * pointer is at the middle of the object. |
672bba3a | 809 | * |
81819f0f CL |
810 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
811 | * 0xa5 (POISON_END) | |
812 | * | |
3b0efdfa | 813 | * object + s->object_size |
81819f0f | 814 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 815 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 816 | * object_size == inuse. |
672bba3a | 817 | * |
81819f0f CL |
818 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
819 | * 0xcc (RED_ACTIVE) for objects in use. | |
820 | * | |
821 | * object + s->inuse | |
672bba3a CL |
822 | * Meta data starts here. |
823 | * | |
81819f0f CL |
824 | * A. Free pointer (if we cannot overwrite object on free) |
825 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 826 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 827 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
828 | * before the word boundary. |
829 | * | |
830 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
831 | * |
832 | * object + s->size | |
672bba3a | 833 | * Nothing is used beyond s->size. |
81819f0f | 834 | * |
3b0efdfa | 835 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 836 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
837 | * may be used with merged slabcaches. |
838 | */ | |
839 | ||
81819f0f CL |
840 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
841 | { | |
cbfc35a4 | 842 | unsigned long off = get_info_end(s); /* The end of info */ |
81819f0f CL |
843 | |
844 | if (s->flags & SLAB_STORE_USER) | |
845 | /* We also have user information there */ | |
846 | off += 2 * sizeof(struct track); | |
847 | ||
80a9201a AP |
848 | off += kasan_metadata_size(s); |
849 | ||
d86bd1be | 850 | if (size_from_object(s) == off) |
81819f0f CL |
851 | return 1; |
852 | ||
24922684 | 853 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 854 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
855 | } |
856 | ||
39b26464 | 857 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
858 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
859 | { | |
24922684 CL |
860 | u8 *start; |
861 | u8 *fault; | |
862 | u8 *end; | |
5d682681 | 863 | u8 *pad; |
24922684 CL |
864 | int length; |
865 | int remainder; | |
81819f0f CL |
866 | |
867 | if (!(s->flags & SLAB_POISON)) | |
868 | return 1; | |
869 | ||
a973e9dd | 870 | start = page_address(page); |
a50b854e | 871 | length = page_size(page); |
39b26464 CL |
872 | end = start + length; |
873 | remainder = length % s->size; | |
81819f0f CL |
874 | if (!remainder) |
875 | return 1; | |
876 | ||
5d682681 | 877 | pad = end - remainder; |
a79316c6 | 878 | metadata_access_enable(); |
aa1ef4d7 | 879 | fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder); |
a79316c6 | 880 | metadata_access_disable(); |
24922684 CL |
881 | if (!fault) |
882 | return 1; | |
883 | while (end > fault && end[-1] == POISON_INUSE) | |
884 | end--; | |
885 | ||
e1b70dd1 MC |
886 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p @offset=%tu", |
887 | fault, end - 1, fault - start); | |
5d682681 | 888 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 889 | |
5d682681 | 890 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
24922684 | 891 | return 0; |
81819f0f CL |
892 | } |
893 | ||
894 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 895 | void *object, u8 val) |
81819f0f CL |
896 | { |
897 | u8 *p = object; | |
3b0efdfa | 898 | u8 *endobject = object + s->object_size; |
81819f0f CL |
899 | |
900 | if (s->flags & SLAB_RED_ZONE) { | |
d86bd1be JK |
901 | if (!check_bytes_and_report(s, page, object, "Redzone", |
902 | object - s->red_left_pad, val, s->red_left_pad)) | |
903 | return 0; | |
904 | ||
24922684 | 905 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 906 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 907 | return 0; |
81819f0f | 908 | } else { |
3b0efdfa | 909 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 910 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
911 | endobject, POISON_INUSE, |
912 | s->inuse - s->object_size); | |
3adbefee | 913 | } |
81819f0f CL |
914 | } |
915 | ||
916 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 917 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 918 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 919 | POISON_FREE, s->object_size - 1) || |
24922684 | 920 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 921 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 922 | return 0; |
81819f0f CL |
923 | /* |
924 | * check_pad_bytes cleans up on its own. | |
925 | */ | |
926 | check_pad_bytes(s, page, p); | |
927 | } | |
928 | ||
cbfc35a4 | 929 | if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE) |
81819f0f CL |
930 | /* |
931 | * Object and freepointer overlap. Cannot check | |
932 | * freepointer while object is allocated. | |
933 | */ | |
934 | return 1; | |
935 | ||
936 | /* Check free pointer validity */ | |
937 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
938 | object_err(s, page, p, "Freepointer corrupt"); | |
939 | /* | |
9f6c708e | 940 | * No choice but to zap it and thus lose the remainder |
81819f0f | 941 | * of the free objects in this slab. May cause |
672bba3a | 942 | * another error because the object count is now wrong. |
81819f0f | 943 | */ |
a973e9dd | 944 | set_freepointer(s, p, NULL); |
81819f0f CL |
945 | return 0; |
946 | } | |
947 | return 1; | |
948 | } | |
949 | ||
950 | static int check_slab(struct kmem_cache *s, struct page *page) | |
951 | { | |
39b26464 CL |
952 | int maxobj; |
953 | ||
81819f0f CL |
954 | VM_BUG_ON(!irqs_disabled()); |
955 | ||
956 | if (!PageSlab(page)) { | |
24922684 | 957 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
958 | return 0; |
959 | } | |
39b26464 | 960 | |
9736d2a9 | 961 | maxobj = order_objects(compound_order(page), s->size); |
39b26464 CL |
962 | if (page->objects > maxobj) { |
963 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 964 | page->objects, maxobj); |
39b26464 CL |
965 | return 0; |
966 | } | |
967 | if (page->inuse > page->objects) { | |
24922684 | 968 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 969 | page->inuse, page->objects); |
81819f0f CL |
970 | return 0; |
971 | } | |
972 | /* Slab_pad_check fixes things up after itself */ | |
973 | slab_pad_check(s, page); | |
974 | return 1; | |
975 | } | |
976 | ||
977 | /* | |
672bba3a CL |
978 | * Determine if a certain object on a page is on the freelist. Must hold the |
979 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
980 | */ |
981 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
982 | { | |
983 | int nr = 0; | |
881db7fb | 984 | void *fp; |
81819f0f | 985 | void *object = NULL; |
f6edde9c | 986 | int max_objects; |
81819f0f | 987 | |
881db7fb | 988 | fp = page->freelist; |
39b26464 | 989 | while (fp && nr <= page->objects) { |
81819f0f CL |
990 | if (fp == search) |
991 | return 1; | |
992 | if (!check_valid_pointer(s, page, fp)) { | |
993 | if (object) { | |
994 | object_err(s, page, object, | |
995 | "Freechain corrupt"); | |
a973e9dd | 996 | set_freepointer(s, object, NULL); |
81819f0f | 997 | } else { |
24922684 | 998 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 999 | page->freelist = NULL; |
39b26464 | 1000 | page->inuse = page->objects; |
24922684 | 1001 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
1002 | return 0; |
1003 | } | |
1004 | break; | |
1005 | } | |
1006 | object = fp; | |
1007 | fp = get_freepointer(s, object); | |
1008 | nr++; | |
1009 | } | |
1010 | ||
9736d2a9 | 1011 | max_objects = order_objects(compound_order(page), s->size); |
210b5c06 CG |
1012 | if (max_objects > MAX_OBJS_PER_PAGE) |
1013 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
1014 | |
1015 | if (page->objects != max_objects) { | |
756a025f JP |
1016 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
1017 | page->objects, max_objects); | |
224a88be CL |
1018 | page->objects = max_objects; |
1019 | slab_fix(s, "Number of objects adjusted."); | |
1020 | } | |
39b26464 | 1021 | if (page->inuse != page->objects - nr) { |
756a025f JP |
1022 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
1023 | page->inuse, page->objects - nr); | |
39b26464 | 1024 | page->inuse = page->objects - nr; |
24922684 | 1025 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
1026 | } |
1027 | return search == NULL; | |
1028 | } | |
1029 | ||
0121c619 CL |
1030 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
1031 | int alloc) | |
3ec09742 CL |
1032 | { |
1033 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 1034 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
1035 | s->name, |
1036 | alloc ? "alloc" : "free", | |
1037 | object, page->inuse, | |
1038 | page->freelist); | |
1039 | ||
1040 | if (!alloc) | |
aa2efd5e | 1041 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1042 | s->object_size); |
3ec09742 CL |
1043 | |
1044 | dump_stack(); | |
1045 | } | |
1046 | } | |
1047 | ||
643b1138 | 1048 | /* |
672bba3a | 1049 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1050 | */ |
5cc6eee8 CL |
1051 | static void add_full(struct kmem_cache *s, |
1052 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1053 | { |
5cc6eee8 CL |
1054 | if (!(s->flags & SLAB_STORE_USER)) |
1055 | return; | |
1056 | ||
255d0884 | 1057 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1058 | list_add(&page->slab_list, &n->full); |
643b1138 CL |
1059 | } |
1060 | ||
c65c1877 | 1061 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1062 | { |
643b1138 CL |
1063 | if (!(s->flags & SLAB_STORE_USER)) |
1064 | return; | |
1065 | ||
255d0884 | 1066 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1067 | list_del(&page->slab_list); |
643b1138 CL |
1068 | } |
1069 | ||
0f389ec6 CL |
1070 | /* Tracking of the number of slabs for debugging purposes */ |
1071 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1072 | { | |
1073 | struct kmem_cache_node *n = get_node(s, node); | |
1074 | ||
1075 | return atomic_long_read(&n->nr_slabs); | |
1076 | } | |
1077 | ||
26c02cf0 AB |
1078 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1079 | { | |
1080 | return atomic_long_read(&n->nr_slabs); | |
1081 | } | |
1082 | ||
205ab99d | 1083 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1084 | { |
1085 | struct kmem_cache_node *n = get_node(s, node); | |
1086 | ||
1087 | /* | |
1088 | * May be called early in order to allocate a slab for the | |
1089 | * kmem_cache_node structure. Solve the chicken-egg | |
1090 | * dilemma by deferring the increment of the count during | |
1091 | * bootstrap (see early_kmem_cache_node_alloc). | |
1092 | */ | |
338b2642 | 1093 | if (likely(n)) { |
0f389ec6 | 1094 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1095 | atomic_long_add(objects, &n->total_objects); |
1096 | } | |
0f389ec6 | 1097 | } |
205ab99d | 1098 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1099 | { |
1100 | struct kmem_cache_node *n = get_node(s, node); | |
1101 | ||
1102 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1103 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1104 | } |
1105 | ||
1106 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1107 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1108 | void *object) | |
1109 | { | |
8fc8d666 | 1110 | if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)) |
3ec09742 CL |
1111 | return; |
1112 | ||
f7cb1933 | 1113 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1114 | init_tracking(s, object); |
1115 | } | |
1116 | ||
a50b854e MWO |
1117 | static |
1118 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) | |
a7101224 | 1119 | { |
8fc8d666 | 1120 | if (!kmem_cache_debug_flags(s, SLAB_POISON)) |
a7101224 AK |
1121 | return; |
1122 | ||
1123 | metadata_access_enable(); | |
aa1ef4d7 | 1124 | memset(kasan_reset_tag(addr), POISON_INUSE, page_size(page)); |
a7101224 AK |
1125 | metadata_access_disable(); |
1126 | } | |
1127 | ||
becfda68 | 1128 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
278d7756 | 1129 | struct page *page, void *object) |
81819f0f CL |
1130 | { |
1131 | if (!check_slab(s, page)) | |
becfda68 | 1132 | return 0; |
81819f0f | 1133 | |
81819f0f CL |
1134 | if (!check_valid_pointer(s, page, object)) { |
1135 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1136 | return 0; |
81819f0f CL |
1137 | } |
1138 | ||
f7cb1933 | 1139 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1140 | return 0; |
1141 | ||
1142 | return 1; | |
1143 | } | |
1144 | ||
1145 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1146 | struct page *page, | |
1147 | void *object, unsigned long addr) | |
1148 | { | |
1149 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
278d7756 | 1150 | if (!alloc_consistency_checks(s, page, object)) |
becfda68 LA |
1151 | goto bad; |
1152 | } | |
81819f0f | 1153 | |
3ec09742 CL |
1154 | /* Success perform special debug activities for allocs */ |
1155 | if (s->flags & SLAB_STORE_USER) | |
1156 | set_track(s, object, TRACK_ALLOC, addr); | |
1157 | trace(s, page, object, 1); | |
f7cb1933 | 1158 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1159 | return 1; |
3ec09742 | 1160 | |
81819f0f CL |
1161 | bad: |
1162 | if (PageSlab(page)) { | |
1163 | /* | |
1164 | * If this is a slab page then lets do the best we can | |
1165 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1166 | * as used avoids touching the remaining objects. |
81819f0f | 1167 | */ |
24922684 | 1168 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1169 | page->inuse = page->objects; |
a973e9dd | 1170 | page->freelist = NULL; |
81819f0f CL |
1171 | } |
1172 | return 0; | |
1173 | } | |
1174 | ||
becfda68 LA |
1175 | static inline int free_consistency_checks(struct kmem_cache *s, |
1176 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1177 | { |
81819f0f | 1178 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1179 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1180 | return 0; |
81819f0f CL |
1181 | } |
1182 | ||
1183 | if (on_freelist(s, page, object)) { | |
24922684 | 1184 | object_err(s, page, object, "Object already free"); |
becfda68 | 1185 | return 0; |
81819f0f CL |
1186 | } |
1187 | ||
f7cb1933 | 1188 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1189 | return 0; |
81819f0f | 1190 | |
1b4f59e3 | 1191 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1192 | if (!PageSlab(page)) { |
756a025f JP |
1193 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1194 | object); | |
1b4f59e3 | 1195 | } else if (!page->slab_cache) { |
f9f58285 FF |
1196 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1197 | object); | |
70d71228 | 1198 | dump_stack(); |
06428780 | 1199 | } else |
24922684 CL |
1200 | object_err(s, page, object, |
1201 | "page slab pointer corrupt."); | |
becfda68 LA |
1202 | return 0; |
1203 | } | |
1204 | return 1; | |
1205 | } | |
1206 | ||
1207 | /* Supports checking bulk free of a constructed freelist */ | |
1208 | static noinline int free_debug_processing( | |
1209 | struct kmem_cache *s, struct page *page, | |
1210 | void *head, void *tail, int bulk_cnt, | |
1211 | unsigned long addr) | |
1212 | { | |
1213 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1214 | void *object = head; | |
1215 | int cnt = 0; | |
3f649ab7 | 1216 | unsigned long flags; |
becfda68 LA |
1217 | int ret = 0; |
1218 | ||
1219 | spin_lock_irqsave(&n->list_lock, flags); | |
1220 | slab_lock(page); | |
1221 | ||
1222 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1223 | if (!check_slab(s, page)) | |
1224 | goto out; | |
1225 | } | |
1226 | ||
1227 | next_object: | |
1228 | cnt++; | |
1229 | ||
1230 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1231 | if (!free_consistency_checks(s, page, object, addr)) | |
1232 | goto out; | |
81819f0f | 1233 | } |
3ec09742 | 1234 | |
3ec09742 CL |
1235 | if (s->flags & SLAB_STORE_USER) |
1236 | set_track(s, object, TRACK_FREE, addr); | |
1237 | trace(s, page, object, 0); | |
81084651 | 1238 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1239 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1240 | |
1241 | /* Reached end of constructed freelist yet? */ | |
1242 | if (object != tail) { | |
1243 | object = get_freepointer(s, object); | |
1244 | goto next_object; | |
1245 | } | |
804aa132 LA |
1246 | ret = 1; |
1247 | ||
5c2e4bbb | 1248 | out: |
81084651 JDB |
1249 | if (cnt != bulk_cnt) |
1250 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1251 | bulk_cnt, cnt); | |
1252 | ||
881db7fb | 1253 | slab_unlock(page); |
282acb43 | 1254 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1255 | if (!ret) |
1256 | slab_fix(s, "Object at 0x%p not freed", object); | |
1257 | return ret; | |
81819f0f CL |
1258 | } |
1259 | ||
e17f1dfb VB |
1260 | /* |
1261 | * Parse a block of slub_debug options. Blocks are delimited by ';' | |
1262 | * | |
1263 | * @str: start of block | |
1264 | * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified | |
1265 | * @slabs: return start of list of slabs, or NULL when there's no list | |
1266 | * @init: assume this is initial parsing and not per-kmem-create parsing | |
1267 | * | |
1268 | * returns the start of next block if there's any, or NULL | |
1269 | */ | |
1270 | static char * | |
1271 | parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init) | |
41ecc55b | 1272 | { |
e17f1dfb | 1273 | bool higher_order_disable = false; |
f0630fff | 1274 | |
e17f1dfb VB |
1275 | /* Skip any completely empty blocks */ |
1276 | while (*str && *str == ';') | |
1277 | str++; | |
1278 | ||
1279 | if (*str == ',') { | |
f0630fff CL |
1280 | /* |
1281 | * No options but restriction on slabs. This means full | |
1282 | * debugging for slabs matching a pattern. | |
1283 | */ | |
e17f1dfb | 1284 | *flags = DEBUG_DEFAULT_FLAGS; |
f0630fff | 1285 | goto check_slabs; |
e17f1dfb VB |
1286 | } |
1287 | *flags = 0; | |
f0630fff | 1288 | |
e17f1dfb VB |
1289 | /* Determine which debug features should be switched on */ |
1290 | for (; *str && *str != ',' && *str != ';'; str++) { | |
f0630fff | 1291 | switch (tolower(*str)) { |
e17f1dfb VB |
1292 | case '-': |
1293 | *flags = 0; | |
1294 | break; | |
f0630fff | 1295 | case 'f': |
e17f1dfb | 1296 | *flags |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1297 | break; |
1298 | case 'z': | |
e17f1dfb | 1299 | *flags |= SLAB_RED_ZONE; |
f0630fff CL |
1300 | break; |
1301 | case 'p': | |
e17f1dfb | 1302 | *flags |= SLAB_POISON; |
f0630fff CL |
1303 | break; |
1304 | case 'u': | |
e17f1dfb | 1305 | *flags |= SLAB_STORE_USER; |
f0630fff CL |
1306 | break; |
1307 | case 't': | |
e17f1dfb | 1308 | *flags |= SLAB_TRACE; |
f0630fff | 1309 | break; |
4c13dd3b | 1310 | case 'a': |
e17f1dfb | 1311 | *flags |= SLAB_FAILSLAB; |
4c13dd3b | 1312 | break; |
08303a73 CA |
1313 | case 'o': |
1314 | /* | |
1315 | * Avoid enabling debugging on caches if its minimum | |
1316 | * order would increase as a result. | |
1317 | */ | |
e17f1dfb | 1318 | higher_order_disable = true; |
08303a73 | 1319 | break; |
f0630fff | 1320 | default: |
e17f1dfb VB |
1321 | if (init) |
1322 | pr_err("slub_debug option '%c' unknown. skipped\n", *str); | |
f0630fff | 1323 | } |
41ecc55b | 1324 | } |
f0630fff | 1325 | check_slabs: |
41ecc55b | 1326 | if (*str == ',') |
e17f1dfb VB |
1327 | *slabs = ++str; |
1328 | else | |
1329 | *slabs = NULL; | |
1330 | ||
1331 | /* Skip over the slab list */ | |
1332 | while (*str && *str != ';') | |
1333 | str++; | |
1334 | ||
1335 | /* Skip any completely empty blocks */ | |
1336 | while (*str && *str == ';') | |
1337 | str++; | |
1338 | ||
1339 | if (init && higher_order_disable) | |
1340 | disable_higher_order_debug = 1; | |
1341 | ||
1342 | if (*str) | |
1343 | return str; | |
1344 | else | |
1345 | return NULL; | |
1346 | } | |
1347 | ||
1348 | static int __init setup_slub_debug(char *str) | |
1349 | { | |
1350 | slab_flags_t flags; | |
1351 | char *saved_str; | |
1352 | char *slab_list; | |
1353 | bool global_slub_debug_changed = false; | |
1354 | bool slab_list_specified = false; | |
1355 | ||
1356 | slub_debug = DEBUG_DEFAULT_FLAGS; | |
1357 | if (*str++ != '=' || !*str) | |
1358 | /* | |
1359 | * No options specified. Switch on full debugging. | |
1360 | */ | |
1361 | goto out; | |
1362 | ||
1363 | saved_str = str; | |
1364 | while (str) { | |
1365 | str = parse_slub_debug_flags(str, &flags, &slab_list, true); | |
1366 | ||
1367 | if (!slab_list) { | |
1368 | slub_debug = flags; | |
1369 | global_slub_debug_changed = true; | |
1370 | } else { | |
1371 | slab_list_specified = true; | |
1372 | } | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * For backwards compatibility, a single list of flags with list of | |
1377 | * slabs means debugging is only enabled for those slabs, so the global | |
1378 | * slub_debug should be 0. We can extended that to multiple lists as | |
1379 | * long as there is no option specifying flags without a slab list. | |
1380 | */ | |
1381 | if (slab_list_specified) { | |
1382 | if (!global_slub_debug_changed) | |
1383 | slub_debug = 0; | |
1384 | slub_debug_string = saved_str; | |
1385 | } | |
f0630fff | 1386 | out: |
ca0cab65 VB |
1387 | if (slub_debug != 0 || slub_debug_string) |
1388 | static_branch_enable(&slub_debug_enabled); | |
6471384a AP |
1389 | if ((static_branch_unlikely(&init_on_alloc) || |
1390 | static_branch_unlikely(&init_on_free)) && | |
1391 | (slub_debug & SLAB_POISON)) | |
1392 | pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n"); | |
41ecc55b CL |
1393 | return 1; |
1394 | } | |
1395 | ||
1396 | __setup("slub_debug", setup_slub_debug); | |
1397 | ||
c5fd3ca0 AT |
1398 | /* |
1399 | * kmem_cache_flags - apply debugging options to the cache | |
1400 | * @object_size: the size of an object without meta data | |
1401 | * @flags: flags to set | |
1402 | * @name: name of the cache | |
c5fd3ca0 AT |
1403 | * |
1404 | * Debug option(s) are applied to @flags. In addition to the debug | |
1405 | * option(s), if a slab name (or multiple) is specified i.e. | |
1406 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1407 | * then only the select slabs will receive the debug option(s). | |
1408 | */ | |
0293d1fd | 1409 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
f505bc0c | 1410 | slab_flags_t flags, const char *name) |
41ecc55b | 1411 | { |
c5fd3ca0 AT |
1412 | char *iter; |
1413 | size_t len; | |
e17f1dfb VB |
1414 | char *next_block; |
1415 | slab_flags_t block_flags; | |
c5fd3ca0 | 1416 | |
c5fd3ca0 | 1417 | len = strlen(name); |
e17f1dfb VB |
1418 | next_block = slub_debug_string; |
1419 | /* Go through all blocks of debug options, see if any matches our slab's name */ | |
1420 | while (next_block) { | |
1421 | next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false); | |
1422 | if (!iter) | |
1423 | continue; | |
1424 | /* Found a block that has a slab list, search it */ | |
1425 | while (*iter) { | |
1426 | char *end, *glob; | |
1427 | size_t cmplen; | |
1428 | ||
1429 | end = strchrnul(iter, ','); | |
1430 | if (next_block && next_block < end) | |
1431 | end = next_block - 1; | |
1432 | ||
1433 | glob = strnchr(iter, end - iter, '*'); | |
1434 | if (glob) | |
1435 | cmplen = glob - iter; | |
1436 | else | |
1437 | cmplen = max_t(size_t, len, (end - iter)); | |
c5fd3ca0 | 1438 | |
e17f1dfb VB |
1439 | if (!strncmp(name, iter, cmplen)) { |
1440 | flags |= block_flags; | |
1441 | return flags; | |
1442 | } | |
c5fd3ca0 | 1443 | |
e17f1dfb VB |
1444 | if (!*end || *end == ';') |
1445 | break; | |
1446 | iter = end + 1; | |
c5fd3ca0 | 1447 | } |
c5fd3ca0 | 1448 | } |
ba0268a8 | 1449 | |
484cfaca | 1450 | return flags | slub_debug; |
41ecc55b | 1451 | } |
b4a64718 | 1452 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1453 | static inline void setup_object_debug(struct kmem_cache *s, |
1454 | struct page *page, void *object) {} | |
a50b854e MWO |
1455 | static inline |
1456 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {} | |
41ecc55b | 1457 | |
3ec09742 | 1458 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1459 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1460 | |
282acb43 | 1461 | static inline int free_debug_processing( |
81084651 JDB |
1462 | struct kmem_cache *s, struct page *page, |
1463 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1464 | unsigned long addr) { return 0; } |
41ecc55b | 1465 | |
41ecc55b CL |
1466 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1467 | { return 1; } | |
1468 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1469 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1470 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1471 | struct page *page) {} | |
c65c1877 PZ |
1472 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1473 | struct page *page) {} | |
0293d1fd | 1474 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
f505bc0c | 1475 | slab_flags_t flags, const char *name) |
ba0268a8 CL |
1476 | { |
1477 | return flags; | |
1478 | } | |
41ecc55b | 1479 | #define slub_debug 0 |
0f389ec6 | 1480 | |
fdaa45e9 IM |
1481 | #define disable_higher_order_debug 0 |
1482 | ||
0f389ec6 CL |
1483 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1484 | { return 0; } | |
26c02cf0 AB |
1485 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1486 | { return 0; } | |
205ab99d CL |
1487 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1488 | int objects) {} | |
1489 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1490 | int objects) {} | |
7d550c56 | 1491 | |
52f23478 | 1492 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 1493 | void **freelist, void *nextfree) |
52f23478 DZ |
1494 | { |
1495 | return false; | |
1496 | } | |
02e72cc6 AR |
1497 | #endif /* CONFIG_SLUB_DEBUG */ |
1498 | ||
1499 | /* | |
1500 | * Hooks for other subsystems that check memory allocations. In a typical | |
1501 | * production configuration these hooks all should produce no code at all. | |
1502 | */ | |
0116523c | 1503 | static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
d56791b3 | 1504 | { |
53128245 | 1505 | ptr = kasan_kmalloc_large(ptr, size, flags); |
a2f77575 | 1506 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ |
d56791b3 | 1507 | kmemleak_alloc(ptr, size, 1, flags); |
53128245 | 1508 | return ptr; |
d56791b3 RB |
1509 | } |
1510 | ||
ee3ce779 | 1511 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1512 | { |
1513 | kmemleak_free(x); | |
ee3ce779 | 1514 | kasan_kfree_large(x, _RET_IP_); |
d56791b3 RB |
1515 | } |
1516 | ||
c3895391 | 1517 | static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x) |
d56791b3 RB |
1518 | { |
1519 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1520 | |
02e72cc6 AR |
1521 | /* |
1522 | * Trouble is that we may no longer disable interrupts in the fast path | |
1523 | * So in order to make the debug calls that expect irqs to be | |
1524 | * disabled we need to disable interrupts temporarily. | |
1525 | */ | |
4675ff05 | 1526 | #ifdef CONFIG_LOCKDEP |
02e72cc6 AR |
1527 | { |
1528 | unsigned long flags; | |
1529 | ||
1530 | local_irq_save(flags); | |
02e72cc6 AR |
1531 | debug_check_no_locks_freed(x, s->object_size); |
1532 | local_irq_restore(flags); | |
1533 | } | |
1534 | #endif | |
1535 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1536 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1537 | |
cfbe1636 ME |
1538 | /* Use KCSAN to help debug racy use-after-free. */ |
1539 | if (!(s->flags & SLAB_TYPESAFE_BY_RCU)) | |
1540 | __kcsan_check_access(x, s->object_size, | |
1541 | KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT); | |
1542 | ||
c3895391 AK |
1543 | /* KASAN might put x into memory quarantine, delaying its reuse */ |
1544 | return kasan_slab_free(s, x, _RET_IP_); | |
02e72cc6 | 1545 | } |
205ab99d | 1546 | |
c3895391 AK |
1547 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
1548 | void **head, void **tail) | |
81084651 | 1549 | { |
6471384a AP |
1550 | |
1551 | void *object; | |
1552 | void *next = *head; | |
1553 | void *old_tail = *tail ? *tail : *head; | |
1554 | int rsize; | |
1555 | ||
aea4df4c LA |
1556 | /* Head and tail of the reconstructed freelist */ |
1557 | *head = NULL; | |
1558 | *tail = NULL; | |
1b7e816f | 1559 | |
aea4df4c LA |
1560 | do { |
1561 | object = next; | |
1562 | next = get_freepointer(s, object); | |
1563 | ||
1564 | if (slab_want_init_on_free(s)) { | |
6471384a AP |
1565 | /* |
1566 | * Clear the object and the metadata, but don't touch | |
1567 | * the redzone. | |
1568 | */ | |
aa1ef4d7 | 1569 | memset(kasan_reset_tag(object), 0, s->object_size); |
6471384a AP |
1570 | rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad |
1571 | : 0; | |
aa1ef4d7 | 1572 | memset((char *)kasan_reset_tag(object) + s->inuse, 0, |
6471384a | 1573 | s->size - s->inuse - rsize); |
81084651 | 1574 | |
aea4df4c | 1575 | } |
c3895391 AK |
1576 | /* If object's reuse doesn't have to be delayed */ |
1577 | if (!slab_free_hook(s, object)) { | |
1578 | /* Move object to the new freelist */ | |
1579 | set_freepointer(s, object, *head); | |
1580 | *head = object; | |
1581 | if (!*tail) | |
1582 | *tail = object; | |
1583 | } | |
1584 | } while (object != old_tail); | |
1585 | ||
1586 | if (*head == *tail) | |
1587 | *tail = NULL; | |
1588 | ||
1589 | return *head != NULL; | |
81084651 JDB |
1590 | } |
1591 | ||
4d176711 | 1592 | static void *setup_object(struct kmem_cache *s, struct page *page, |
588f8ba9 TG |
1593 | void *object) |
1594 | { | |
1595 | setup_object_debug(s, page, object); | |
4d176711 | 1596 | object = kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1597 | if (unlikely(s->ctor)) { |
1598 | kasan_unpoison_object_data(s, object); | |
1599 | s->ctor(object); | |
1600 | kasan_poison_object_data(s, object); | |
1601 | } | |
4d176711 | 1602 | return object; |
588f8ba9 TG |
1603 | } |
1604 | ||
81819f0f CL |
1605 | /* |
1606 | * Slab allocation and freeing | |
1607 | */ | |
5dfb4175 VD |
1608 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1609 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1610 | { |
5dfb4175 | 1611 | struct page *page; |
19af27af | 1612 | unsigned int order = oo_order(oo); |
65c3376a | 1613 | |
2154a336 | 1614 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1615 | page = alloc_pages(flags, order); |
65c3376a | 1616 | else |
96db800f | 1617 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1618 | |
5dfb4175 | 1619 | return page; |
65c3376a CL |
1620 | } |
1621 | ||
210e7a43 TG |
1622 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1623 | /* Pre-initialize the random sequence cache */ | |
1624 | static int init_cache_random_seq(struct kmem_cache *s) | |
1625 | { | |
19af27af | 1626 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 1627 | int err; |
210e7a43 | 1628 | |
a810007a SR |
1629 | /* Bailout if already initialised */ |
1630 | if (s->random_seq) | |
1631 | return 0; | |
1632 | ||
210e7a43 TG |
1633 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1634 | if (err) { | |
1635 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1636 | s->name); | |
1637 | return err; | |
1638 | } | |
1639 | ||
1640 | /* Transform to an offset on the set of pages */ | |
1641 | if (s->random_seq) { | |
19af27af AD |
1642 | unsigned int i; |
1643 | ||
210e7a43 TG |
1644 | for (i = 0; i < count; i++) |
1645 | s->random_seq[i] *= s->size; | |
1646 | } | |
1647 | return 0; | |
1648 | } | |
1649 | ||
1650 | /* Initialize each random sequence freelist per cache */ | |
1651 | static void __init init_freelist_randomization(void) | |
1652 | { | |
1653 | struct kmem_cache *s; | |
1654 | ||
1655 | mutex_lock(&slab_mutex); | |
1656 | ||
1657 | list_for_each_entry(s, &slab_caches, list) | |
1658 | init_cache_random_seq(s); | |
1659 | ||
1660 | mutex_unlock(&slab_mutex); | |
1661 | } | |
1662 | ||
1663 | /* Get the next entry on the pre-computed freelist randomized */ | |
1664 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1665 | unsigned long *pos, void *start, | |
1666 | unsigned long page_limit, | |
1667 | unsigned long freelist_count) | |
1668 | { | |
1669 | unsigned int idx; | |
1670 | ||
1671 | /* | |
1672 | * If the target page allocation failed, the number of objects on the | |
1673 | * page might be smaller than the usual size defined by the cache. | |
1674 | */ | |
1675 | do { | |
1676 | idx = s->random_seq[*pos]; | |
1677 | *pos += 1; | |
1678 | if (*pos >= freelist_count) | |
1679 | *pos = 0; | |
1680 | } while (unlikely(idx >= page_limit)); | |
1681 | ||
1682 | return (char *)start + idx; | |
1683 | } | |
1684 | ||
1685 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1686 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1687 | { | |
1688 | void *start; | |
1689 | void *cur; | |
1690 | void *next; | |
1691 | unsigned long idx, pos, page_limit, freelist_count; | |
1692 | ||
1693 | if (page->objects < 2 || !s->random_seq) | |
1694 | return false; | |
1695 | ||
1696 | freelist_count = oo_objects(s->oo); | |
1697 | pos = get_random_int() % freelist_count; | |
1698 | ||
1699 | page_limit = page->objects * s->size; | |
1700 | start = fixup_red_left(s, page_address(page)); | |
1701 | ||
1702 | /* First entry is used as the base of the freelist */ | |
1703 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1704 | freelist_count); | |
4d176711 | 1705 | cur = setup_object(s, page, cur); |
210e7a43 TG |
1706 | page->freelist = cur; |
1707 | ||
1708 | for (idx = 1; idx < page->objects; idx++) { | |
210e7a43 TG |
1709 | next = next_freelist_entry(s, page, &pos, start, page_limit, |
1710 | freelist_count); | |
4d176711 | 1711 | next = setup_object(s, page, next); |
210e7a43 TG |
1712 | set_freepointer(s, cur, next); |
1713 | cur = next; | |
1714 | } | |
210e7a43 TG |
1715 | set_freepointer(s, cur, NULL); |
1716 | ||
1717 | return true; | |
1718 | } | |
1719 | #else | |
1720 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1721 | { | |
1722 | return 0; | |
1723 | } | |
1724 | static inline void init_freelist_randomization(void) { } | |
1725 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1726 | { | |
1727 | return false; | |
1728 | } | |
1729 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1730 | ||
81819f0f CL |
1731 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1732 | { | |
06428780 | 1733 | struct page *page; |
834f3d11 | 1734 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1735 | gfp_t alloc_gfp; |
4d176711 | 1736 | void *start, *p, *next; |
a50b854e | 1737 | int idx; |
210e7a43 | 1738 | bool shuffle; |
81819f0f | 1739 | |
7e0528da CL |
1740 | flags &= gfp_allowed_mask; |
1741 | ||
d0164adc | 1742 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1743 | local_irq_enable(); |
1744 | ||
b7a49f0d | 1745 | flags |= s->allocflags; |
e12ba74d | 1746 | |
ba52270d PE |
1747 | /* |
1748 | * Let the initial higher-order allocation fail under memory pressure | |
1749 | * so we fall-back to the minimum order allocation. | |
1750 | */ | |
1751 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1752 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1753 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1754 | |
5dfb4175 | 1755 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1756 | if (unlikely(!page)) { |
1757 | oo = s->min; | |
80c3a998 | 1758 | alloc_gfp = flags; |
65c3376a CL |
1759 | /* |
1760 | * Allocation may have failed due to fragmentation. | |
1761 | * Try a lower order alloc if possible | |
1762 | */ | |
5dfb4175 | 1763 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1764 | if (unlikely(!page)) |
1765 | goto out; | |
1766 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1767 | } |
5a896d9e | 1768 | |
834f3d11 | 1769 | page->objects = oo_objects(oo); |
81819f0f | 1770 | |
1f3147b4 RG |
1771 | account_slab_page(page, oo_order(oo), s); |
1772 | ||
1b4f59e3 | 1773 | page->slab_cache = s; |
c03f94cc | 1774 | __SetPageSlab(page); |
2f064f34 | 1775 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1776 | SetPageSlabPfmemalloc(page); |
81819f0f | 1777 | |
a7101224 | 1778 | kasan_poison_slab(page); |
81819f0f | 1779 | |
a7101224 | 1780 | start = page_address(page); |
81819f0f | 1781 | |
a50b854e | 1782 | setup_page_debug(s, page, start); |
0316bec2 | 1783 | |
210e7a43 TG |
1784 | shuffle = shuffle_freelist(s, page); |
1785 | ||
1786 | if (!shuffle) { | |
4d176711 AK |
1787 | start = fixup_red_left(s, start); |
1788 | start = setup_object(s, page, start); | |
1789 | page->freelist = start; | |
18e50661 AK |
1790 | for (idx = 0, p = start; idx < page->objects - 1; idx++) { |
1791 | next = p + s->size; | |
1792 | next = setup_object(s, page, next); | |
1793 | set_freepointer(s, p, next); | |
1794 | p = next; | |
1795 | } | |
1796 | set_freepointer(s, p, NULL); | |
81819f0f | 1797 | } |
81819f0f | 1798 | |
e6e82ea1 | 1799 | page->inuse = page->objects; |
8cb0a506 | 1800 | page->frozen = 1; |
588f8ba9 | 1801 | |
81819f0f | 1802 | out: |
d0164adc | 1803 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1804 | local_irq_disable(); |
1805 | if (!page) | |
1806 | return NULL; | |
1807 | ||
588f8ba9 TG |
1808 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1809 | ||
81819f0f CL |
1810 | return page; |
1811 | } | |
1812 | ||
588f8ba9 TG |
1813 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1814 | { | |
44405099 LL |
1815 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
1816 | flags = kmalloc_fix_flags(flags); | |
588f8ba9 TG |
1817 | |
1818 | return allocate_slab(s, | |
1819 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1820 | } | |
1821 | ||
81819f0f CL |
1822 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1823 | { | |
834f3d11 CL |
1824 | int order = compound_order(page); |
1825 | int pages = 1 << order; | |
81819f0f | 1826 | |
8fc8d666 | 1827 | if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) { |
81819f0f CL |
1828 | void *p; |
1829 | ||
1830 | slab_pad_check(s, page); | |
224a88be CL |
1831 | for_each_object(p, s, page_address(page), |
1832 | page->objects) | |
f7cb1933 | 1833 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1834 | } |
1835 | ||
072bb0aa | 1836 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1837 | __ClearPageSlab(page); |
0c06dd75 VB |
1838 | /* In union with page->mapping where page allocator expects NULL */ |
1839 | page->slab_cache = NULL; | |
1eb5ac64 NP |
1840 | if (current->reclaim_state) |
1841 | current->reclaim_state->reclaimed_slab += pages; | |
74d555be | 1842 | unaccount_slab_page(page, order, s); |
27ee57c9 | 1843 | __free_pages(page, order); |
81819f0f CL |
1844 | } |
1845 | ||
1846 | static void rcu_free_slab(struct rcu_head *h) | |
1847 | { | |
bf68c214 | 1848 | struct page *page = container_of(h, struct page, rcu_head); |
da9a638c | 1849 | |
1b4f59e3 | 1850 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1851 | } |
1852 | ||
1853 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1854 | { | |
5f0d5a3a | 1855 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bf68c214 | 1856 | call_rcu(&page->rcu_head, rcu_free_slab); |
81819f0f CL |
1857 | } else |
1858 | __free_slab(s, page); | |
1859 | } | |
1860 | ||
1861 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1862 | { | |
205ab99d | 1863 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1864 | free_slab(s, page); |
1865 | } | |
1866 | ||
1867 | /* | |
5cc6eee8 | 1868 | * Management of partially allocated slabs. |
81819f0f | 1869 | */ |
1e4dd946 SR |
1870 | static inline void |
1871 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1872 | { |
e95eed57 | 1873 | n->nr_partial++; |
136333d1 | 1874 | if (tail == DEACTIVATE_TO_TAIL) |
916ac052 | 1875 | list_add_tail(&page->slab_list, &n->partial); |
7c2e132c | 1876 | else |
916ac052 | 1877 | list_add(&page->slab_list, &n->partial); |
81819f0f CL |
1878 | } |
1879 | ||
1e4dd946 SR |
1880 | static inline void add_partial(struct kmem_cache_node *n, |
1881 | struct page *page, int tail) | |
62e346a8 | 1882 | { |
c65c1877 | 1883 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1884 | __add_partial(n, page, tail); |
1885 | } | |
c65c1877 | 1886 | |
1e4dd946 SR |
1887 | static inline void remove_partial(struct kmem_cache_node *n, |
1888 | struct page *page) | |
1889 | { | |
1890 | lockdep_assert_held(&n->list_lock); | |
916ac052 | 1891 | list_del(&page->slab_list); |
52b4b950 | 1892 | n->nr_partial--; |
1e4dd946 SR |
1893 | } |
1894 | ||
81819f0f | 1895 | /* |
7ced3719 CL |
1896 | * Remove slab from the partial list, freeze it and |
1897 | * return the pointer to the freelist. | |
81819f0f | 1898 | * |
497b66f2 | 1899 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1900 | */ |
497b66f2 | 1901 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1902 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1903 | int mode, int *objects) |
81819f0f | 1904 | { |
2cfb7455 CL |
1905 | void *freelist; |
1906 | unsigned long counters; | |
1907 | struct page new; | |
1908 | ||
c65c1877 PZ |
1909 | lockdep_assert_held(&n->list_lock); |
1910 | ||
2cfb7455 CL |
1911 | /* |
1912 | * Zap the freelist and set the frozen bit. | |
1913 | * The old freelist is the list of objects for the | |
1914 | * per cpu allocation list. | |
1915 | */ | |
7ced3719 CL |
1916 | freelist = page->freelist; |
1917 | counters = page->counters; | |
1918 | new.counters = counters; | |
633b0764 | 1919 | *objects = new.objects - new.inuse; |
23910c50 | 1920 | if (mode) { |
7ced3719 | 1921 | new.inuse = page->objects; |
23910c50 PE |
1922 | new.freelist = NULL; |
1923 | } else { | |
1924 | new.freelist = freelist; | |
1925 | } | |
2cfb7455 | 1926 | |
a0132ac0 | 1927 | VM_BUG_ON(new.frozen); |
7ced3719 | 1928 | new.frozen = 1; |
2cfb7455 | 1929 | |
7ced3719 | 1930 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1931 | freelist, counters, |
02d7633f | 1932 | new.freelist, new.counters, |
7ced3719 | 1933 | "acquire_slab")) |
7ced3719 | 1934 | return NULL; |
2cfb7455 CL |
1935 | |
1936 | remove_partial(n, page); | |
7ced3719 | 1937 | WARN_ON(!freelist); |
49e22585 | 1938 | return freelist; |
81819f0f CL |
1939 | } |
1940 | ||
633b0764 | 1941 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1942 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1943 | |
81819f0f | 1944 | /* |
672bba3a | 1945 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1946 | */ |
8ba00bb6 JK |
1947 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1948 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1949 | { |
49e22585 CL |
1950 | struct page *page, *page2; |
1951 | void *object = NULL; | |
e5d9998f | 1952 | unsigned int available = 0; |
633b0764 | 1953 | int objects; |
81819f0f CL |
1954 | |
1955 | /* | |
1956 | * Racy check. If we mistakenly see no partial slabs then we | |
1957 | * just allocate an empty slab. If we mistakenly try to get a | |
70b6d25e | 1958 | * partial slab and there is none available then get_partial() |
672bba3a | 1959 | * will return NULL. |
81819f0f CL |
1960 | */ |
1961 | if (!n || !n->nr_partial) | |
1962 | return NULL; | |
1963 | ||
1964 | spin_lock(&n->list_lock); | |
916ac052 | 1965 | list_for_each_entry_safe(page, page2, &n->partial, slab_list) { |
8ba00bb6 | 1966 | void *t; |
49e22585 | 1967 | |
8ba00bb6 JK |
1968 | if (!pfmemalloc_match(page, flags)) |
1969 | continue; | |
1970 | ||
633b0764 | 1971 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 | 1972 | if (!t) |
3555c14a | 1973 | break; |
49e22585 | 1974 | |
633b0764 | 1975 | available += objects; |
12d79634 | 1976 | if (!object) { |
49e22585 | 1977 | c->page = page; |
49e22585 | 1978 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1979 | object = t; |
49e22585 | 1980 | } else { |
633b0764 | 1981 | put_cpu_partial(s, page, 0); |
8028dcea | 1982 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1983 | } |
345c905d | 1984 | if (!kmem_cache_has_cpu_partial(s) |
e6d0e1dc | 1985 | || available > slub_cpu_partial(s) / 2) |
49e22585 CL |
1986 | break; |
1987 | ||
497b66f2 | 1988 | } |
81819f0f | 1989 | spin_unlock(&n->list_lock); |
497b66f2 | 1990 | return object; |
81819f0f CL |
1991 | } |
1992 | ||
1993 | /* | |
672bba3a | 1994 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1995 | */ |
de3ec035 | 1996 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1997 | struct kmem_cache_cpu *c) |
81819f0f CL |
1998 | { |
1999 | #ifdef CONFIG_NUMA | |
2000 | struct zonelist *zonelist; | |
dd1a239f | 2001 | struct zoneref *z; |
54a6eb5c | 2002 | struct zone *zone; |
97a225e6 | 2003 | enum zone_type highest_zoneidx = gfp_zone(flags); |
497b66f2 | 2004 | void *object; |
cc9a6c87 | 2005 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
2006 | |
2007 | /* | |
672bba3a CL |
2008 | * The defrag ratio allows a configuration of the tradeoffs between |
2009 | * inter node defragmentation and node local allocations. A lower | |
2010 | * defrag_ratio increases the tendency to do local allocations | |
2011 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 2012 | * |
672bba3a CL |
2013 | * If the defrag_ratio is set to 0 then kmalloc() always |
2014 | * returns node local objects. If the ratio is higher then kmalloc() | |
2015 | * may return off node objects because partial slabs are obtained | |
2016 | * from other nodes and filled up. | |
81819f0f | 2017 | * |
43efd3ea LP |
2018 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
2019 | * (which makes defrag_ratio = 1000) then every (well almost) | |
2020 | * allocation will first attempt to defrag slab caches on other nodes. | |
2021 | * This means scanning over all nodes to look for partial slabs which | |
2022 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 2023 | * with available objects. |
81819f0f | 2024 | */ |
9824601e CL |
2025 | if (!s->remote_node_defrag_ratio || |
2026 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
2027 | return NULL; |
2028 | ||
cc9a6c87 | 2029 | do { |
d26914d1 | 2030 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 2031 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
97a225e6 | 2032 | for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) { |
cc9a6c87 MG |
2033 | struct kmem_cache_node *n; |
2034 | ||
2035 | n = get_node(s, zone_to_nid(zone)); | |
2036 | ||
dee2f8aa | 2037 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 2038 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 2039 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
2040 | if (object) { |
2041 | /* | |
d26914d1 MG |
2042 | * Don't check read_mems_allowed_retry() |
2043 | * here - if mems_allowed was updated in | |
2044 | * parallel, that was a harmless race | |
2045 | * between allocation and the cpuset | |
2046 | * update | |
cc9a6c87 | 2047 | */ |
cc9a6c87 MG |
2048 | return object; |
2049 | } | |
c0ff7453 | 2050 | } |
81819f0f | 2051 | } |
d26914d1 | 2052 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
6dfd1b65 | 2053 | #endif /* CONFIG_NUMA */ |
81819f0f CL |
2054 | return NULL; |
2055 | } | |
2056 | ||
2057 | /* | |
2058 | * Get a partial page, lock it and return it. | |
2059 | */ | |
497b66f2 | 2060 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 2061 | struct kmem_cache_cpu *c) |
81819f0f | 2062 | { |
497b66f2 | 2063 | void *object; |
a561ce00 JK |
2064 | int searchnode = node; |
2065 | ||
2066 | if (node == NUMA_NO_NODE) | |
2067 | searchnode = numa_mem_id(); | |
81819f0f | 2068 | |
8ba00bb6 | 2069 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
2070 | if (object || node != NUMA_NO_NODE) |
2071 | return object; | |
81819f0f | 2072 | |
acd19fd1 | 2073 | return get_any_partial(s, flags, c); |
81819f0f CL |
2074 | } |
2075 | ||
923717cb | 2076 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2077 | /* |
0d645ed1 | 2078 | * Calculate the next globally unique transaction for disambiguation |
8a5ec0ba CL |
2079 | * during cmpxchg. The transactions start with the cpu number and are then |
2080 | * incremented by CONFIG_NR_CPUS. | |
2081 | */ | |
2082 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
2083 | #else | |
2084 | /* | |
2085 | * No preemption supported therefore also no need to check for | |
2086 | * different cpus. | |
2087 | */ | |
2088 | #define TID_STEP 1 | |
2089 | #endif | |
2090 | ||
2091 | static inline unsigned long next_tid(unsigned long tid) | |
2092 | { | |
2093 | return tid + TID_STEP; | |
2094 | } | |
2095 | ||
9d5f0be0 | 2096 | #ifdef SLUB_DEBUG_CMPXCHG |
8a5ec0ba CL |
2097 | static inline unsigned int tid_to_cpu(unsigned long tid) |
2098 | { | |
2099 | return tid % TID_STEP; | |
2100 | } | |
2101 | ||
2102 | static inline unsigned long tid_to_event(unsigned long tid) | |
2103 | { | |
2104 | return tid / TID_STEP; | |
2105 | } | |
9d5f0be0 | 2106 | #endif |
8a5ec0ba CL |
2107 | |
2108 | static inline unsigned int init_tid(int cpu) | |
2109 | { | |
2110 | return cpu; | |
2111 | } | |
2112 | ||
2113 | static inline void note_cmpxchg_failure(const char *n, | |
2114 | const struct kmem_cache *s, unsigned long tid) | |
2115 | { | |
2116 | #ifdef SLUB_DEBUG_CMPXCHG | |
2117 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2118 | ||
f9f58285 | 2119 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba | 2120 | |
923717cb | 2121 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2122 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) |
f9f58285 | 2123 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2124 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2125 | else | |
2126 | #endif | |
2127 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2128 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2129 | tid_to_event(tid), tid_to_event(actual_tid)); |
2130 | else | |
f9f58285 | 2131 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2132 | actual_tid, tid, next_tid(tid)); |
2133 | #endif | |
4fdccdfb | 2134 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2135 | } |
2136 | ||
788e1aad | 2137 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2138 | { |
8a5ec0ba CL |
2139 | int cpu; |
2140 | ||
2141 | for_each_possible_cpu(cpu) | |
2142 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 2143 | } |
2cfb7455 | 2144 | |
81819f0f CL |
2145 | /* |
2146 | * Remove the cpu slab | |
2147 | */ | |
d0e0ac97 | 2148 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
d4ff6d35 | 2149 | void *freelist, struct kmem_cache_cpu *c) |
81819f0f | 2150 | { |
2cfb7455 | 2151 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
2152 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
2153 | int lock = 0; | |
2154 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 2155 | void *nextfree; |
136333d1 | 2156 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2157 | struct page new; |
2158 | struct page old; | |
2159 | ||
2160 | if (page->freelist) { | |
84e554e6 | 2161 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2162 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2163 | } |
2164 | ||
894b8788 | 2165 | /* |
2cfb7455 CL |
2166 | * Stage one: Free all available per cpu objects back |
2167 | * to the page freelist while it is still frozen. Leave the | |
2168 | * last one. | |
2169 | * | |
2170 | * There is no need to take the list->lock because the page | |
2171 | * is still frozen. | |
2172 | */ | |
2173 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2174 | void *prior; | |
2175 | unsigned long counters; | |
2176 | ||
52f23478 DZ |
2177 | /* |
2178 | * If 'nextfree' is invalid, it is possible that the object at | |
2179 | * 'freelist' is already corrupted. So isolate all objects | |
2180 | * starting at 'freelist'. | |
2181 | */ | |
dc07a728 | 2182 | if (freelist_corrupted(s, page, &freelist, nextfree)) |
52f23478 DZ |
2183 | break; |
2184 | ||
2cfb7455 CL |
2185 | do { |
2186 | prior = page->freelist; | |
2187 | counters = page->counters; | |
2188 | set_freepointer(s, freelist, prior); | |
2189 | new.counters = counters; | |
2190 | new.inuse--; | |
a0132ac0 | 2191 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 2192 | |
1d07171c | 2193 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2194 | prior, counters, |
2195 | freelist, new.counters, | |
2196 | "drain percpu freelist")); | |
2197 | ||
2198 | freelist = nextfree; | |
2199 | } | |
2200 | ||
894b8788 | 2201 | /* |
2cfb7455 CL |
2202 | * Stage two: Ensure that the page is unfrozen while the |
2203 | * list presence reflects the actual number of objects | |
2204 | * during unfreeze. | |
2205 | * | |
2206 | * We setup the list membership and then perform a cmpxchg | |
2207 | * with the count. If there is a mismatch then the page | |
2208 | * is not unfrozen but the page is on the wrong list. | |
2209 | * | |
2210 | * Then we restart the process which may have to remove | |
2211 | * the page from the list that we just put it on again | |
2212 | * because the number of objects in the slab may have | |
2213 | * changed. | |
894b8788 | 2214 | */ |
2cfb7455 | 2215 | redo: |
894b8788 | 2216 | |
2cfb7455 CL |
2217 | old.freelist = page->freelist; |
2218 | old.counters = page->counters; | |
a0132ac0 | 2219 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2220 | |
2cfb7455 CL |
2221 | /* Determine target state of the slab */ |
2222 | new.counters = old.counters; | |
2223 | if (freelist) { | |
2224 | new.inuse--; | |
2225 | set_freepointer(s, freelist, old.freelist); | |
2226 | new.freelist = freelist; | |
2227 | } else | |
2228 | new.freelist = old.freelist; | |
2229 | ||
2230 | new.frozen = 0; | |
2231 | ||
8a5b20ae | 2232 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2233 | m = M_FREE; |
2234 | else if (new.freelist) { | |
2235 | m = M_PARTIAL; | |
2236 | if (!lock) { | |
2237 | lock = 1; | |
2238 | /* | |
8bb4e7a2 | 2239 | * Taking the spinlock removes the possibility |
2cfb7455 CL |
2240 | * that acquire_slab() will see a slab page that |
2241 | * is frozen | |
2242 | */ | |
2243 | spin_lock(&n->list_lock); | |
2244 | } | |
2245 | } else { | |
2246 | m = M_FULL; | |
965c4848 | 2247 | if (kmem_cache_debug_flags(s, SLAB_STORE_USER) && !lock) { |
2cfb7455 CL |
2248 | lock = 1; |
2249 | /* | |
2250 | * This also ensures that the scanning of full | |
2251 | * slabs from diagnostic functions will not see | |
2252 | * any frozen slabs. | |
2253 | */ | |
2254 | spin_lock(&n->list_lock); | |
2255 | } | |
2256 | } | |
2257 | ||
2258 | if (l != m) { | |
2cfb7455 | 2259 | if (l == M_PARTIAL) |
2cfb7455 | 2260 | remove_partial(n, page); |
2cfb7455 | 2261 | else if (l == M_FULL) |
c65c1877 | 2262 | remove_full(s, n, page); |
2cfb7455 | 2263 | |
88349a28 | 2264 | if (m == M_PARTIAL) |
2cfb7455 | 2265 | add_partial(n, page, tail); |
88349a28 | 2266 | else if (m == M_FULL) |
2cfb7455 | 2267 | add_full(s, n, page); |
2cfb7455 CL |
2268 | } |
2269 | ||
2270 | l = m; | |
1d07171c | 2271 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2272 | old.freelist, old.counters, |
2273 | new.freelist, new.counters, | |
2274 | "unfreezing slab")) | |
2275 | goto redo; | |
2276 | ||
2cfb7455 CL |
2277 | if (lock) |
2278 | spin_unlock(&n->list_lock); | |
2279 | ||
88349a28 WY |
2280 | if (m == M_PARTIAL) |
2281 | stat(s, tail); | |
2282 | else if (m == M_FULL) | |
2283 | stat(s, DEACTIVATE_FULL); | |
2284 | else if (m == M_FREE) { | |
2cfb7455 CL |
2285 | stat(s, DEACTIVATE_EMPTY); |
2286 | discard_slab(s, page); | |
2287 | stat(s, FREE_SLAB); | |
894b8788 | 2288 | } |
d4ff6d35 WY |
2289 | |
2290 | c->page = NULL; | |
2291 | c->freelist = NULL; | |
81819f0f CL |
2292 | } |
2293 | ||
d24ac77f JK |
2294 | /* |
2295 | * Unfreeze all the cpu partial slabs. | |
2296 | * | |
59a09917 CL |
2297 | * This function must be called with interrupts disabled |
2298 | * for the cpu using c (or some other guarantee must be there | |
2299 | * to guarantee no concurrent accesses). | |
d24ac77f | 2300 | */ |
59a09917 CL |
2301 | static void unfreeze_partials(struct kmem_cache *s, |
2302 | struct kmem_cache_cpu *c) | |
49e22585 | 2303 | { |
345c905d | 2304 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2305 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2306 | struct page *page, *discard_page = NULL; |
49e22585 | 2307 | |
4c7ba22e | 2308 | while ((page = slub_percpu_partial(c))) { |
49e22585 CL |
2309 | struct page new; |
2310 | struct page old; | |
2311 | ||
4c7ba22e | 2312 | slub_set_percpu_partial(c, page); |
43d77867 JK |
2313 | |
2314 | n2 = get_node(s, page_to_nid(page)); | |
2315 | if (n != n2) { | |
2316 | if (n) | |
2317 | spin_unlock(&n->list_lock); | |
2318 | ||
2319 | n = n2; | |
2320 | spin_lock(&n->list_lock); | |
2321 | } | |
49e22585 CL |
2322 | |
2323 | do { | |
2324 | ||
2325 | old.freelist = page->freelist; | |
2326 | old.counters = page->counters; | |
a0132ac0 | 2327 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2328 | |
2329 | new.counters = old.counters; | |
2330 | new.freelist = old.freelist; | |
2331 | ||
2332 | new.frozen = 0; | |
2333 | ||
d24ac77f | 2334 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2335 | old.freelist, old.counters, |
2336 | new.freelist, new.counters, | |
2337 | "unfreezing slab")); | |
2338 | ||
8a5b20ae | 2339 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2340 | page->next = discard_page; |
2341 | discard_page = page; | |
43d77867 JK |
2342 | } else { |
2343 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2344 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2345 | } |
2346 | } | |
2347 | ||
2348 | if (n) | |
2349 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2350 | |
2351 | while (discard_page) { | |
2352 | page = discard_page; | |
2353 | discard_page = discard_page->next; | |
2354 | ||
2355 | stat(s, DEACTIVATE_EMPTY); | |
2356 | discard_slab(s, page); | |
2357 | stat(s, FREE_SLAB); | |
2358 | } | |
6dfd1b65 | 2359 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2360 | } |
2361 | ||
2362 | /* | |
9234bae9 WY |
2363 | * Put a page that was just frozen (in __slab_free|get_partial_node) into a |
2364 | * partial page slot if available. | |
49e22585 CL |
2365 | * |
2366 | * If we did not find a slot then simply move all the partials to the | |
2367 | * per node partial list. | |
2368 | */ | |
633b0764 | 2369 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2370 | { |
345c905d | 2371 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2372 | struct page *oldpage; |
2373 | int pages; | |
2374 | int pobjects; | |
2375 | ||
d6e0b7fa | 2376 | preempt_disable(); |
49e22585 CL |
2377 | do { |
2378 | pages = 0; | |
2379 | pobjects = 0; | |
2380 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2381 | ||
2382 | if (oldpage) { | |
2383 | pobjects = oldpage->pobjects; | |
2384 | pages = oldpage->pages; | |
bbd4e305 | 2385 | if (drain && pobjects > slub_cpu_partial(s)) { |
49e22585 CL |
2386 | unsigned long flags; |
2387 | /* | |
2388 | * partial array is full. Move the existing | |
2389 | * set to the per node partial list. | |
2390 | */ | |
2391 | local_irq_save(flags); | |
59a09917 | 2392 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2393 | local_irq_restore(flags); |
e24fc410 | 2394 | oldpage = NULL; |
49e22585 CL |
2395 | pobjects = 0; |
2396 | pages = 0; | |
8028dcea | 2397 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2398 | } |
2399 | } | |
2400 | ||
2401 | pages++; | |
2402 | pobjects += page->objects - page->inuse; | |
2403 | ||
2404 | page->pages = pages; | |
2405 | page->pobjects = pobjects; | |
2406 | page->next = oldpage; | |
2407 | ||
d0e0ac97 CG |
2408 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2409 | != oldpage); | |
bbd4e305 | 2410 | if (unlikely(!slub_cpu_partial(s))) { |
d6e0b7fa VD |
2411 | unsigned long flags; |
2412 | ||
2413 | local_irq_save(flags); | |
2414 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2415 | local_irq_restore(flags); | |
2416 | } | |
2417 | preempt_enable(); | |
6dfd1b65 | 2418 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2419 | } |
2420 | ||
dfb4f096 | 2421 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2422 | { |
84e554e6 | 2423 | stat(s, CPUSLAB_FLUSH); |
d4ff6d35 | 2424 | deactivate_slab(s, c->page, c->freelist, c); |
c17dda40 CL |
2425 | |
2426 | c->tid = next_tid(c->tid); | |
81819f0f CL |
2427 | } |
2428 | ||
2429 | /* | |
2430 | * Flush cpu slab. | |
6446faa2 | 2431 | * |
81819f0f CL |
2432 | * Called from IPI handler with interrupts disabled. |
2433 | */ | |
0c710013 | 2434 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2435 | { |
9dfc6e68 | 2436 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2437 | |
1265ef2d WY |
2438 | if (c->page) |
2439 | flush_slab(s, c); | |
49e22585 | 2440 | |
1265ef2d | 2441 | unfreeze_partials(s, c); |
81819f0f CL |
2442 | } |
2443 | ||
2444 | static void flush_cpu_slab(void *d) | |
2445 | { | |
2446 | struct kmem_cache *s = d; | |
81819f0f | 2447 | |
dfb4f096 | 2448 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2449 | } |
2450 | ||
a8364d55 GBY |
2451 | static bool has_cpu_slab(int cpu, void *info) |
2452 | { | |
2453 | struct kmem_cache *s = info; | |
2454 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2455 | ||
a93cf07b | 2456 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2457 | } |
2458 | ||
81819f0f CL |
2459 | static void flush_all(struct kmem_cache *s) |
2460 | { | |
cb923159 | 2461 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1); |
81819f0f CL |
2462 | } |
2463 | ||
a96a87bf SAS |
2464 | /* |
2465 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2466 | * necessary. | |
2467 | */ | |
2468 | static int slub_cpu_dead(unsigned int cpu) | |
2469 | { | |
2470 | struct kmem_cache *s; | |
2471 | unsigned long flags; | |
2472 | ||
2473 | mutex_lock(&slab_mutex); | |
2474 | list_for_each_entry(s, &slab_caches, list) { | |
2475 | local_irq_save(flags); | |
2476 | __flush_cpu_slab(s, cpu); | |
2477 | local_irq_restore(flags); | |
2478 | } | |
2479 | mutex_unlock(&slab_mutex); | |
2480 | return 0; | |
2481 | } | |
2482 | ||
dfb4f096 CL |
2483 | /* |
2484 | * Check if the objects in a per cpu structure fit numa | |
2485 | * locality expectations. | |
2486 | */ | |
57d437d2 | 2487 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2488 | { |
2489 | #ifdef CONFIG_NUMA | |
6159d0f5 | 2490 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2491 | return 0; |
2492 | #endif | |
2493 | return 1; | |
2494 | } | |
2495 | ||
9a02d699 | 2496 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2497 | static int count_free(struct page *page) |
2498 | { | |
2499 | return page->objects - page->inuse; | |
2500 | } | |
2501 | ||
9a02d699 DR |
2502 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2503 | { | |
2504 | return atomic_long_read(&n->total_objects); | |
2505 | } | |
2506 | #endif /* CONFIG_SLUB_DEBUG */ | |
2507 | ||
2508 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2509 | static unsigned long count_partial(struct kmem_cache_node *n, |
2510 | int (*get_count)(struct page *)) | |
2511 | { | |
2512 | unsigned long flags; | |
2513 | unsigned long x = 0; | |
2514 | struct page *page; | |
2515 | ||
2516 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 2517 | list_for_each_entry(page, &n->partial, slab_list) |
781b2ba6 PE |
2518 | x += get_count(page); |
2519 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2520 | return x; | |
2521 | } | |
9a02d699 | 2522 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2523 | |
781b2ba6 PE |
2524 | static noinline void |
2525 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2526 | { | |
9a02d699 DR |
2527 | #ifdef CONFIG_SLUB_DEBUG |
2528 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2529 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2530 | int node; |
fa45dc25 | 2531 | struct kmem_cache_node *n; |
781b2ba6 | 2532 | |
9a02d699 DR |
2533 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2534 | return; | |
2535 | ||
5b3810e5 VB |
2536 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2537 | nid, gfpflags, &gfpflags); | |
19af27af | 2538 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2539 | s->name, s->object_size, s->size, oo_order(s->oo), |
2540 | oo_order(s->min)); | |
781b2ba6 | 2541 | |
3b0efdfa | 2542 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2543 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2544 | s->name); | |
fa5ec8a1 | 2545 | |
fa45dc25 | 2546 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2547 | unsigned long nr_slabs; |
2548 | unsigned long nr_objs; | |
2549 | unsigned long nr_free; | |
2550 | ||
26c02cf0 AB |
2551 | nr_free = count_partial(n, count_free); |
2552 | nr_slabs = node_nr_slabs(n); | |
2553 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2554 | |
f9f58285 | 2555 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2556 | node, nr_slabs, nr_objs, nr_free); |
2557 | } | |
9a02d699 | 2558 | #endif |
781b2ba6 PE |
2559 | } |
2560 | ||
497b66f2 CL |
2561 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2562 | int node, struct kmem_cache_cpu **pc) | |
2563 | { | |
6faa6833 | 2564 | void *freelist; |
188fd063 CL |
2565 | struct kmem_cache_cpu *c = *pc; |
2566 | struct page *page; | |
497b66f2 | 2567 | |
128227e7 MW |
2568 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
2569 | ||
188fd063 | 2570 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2571 | |
188fd063 CL |
2572 | if (freelist) |
2573 | return freelist; | |
2574 | ||
2575 | page = new_slab(s, flags, node); | |
497b66f2 | 2576 | if (page) { |
7c8e0181 | 2577 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2578 | if (c->page) |
2579 | flush_slab(s, c); | |
2580 | ||
2581 | /* | |
2582 | * No other reference to the page yet so we can | |
2583 | * muck around with it freely without cmpxchg | |
2584 | */ | |
6faa6833 | 2585 | freelist = page->freelist; |
497b66f2 CL |
2586 | page->freelist = NULL; |
2587 | ||
2588 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2589 | c->page = page; |
2590 | *pc = c; | |
edde82b6 | 2591 | } |
497b66f2 | 2592 | |
6faa6833 | 2593 | return freelist; |
497b66f2 CL |
2594 | } |
2595 | ||
072bb0aa MG |
2596 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2597 | { | |
2598 | if (unlikely(PageSlabPfmemalloc(page))) | |
2599 | return gfp_pfmemalloc_allowed(gfpflags); | |
2600 | ||
2601 | return true; | |
2602 | } | |
2603 | ||
213eeb9f | 2604 | /* |
d0e0ac97 CG |
2605 | * Check the page->freelist of a page and either transfer the freelist to the |
2606 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2607 | * |
2608 | * The page is still frozen if the return value is not NULL. | |
2609 | * | |
2610 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2611 | * |
2612 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2613 | */ |
2614 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2615 | { | |
2616 | struct page new; | |
2617 | unsigned long counters; | |
2618 | void *freelist; | |
2619 | ||
2620 | do { | |
2621 | freelist = page->freelist; | |
2622 | counters = page->counters; | |
6faa6833 | 2623 | |
213eeb9f | 2624 | new.counters = counters; |
a0132ac0 | 2625 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2626 | |
2627 | new.inuse = page->objects; | |
2628 | new.frozen = freelist != NULL; | |
2629 | ||
d24ac77f | 2630 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2631 | freelist, counters, |
2632 | NULL, new.counters, | |
2633 | "get_freelist")); | |
2634 | ||
2635 | return freelist; | |
2636 | } | |
2637 | ||
81819f0f | 2638 | /* |
894b8788 CL |
2639 | * Slow path. The lockless freelist is empty or we need to perform |
2640 | * debugging duties. | |
2641 | * | |
894b8788 CL |
2642 | * Processing is still very fast if new objects have been freed to the |
2643 | * regular freelist. In that case we simply take over the regular freelist | |
2644 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2645 | * |
894b8788 CL |
2646 | * If that is not working then we fall back to the partial lists. We take the |
2647 | * first element of the freelist as the object to allocate now and move the | |
2648 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2649 | * |
894b8788 | 2650 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2651 | * we need to allocate a new slab. This is the slowest path since it involves |
2652 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2653 | * |
2654 | * Version of __slab_alloc to use when we know that interrupts are | |
2655 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2656 | */ |
a380a3c7 | 2657 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2658 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2659 | { |
6faa6833 | 2660 | void *freelist; |
f6e7def7 | 2661 | struct page *page; |
81819f0f | 2662 | |
9f986d99 AW |
2663 | stat(s, ALLOC_SLOWPATH); |
2664 | ||
f6e7def7 | 2665 | page = c->page; |
0715e6c5 VB |
2666 | if (!page) { |
2667 | /* | |
2668 | * if the node is not online or has no normal memory, just | |
2669 | * ignore the node constraint | |
2670 | */ | |
2671 | if (unlikely(node != NUMA_NO_NODE && | |
2672 | !node_state(node, N_NORMAL_MEMORY))) | |
2673 | node = NUMA_NO_NODE; | |
81819f0f | 2674 | goto new_slab; |
0715e6c5 | 2675 | } |
49e22585 | 2676 | redo: |
6faa6833 | 2677 | |
57d437d2 | 2678 | if (unlikely(!node_match(page, node))) { |
0715e6c5 VB |
2679 | /* |
2680 | * same as above but node_match() being false already | |
2681 | * implies node != NUMA_NO_NODE | |
2682 | */ | |
2683 | if (!node_state(node, N_NORMAL_MEMORY)) { | |
2684 | node = NUMA_NO_NODE; | |
2685 | goto redo; | |
2686 | } else { | |
a561ce00 | 2687 | stat(s, ALLOC_NODE_MISMATCH); |
d4ff6d35 | 2688 | deactivate_slab(s, page, c->freelist, c); |
a561ce00 JK |
2689 | goto new_slab; |
2690 | } | |
fc59c053 | 2691 | } |
6446faa2 | 2692 | |
072bb0aa MG |
2693 | /* |
2694 | * By rights, we should be searching for a slab page that was | |
2695 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2696 | * information when the page leaves the per-cpu allocator | |
2697 | */ | |
2698 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
d4ff6d35 | 2699 | deactivate_slab(s, page, c->freelist, c); |
072bb0aa MG |
2700 | goto new_slab; |
2701 | } | |
2702 | ||
73736e03 | 2703 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2704 | freelist = c->freelist; |
2705 | if (freelist) | |
73736e03 | 2706 | goto load_freelist; |
03e404af | 2707 | |
f6e7def7 | 2708 | freelist = get_freelist(s, page); |
6446faa2 | 2709 | |
6faa6833 | 2710 | if (!freelist) { |
03e404af CL |
2711 | c->page = NULL; |
2712 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2713 | goto new_slab; |
03e404af | 2714 | } |
6446faa2 | 2715 | |
84e554e6 | 2716 | stat(s, ALLOC_REFILL); |
6446faa2 | 2717 | |
894b8788 | 2718 | load_freelist: |
507effea CL |
2719 | /* |
2720 | * freelist is pointing to the list of objects to be used. | |
2721 | * page is pointing to the page from which the objects are obtained. | |
2722 | * That page must be frozen for per cpu allocations to work. | |
2723 | */ | |
a0132ac0 | 2724 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2725 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2726 | c->tid = next_tid(c->tid); |
6faa6833 | 2727 | return freelist; |
81819f0f | 2728 | |
81819f0f | 2729 | new_slab: |
2cfb7455 | 2730 | |
a93cf07b WY |
2731 | if (slub_percpu_partial(c)) { |
2732 | page = c->page = slub_percpu_partial(c); | |
2733 | slub_set_percpu_partial(c, page); | |
49e22585 | 2734 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2735 | goto redo; |
81819f0f CL |
2736 | } |
2737 | ||
188fd063 | 2738 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2739 | |
f4697436 | 2740 | if (unlikely(!freelist)) { |
9a02d699 | 2741 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2742 | return NULL; |
81819f0f | 2743 | } |
2cfb7455 | 2744 | |
f6e7def7 | 2745 | page = c->page; |
5091b74a | 2746 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2747 | goto load_freelist; |
2cfb7455 | 2748 | |
497b66f2 | 2749 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2750 | if (kmem_cache_debug(s) && |
2751 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2752 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2753 | |
d4ff6d35 | 2754 | deactivate_slab(s, page, get_freepointer(s, freelist), c); |
6faa6833 | 2755 | return freelist; |
894b8788 CL |
2756 | } |
2757 | ||
a380a3c7 CL |
2758 | /* |
2759 | * Another one that disabled interrupt and compensates for possible | |
2760 | * cpu changes by refetching the per cpu area pointer. | |
2761 | */ | |
2762 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2763 | unsigned long addr, struct kmem_cache_cpu *c) | |
2764 | { | |
2765 | void *p; | |
2766 | unsigned long flags; | |
2767 | ||
2768 | local_irq_save(flags); | |
923717cb | 2769 | #ifdef CONFIG_PREEMPTION |
a380a3c7 CL |
2770 | /* |
2771 | * We may have been preempted and rescheduled on a different | |
2772 | * cpu before disabling interrupts. Need to reload cpu area | |
2773 | * pointer. | |
2774 | */ | |
2775 | c = this_cpu_ptr(s->cpu_slab); | |
2776 | #endif | |
2777 | ||
2778 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2779 | local_irq_restore(flags); | |
2780 | return p; | |
2781 | } | |
2782 | ||
0f181f9f AP |
2783 | /* |
2784 | * If the object has been wiped upon free, make sure it's fully initialized by | |
2785 | * zeroing out freelist pointer. | |
2786 | */ | |
2787 | static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s, | |
2788 | void *obj) | |
2789 | { | |
2790 | if (unlikely(slab_want_init_on_free(s)) && obj) | |
ce5716c6 AK |
2791 | memset((void *)((char *)kasan_reset_tag(obj) + s->offset), |
2792 | 0, sizeof(void *)); | |
0f181f9f AP |
2793 | } |
2794 | ||
894b8788 CL |
2795 | /* |
2796 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2797 | * have the fastpath folded into their functions. So no function call | |
2798 | * overhead for requests that can be satisfied on the fastpath. | |
2799 | * | |
2800 | * The fastpath works by first checking if the lockless freelist can be used. | |
2801 | * If not then __slab_alloc is called for slow processing. | |
2802 | * | |
2803 | * Otherwise we can simply pick the next object from the lockless free list. | |
2804 | */ | |
2b847c3c | 2805 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2806 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2807 | { |
03ec0ed5 | 2808 | void *object; |
dfb4f096 | 2809 | struct kmem_cache_cpu *c; |
57d437d2 | 2810 | struct page *page; |
8a5ec0ba | 2811 | unsigned long tid; |
964d4bd3 | 2812 | struct obj_cgroup *objcg = NULL; |
1f84260c | 2813 | |
964d4bd3 | 2814 | s = slab_pre_alloc_hook(s, &objcg, 1, gfpflags); |
8135be5a | 2815 | if (!s) |
773ff60e | 2816 | return NULL; |
8a5ec0ba | 2817 | redo: |
8a5ec0ba CL |
2818 | /* |
2819 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2820 | * enabled. We may switch back and forth between cpus while | |
2821 | * reading from one cpu area. That does not matter as long | |
2822 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2823 | * |
9aabf810 | 2824 | * We should guarantee that tid and kmem_cache are retrieved on |
923717cb | 2825 | * the same cpu. It could be different if CONFIG_PREEMPTION so we need |
9aabf810 | 2826 | * to check if it is matched or not. |
8a5ec0ba | 2827 | */ |
9aabf810 JK |
2828 | do { |
2829 | tid = this_cpu_read(s->cpu_slab->tid); | |
2830 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 2831 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 2832 | unlikely(tid != READ_ONCE(c->tid))); |
9aabf810 JK |
2833 | |
2834 | /* | |
2835 | * Irqless object alloc/free algorithm used here depends on sequence | |
2836 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2837 | * on c to guarantee that object and page associated with previous tid | |
2838 | * won't be used with current tid. If we fetch tid first, object and | |
2839 | * page could be one associated with next tid and our alloc/free | |
2840 | * request will be failed. In this case, we will retry. So, no problem. | |
2841 | */ | |
2842 | barrier(); | |
8a5ec0ba | 2843 | |
8a5ec0ba CL |
2844 | /* |
2845 | * The transaction ids are globally unique per cpu and per operation on | |
2846 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2847 | * occurs on the right processor and that there was no operation on the | |
2848 | * linked list in between. | |
2849 | */ | |
8a5ec0ba | 2850 | |
9dfc6e68 | 2851 | object = c->freelist; |
57d437d2 | 2852 | page = c->page; |
22e4663e | 2853 | if (unlikely(!object || !page || !node_match(page, node))) { |
dfb4f096 | 2854 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 | 2855 | } else { |
0ad9500e ED |
2856 | void *next_object = get_freepointer_safe(s, object); |
2857 | ||
8a5ec0ba | 2858 | /* |
25985edc | 2859 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2860 | * operation and if we are on the right processor. |
2861 | * | |
d0e0ac97 CG |
2862 | * The cmpxchg does the following atomically (without lock |
2863 | * semantics!) | |
8a5ec0ba CL |
2864 | * 1. Relocate first pointer to the current per cpu area. |
2865 | * 2. Verify that tid and freelist have not been changed | |
2866 | * 3. If they were not changed replace tid and freelist | |
2867 | * | |
d0e0ac97 CG |
2868 | * Since this is without lock semantics the protection is only |
2869 | * against code executing on this cpu *not* from access by | |
2870 | * other cpus. | |
8a5ec0ba | 2871 | */ |
933393f5 | 2872 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2873 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2874 | object, tid, | |
0ad9500e | 2875 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2876 | |
2877 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2878 | goto redo; | |
2879 | } | |
0ad9500e | 2880 | prefetch_freepointer(s, next_object); |
84e554e6 | 2881 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2882 | } |
0f181f9f | 2883 | |
ce5716c6 | 2884 | maybe_wipe_obj_freeptr(s, object); |
8a5ec0ba | 2885 | |
6471384a | 2886 | if (unlikely(slab_want_init_on_alloc(gfpflags, s)) && object) |
aa1ef4d7 | 2887 | memset(kasan_reset_tag(object), 0, s->object_size); |
d07dbea4 | 2888 | |
964d4bd3 | 2889 | slab_post_alloc_hook(s, objcg, gfpflags, 1, &object); |
5a896d9e | 2890 | |
894b8788 | 2891 | return object; |
81819f0f CL |
2892 | } |
2893 | ||
2b847c3c EG |
2894 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2895 | gfp_t gfpflags, unsigned long addr) | |
2896 | { | |
2897 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2898 | } | |
2899 | ||
81819f0f CL |
2900 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2901 | { | |
2b847c3c | 2902 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2903 | |
d0e0ac97 CG |
2904 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2905 | s->size, gfpflags); | |
5b882be4 EGM |
2906 | |
2907 | return ret; | |
81819f0f CL |
2908 | } |
2909 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2910 | ||
0f24f128 | 2911 | #ifdef CONFIG_TRACING |
4a92379b RK |
2912 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2913 | { | |
2b847c3c | 2914 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2915 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0116523c | 2916 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2917 | return ret; |
2918 | } | |
2919 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2920 | #endif |
2921 | ||
81819f0f CL |
2922 | #ifdef CONFIG_NUMA |
2923 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2924 | { | |
2b847c3c | 2925 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2926 | |
ca2b84cb | 2927 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2928 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2929 | |
2930 | return ret; | |
81819f0f CL |
2931 | } |
2932 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2933 | |
0f24f128 | 2934 | #ifdef CONFIG_TRACING |
4a92379b | 2935 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2936 | gfp_t gfpflags, |
4a92379b | 2937 | int node, size_t size) |
5b882be4 | 2938 | { |
2b847c3c | 2939 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2940 | |
2941 | trace_kmalloc_node(_RET_IP_, ret, | |
2942 | size, s->size, gfpflags, node); | |
0316bec2 | 2943 | |
0116523c | 2944 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2945 | return ret; |
5b882be4 | 2946 | } |
4a92379b | 2947 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2948 | #endif |
6dfd1b65 | 2949 | #endif /* CONFIG_NUMA */ |
5b882be4 | 2950 | |
81819f0f | 2951 | /* |
94e4d712 | 2952 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2953 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2954 | * |
894b8788 CL |
2955 | * So we still attempt to reduce cache line usage. Just take the slab |
2956 | * lock and free the item. If there is no additional partial page | |
2957 | * handling required then we can return immediately. | |
81819f0f | 2958 | */ |
894b8788 | 2959 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2960 | void *head, void *tail, int cnt, |
2961 | unsigned long addr) | |
2962 | ||
81819f0f CL |
2963 | { |
2964 | void *prior; | |
2cfb7455 | 2965 | int was_frozen; |
2cfb7455 CL |
2966 | struct page new; |
2967 | unsigned long counters; | |
2968 | struct kmem_cache_node *n = NULL; | |
3f649ab7 | 2969 | unsigned long flags; |
81819f0f | 2970 | |
8a5ec0ba | 2971 | stat(s, FREE_SLOWPATH); |
81819f0f | 2972 | |
19c7ff9e | 2973 | if (kmem_cache_debug(s) && |
282acb43 | 2974 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2975 | return; |
6446faa2 | 2976 | |
2cfb7455 | 2977 | do { |
837d678d JK |
2978 | if (unlikely(n)) { |
2979 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2980 | n = NULL; | |
2981 | } | |
2cfb7455 CL |
2982 | prior = page->freelist; |
2983 | counters = page->counters; | |
81084651 | 2984 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2985 | new.counters = counters; |
2986 | was_frozen = new.frozen; | |
81084651 | 2987 | new.inuse -= cnt; |
837d678d | 2988 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2989 | |
c65c1877 | 2990 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2991 | |
2992 | /* | |
d0e0ac97 CG |
2993 | * Slab was on no list before and will be |
2994 | * partially empty | |
2995 | * We can defer the list move and instead | |
2996 | * freeze it. | |
49e22585 CL |
2997 | */ |
2998 | new.frozen = 1; | |
2999 | ||
c65c1877 | 3000 | } else { /* Needs to be taken off a list */ |
49e22585 | 3001 | |
b455def2 | 3002 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
3003 | /* |
3004 | * Speculatively acquire the list_lock. | |
3005 | * If the cmpxchg does not succeed then we may | |
3006 | * drop the list_lock without any processing. | |
3007 | * | |
3008 | * Otherwise the list_lock will synchronize with | |
3009 | * other processors updating the list of slabs. | |
3010 | */ | |
3011 | spin_lock_irqsave(&n->list_lock, flags); | |
3012 | ||
3013 | } | |
2cfb7455 | 3014 | } |
81819f0f | 3015 | |
2cfb7455 CL |
3016 | } while (!cmpxchg_double_slab(s, page, |
3017 | prior, counters, | |
81084651 | 3018 | head, new.counters, |
2cfb7455 | 3019 | "__slab_free")); |
81819f0f | 3020 | |
2cfb7455 | 3021 | if (likely(!n)) { |
49e22585 | 3022 | |
c270cf30 AW |
3023 | if (likely(was_frozen)) { |
3024 | /* | |
3025 | * The list lock was not taken therefore no list | |
3026 | * activity can be necessary. | |
3027 | */ | |
3028 | stat(s, FREE_FROZEN); | |
3029 | } else if (new.frozen) { | |
3030 | /* | |
3031 | * If we just froze the page then put it onto the | |
3032 | * per cpu partial list. | |
3033 | */ | |
49e22585 | 3034 | put_cpu_partial(s, page, 1); |
8028dcea AS |
3035 | stat(s, CPU_PARTIAL_FREE); |
3036 | } | |
c270cf30 | 3037 | |
b455def2 L |
3038 | return; |
3039 | } | |
81819f0f | 3040 | |
8a5b20ae | 3041 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
3042 | goto slab_empty; |
3043 | ||
81819f0f | 3044 | /* |
837d678d JK |
3045 | * Objects left in the slab. If it was not on the partial list before |
3046 | * then add it. | |
81819f0f | 3047 | */ |
345c905d | 3048 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
a4d3f891 | 3049 | remove_full(s, n, page); |
837d678d JK |
3050 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
3051 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 3052 | } |
80f08c19 | 3053 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
3054 | return; |
3055 | ||
3056 | slab_empty: | |
a973e9dd | 3057 | if (prior) { |
81819f0f | 3058 | /* |
6fbabb20 | 3059 | * Slab on the partial list. |
81819f0f | 3060 | */ |
5cc6eee8 | 3061 | remove_partial(n, page); |
84e554e6 | 3062 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 3063 | } else { |
6fbabb20 | 3064 | /* Slab must be on the full list */ |
c65c1877 PZ |
3065 | remove_full(s, n, page); |
3066 | } | |
2cfb7455 | 3067 | |
80f08c19 | 3068 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 3069 | stat(s, FREE_SLAB); |
81819f0f | 3070 | discard_slab(s, page); |
81819f0f CL |
3071 | } |
3072 | ||
894b8788 CL |
3073 | /* |
3074 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
3075 | * can perform fastpath freeing without additional function calls. | |
3076 | * | |
3077 | * The fastpath is only possible if we are freeing to the current cpu slab | |
3078 | * of this processor. This typically the case if we have just allocated | |
3079 | * the item before. | |
3080 | * | |
3081 | * If fastpath is not possible then fall back to __slab_free where we deal | |
3082 | * with all sorts of special processing. | |
81084651 JDB |
3083 | * |
3084 | * Bulk free of a freelist with several objects (all pointing to the | |
3085 | * same page) possible by specifying head and tail ptr, plus objects | |
3086 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 3087 | */ |
80a9201a AP |
3088 | static __always_inline void do_slab_free(struct kmem_cache *s, |
3089 | struct page *page, void *head, void *tail, | |
3090 | int cnt, unsigned long addr) | |
894b8788 | 3091 | { |
81084651 | 3092 | void *tail_obj = tail ? : head; |
dfb4f096 | 3093 | struct kmem_cache_cpu *c; |
8a5ec0ba | 3094 | unsigned long tid; |
964d4bd3 | 3095 | |
d1b2cf6c | 3096 | memcg_slab_free_hook(s, &head, 1); |
8a5ec0ba CL |
3097 | redo: |
3098 | /* | |
3099 | * Determine the currently cpus per cpu slab. | |
3100 | * The cpu may change afterward. However that does not matter since | |
3101 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 3102 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 3103 | */ |
9aabf810 JK |
3104 | do { |
3105 | tid = this_cpu_read(s->cpu_slab->tid); | |
3106 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 3107 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 3108 | unlikely(tid != READ_ONCE(c->tid))); |
c016b0bd | 3109 | |
9aabf810 JK |
3110 | /* Same with comment on barrier() in slab_alloc_node() */ |
3111 | barrier(); | |
c016b0bd | 3112 | |
442b06bc | 3113 | if (likely(page == c->page)) { |
5076190d LT |
3114 | void **freelist = READ_ONCE(c->freelist); |
3115 | ||
3116 | set_freepointer(s, tail_obj, freelist); | |
8a5ec0ba | 3117 | |
933393f5 | 3118 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba | 3119 | s->cpu_slab->freelist, s->cpu_slab->tid, |
5076190d | 3120 | freelist, tid, |
81084651 | 3121 | head, next_tid(tid)))) { |
8a5ec0ba CL |
3122 | |
3123 | note_cmpxchg_failure("slab_free", s, tid); | |
3124 | goto redo; | |
3125 | } | |
84e554e6 | 3126 | stat(s, FREE_FASTPATH); |
894b8788 | 3127 | } else |
81084651 | 3128 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 3129 | |
894b8788 CL |
3130 | } |
3131 | ||
80a9201a AP |
3132 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
3133 | void *head, void *tail, int cnt, | |
3134 | unsigned long addr) | |
3135 | { | |
80a9201a | 3136 | /* |
c3895391 AK |
3137 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
3138 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 3139 | */ |
c3895391 AK |
3140 | if (slab_free_freelist_hook(s, &head, &tail)) |
3141 | do_slab_free(s, page, head, tail, cnt, addr); | |
80a9201a AP |
3142 | } |
3143 | ||
2bd926b4 | 3144 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
3145 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
3146 | { | |
3147 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
3148 | } | |
3149 | #endif | |
3150 | ||
81819f0f CL |
3151 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3152 | { | |
b9ce5ef4 GC |
3153 | s = cache_from_obj(s, x); |
3154 | if (!s) | |
79576102 | 3155 | return; |
81084651 | 3156 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 3157 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
3158 | } |
3159 | EXPORT_SYMBOL(kmem_cache_free); | |
3160 | ||
d0ecd894 | 3161 | struct detached_freelist { |
fbd02630 | 3162 | struct page *page; |
d0ecd894 JDB |
3163 | void *tail; |
3164 | void *freelist; | |
3165 | int cnt; | |
376bf125 | 3166 | struct kmem_cache *s; |
d0ecd894 | 3167 | }; |
fbd02630 | 3168 | |
d0ecd894 JDB |
3169 | /* |
3170 | * This function progressively scans the array with free objects (with | |
3171 | * a limited look ahead) and extract objects belonging to the same | |
3172 | * page. It builds a detached freelist directly within the given | |
3173 | * page/objects. This can happen without any need for | |
3174 | * synchronization, because the objects are owned by running process. | |
3175 | * The freelist is build up as a single linked list in the objects. | |
3176 | * The idea is, that this detached freelist can then be bulk | |
3177 | * transferred to the real freelist(s), but only requiring a single | |
3178 | * synchronization primitive. Look ahead in the array is limited due | |
3179 | * to performance reasons. | |
3180 | */ | |
376bf125 JDB |
3181 | static inline |
3182 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3183 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3184 | { |
3185 | size_t first_skipped_index = 0; | |
3186 | int lookahead = 3; | |
3187 | void *object; | |
ca257195 | 3188 | struct page *page; |
fbd02630 | 3189 | |
d0ecd894 JDB |
3190 | /* Always re-init detached_freelist */ |
3191 | df->page = NULL; | |
fbd02630 | 3192 | |
d0ecd894 JDB |
3193 | do { |
3194 | object = p[--size]; | |
ca257195 | 3195 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3196 | } while (!object && size); |
3eed034d | 3197 | |
d0ecd894 JDB |
3198 | if (!object) |
3199 | return 0; | |
fbd02630 | 3200 | |
ca257195 JDB |
3201 | page = virt_to_head_page(object); |
3202 | if (!s) { | |
3203 | /* Handle kalloc'ed objects */ | |
3204 | if (unlikely(!PageSlab(page))) { | |
3205 | BUG_ON(!PageCompound(page)); | |
3206 | kfree_hook(object); | |
4949148a | 3207 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3208 | p[size] = NULL; /* mark object processed */ |
3209 | return size; | |
3210 | } | |
3211 | /* Derive kmem_cache from object */ | |
3212 | df->s = page->slab_cache; | |
3213 | } else { | |
3214 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3215 | } | |
376bf125 | 3216 | |
d0ecd894 | 3217 | /* Start new detached freelist */ |
ca257195 | 3218 | df->page = page; |
376bf125 | 3219 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3220 | df->tail = object; |
3221 | df->freelist = object; | |
3222 | p[size] = NULL; /* mark object processed */ | |
3223 | df->cnt = 1; | |
3224 | ||
3225 | while (size) { | |
3226 | object = p[--size]; | |
3227 | if (!object) | |
3228 | continue; /* Skip processed objects */ | |
3229 | ||
3230 | /* df->page is always set at this point */ | |
3231 | if (df->page == virt_to_head_page(object)) { | |
3232 | /* Opportunity build freelist */ | |
376bf125 | 3233 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3234 | df->freelist = object; |
3235 | df->cnt++; | |
3236 | p[size] = NULL; /* mark object processed */ | |
3237 | ||
3238 | continue; | |
fbd02630 | 3239 | } |
d0ecd894 JDB |
3240 | |
3241 | /* Limit look ahead search */ | |
3242 | if (!--lookahead) | |
3243 | break; | |
3244 | ||
3245 | if (!first_skipped_index) | |
3246 | first_skipped_index = size + 1; | |
fbd02630 | 3247 | } |
d0ecd894 JDB |
3248 | |
3249 | return first_skipped_index; | |
3250 | } | |
3251 | ||
d0ecd894 | 3252 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3253 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3254 | { |
3255 | if (WARN_ON(!size)) | |
3256 | return; | |
3257 | ||
d1b2cf6c | 3258 | memcg_slab_free_hook(s, p, size); |
d0ecd894 JDB |
3259 | do { |
3260 | struct detached_freelist df; | |
3261 | ||
3262 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3263 | if (!df.page) |
d0ecd894 JDB |
3264 | continue; |
3265 | ||
376bf125 | 3266 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 3267 | } while (likely(size)); |
484748f0 CL |
3268 | } |
3269 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3270 | ||
994eb764 | 3271 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3272 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3273 | void **p) | |
484748f0 | 3274 | { |
994eb764 JDB |
3275 | struct kmem_cache_cpu *c; |
3276 | int i; | |
964d4bd3 | 3277 | struct obj_cgroup *objcg = NULL; |
994eb764 | 3278 | |
03ec0ed5 | 3279 | /* memcg and kmem_cache debug support */ |
964d4bd3 | 3280 | s = slab_pre_alloc_hook(s, &objcg, size, flags); |
03ec0ed5 JDB |
3281 | if (unlikely(!s)) |
3282 | return false; | |
994eb764 JDB |
3283 | /* |
3284 | * Drain objects in the per cpu slab, while disabling local | |
3285 | * IRQs, which protects against PREEMPT and interrupts | |
3286 | * handlers invoking normal fastpath. | |
3287 | */ | |
3288 | local_irq_disable(); | |
3289 | c = this_cpu_ptr(s->cpu_slab); | |
3290 | ||
3291 | for (i = 0; i < size; i++) { | |
3292 | void *object = c->freelist; | |
3293 | ||
ebe909e0 | 3294 | if (unlikely(!object)) { |
fd4d9c7d JH |
3295 | /* |
3296 | * We may have removed an object from c->freelist using | |
3297 | * the fastpath in the previous iteration; in that case, | |
3298 | * c->tid has not been bumped yet. | |
3299 | * Since ___slab_alloc() may reenable interrupts while | |
3300 | * allocating memory, we should bump c->tid now. | |
3301 | */ | |
3302 | c->tid = next_tid(c->tid); | |
3303 | ||
ebe909e0 JDB |
3304 | /* |
3305 | * Invoking slow path likely have side-effect | |
3306 | * of re-populating per CPU c->freelist | |
3307 | */ | |
87098373 | 3308 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3309 | _RET_IP_, c); |
87098373 CL |
3310 | if (unlikely(!p[i])) |
3311 | goto error; | |
3312 | ||
ebe909e0 | 3313 | c = this_cpu_ptr(s->cpu_slab); |
0f181f9f AP |
3314 | maybe_wipe_obj_freeptr(s, p[i]); |
3315 | ||
ebe909e0 JDB |
3316 | continue; /* goto for-loop */ |
3317 | } | |
994eb764 JDB |
3318 | c->freelist = get_freepointer(s, object); |
3319 | p[i] = object; | |
0f181f9f | 3320 | maybe_wipe_obj_freeptr(s, p[i]); |
994eb764 JDB |
3321 | } |
3322 | c->tid = next_tid(c->tid); | |
3323 | local_irq_enable(); | |
3324 | ||
3325 | /* Clear memory outside IRQ disabled fastpath loop */ | |
6471384a | 3326 | if (unlikely(slab_want_init_on_alloc(flags, s))) { |
994eb764 JDB |
3327 | int j; |
3328 | ||
3329 | for (j = 0; j < i; j++) | |
ce5716c6 | 3330 | memset(kasan_reset_tag(p[j]), 0, s->object_size); |
994eb764 JDB |
3331 | } |
3332 | ||
03ec0ed5 | 3333 | /* memcg and kmem_cache debug support */ |
964d4bd3 | 3334 | slab_post_alloc_hook(s, objcg, flags, size, p); |
865762a8 | 3335 | return i; |
87098373 | 3336 | error: |
87098373 | 3337 | local_irq_enable(); |
964d4bd3 | 3338 | slab_post_alloc_hook(s, objcg, flags, i, p); |
03ec0ed5 | 3339 | __kmem_cache_free_bulk(s, i, p); |
865762a8 | 3340 | return 0; |
484748f0 CL |
3341 | } |
3342 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3343 | ||
3344 | ||
81819f0f | 3345 | /* |
672bba3a CL |
3346 | * Object placement in a slab is made very easy because we always start at |
3347 | * offset 0. If we tune the size of the object to the alignment then we can | |
3348 | * get the required alignment by putting one properly sized object after | |
3349 | * another. | |
81819f0f CL |
3350 | * |
3351 | * Notice that the allocation order determines the sizes of the per cpu | |
3352 | * caches. Each processor has always one slab available for allocations. | |
3353 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3354 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3355 | * locking overhead. |
81819f0f CL |
3356 | */ |
3357 | ||
3358 | /* | |
3359 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3360 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3361 | * and increases the number of allocations possible without having to | |
3362 | * take the list_lock. | |
3363 | */ | |
19af27af AD |
3364 | static unsigned int slub_min_order; |
3365 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3366 | static unsigned int slub_min_objects; | |
81819f0f | 3367 | |
81819f0f CL |
3368 | /* |
3369 | * Calculate the order of allocation given an slab object size. | |
3370 | * | |
672bba3a CL |
3371 | * The order of allocation has significant impact on performance and other |
3372 | * system components. Generally order 0 allocations should be preferred since | |
3373 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3374 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3375 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3376 | * would be wasted. |
3377 | * | |
3378 | * In order to reach satisfactory performance we must ensure that a minimum | |
3379 | * number of objects is in one slab. Otherwise we may generate too much | |
3380 | * activity on the partial lists which requires taking the list_lock. This is | |
3381 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3382 | * |
672bba3a CL |
3383 | * slub_max_order specifies the order where we begin to stop considering the |
3384 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3385 | * we try to keep the page order as low as possible. So we accept more waste | |
3386 | * of space in favor of a small page order. | |
81819f0f | 3387 | * |
672bba3a CL |
3388 | * Higher order allocations also allow the placement of more objects in a |
3389 | * slab and thereby reduce object handling overhead. If the user has | |
3390 | * requested a higher mininum order then we start with that one instead of | |
3391 | * the smallest order which will fit the object. | |
81819f0f | 3392 | */ |
19af27af AD |
3393 | static inline unsigned int slab_order(unsigned int size, |
3394 | unsigned int min_objects, unsigned int max_order, | |
9736d2a9 | 3395 | unsigned int fract_leftover) |
81819f0f | 3396 | { |
19af27af AD |
3397 | unsigned int min_order = slub_min_order; |
3398 | unsigned int order; | |
81819f0f | 3399 | |
9736d2a9 | 3400 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3401 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3402 | |
9736d2a9 | 3403 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3404 | order <= max_order; order++) { |
81819f0f | 3405 | |
19af27af AD |
3406 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3407 | unsigned int rem; | |
81819f0f | 3408 | |
9736d2a9 | 3409 | rem = slab_size % size; |
81819f0f | 3410 | |
5e6d444e | 3411 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3412 | break; |
81819f0f | 3413 | } |
672bba3a | 3414 | |
81819f0f CL |
3415 | return order; |
3416 | } | |
3417 | ||
9736d2a9 | 3418 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3419 | { |
19af27af AD |
3420 | unsigned int order; |
3421 | unsigned int min_objects; | |
3422 | unsigned int max_objects; | |
3286222f | 3423 | unsigned int nr_cpus; |
5e6d444e CL |
3424 | |
3425 | /* | |
3426 | * Attempt to find best configuration for a slab. This | |
3427 | * works by first attempting to generate a layout with | |
3428 | * the best configuration and backing off gradually. | |
3429 | * | |
422ff4d7 | 3430 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3431 | * we reduce the minimum objects required in a slab. |
3432 | */ | |
3433 | min_objects = slub_min_objects; | |
3286222f VB |
3434 | if (!min_objects) { |
3435 | /* | |
3436 | * Some architectures will only update present cpus when | |
3437 | * onlining them, so don't trust the number if it's just 1. But | |
3438 | * we also don't want to use nr_cpu_ids always, as on some other | |
3439 | * architectures, there can be many possible cpus, but never | |
3440 | * onlined. Here we compromise between trying to avoid too high | |
3441 | * order on systems that appear larger than they are, and too | |
3442 | * low order on systems that appear smaller than they are. | |
3443 | */ | |
3444 | nr_cpus = num_present_cpus(); | |
3445 | if (nr_cpus <= 1) | |
3446 | nr_cpus = nr_cpu_ids; | |
3447 | min_objects = 4 * (fls(nr_cpus) + 1); | |
3448 | } | |
9736d2a9 | 3449 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3450 | min_objects = min(min_objects, max_objects); |
3451 | ||
5e6d444e | 3452 | while (min_objects > 1) { |
19af27af AD |
3453 | unsigned int fraction; |
3454 | ||
c124f5b5 | 3455 | fraction = 16; |
5e6d444e CL |
3456 | while (fraction >= 4) { |
3457 | order = slab_order(size, min_objects, | |
9736d2a9 | 3458 | slub_max_order, fraction); |
5e6d444e CL |
3459 | if (order <= slub_max_order) |
3460 | return order; | |
3461 | fraction /= 2; | |
3462 | } | |
5086c389 | 3463 | min_objects--; |
5e6d444e CL |
3464 | } |
3465 | ||
3466 | /* | |
3467 | * We were unable to place multiple objects in a slab. Now | |
3468 | * lets see if we can place a single object there. | |
3469 | */ | |
9736d2a9 | 3470 | order = slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3471 | if (order <= slub_max_order) |
3472 | return order; | |
3473 | ||
3474 | /* | |
3475 | * Doh this slab cannot be placed using slub_max_order. | |
3476 | */ | |
9736d2a9 | 3477 | order = slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3478 | if (order < MAX_ORDER) |
5e6d444e CL |
3479 | return order; |
3480 | return -ENOSYS; | |
3481 | } | |
3482 | ||
5595cffc | 3483 | static void |
4053497d | 3484 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3485 | { |
3486 | n->nr_partial = 0; | |
81819f0f CL |
3487 | spin_lock_init(&n->list_lock); |
3488 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3489 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3490 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3491 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3492 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3493 | #endif |
81819f0f CL |
3494 | } |
3495 | ||
55136592 | 3496 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3497 | { |
6c182dc0 | 3498 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3499 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3500 | |
8a5ec0ba | 3501 | /* |
d4d84fef CM |
3502 | * Must align to double word boundary for the double cmpxchg |
3503 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3504 | */ |
d4d84fef CM |
3505 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3506 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3507 | |
3508 | if (!s->cpu_slab) | |
3509 | return 0; | |
3510 | ||
3511 | init_kmem_cache_cpus(s); | |
4c93c355 | 3512 | |
8a5ec0ba | 3513 | return 1; |
4c93c355 | 3514 | } |
4c93c355 | 3515 | |
51df1142 CL |
3516 | static struct kmem_cache *kmem_cache_node; |
3517 | ||
81819f0f CL |
3518 | /* |
3519 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3520 | * slab on the node for this slabcache. There are no concurrent accesses | |
3521 | * possible. | |
3522 | * | |
721ae22a ZYW |
3523 | * Note that this function only works on the kmem_cache_node |
3524 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3525 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3526 | */ |
55136592 | 3527 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3528 | { |
3529 | struct page *page; | |
3530 | struct kmem_cache_node *n; | |
3531 | ||
51df1142 | 3532 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3533 | |
51df1142 | 3534 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3535 | |
3536 | BUG_ON(!page); | |
a2f92ee7 | 3537 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3538 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3539 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3540 | } |
3541 | ||
81819f0f CL |
3542 | n = page->freelist; |
3543 | BUG_ON(!n); | |
8ab1372f | 3544 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3545 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3546 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3547 | #endif |
12b22386 | 3548 | n = kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
505f5dcb | 3549 | GFP_KERNEL); |
12b22386 AK |
3550 | page->freelist = get_freepointer(kmem_cache_node, n); |
3551 | page->inuse = 1; | |
3552 | page->frozen = 0; | |
3553 | kmem_cache_node->node[node] = n; | |
4053497d | 3554 | init_kmem_cache_node(n); |
51df1142 | 3555 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3556 | |
67b6c900 | 3557 | /* |
1e4dd946 SR |
3558 | * No locks need to be taken here as it has just been |
3559 | * initialized and there is no concurrent access. | |
67b6c900 | 3560 | */ |
1e4dd946 | 3561 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3562 | } |
3563 | ||
3564 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3565 | { | |
3566 | int node; | |
fa45dc25 | 3567 | struct kmem_cache_node *n; |
81819f0f | 3568 | |
fa45dc25 | 3569 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3570 | s->node[node] = NULL; |
ea37df54 | 3571 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3572 | } |
3573 | } | |
3574 | ||
52b4b950 DS |
3575 | void __kmem_cache_release(struct kmem_cache *s) |
3576 | { | |
210e7a43 | 3577 | cache_random_seq_destroy(s); |
52b4b950 DS |
3578 | free_percpu(s->cpu_slab); |
3579 | free_kmem_cache_nodes(s); | |
3580 | } | |
3581 | ||
55136592 | 3582 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3583 | { |
3584 | int node; | |
81819f0f | 3585 | |
f64dc58c | 3586 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3587 | struct kmem_cache_node *n; |
3588 | ||
73367bd8 | 3589 | if (slab_state == DOWN) { |
55136592 | 3590 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3591 | continue; |
3592 | } | |
51df1142 | 3593 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3594 | GFP_KERNEL, node); |
81819f0f | 3595 | |
73367bd8 AD |
3596 | if (!n) { |
3597 | free_kmem_cache_nodes(s); | |
3598 | return 0; | |
81819f0f | 3599 | } |
73367bd8 | 3600 | |
4053497d | 3601 | init_kmem_cache_node(n); |
ea37df54 | 3602 | s->node[node] = n; |
81819f0f CL |
3603 | } |
3604 | return 1; | |
3605 | } | |
81819f0f | 3606 | |
c0bdb232 | 3607 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3608 | { |
3609 | if (min < MIN_PARTIAL) | |
3610 | min = MIN_PARTIAL; | |
3611 | else if (min > MAX_PARTIAL) | |
3612 | min = MAX_PARTIAL; | |
3613 | s->min_partial = min; | |
3614 | } | |
3615 | ||
e6d0e1dc WY |
3616 | static void set_cpu_partial(struct kmem_cache *s) |
3617 | { | |
3618 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3619 | /* | |
3620 | * cpu_partial determined the maximum number of objects kept in the | |
3621 | * per cpu partial lists of a processor. | |
3622 | * | |
3623 | * Per cpu partial lists mainly contain slabs that just have one | |
3624 | * object freed. If they are used for allocation then they can be | |
3625 | * filled up again with minimal effort. The slab will never hit the | |
3626 | * per node partial lists and therefore no locking will be required. | |
3627 | * | |
3628 | * This setting also determines | |
3629 | * | |
3630 | * A) The number of objects from per cpu partial slabs dumped to the | |
3631 | * per node list when we reach the limit. | |
3632 | * B) The number of objects in cpu partial slabs to extract from the | |
3633 | * per node list when we run out of per cpu objects. We only fetch | |
3634 | * 50% to keep some capacity around for frees. | |
3635 | */ | |
3636 | if (!kmem_cache_has_cpu_partial(s)) | |
bbd4e305 | 3637 | slub_set_cpu_partial(s, 0); |
e6d0e1dc | 3638 | else if (s->size >= PAGE_SIZE) |
bbd4e305 | 3639 | slub_set_cpu_partial(s, 2); |
e6d0e1dc | 3640 | else if (s->size >= 1024) |
bbd4e305 | 3641 | slub_set_cpu_partial(s, 6); |
e6d0e1dc | 3642 | else if (s->size >= 256) |
bbd4e305 | 3643 | slub_set_cpu_partial(s, 13); |
e6d0e1dc | 3644 | else |
bbd4e305 | 3645 | slub_set_cpu_partial(s, 30); |
e6d0e1dc WY |
3646 | #endif |
3647 | } | |
3648 | ||
81819f0f CL |
3649 | /* |
3650 | * calculate_sizes() determines the order and the distribution of data within | |
3651 | * a slab object. | |
3652 | */ | |
06b285dc | 3653 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 3654 | { |
d50112ed | 3655 | slab_flags_t flags = s->flags; |
be4a7988 | 3656 | unsigned int size = s->object_size; |
89b83f28 | 3657 | unsigned int freepointer_area; |
19af27af | 3658 | unsigned int order; |
81819f0f | 3659 | |
d8b42bf5 CL |
3660 | /* |
3661 | * Round up object size to the next word boundary. We can only | |
3662 | * place the free pointer at word boundaries and this determines | |
3663 | * the possible location of the free pointer. | |
3664 | */ | |
3665 | size = ALIGN(size, sizeof(void *)); | |
89b83f28 KC |
3666 | /* |
3667 | * This is the area of the object where a freepointer can be | |
3668 | * safely written. If redzoning adds more to the inuse size, we | |
3669 | * can't use that portion for writing the freepointer, so | |
3670 | * s->offset must be limited within this for the general case. | |
3671 | */ | |
3672 | freepointer_area = size; | |
d8b42bf5 CL |
3673 | |
3674 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3675 | /* |
3676 | * Determine if we can poison the object itself. If the user of | |
3677 | * the slab may touch the object after free or before allocation | |
3678 | * then we should never poison the object itself. | |
3679 | */ | |
5f0d5a3a | 3680 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3681 | !s->ctor) |
81819f0f CL |
3682 | s->flags |= __OBJECT_POISON; |
3683 | else | |
3684 | s->flags &= ~__OBJECT_POISON; | |
3685 | ||
81819f0f CL |
3686 | |
3687 | /* | |
672bba3a | 3688 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3689 | * end of the object and the free pointer. If not then add an |
672bba3a | 3690 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3691 | */ |
3b0efdfa | 3692 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3693 | size += sizeof(void *); |
41ecc55b | 3694 | #endif |
81819f0f CL |
3695 | |
3696 | /* | |
672bba3a CL |
3697 | * With that we have determined the number of bytes in actual use |
3698 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3699 | */ |
3700 | s->inuse = size; | |
3701 | ||
5f0d5a3a | 3702 | if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
c59def9f | 3703 | s->ctor)) { |
81819f0f CL |
3704 | /* |
3705 | * Relocate free pointer after the object if it is not | |
3706 | * permitted to overwrite the first word of the object on | |
3707 | * kmem_cache_free. | |
3708 | * | |
3709 | * This is the case if we do RCU, have a constructor or | |
3710 | * destructor or are poisoning the objects. | |
cbfc35a4 WL |
3711 | * |
3712 | * The assumption that s->offset >= s->inuse means free | |
3713 | * pointer is outside of the object is used in the | |
3714 | * freeptr_outside_object() function. If that is no | |
3715 | * longer true, the function needs to be modified. | |
81819f0f CL |
3716 | */ |
3717 | s->offset = size; | |
3718 | size += sizeof(void *); | |
89b83f28 | 3719 | } else if (freepointer_area > sizeof(void *)) { |
3202fa62 KC |
3720 | /* |
3721 | * Store freelist pointer near middle of object to keep | |
3722 | * it away from the edges of the object to avoid small | |
3723 | * sized over/underflows from neighboring allocations. | |
3724 | */ | |
89b83f28 | 3725 | s->offset = ALIGN(freepointer_area / 2, sizeof(void *)); |
81819f0f CL |
3726 | } |
3727 | ||
c12b3c62 | 3728 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3729 | if (flags & SLAB_STORE_USER) |
3730 | /* | |
3731 | * Need to store information about allocs and frees after | |
3732 | * the object. | |
3733 | */ | |
3734 | size += 2 * sizeof(struct track); | |
80a9201a | 3735 | #endif |
81819f0f | 3736 | |
80a9201a AP |
3737 | kasan_cache_create(s, &size, &s->flags); |
3738 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3739 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3740 | /* |
3741 | * Add some empty padding so that we can catch | |
3742 | * overwrites from earlier objects rather than let | |
3743 | * tracking information or the free pointer be | |
0211a9c8 | 3744 | * corrupted if a user writes before the start |
81819f0f CL |
3745 | * of the object. |
3746 | */ | |
3747 | size += sizeof(void *); | |
d86bd1be JK |
3748 | |
3749 | s->red_left_pad = sizeof(void *); | |
3750 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3751 | size += s->red_left_pad; | |
3752 | } | |
41ecc55b | 3753 | #endif |
672bba3a | 3754 | |
81819f0f CL |
3755 | /* |
3756 | * SLUB stores one object immediately after another beginning from | |
3757 | * offset 0. In order to align the objects we have to simply size | |
3758 | * each object to conform to the alignment. | |
3759 | */ | |
45906855 | 3760 | size = ALIGN(size, s->align); |
81819f0f | 3761 | s->size = size; |
4138fdfc | 3762 | s->reciprocal_size = reciprocal_value(size); |
06b285dc CL |
3763 | if (forced_order >= 0) |
3764 | order = forced_order; | |
3765 | else | |
9736d2a9 | 3766 | order = calculate_order(size); |
81819f0f | 3767 | |
19af27af | 3768 | if ((int)order < 0) |
81819f0f CL |
3769 | return 0; |
3770 | ||
b7a49f0d | 3771 | s->allocflags = 0; |
834f3d11 | 3772 | if (order) |
b7a49f0d CL |
3773 | s->allocflags |= __GFP_COMP; |
3774 | ||
3775 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3776 | s->allocflags |= GFP_DMA; |
b7a49f0d | 3777 | |
6d6ea1e9 NB |
3778 | if (s->flags & SLAB_CACHE_DMA32) |
3779 | s->allocflags |= GFP_DMA32; | |
3780 | ||
b7a49f0d CL |
3781 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
3782 | s->allocflags |= __GFP_RECLAIMABLE; | |
3783 | ||
81819f0f CL |
3784 | /* |
3785 | * Determine the number of objects per slab | |
3786 | */ | |
9736d2a9 MW |
3787 | s->oo = oo_make(order, size); |
3788 | s->min = oo_make(get_order(size), size); | |
205ab99d CL |
3789 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3790 | s->max = s->oo; | |
81819f0f | 3791 | |
834f3d11 | 3792 | return !!oo_objects(s->oo); |
81819f0f CL |
3793 | } |
3794 | ||
d50112ed | 3795 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 3796 | { |
f505bc0c | 3797 | s->flags = kmem_cache_flags(s->size, flags, s->name); |
2482ddec KC |
3798 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
3799 | s->random = get_random_long(); | |
3800 | #endif | |
81819f0f | 3801 | |
06b285dc | 3802 | if (!calculate_sizes(s, -1)) |
81819f0f | 3803 | goto error; |
3de47213 DR |
3804 | if (disable_higher_order_debug) { |
3805 | /* | |
3806 | * Disable debugging flags that store metadata if the min slab | |
3807 | * order increased. | |
3808 | */ | |
3b0efdfa | 3809 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3810 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3811 | s->offset = 0; | |
3812 | if (!calculate_sizes(s, -1)) | |
3813 | goto error; | |
3814 | } | |
3815 | } | |
81819f0f | 3816 | |
2565409f HC |
3817 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3818 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3819 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3820 | /* Enable fast mode */ |
3821 | s->flags |= __CMPXCHG_DOUBLE; | |
3822 | #endif | |
3823 | ||
3b89d7d8 DR |
3824 | /* |
3825 | * The larger the object size is, the more pages we want on the partial | |
3826 | * list to avoid pounding the page allocator excessively. | |
3827 | */ | |
49e22585 CL |
3828 | set_min_partial(s, ilog2(s->size) / 2); |
3829 | ||
e6d0e1dc | 3830 | set_cpu_partial(s); |
49e22585 | 3831 | |
81819f0f | 3832 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3833 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3834 | #endif |
210e7a43 TG |
3835 | |
3836 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3837 | if (slab_state >= UP) { | |
3838 | if (init_cache_random_seq(s)) | |
3839 | goto error; | |
3840 | } | |
3841 | ||
55136592 | 3842 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3843 | goto error; |
81819f0f | 3844 | |
55136592 | 3845 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3846 | return 0; |
ff12059e | 3847 | |
4c93c355 | 3848 | free_kmem_cache_nodes(s); |
81819f0f | 3849 | error: |
278b1bb1 | 3850 | return -EINVAL; |
81819f0f | 3851 | } |
81819f0f | 3852 | |
33b12c38 | 3853 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
55860d96 | 3854 | const char *text) |
33b12c38 CL |
3855 | { |
3856 | #ifdef CONFIG_SLUB_DEBUG | |
3857 | void *addr = page_address(page); | |
55860d96 | 3858 | unsigned long *map; |
33b12c38 | 3859 | void *p; |
aa456c7a | 3860 | |
945cf2b6 | 3861 | slab_err(s, page, text, s->name); |
33b12c38 | 3862 | slab_lock(page); |
33b12c38 | 3863 | |
90e9f6a6 | 3864 | map = get_map(s, page); |
33b12c38 CL |
3865 | for_each_object(p, s, addr, page->objects) { |
3866 | ||
4138fdfc | 3867 | if (!test_bit(__obj_to_index(s, addr, p), map)) { |
f9f58285 | 3868 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3869 | print_tracking(s, p); |
3870 | } | |
3871 | } | |
55860d96 | 3872 | put_map(map); |
33b12c38 CL |
3873 | slab_unlock(page); |
3874 | #endif | |
3875 | } | |
3876 | ||
81819f0f | 3877 | /* |
599870b1 | 3878 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3879 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3880 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3881 | */ |
599870b1 | 3882 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3883 | { |
60398923 | 3884 | LIST_HEAD(discard); |
81819f0f CL |
3885 | struct page *page, *h; |
3886 | ||
52b4b950 DS |
3887 | BUG_ON(irqs_disabled()); |
3888 | spin_lock_irq(&n->list_lock); | |
916ac052 | 3889 | list_for_each_entry_safe(page, h, &n->partial, slab_list) { |
81819f0f | 3890 | if (!page->inuse) { |
52b4b950 | 3891 | remove_partial(n, page); |
916ac052 | 3892 | list_add(&page->slab_list, &discard); |
33b12c38 CL |
3893 | } else { |
3894 | list_slab_objects(s, page, | |
55860d96 | 3895 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3896 | } |
33b12c38 | 3897 | } |
52b4b950 | 3898 | spin_unlock_irq(&n->list_lock); |
60398923 | 3899 | |
916ac052 | 3900 | list_for_each_entry_safe(page, h, &discard, slab_list) |
60398923 | 3901 | discard_slab(s, page); |
81819f0f CL |
3902 | } |
3903 | ||
f9e13c0a SB |
3904 | bool __kmem_cache_empty(struct kmem_cache *s) |
3905 | { | |
3906 | int node; | |
3907 | struct kmem_cache_node *n; | |
3908 | ||
3909 | for_each_kmem_cache_node(s, node, n) | |
3910 | if (n->nr_partial || slabs_node(s, node)) | |
3911 | return false; | |
3912 | return true; | |
3913 | } | |
3914 | ||
81819f0f | 3915 | /* |
672bba3a | 3916 | * Release all resources used by a slab cache. |
81819f0f | 3917 | */ |
52b4b950 | 3918 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3919 | { |
3920 | int node; | |
fa45dc25 | 3921 | struct kmem_cache_node *n; |
81819f0f CL |
3922 | |
3923 | flush_all(s); | |
81819f0f | 3924 | /* Attempt to free all objects */ |
fa45dc25 | 3925 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3926 | free_partial(s, n); |
3927 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3928 | return 1; |
3929 | } | |
81819f0f CL |
3930 | return 0; |
3931 | } | |
3932 | ||
81819f0f CL |
3933 | /******************************************************************** |
3934 | * Kmalloc subsystem | |
3935 | *******************************************************************/ | |
3936 | ||
81819f0f CL |
3937 | static int __init setup_slub_min_order(char *str) |
3938 | { | |
19af27af | 3939 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
3940 | |
3941 | return 1; | |
3942 | } | |
3943 | ||
3944 | __setup("slub_min_order=", setup_slub_min_order); | |
3945 | ||
3946 | static int __init setup_slub_max_order(char *str) | |
3947 | { | |
19af27af AD |
3948 | get_option(&str, (int *)&slub_max_order); |
3949 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
3950 | |
3951 | return 1; | |
3952 | } | |
3953 | ||
3954 | __setup("slub_max_order=", setup_slub_max_order); | |
3955 | ||
3956 | static int __init setup_slub_min_objects(char *str) | |
3957 | { | |
19af27af | 3958 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
3959 | |
3960 | return 1; | |
3961 | } | |
3962 | ||
3963 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3964 | ||
81819f0f CL |
3965 | void *__kmalloc(size_t size, gfp_t flags) |
3966 | { | |
aadb4bc4 | 3967 | struct kmem_cache *s; |
5b882be4 | 3968 | void *ret; |
81819f0f | 3969 | |
95a05b42 | 3970 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3971 | return kmalloc_large(size, flags); |
aadb4bc4 | 3972 | |
2c59dd65 | 3973 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3974 | |
3975 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3976 | return s; |
3977 | ||
2b847c3c | 3978 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3979 | |
ca2b84cb | 3980 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3981 | |
0116523c | 3982 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3983 | |
5b882be4 | 3984 | return ret; |
81819f0f CL |
3985 | } |
3986 | EXPORT_SYMBOL(__kmalloc); | |
3987 | ||
5d1f57e4 | 3988 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3989 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3990 | { | |
b1eeab67 | 3991 | struct page *page; |
e4f7c0b4 | 3992 | void *ptr = NULL; |
6a486c0a | 3993 | unsigned int order = get_order(size); |
f619cfe1 | 3994 | |
75f296d9 | 3995 | flags |= __GFP_COMP; |
6a486c0a VB |
3996 | page = alloc_pages_node(node, flags, order); |
3997 | if (page) { | |
e4f7c0b4 | 3998 | ptr = page_address(page); |
ca3c4f41 MS |
3999 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, |
4000 | PAGE_SIZE << order); | |
6a486c0a | 4001 | } |
e4f7c0b4 | 4002 | |
0116523c | 4003 | return kmalloc_large_node_hook(ptr, size, flags); |
f619cfe1 CL |
4004 | } |
4005 | ||
81819f0f CL |
4006 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
4007 | { | |
aadb4bc4 | 4008 | struct kmem_cache *s; |
5b882be4 | 4009 | void *ret; |
81819f0f | 4010 | |
95a05b42 | 4011 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
4012 | ret = kmalloc_large_node(size, flags, node); |
4013 | ||
ca2b84cb EGM |
4014 | trace_kmalloc_node(_RET_IP_, ret, |
4015 | size, PAGE_SIZE << get_order(size), | |
4016 | flags, node); | |
5b882be4 EGM |
4017 | |
4018 | return ret; | |
4019 | } | |
aadb4bc4 | 4020 | |
2c59dd65 | 4021 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4022 | |
4023 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4024 | return s; |
4025 | ||
2b847c3c | 4026 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 4027 | |
ca2b84cb | 4028 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 4029 | |
0116523c | 4030 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4031 | |
5b882be4 | 4032 | return ret; |
81819f0f CL |
4033 | } |
4034 | EXPORT_SYMBOL(__kmalloc_node); | |
6dfd1b65 | 4035 | #endif /* CONFIG_NUMA */ |
81819f0f | 4036 | |
ed18adc1 KC |
4037 | #ifdef CONFIG_HARDENED_USERCOPY |
4038 | /* | |
afcc90f8 KC |
4039 | * Rejects incorrectly sized objects and objects that are to be copied |
4040 | * to/from userspace but do not fall entirely within the containing slab | |
4041 | * cache's usercopy region. | |
ed18adc1 KC |
4042 | * |
4043 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4044 | * to indicate an error. | |
4045 | */ | |
f4e6e289 KC |
4046 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
4047 | bool to_user) | |
ed18adc1 KC |
4048 | { |
4049 | struct kmem_cache *s; | |
44065b2e | 4050 | unsigned int offset; |
ed18adc1 KC |
4051 | size_t object_size; |
4052 | ||
96fedce2 AK |
4053 | ptr = kasan_reset_tag(ptr); |
4054 | ||
ed18adc1 KC |
4055 | /* Find object and usable object size. */ |
4056 | s = page->slab_cache; | |
ed18adc1 KC |
4057 | |
4058 | /* Reject impossible pointers. */ | |
4059 | if (ptr < page_address(page)) | |
f4e6e289 KC |
4060 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
4061 | to_user, 0, n); | |
ed18adc1 KC |
4062 | |
4063 | /* Find offset within object. */ | |
4064 | offset = (ptr - page_address(page)) % s->size; | |
4065 | ||
4066 | /* Adjust for redzone and reject if within the redzone. */ | |
59052e89 | 4067 | if (kmem_cache_debug_flags(s, SLAB_RED_ZONE)) { |
ed18adc1 | 4068 | if (offset < s->red_left_pad) |
f4e6e289 KC |
4069 | usercopy_abort("SLUB object in left red zone", |
4070 | s->name, to_user, offset, n); | |
ed18adc1 KC |
4071 | offset -= s->red_left_pad; |
4072 | } | |
4073 | ||
afcc90f8 KC |
4074 | /* Allow address range falling entirely within usercopy region. */ |
4075 | if (offset >= s->useroffset && | |
4076 | offset - s->useroffset <= s->usersize && | |
4077 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 4078 | return; |
ed18adc1 | 4079 | |
afcc90f8 KC |
4080 | /* |
4081 | * If the copy is still within the allocated object, produce | |
4082 | * a warning instead of rejecting the copy. This is intended | |
4083 | * to be a temporary method to find any missing usercopy | |
4084 | * whitelists. | |
4085 | */ | |
4086 | object_size = slab_ksize(s); | |
2d891fbc KC |
4087 | if (usercopy_fallback && |
4088 | offset <= object_size && n <= object_size - offset) { | |
afcc90f8 KC |
4089 | usercopy_warn("SLUB object", s->name, to_user, offset, n); |
4090 | return; | |
4091 | } | |
ed18adc1 | 4092 | |
f4e6e289 | 4093 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
4094 | } |
4095 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4096 | ||
10d1f8cb | 4097 | size_t __ksize(const void *object) |
81819f0f | 4098 | { |
272c1d21 | 4099 | struct page *page; |
81819f0f | 4100 | |
ef8b4520 | 4101 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
4102 | return 0; |
4103 | ||
294a80a8 | 4104 | page = virt_to_head_page(object); |
294a80a8 | 4105 | |
76994412 PE |
4106 | if (unlikely(!PageSlab(page))) { |
4107 | WARN_ON(!PageCompound(page)); | |
a50b854e | 4108 | return page_size(page); |
76994412 | 4109 | } |
81819f0f | 4110 | |
1b4f59e3 | 4111 | return slab_ksize(page->slab_cache); |
81819f0f | 4112 | } |
10d1f8cb | 4113 | EXPORT_SYMBOL(__ksize); |
81819f0f CL |
4114 | |
4115 | void kfree(const void *x) | |
4116 | { | |
81819f0f | 4117 | struct page *page; |
5bb983b0 | 4118 | void *object = (void *)x; |
81819f0f | 4119 | |
2121db74 PE |
4120 | trace_kfree(_RET_IP_, x); |
4121 | ||
2408c550 | 4122 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
4123 | return; |
4124 | ||
b49af68f | 4125 | page = virt_to_head_page(x); |
aadb4bc4 | 4126 | if (unlikely(!PageSlab(page))) { |
6a486c0a VB |
4127 | unsigned int order = compound_order(page); |
4128 | ||
0937502a | 4129 | BUG_ON(!PageCompound(page)); |
47adccce | 4130 | kfree_hook(object); |
ca3c4f41 MS |
4131 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, |
4132 | -(PAGE_SIZE << order)); | |
6a486c0a | 4133 | __free_pages(page, order); |
aadb4bc4 CL |
4134 | return; |
4135 | } | |
81084651 | 4136 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
4137 | } |
4138 | EXPORT_SYMBOL(kfree); | |
4139 | ||
832f37f5 VD |
4140 | #define SHRINK_PROMOTE_MAX 32 |
4141 | ||
2086d26a | 4142 | /* |
832f37f5 VD |
4143 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
4144 | * up most to the head of the partial lists. New allocations will then | |
4145 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
4146 | * |
4147 | * The slabs with the least items are placed last. This results in them | |
4148 | * being allocated from last increasing the chance that the last objects | |
4149 | * are freed in them. | |
2086d26a | 4150 | */ |
c9fc5864 | 4151 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
4152 | { |
4153 | int node; | |
4154 | int i; | |
4155 | struct kmem_cache_node *n; | |
4156 | struct page *page; | |
4157 | struct page *t; | |
832f37f5 VD |
4158 | struct list_head discard; |
4159 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 4160 | unsigned long flags; |
ce3712d7 | 4161 | int ret = 0; |
2086d26a | 4162 | |
2086d26a | 4163 | flush_all(s); |
fa45dc25 | 4164 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
4165 | INIT_LIST_HEAD(&discard); |
4166 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
4167 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
4168 | |
4169 | spin_lock_irqsave(&n->list_lock, flags); | |
4170 | ||
4171 | /* | |
832f37f5 | 4172 | * Build lists of slabs to discard or promote. |
2086d26a | 4173 | * |
672bba3a CL |
4174 | * Note that concurrent frees may occur while we hold the |
4175 | * list_lock. page->inuse here is the upper limit. | |
2086d26a | 4176 | */ |
916ac052 | 4177 | list_for_each_entry_safe(page, t, &n->partial, slab_list) { |
832f37f5 VD |
4178 | int free = page->objects - page->inuse; |
4179 | ||
4180 | /* Do not reread page->inuse */ | |
4181 | barrier(); | |
4182 | ||
4183 | /* We do not keep full slabs on the list */ | |
4184 | BUG_ON(free <= 0); | |
4185 | ||
4186 | if (free == page->objects) { | |
916ac052 | 4187 | list_move(&page->slab_list, &discard); |
69cb8e6b | 4188 | n->nr_partial--; |
832f37f5 | 4189 | } else if (free <= SHRINK_PROMOTE_MAX) |
916ac052 | 4190 | list_move(&page->slab_list, promote + free - 1); |
2086d26a CL |
4191 | } |
4192 | ||
2086d26a | 4193 | /* |
832f37f5 VD |
4194 | * Promote the slabs filled up most to the head of the |
4195 | * partial list. | |
2086d26a | 4196 | */ |
832f37f5 VD |
4197 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
4198 | list_splice(promote + i, &n->partial); | |
2086d26a | 4199 | |
2086d26a | 4200 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
4201 | |
4202 | /* Release empty slabs */ | |
916ac052 | 4203 | list_for_each_entry_safe(page, t, &discard, slab_list) |
69cb8e6b | 4204 | discard_slab(s, page); |
ce3712d7 VD |
4205 | |
4206 | if (slabs_node(s, node)) | |
4207 | ret = 1; | |
2086d26a CL |
4208 | } |
4209 | ||
ce3712d7 | 4210 | return ret; |
2086d26a | 4211 | } |
2086d26a | 4212 | |
b9049e23 YG |
4213 | static int slab_mem_going_offline_callback(void *arg) |
4214 | { | |
4215 | struct kmem_cache *s; | |
4216 | ||
18004c5d | 4217 | mutex_lock(&slab_mutex); |
b9049e23 | 4218 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4219 | __kmem_cache_shrink(s); |
18004c5d | 4220 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4221 | |
4222 | return 0; | |
4223 | } | |
4224 | ||
4225 | static void slab_mem_offline_callback(void *arg) | |
4226 | { | |
4227 | struct kmem_cache_node *n; | |
4228 | struct kmem_cache *s; | |
4229 | struct memory_notify *marg = arg; | |
4230 | int offline_node; | |
4231 | ||
b9d5ab25 | 4232 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4233 | |
4234 | /* | |
4235 | * If the node still has available memory. we need kmem_cache_node | |
4236 | * for it yet. | |
4237 | */ | |
4238 | if (offline_node < 0) | |
4239 | return; | |
4240 | ||
18004c5d | 4241 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4242 | list_for_each_entry(s, &slab_caches, list) { |
4243 | n = get_node(s, offline_node); | |
4244 | if (n) { | |
4245 | /* | |
4246 | * if n->nr_slabs > 0, slabs still exist on the node | |
4247 | * that is going down. We were unable to free them, | |
c9404c9c | 4248 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
4249 | * callback. So, we must fail. |
4250 | */ | |
0f389ec6 | 4251 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
4252 | |
4253 | s->node[offline_node] = NULL; | |
8de66a0c | 4254 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
4255 | } |
4256 | } | |
18004c5d | 4257 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4258 | } |
4259 | ||
4260 | static int slab_mem_going_online_callback(void *arg) | |
4261 | { | |
4262 | struct kmem_cache_node *n; | |
4263 | struct kmem_cache *s; | |
4264 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4265 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4266 | int ret = 0; |
4267 | ||
4268 | /* | |
4269 | * If the node's memory is already available, then kmem_cache_node is | |
4270 | * already created. Nothing to do. | |
4271 | */ | |
4272 | if (nid < 0) | |
4273 | return 0; | |
4274 | ||
4275 | /* | |
0121c619 | 4276 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4277 | * allocate a kmem_cache_node structure in order to bring the node |
4278 | * online. | |
4279 | */ | |
18004c5d | 4280 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4281 | list_for_each_entry(s, &slab_caches, list) { |
4282 | /* | |
4283 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4284 | * since memory is not yet available from the node that | |
4285 | * is brought up. | |
4286 | */ | |
8de66a0c | 4287 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4288 | if (!n) { |
4289 | ret = -ENOMEM; | |
4290 | goto out; | |
4291 | } | |
4053497d | 4292 | init_kmem_cache_node(n); |
b9049e23 YG |
4293 | s->node[nid] = n; |
4294 | } | |
4295 | out: | |
18004c5d | 4296 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4297 | return ret; |
4298 | } | |
4299 | ||
4300 | static int slab_memory_callback(struct notifier_block *self, | |
4301 | unsigned long action, void *arg) | |
4302 | { | |
4303 | int ret = 0; | |
4304 | ||
4305 | switch (action) { | |
4306 | case MEM_GOING_ONLINE: | |
4307 | ret = slab_mem_going_online_callback(arg); | |
4308 | break; | |
4309 | case MEM_GOING_OFFLINE: | |
4310 | ret = slab_mem_going_offline_callback(arg); | |
4311 | break; | |
4312 | case MEM_OFFLINE: | |
4313 | case MEM_CANCEL_ONLINE: | |
4314 | slab_mem_offline_callback(arg); | |
4315 | break; | |
4316 | case MEM_ONLINE: | |
4317 | case MEM_CANCEL_OFFLINE: | |
4318 | break; | |
4319 | } | |
dc19f9db KH |
4320 | if (ret) |
4321 | ret = notifier_from_errno(ret); | |
4322 | else | |
4323 | ret = NOTIFY_OK; | |
b9049e23 YG |
4324 | return ret; |
4325 | } | |
4326 | ||
3ac38faa AM |
4327 | static struct notifier_block slab_memory_callback_nb = { |
4328 | .notifier_call = slab_memory_callback, | |
4329 | .priority = SLAB_CALLBACK_PRI, | |
4330 | }; | |
b9049e23 | 4331 | |
81819f0f CL |
4332 | /******************************************************************** |
4333 | * Basic setup of slabs | |
4334 | *******************************************************************/ | |
4335 | ||
51df1142 CL |
4336 | /* |
4337 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4338 | * the page allocator. Allocate them properly then fix up the pointers |
4339 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4340 | */ |
4341 | ||
dffb4d60 | 4342 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4343 | { |
4344 | int node; | |
dffb4d60 | 4345 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4346 | struct kmem_cache_node *n; |
51df1142 | 4347 | |
dffb4d60 | 4348 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4349 | |
7d557b3c GC |
4350 | /* |
4351 | * This runs very early, and only the boot processor is supposed to be | |
4352 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4353 | * IPIs around. | |
4354 | */ | |
4355 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4356 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4357 | struct page *p; |
4358 | ||
916ac052 | 4359 | list_for_each_entry(p, &n->partial, slab_list) |
fa45dc25 | 4360 | p->slab_cache = s; |
51df1142 | 4361 | |
607bf324 | 4362 | #ifdef CONFIG_SLUB_DEBUG |
916ac052 | 4363 | list_for_each_entry(p, &n->full, slab_list) |
fa45dc25 | 4364 | p->slab_cache = s; |
51df1142 | 4365 | #endif |
51df1142 | 4366 | } |
dffb4d60 CL |
4367 | list_add(&s->list, &slab_caches); |
4368 | return s; | |
51df1142 CL |
4369 | } |
4370 | ||
81819f0f CL |
4371 | void __init kmem_cache_init(void) |
4372 | { | |
dffb4d60 CL |
4373 | static __initdata struct kmem_cache boot_kmem_cache, |
4374 | boot_kmem_cache_node; | |
51df1142 | 4375 | |
fc8d8620 SG |
4376 | if (debug_guardpage_minorder()) |
4377 | slub_max_order = 0; | |
4378 | ||
dffb4d60 CL |
4379 | kmem_cache_node = &boot_kmem_cache_node; |
4380 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4381 | |
dffb4d60 | 4382 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4383 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4384 | |
3ac38faa | 4385 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4386 | |
4387 | /* Able to allocate the per node structures */ | |
4388 | slab_state = PARTIAL; | |
4389 | ||
dffb4d60 CL |
4390 | create_boot_cache(kmem_cache, "kmem_cache", |
4391 | offsetof(struct kmem_cache, node) + | |
4392 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4393 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4394 | |
dffb4d60 | 4395 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4396 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4397 | |
4398 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4399 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4400 | create_kmalloc_caches(0); |
81819f0f | 4401 | |
210e7a43 TG |
4402 | /* Setup random freelists for each cache */ |
4403 | init_freelist_randomization(); | |
4404 | ||
a96a87bf SAS |
4405 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4406 | slub_cpu_dead); | |
81819f0f | 4407 | |
b9726c26 | 4408 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", |
f97d5f63 | 4409 | cache_line_size(), |
81819f0f CL |
4410 | slub_min_order, slub_max_order, slub_min_objects, |
4411 | nr_cpu_ids, nr_node_ids); | |
4412 | } | |
4413 | ||
7e85ee0c PE |
4414 | void __init kmem_cache_init_late(void) |
4415 | { | |
7e85ee0c PE |
4416 | } |
4417 | ||
2633d7a0 | 4418 | struct kmem_cache * |
f4957d5b | 4419 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4420 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4421 | { |
10befea9 | 4422 | struct kmem_cache *s; |
81819f0f | 4423 | |
a44cb944 | 4424 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4425 | if (s) { |
4426 | s->refcount++; | |
84d0ddd6 | 4427 | |
81819f0f CL |
4428 | /* |
4429 | * Adjust the object sizes so that we clear | |
4430 | * the complete object on kzalloc. | |
4431 | */ | |
1b473f29 | 4432 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4433 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4434 | |
7b8f3b66 | 4435 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4436 | s->refcount--; |
cbb79694 | 4437 | s = NULL; |
7b8f3b66 | 4438 | } |
a0e1d1be | 4439 | } |
6446faa2 | 4440 | |
cbb79694 CL |
4441 | return s; |
4442 | } | |
84c1cf62 | 4443 | |
d50112ed | 4444 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4445 | { |
aac3a166 PE |
4446 | int err; |
4447 | ||
4448 | err = kmem_cache_open(s, flags); | |
4449 | if (err) | |
4450 | return err; | |
20cea968 | 4451 | |
45530c44 CL |
4452 | /* Mutex is not taken during early boot */ |
4453 | if (slab_state <= UP) | |
4454 | return 0; | |
4455 | ||
aac3a166 | 4456 | err = sysfs_slab_add(s); |
aac3a166 | 4457 | if (err) |
52b4b950 | 4458 | __kmem_cache_release(s); |
20cea968 | 4459 | |
aac3a166 | 4460 | return err; |
81819f0f | 4461 | } |
81819f0f | 4462 | |
ce71e27c | 4463 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4464 | { |
aadb4bc4 | 4465 | struct kmem_cache *s; |
94b528d0 | 4466 | void *ret; |
aadb4bc4 | 4467 | |
95a05b42 | 4468 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4469 | return kmalloc_large(size, gfpflags); |
4470 | ||
2c59dd65 | 4471 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4472 | |
2408c550 | 4473 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4474 | return s; |
81819f0f | 4475 | |
2b847c3c | 4476 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4477 | |
25985edc | 4478 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4479 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4480 | |
4481 | return ret; | |
81819f0f | 4482 | } |
fd7cb575 | 4483 | EXPORT_SYMBOL(__kmalloc_track_caller); |
81819f0f | 4484 | |
5d1f57e4 | 4485 | #ifdef CONFIG_NUMA |
81819f0f | 4486 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4487 | int node, unsigned long caller) |
81819f0f | 4488 | { |
aadb4bc4 | 4489 | struct kmem_cache *s; |
94b528d0 | 4490 | void *ret; |
aadb4bc4 | 4491 | |
95a05b42 | 4492 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4493 | ret = kmalloc_large_node(size, gfpflags, node); |
4494 | ||
4495 | trace_kmalloc_node(caller, ret, | |
4496 | size, PAGE_SIZE << get_order(size), | |
4497 | gfpflags, node); | |
4498 | ||
4499 | return ret; | |
4500 | } | |
eada35ef | 4501 | |
2c59dd65 | 4502 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4503 | |
2408c550 | 4504 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4505 | return s; |
81819f0f | 4506 | |
2b847c3c | 4507 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4508 | |
25985edc | 4509 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4510 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4511 | |
4512 | return ret; | |
81819f0f | 4513 | } |
fd7cb575 | 4514 | EXPORT_SYMBOL(__kmalloc_node_track_caller); |
5d1f57e4 | 4515 | #endif |
81819f0f | 4516 | |
ab4d5ed5 | 4517 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4518 | static int count_inuse(struct page *page) |
4519 | { | |
4520 | return page->inuse; | |
4521 | } | |
4522 | ||
4523 | static int count_total(struct page *page) | |
4524 | { | |
4525 | return page->objects; | |
4526 | } | |
ab4d5ed5 | 4527 | #endif |
205ab99d | 4528 | |
ab4d5ed5 | 4529 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 | 4530 | static void validate_slab(struct kmem_cache *s, struct page *page) |
53e15af0 CL |
4531 | { |
4532 | void *p; | |
a973e9dd | 4533 | void *addr = page_address(page); |
90e9f6a6 YZ |
4534 | unsigned long *map; |
4535 | ||
4536 | slab_lock(page); | |
53e15af0 | 4537 | |
dd98afd4 | 4538 | if (!check_slab(s, page) || !on_freelist(s, page, NULL)) |
90e9f6a6 | 4539 | goto unlock; |
53e15af0 CL |
4540 | |
4541 | /* Now we know that a valid freelist exists */ | |
90e9f6a6 | 4542 | map = get_map(s, page); |
5f80b13a | 4543 | for_each_object(p, s, addr, page->objects) { |
4138fdfc | 4544 | u8 val = test_bit(__obj_to_index(s, addr, p), map) ? |
dd98afd4 | 4545 | SLUB_RED_INACTIVE : SLUB_RED_ACTIVE; |
53e15af0 | 4546 | |
dd98afd4 YZ |
4547 | if (!check_object(s, page, p, val)) |
4548 | break; | |
4549 | } | |
90e9f6a6 YZ |
4550 | put_map(map); |
4551 | unlock: | |
881db7fb | 4552 | slab_unlock(page); |
53e15af0 CL |
4553 | } |
4554 | ||
434e245d | 4555 | static int validate_slab_node(struct kmem_cache *s, |
90e9f6a6 | 4556 | struct kmem_cache_node *n) |
53e15af0 CL |
4557 | { |
4558 | unsigned long count = 0; | |
4559 | struct page *page; | |
4560 | unsigned long flags; | |
4561 | ||
4562 | spin_lock_irqsave(&n->list_lock, flags); | |
4563 | ||
916ac052 | 4564 | list_for_each_entry(page, &n->partial, slab_list) { |
90e9f6a6 | 4565 | validate_slab(s, page); |
53e15af0 CL |
4566 | count++; |
4567 | } | |
4568 | if (count != n->nr_partial) | |
f9f58285 FF |
4569 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4570 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4571 | |
4572 | if (!(s->flags & SLAB_STORE_USER)) | |
4573 | goto out; | |
4574 | ||
916ac052 | 4575 | list_for_each_entry(page, &n->full, slab_list) { |
90e9f6a6 | 4576 | validate_slab(s, page); |
53e15af0 CL |
4577 | count++; |
4578 | } | |
4579 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4580 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4581 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4582 | |
4583 | out: | |
4584 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4585 | return count; | |
4586 | } | |
4587 | ||
434e245d | 4588 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4589 | { |
4590 | int node; | |
4591 | unsigned long count = 0; | |
fa45dc25 | 4592 | struct kmem_cache_node *n; |
53e15af0 CL |
4593 | |
4594 | flush_all(s); | |
fa45dc25 | 4595 | for_each_kmem_cache_node(s, node, n) |
90e9f6a6 YZ |
4596 | count += validate_slab_node(s, n); |
4597 | ||
53e15af0 CL |
4598 | return count; |
4599 | } | |
88a420e4 | 4600 | /* |
672bba3a | 4601 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4602 | * and freed. |
4603 | */ | |
4604 | ||
4605 | struct location { | |
4606 | unsigned long count; | |
ce71e27c | 4607 | unsigned long addr; |
45edfa58 CL |
4608 | long long sum_time; |
4609 | long min_time; | |
4610 | long max_time; | |
4611 | long min_pid; | |
4612 | long max_pid; | |
174596a0 | 4613 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4614 | nodemask_t nodes; |
88a420e4 CL |
4615 | }; |
4616 | ||
4617 | struct loc_track { | |
4618 | unsigned long max; | |
4619 | unsigned long count; | |
4620 | struct location *loc; | |
4621 | }; | |
4622 | ||
4623 | static void free_loc_track(struct loc_track *t) | |
4624 | { | |
4625 | if (t->max) | |
4626 | free_pages((unsigned long)t->loc, | |
4627 | get_order(sizeof(struct location) * t->max)); | |
4628 | } | |
4629 | ||
68dff6a9 | 4630 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4631 | { |
4632 | struct location *l; | |
4633 | int order; | |
4634 | ||
88a420e4 CL |
4635 | order = get_order(sizeof(struct location) * max); |
4636 | ||
68dff6a9 | 4637 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4638 | if (!l) |
4639 | return 0; | |
4640 | ||
4641 | if (t->count) { | |
4642 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4643 | free_loc_track(t); | |
4644 | } | |
4645 | t->max = max; | |
4646 | t->loc = l; | |
4647 | return 1; | |
4648 | } | |
4649 | ||
4650 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4651 | const struct track *track) |
88a420e4 CL |
4652 | { |
4653 | long start, end, pos; | |
4654 | struct location *l; | |
ce71e27c | 4655 | unsigned long caddr; |
45edfa58 | 4656 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4657 | |
4658 | start = -1; | |
4659 | end = t->count; | |
4660 | ||
4661 | for ( ; ; ) { | |
4662 | pos = start + (end - start + 1) / 2; | |
4663 | ||
4664 | /* | |
4665 | * There is nothing at "end". If we end up there | |
4666 | * we need to add something to before end. | |
4667 | */ | |
4668 | if (pos == end) | |
4669 | break; | |
4670 | ||
4671 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4672 | if (track->addr == caddr) { |
4673 | ||
4674 | l = &t->loc[pos]; | |
4675 | l->count++; | |
4676 | if (track->when) { | |
4677 | l->sum_time += age; | |
4678 | if (age < l->min_time) | |
4679 | l->min_time = age; | |
4680 | if (age > l->max_time) | |
4681 | l->max_time = age; | |
4682 | ||
4683 | if (track->pid < l->min_pid) | |
4684 | l->min_pid = track->pid; | |
4685 | if (track->pid > l->max_pid) | |
4686 | l->max_pid = track->pid; | |
4687 | ||
174596a0 RR |
4688 | cpumask_set_cpu(track->cpu, |
4689 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4690 | } |
4691 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4692 | return 1; |
4693 | } | |
4694 | ||
45edfa58 | 4695 | if (track->addr < caddr) |
88a420e4 CL |
4696 | end = pos; |
4697 | else | |
4698 | start = pos; | |
4699 | } | |
4700 | ||
4701 | /* | |
672bba3a | 4702 | * Not found. Insert new tracking element. |
88a420e4 | 4703 | */ |
68dff6a9 | 4704 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4705 | return 0; |
4706 | ||
4707 | l = t->loc + pos; | |
4708 | if (pos < t->count) | |
4709 | memmove(l + 1, l, | |
4710 | (t->count - pos) * sizeof(struct location)); | |
4711 | t->count++; | |
4712 | l->count = 1; | |
45edfa58 CL |
4713 | l->addr = track->addr; |
4714 | l->sum_time = age; | |
4715 | l->min_time = age; | |
4716 | l->max_time = age; | |
4717 | l->min_pid = track->pid; | |
4718 | l->max_pid = track->pid; | |
174596a0 RR |
4719 | cpumask_clear(to_cpumask(l->cpus)); |
4720 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4721 | nodes_clear(l->nodes); |
4722 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4723 | return 1; |
4724 | } | |
4725 | ||
4726 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
90e9f6a6 | 4727 | struct page *page, enum track_item alloc) |
88a420e4 | 4728 | { |
a973e9dd | 4729 | void *addr = page_address(page); |
88a420e4 | 4730 | void *p; |
90e9f6a6 | 4731 | unsigned long *map; |
88a420e4 | 4732 | |
90e9f6a6 | 4733 | map = get_map(s, page); |
224a88be | 4734 | for_each_object(p, s, addr, page->objects) |
4138fdfc | 4735 | if (!test_bit(__obj_to_index(s, addr, p), map)) |
45edfa58 | 4736 | add_location(t, s, get_track(s, p, alloc)); |
90e9f6a6 | 4737 | put_map(map); |
88a420e4 CL |
4738 | } |
4739 | ||
4740 | static int list_locations(struct kmem_cache *s, char *buf, | |
bf16d19a | 4741 | enum track_item alloc) |
88a420e4 | 4742 | { |
e374d483 | 4743 | int len = 0; |
88a420e4 | 4744 | unsigned long i; |
68dff6a9 | 4745 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4746 | int node; |
fa45dc25 | 4747 | struct kmem_cache_node *n; |
88a420e4 | 4748 | |
90e9f6a6 YZ |
4749 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4750 | GFP_KERNEL)) { | |
bf16d19a | 4751 | return sysfs_emit(buf, "Out of memory\n"); |
bbd7d57b | 4752 | } |
88a420e4 CL |
4753 | /* Push back cpu slabs */ |
4754 | flush_all(s); | |
4755 | ||
fa45dc25 | 4756 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4757 | unsigned long flags; |
4758 | struct page *page; | |
4759 | ||
9e86943b | 4760 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4761 | continue; |
4762 | ||
4763 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 4764 | list_for_each_entry(page, &n->partial, slab_list) |
90e9f6a6 | 4765 | process_slab(&t, s, page, alloc); |
916ac052 | 4766 | list_for_each_entry(page, &n->full, slab_list) |
90e9f6a6 | 4767 | process_slab(&t, s, page, alloc); |
88a420e4 CL |
4768 | spin_unlock_irqrestore(&n->list_lock, flags); |
4769 | } | |
4770 | ||
4771 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4772 | struct location *l = &t.loc[i]; |
88a420e4 | 4773 | |
bf16d19a | 4774 | len += sysfs_emit_at(buf, len, "%7ld ", l->count); |
45edfa58 CL |
4775 | |
4776 | if (l->addr) | |
bf16d19a | 4777 | len += sysfs_emit_at(buf, len, "%pS", (void *)l->addr); |
88a420e4 | 4778 | else |
bf16d19a JP |
4779 | len += sysfs_emit_at(buf, len, "<not-available>"); |
4780 | ||
4781 | if (l->sum_time != l->min_time) | |
4782 | len += sysfs_emit_at(buf, len, " age=%ld/%ld/%ld", | |
4783 | l->min_time, | |
4784 | (long)div_u64(l->sum_time, | |
4785 | l->count), | |
4786 | l->max_time); | |
4787 | else | |
4788 | len += sysfs_emit_at(buf, len, " age=%ld", l->min_time); | |
45edfa58 CL |
4789 | |
4790 | if (l->min_pid != l->max_pid) | |
bf16d19a JP |
4791 | len += sysfs_emit_at(buf, len, " pid=%ld-%ld", |
4792 | l->min_pid, l->max_pid); | |
45edfa58 | 4793 | else |
bf16d19a JP |
4794 | len += sysfs_emit_at(buf, len, " pid=%ld", |
4795 | l->min_pid); | |
45edfa58 | 4796 | |
174596a0 | 4797 | if (num_online_cpus() > 1 && |
bf16d19a JP |
4798 | !cpumask_empty(to_cpumask(l->cpus))) |
4799 | len += sysfs_emit_at(buf, len, " cpus=%*pbl", | |
4800 | cpumask_pr_args(to_cpumask(l->cpus))); | |
4801 | ||
4802 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes)) | |
4803 | len += sysfs_emit_at(buf, len, " nodes=%*pbl", | |
4804 | nodemask_pr_args(&l->nodes)); | |
4805 | ||
4806 | len += sysfs_emit_at(buf, len, "\n"); | |
88a420e4 CL |
4807 | } |
4808 | ||
4809 | free_loc_track(&t); | |
4810 | if (!t.count) | |
bf16d19a JP |
4811 | len += sysfs_emit_at(buf, len, "No data\n"); |
4812 | ||
e374d483 | 4813 | return len; |
88a420e4 | 4814 | } |
6dfd1b65 | 4815 | #endif /* CONFIG_SLUB_DEBUG */ |
88a420e4 | 4816 | |
a5a84755 | 4817 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4818 | static void __init resiliency_test(void) |
a5a84755 CL |
4819 | { |
4820 | u8 *p; | |
cc252eae | 4821 | int type = KMALLOC_NORMAL; |
a5a84755 | 4822 | |
95a05b42 | 4823 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4824 | |
f9f58285 FF |
4825 | pr_err("SLUB resiliency testing\n"); |
4826 | pr_err("-----------------------\n"); | |
4827 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4828 | |
4829 | p = kzalloc(16, GFP_KERNEL); | |
4830 | p[16] = 0x12; | |
f9f58285 FF |
4831 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4832 | p + 16); | |
a5a84755 | 4833 | |
cc252eae | 4834 | validate_slab_cache(kmalloc_caches[type][4]); |
a5a84755 CL |
4835 | |
4836 | /* Hmmm... The next two are dangerous */ | |
4837 | p = kzalloc(32, GFP_KERNEL); | |
4838 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4839 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4840 | p); | |
4841 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 | 4842 | |
cc252eae | 4843 | validate_slab_cache(kmalloc_caches[type][5]); |
a5a84755 CL |
4844 | p = kzalloc(64, GFP_KERNEL); |
4845 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4846 | *p = 0x56; | |
f9f58285 FF |
4847 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4848 | p); | |
4849 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
cc252eae | 4850 | validate_slab_cache(kmalloc_caches[type][6]); |
a5a84755 | 4851 | |
f9f58285 | 4852 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4853 | p = kzalloc(128, GFP_KERNEL); |
4854 | kfree(p); | |
4855 | *p = 0x78; | |
f9f58285 | 4856 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
cc252eae | 4857 | validate_slab_cache(kmalloc_caches[type][7]); |
a5a84755 CL |
4858 | |
4859 | p = kzalloc(256, GFP_KERNEL); | |
4860 | kfree(p); | |
4861 | p[50] = 0x9a; | |
f9f58285 | 4862 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
cc252eae | 4863 | validate_slab_cache(kmalloc_caches[type][8]); |
a5a84755 CL |
4864 | |
4865 | p = kzalloc(512, GFP_KERNEL); | |
4866 | kfree(p); | |
4867 | p[512] = 0xab; | |
f9f58285 | 4868 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
cc252eae | 4869 | validate_slab_cache(kmalloc_caches[type][9]); |
a5a84755 CL |
4870 | } |
4871 | #else | |
4872 | #ifdef CONFIG_SYSFS | |
4873 | static void resiliency_test(void) {}; | |
4874 | #endif | |
6dfd1b65 | 4875 | #endif /* SLUB_RESILIENCY_TEST */ |
a5a84755 | 4876 | |
ab4d5ed5 | 4877 | #ifdef CONFIG_SYSFS |
81819f0f | 4878 | enum slab_stat_type { |
205ab99d CL |
4879 | SL_ALL, /* All slabs */ |
4880 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4881 | SL_CPU, /* Only slabs used for cpu caches */ | |
4882 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4883 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4884 | }; |
4885 | ||
205ab99d | 4886 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4887 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4888 | #define SO_CPU (1 << SL_CPU) | |
4889 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4890 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4891 | |
1663f26d TH |
4892 | #ifdef CONFIG_MEMCG |
4893 | static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON); | |
4894 | ||
4895 | static int __init setup_slub_memcg_sysfs(char *str) | |
4896 | { | |
4897 | int v; | |
4898 | ||
4899 | if (get_option(&str, &v) > 0) | |
4900 | memcg_sysfs_enabled = v; | |
4901 | ||
4902 | return 1; | |
4903 | } | |
4904 | ||
4905 | __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs); | |
4906 | #endif | |
4907 | ||
62e5c4b4 | 4908 | static ssize_t show_slab_objects(struct kmem_cache *s, |
bf16d19a | 4909 | char *buf, unsigned long flags) |
81819f0f CL |
4910 | { |
4911 | unsigned long total = 0; | |
81819f0f CL |
4912 | int node; |
4913 | int x; | |
4914 | unsigned long *nodes; | |
bf16d19a | 4915 | int len = 0; |
81819f0f | 4916 | |
6396bb22 | 4917 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
4918 | if (!nodes) |
4919 | return -ENOMEM; | |
81819f0f | 4920 | |
205ab99d CL |
4921 | if (flags & SO_CPU) { |
4922 | int cpu; | |
81819f0f | 4923 | |
205ab99d | 4924 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4925 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4926 | cpu); | |
ec3ab083 | 4927 | int node; |
49e22585 | 4928 | struct page *page; |
dfb4f096 | 4929 | |
4db0c3c2 | 4930 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4931 | if (!page) |
4932 | continue; | |
205ab99d | 4933 | |
ec3ab083 CL |
4934 | node = page_to_nid(page); |
4935 | if (flags & SO_TOTAL) | |
4936 | x = page->objects; | |
4937 | else if (flags & SO_OBJECTS) | |
4938 | x = page->inuse; | |
4939 | else | |
4940 | x = 1; | |
49e22585 | 4941 | |
ec3ab083 CL |
4942 | total += x; |
4943 | nodes[node] += x; | |
4944 | ||
a93cf07b | 4945 | page = slub_percpu_partial_read_once(c); |
49e22585 | 4946 | if (page) { |
8afb1474 LZ |
4947 | node = page_to_nid(page); |
4948 | if (flags & SO_TOTAL) | |
4949 | WARN_ON_ONCE(1); | |
4950 | else if (flags & SO_OBJECTS) | |
4951 | WARN_ON_ONCE(1); | |
4952 | else | |
4953 | x = page->pages; | |
bc6697d8 ED |
4954 | total += x; |
4955 | nodes[node] += x; | |
49e22585 | 4956 | } |
81819f0f CL |
4957 | } |
4958 | } | |
4959 | ||
e4f8e513 QC |
4960 | /* |
4961 | * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex" | |
4962 | * already held which will conflict with an existing lock order: | |
4963 | * | |
4964 | * mem_hotplug_lock->slab_mutex->kernfs_mutex | |
4965 | * | |
4966 | * We don't really need mem_hotplug_lock (to hold off | |
4967 | * slab_mem_going_offline_callback) here because slab's memory hot | |
4968 | * unplug code doesn't destroy the kmem_cache->node[] data. | |
4969 | */ | |
4970 | ||
ab4d5ed5 | 4971 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4972 | if (flags & SO_ALL) { |
fa45dc25 CL |
4973 | struct kmem_cache_node *n; |
4974 | ||
4975 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4976 | |
d0e0ac97 CG |
4977 | if (flags & SO_TOTAL) |
4978 | x = atomic_long_read(&n->total_objects); | |
4979 | else if (flags & SO_OBJECTS) | |
4980 | x = atomic_long_read(&n->total_objects) - | |
4981 | count_partial(n, count_free); | |
81819f0f | 4982 | else |
205ab99d | 4983 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4984 | total += x; |
4985 | nodes[node] += x; | |
4986 | } | |
4987 | ||
ab4d5ed5 CL |
4988 | } else |
4989 | #endif | |
4990 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4991 | struct kmem_cache_node *n; |
81819f0f | 4992 | |
fa45dc25 | 4993 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4994 | if (flags & SO_TOTAL) |
4995 | x = count_partial(n, count_total); | |
4996 | else if (flags & SO_OBJECTS) | |
4997 | x = count_partial(n, count_inuse); | |
81819f0f | 4998 | else |
205ab99d | 4999 | x = n->nr_partial; |
81819f0f CL |
5000 | total += x; |
5001 | nodes[node] += x; | |
5002 | } | |
5003 | } | |
bf16d19a JP |
5004 | |
5005 | len += sysfs_emit_at(buf, len, "%lu", total); | |
81819f0f | 5006 | #ifdef CONFIG_NUMA |
bf16d19a | 5007 | for (node = 0; node < nr_node_ids; node++) { |
81819f0f | 5008 | if (nodes[node]) |
bf16d19a JP |
5009 | len += sysfs_emit_at(buf, len, " N%d=%lu", |
5010 | node, nodes[node]); | |
5011 | } | |
81819f0f | 5012 | #endif |
bf16d19a | 5013 | len += sysfs_emit_at(buf, len, "\n"); |
81819f0f | 5014 | kfree(nodes); |
bf16d19a JP |
5015 | |
5016 | return len; | |
81819f0f CL |
5017 | } |
5018 | ||
81819f0f | 5019 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) |
497888cf | 5020 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
5021 | |
5022 | struct slab_attribute { | |
5023 | struct attribute attr; | |
5024 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
5025 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
5026 | }; | |
5027 | ||
5028 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
5029 | static struct slab_attribute _name##_attr = \ |
5030 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
5031 | |
5032 | #define SLAB_ATTR(_name) \ | |
5033 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 5034 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 5035 | |
81819f0f CL |
5036 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
5037 | { | |
bf16d19a | 5038 | return sysfs_emit(buf, "%u\n", s->size); |
81819f0f CL |
5039 | } |
5040 | SLAB_ATTR_RO(slab_size); | |
5041 | ||
5042 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
5043 | { | |
bf16d19a | 5044 | return sysfs_emit(buf, "%u\n", s->align); |
81819f0f CL |
5045 | } |
5046 | SLAB_ATTR_RO(align); | |
5047 | ||
5048 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
5049 | { | |
bf16d19a | 5050 | return sysfs_emit(buf, "%u\n", s->object_size); |
81819f0f CL |
5051 | } |
5052 | SLAB_ATTR_RO(object_size); | |
5053 | ||
5054 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
5055 | { | |
bf16d19a | 5056 | return sysfs_emit(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
5057 | } |
5058 | SLAB_ATTR_RO(objs_per_slab); | |
5059 | ||
5060 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
5061 | { | |
bf16d19a | 5062 | return sysfs_emit(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 5063 | } |
32a6f409 | 5064 | SLAB_ATTR_RO(order); |
81819f0f | 5065 | |
73d342b1 DR |
5066 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
5067 | { | |
bf16d19a | 5068 | return sysfs_emit(buf, "%lu\n", s->min_partial); |
73d342b1 DR |
5069 | } |
5070 | ||
5071 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
5072 | size_t length) | |
5073 | { | |
5074 | unsigned long min; | |
5075 | int err; | |
5076 | ||
3dbb95f7 | 5077 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
5078 | if (err) |
5079 | return err; | |
5080 | ||
c0bdb232 | 5081 | set_min_partial(s, min); |
73d342b1 DR |
5082 | return length; |
5083 | } | |
5084 | SLAB_ATTR(min_partial); | |
5085 | ||
49e22585 CL |
5086 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
5087 | { | |
bf16d19a | 5088 | return sysfs_emit(buf, "%u\n", slub_cpu_partial(s)); |
49e22585 CL |
5089 | } |
5090 | ||
5091 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
5092 | size_t length) | |
5093 | { | |
e5d9998f | 5094 | unsigned int objects; |
49e22585 CL |
5095 | int err; |
5096 | ||
e5d9998f | 5097 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
5098 | if (err) |
5099 | return err; | |
345c905d | 5100 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 5101 | return -EINVAL; |
49e22585 | 5102 | |
e6d0e1dc | 5103 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
5104 | flush_all(s); |
5105 | return length; | |
5106 | } | |
5107 | SLAB_ATTR(cpu_partial); | |
5108 | ||
81819f0f CL |
5109 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
5110 | { | |
62c70bce JP |
5111 | if (!s->ctor) |
5112 | return 0; | |
bf16d19a | 5113 | return sysfs_emit(buf, "%pS\n", s->ctor); |
81819f0f CL |
5114 | } |
5115 | SLAB_ATTR_RO(ctor); | |
5116 | ||
81819f0f CL |
5117 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
5118 | { | |
bf16d19a | 5119 | return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
5120 | } |
5121 | SLAB_ATTR_RO(aliases); | |
5122 | ||
81819f0f CL |
5123 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5124 | { | |
d9acf4b7 | 5125 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5126 | } |
5127 | SLAB_ATTR_RO(partial); | |
5128 | ||
5129 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5130 | { | |
d9acf4b7 | 5131 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5132 | } |
5133 | SLAB_ATTR_RO(cpu_slabs); | |
5134 | ||
5135 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5136 | { | |
205ab99d | 5137 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5138 | } |
5139 | SLAB_ATTR_RO(objects); | |
5140 | ||
205ab99d CL |
5141 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5142 | { | |
5143 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5144 | } | |
5145 | SLAB_ATTR_RO(objects_partial); | |
5146 | ||
49e22585 CL |
5147 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5148 | { | |
5149 | int objects = 0; | |
5150 | int pages = 0; | |
5151 | int cpu; | |
bf16d19a | 5152 | int len = 0; |
49e22585 CL |
5153 | |
5154 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5155 | struct page *page; |
5156 | ||
5157 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5158 | |
5159 | if (page) { | |
5160 | pages += page->pages; | |
5161 | objects += page->pobjects; | |
5162 | } | |
5163 | } | |
5164 | ||
bf16d19a | 5165 | len += sysfs_emit_at(buf, len, "%d(%d)", objects, pages); |
49e22585 CL |
5166 | |
5167 | #ifdef CONFIG_SMP | |
5168 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5169 | struct page *page; |
5170 | ||
5171 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
bf16d19a JP |
5172 | if (page) |
5173 | len += sysfs_emit_at(buf, len, " C%d=%d(%d)", | |
5174 | cpu, page->pobjects, page->pages); | |
49e22585 CL |
5175 | } |
5176 | #endif | |
bf16d19a JP |
5177 | len += sysfs_emit_at(buf, len, "\n"); |
5178 | ||
5179 | return len; | |
49e22585 CL |
5180 | } |
5181 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5182 | ||
a5a84755 CL |
5183 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5184 | { | |
bf16d19a | 5185 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); |
a5a84755 | 5186 | } |
8f58119a | 5187 | SLAB_ATTR_RO(reclaim_account); |
a5a84755 CL |
5188 | |
5189 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5190 | { | |
bf16d19a | 5191 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
a5a84755 CL |
5192 | } |
5193 | SLAB_ATTR_RO(hwcache_align); | |
5194 | ||
5195 | #ifdef CONFIG_ZONE_DMA | |
5196 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5197 | { | |
bf16d19a | 5198 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); |
a5a84755 CL |
5199 | } |
5200 | SLAB_ATTR_RO(cache_dma); | |
5201 | #endif | |
5202 | ||
8eb8284b DW |
5203 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5204 | { | |
bf16d19a | 5205 | return sysfs_emit(buf, "%u\n", s->usersize); |
8eb8284b DW |
5206 | } |
5207 | SLAB_ATTR_RO(usersize); | |
5208 | ||
a5a84755 CL |
5209 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5210 | { | |
bf16d19a | 5211 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5212 | } |
5213 | SLAB_ATTR_RO(destroy_by_rcu); | |
5214 | ||
ab4d5ed5 | 5215 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5216 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5217 | { | |
5218 | return show_slab_objects(s, buf, SO_ALL); | |
5219 | } | |
5220 | SLAB_ATTR_RO(slabs); | |
5221 | ||
205ab99d CL |
5222 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5223 | { | |
5224 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5225 | } | |
5226 | SLAB_ATTR_RO(total_objects); | |
5227 | ||
81819f0f CL |
5228 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5229 | { | |
bf16d19a | 5230 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f | 5231 | } |
060807f8 | 5232 | SLAB_ATTR_RO(sanity_checks); |
81819f0f CL |
5233 | |
5234 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5235 | { | |
bf16d19a | 5236 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE)); |
81819f0f | 5237 | } |
060807f8 | 5238 | SLAB_ATTR_RO(trace); |
81819f0f | 5239 | |
81819f0f CL |
5240 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5241 | { | |
bf16d19a | 5242 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); |
81819f0f CL |
5243 | } |
5244 | ||
ad38b5b1 | 5245 | SLAB_ATTR_RO(red_zone); |
81819f0f CL |
5246 | |
5247 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5248 | { | |
bf16d19a | 5249 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON)); |
81819f0f CL |
5250 | } |
5251 | ||
ad38b5b1 | 5252 | SLAB_ATTR_RO(poison); |
81819f0f CL |
5253 | |
5254 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5255 | { | |
bf16d19a | 5256 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); |
81819f0f CL |
5257 | } |
5258 | ||
ad38b5b1 | 5259 | SLAB_ATTR_RO(store_user); |
81819f0f | 5260 | |
53e15af0 CL |
5261 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5262 | { | |
5263 | return 0; | |
5264 | } | |
5265 | ||
5266 | static ssize_t validate_store(struct kmem_cache *s, | |
5267 | const char *buf, size_t length) | |
5268 | { | |
434e245d CL |
5269 | int ret = -EINVAL; |
5270 | ||
5271 | if (buf[0] == '1') { | |
5272 | ret = validate_slab_cache(s); | |
5273 | if (ret >= 0) | |
5274 | ret = length; | |
5275 | } | |
5276 | return ret; | |
53e15af0 CL |
5277 | } |
5278 | SLAB_ATTR(validate); | |
a5a84755 CL |
5279 | |
5280 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5281 | { | |
5282 | if (!(s->flags & SLAB_STORE_USER)) | |
5283 | return -ENOSYS; | |
5284 | return list_locations(s, buf, TRACK_ALLOC); | |
5285 | } | |
5286 | SLAB_ATTR_RO(alloc_calls); | |
5287 | ||
5288 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5289 | { | |
5290 | if (!(s->flags & SLAB_STORE_USER)) | |
5291 | return -ENOSYS; | |
5292 | return list_locations(s, buf, TRACK_FREE); | |
5293 | } | |
5294 | SLAB_ATTR_RO(free_calls); | |
5295 | #endif /* CONFIG_SLUB_DEBUG */ | |
5296 | ||
5297 | #ifdef CONFIG_FAILSLAB | |
5298 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5299 | { | |
bf16d19a | 5300 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); |
a5a84755 | 5301 | } |
060807f8 | 5302 | SLAB_ATTR_RO(failslab); |
ab4d5ed5 | 5303 | #endif |
53e15af0 | 5304 | |
2086d26a CL |
5305 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5306 | { | |
5307 | return 0; | |
5308 | } | |
5309 | ||
5310 | static ssize_t shrink_store(struct kmem_cache *s, | |
5311 | const char *buf, size_t length) | |
5312 | { | |
832f37f5 | 5313 | if (buf[0] == '1') |
10befea9 | 5314 | kmem_cache_shrink(s); |
832f37f5 | 5315 | else |
2086d26a CL |
5316 | return -EINVAL; |
5317 | return length; | |
5318 | } | |
5319 | SLAB_ATTR(shrink); | |
5320 | ||
81819f0f | 5321 | #ifdef CONFIG_NUMA |
9824601e | 5322 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5323 | { |
bf16d19a | 5324 | return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5325 | } |
5326 | ||
9824601e | 5327 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5328 | const char *buf, size_t length) |
5329 | { | |
eb7235eb | 5330 | unsigned int ratio; |
0121c619 CL |
5331 | int err; |
5332 | ||
eb7235eb | 5333 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5334 | if (err) |
5335 | return err; | |
eb7235eb AD |
5336 | if (ratio > 100) |
5337 | return -ERANGE; | |
0121c619 | 5338 | |
eb7235eb | 5339 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5340 | |
81819f0f CL |
5341 | return length; |
5342 | } | |
9824601e | 5343 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5344 | #endif |
5345 | ||
8ff12cfc | 5346 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5347 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5348 | { | |
5349 | unsigned long sum = 0; | |
5350 | int cpu; | |
bf16d19a | 5351 | int len = 0; |
6da2ec56 | 5352 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5353 | |
5354 | if (!data) | |
5355 | return -ENOMEM; | |
5356 | ||
5357 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5358 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5359 | |
5360 | data[cpu] = x; | |
5361 | sum += x; | |
5362 | } | |
5363 | ||
bf16d19a | 5364 | len += sysfs_emit_at(buf, len, "%lu", sum); |
8ff12cfc | 5365 | |
50ef37b9 | 5366 | #ifdef CONFIG_SMP |
8ff12cfc | 5367 | for_each_online_cpu(cpu) { |
bf16d19a JP |
5368 | if (data[cpu]) |
5369 | len += sysfs_emit_at(buf, len, " C%d=%u", | |
5370 | cpu, data[cpu]); | |
8ff12cfc | 5371 | } |
50ef37b9 | 5372 | #endif |
8ff12cfc | 5373 | kfree(data); |
bf16d19a JP |
5374 | len += sysfs_emit_at(buf, len, "\n"); |
5375 | ||
5376 | return len; | |
8ff12cfc CL |
5377 | } |
5378 | ||
78eb00cc DR |
5379 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5380 | { | |
5381 | int cpu; | |
5382 | ||
5383 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5384 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5385 | } |
5386 | ||
8ff12cfc CL |
5387 | #define STAT_ATTR(si, text) \ |
5388 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5389 | { \ | |
5390 | return show_stat(s, buf, si); \ | |
5391 | } \ | |
78eb00cc DR |
5392 | static ssize_t text##_store(struct kmem_cache *s, \ |
5393 | const char *buf, size_t length) \ | |
5394 | { \ | |
5395 | if (buf[0] != '0') \ | |
5396 | return -EINVAL; \ | |
5397 | clear_stat(s, si); \ | |
5398 | return length; \ | |
5399 | } \ | |
5400 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5401 | |
5402 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5403 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5404 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5405 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5406 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5407 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5408 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5409 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5410 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5411 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5412 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5413 | STAT_ATTR(FREE_SLAB, free_slab); |
5414 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5415 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5416 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5417 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5418 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5419 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5420 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5421 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5422 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5423 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5424 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5425 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5426 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5427 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
6dfd1b65 | 5428 | #endif /* CONFIG_SLUB_STATS */ |
8ff12cfc | 5429 | |
06428780 | 5430 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5431 | &slab_size_attr.attr, |
5432 | &object_size_attr.attr, | |
5433 | &objs_per_slab_attr.attr, | |
5434 | &order_attr.attr, | |
73d342b1 | 5435 | &min_partial_attr.attr, |
49e22585 | 5436 | &cpu_partial_attr.attr, |
81819f0f | 5437 | &objects_attr.attr, |
205ab99d | 5438 | &objects_partial_attr.attr, |
81819f0f CL |
5439 | &partial_attr.attr, |
5440 | &cpu_slabs_attr.attr, | |
5441 | &ctor_attr.attr, | |
81819f0f CL |
5442 | &aliases_attr.attr, |
5443 | &align_attr.attr, | |
81819f0f CL |
5444 | &hwcache_align_attr.attr, |
5445 | &reclaim_account_attr.attr, | |
5446 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5447 | &shrink_attr.attr, |
49e22585 | 5448 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5449 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5450 | &total_objects_attr.attr, |
5451 | &slabs_attr.attr, | |
5452 | &sanity_checks_attr.attr, | |
5453 | &trace_attr.attr, | |
81819f0f CL |
5454 | &red_zone_attr.attr, |
5455 | &poison_attr.attr, | |
5456 | &store_user_attr.attr, | |
53e15af0 | 5457 | &validate_attr.attr, |
88a420e4 CL |
5458 | &alloc_calls_attr.attr, |
5459 | &free_calls_attr.attr, | |
ab4d5ed5 | 5460 | #endif |
81819f0f CL |
5461 | #ifdef CONFIG_ZONE_DMA |
5462 | &cache_dma_attr.attr, | |
5463 | #endif | |
5464 | #ifdef CONFIG_NUMA | |
9824601e | 5465 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5466 | #endif |
5467 | #ifdef CONFIG_SLUB_STATS | |
5468 | &alloc_fastpath_attr.attr, | |
5469 | &alloc_slowpath_attr.attr, | |
5470 | &free_fastpath_attr.attr, | |
5471 | &free_slowpath_attr.attr, | |
5472 | &free_frozen_attr.attr, | |
5473 | &free_add_partial_attr.attr, | |
5474 | &free_remove_partial_attr.attr, | |
5475 | &alloc_from_partial_attr.attr, | |
5476 | &alloc_slab_attr.attr, | |
5477 | &alloc_refill_attr.attr, | |
e36a2652 | 5478 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5479 | &free_slab_attr.attr, |
5480 | &cpuslab_flush_attr.attr, | |
5481 | &deactivate_full_attr.attr, | |
5482 | &deactivate_empty_attr.attr, | |
5483 | &deactivate_to_head_attr.attr, | |
5484 | &deactivate_to_tail_attr.attr, | |
5485 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5486 | &deactivate_bypass_attr.attr, |
65c3376a | 5487 | &order_fallback_attr.attr, |
b789ef51 CL |
5488 | &cmpxchg_double_fail_attr.attr, |
5489 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5490 | &cpu_partial_alloc_attr.attr, |
5491 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5492 | &cpu_partial_node_attr.attr, |
5493 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5494 | #endif |
4c13dd3b DM |
5495 | #ifdef CONFIG_FAILSLAB |
5496 | &failslab_attr.attr, | |
5497 | #endif | |
8eb8284b | 5498 | &usersize_attr.attr, |
4c13dd3b | 5499 | |
81819f0f CL |
5500 | NULL |
5501 | }; | |
5502 | ||
1fdaaa23 | 5503 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5504 | .attrs = slab_attrs, |
5505 | }; | |
5506 | ||
5507 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5508 | struct attribute *attr, | |
5509 | char *buf) | |
5510 | { | |
5511 | struct slab_attribute *attribute; | |
5512 | struct kmem_cache *s; | |
5513 | int err; | |
5514 | ||
5515 | attribute = to_slab_attr(attr); | |
5516 | s = to_slab(kobj); | |
5517 | ||
5518 | if (!attribute->show) | |
5519 | return -EIO; | |
5520 | ||
5521 | err = attribute->show(s, buf); | |
5522 | ||
5523 | return err; | |
5524 | } | |
5525 | ||
5526 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5527 | struct attribute *attr, | |
5528 | const char *buf, size_t len) | |
5529 | { | |
5530 | struct slab_attribute *attribute; | |
5531 | struct kmem_cache *s; | |
5532 | int err; | |
5533 | ||
5534 | attribute = to_slab_attr(attr); | |
5535 | s = to_slab(kobj); | |
5536 | ||
5537 | if (!attribute->store) | |
5538 | return -EIO; | |
5539 | ||
5540 | err = attribute->store(s, buf, len); | |
81819f0f CL |
5541 | return err; |
5542 | } | |
5543 | ||
41a21285 CL |
5544 | static void kmem_cache_release(struct kobject *k) |
5545 | { | |
5546 | slab_kmem_cache_release(to_slab(k)); | |
5547 | } | |
5548 | ||
52cf25d0 | 5549 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5550 | .show = slab_attr_show, |
5551 | .store = slab_attr_store, | |
5552 | }; | |
5553 | ||
5554 | static struct kobj_type slab_ktype = { | |
5555 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5556 | .release = kmem_cache_release, |
81819f0f CL |
5557 | }; |
5558 | ||
27c3a314 | 5559 | static struct kset *slab_kset; |
81819f0f | 5560 | |
9a41707b VD |
5561 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5562 | { | |
9a41707b VD |
5563 | return slab_kset; |
5564 | } | |
5565 | ||
81819f0f CL |
5566 | #define ID_STR_LENGTH 64 |
5567 | ||
5568 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5569 | * |
5570 | * Format :[flags-]size | |
81819f0f CL |
5571 | */ |
5572 | static char *create_unique_id(struct kmem_cache *s) | |
5573 | { | |
5574 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5575 | char *p = name; | |
5576 | ||
5577 | BUG_ON(!name); | |
5578 | ||
5579 | *p++ = ':'; | |
5580 | /* | |
5581 | * First flags affecting slabcache operations. We will only | |
5582 | * get here for aliasable slabs so we do not need to support | |
5583 | * too many flags. The flags here must cover all flags that | |
5584 | * are matched during merging to guarantee that the id is | |
5585 | * unique. | |
5586 | */ | |
5587 | if (s->flags & SLAB_CACHE_DMA) | |
5588 | *p++ = 'd'; | |
6d6ea1e9 NB |
5589 | if (s->flags & SLAB_CACHE_DMA32) |
5590 | *p++ = 'D'; | |
81819f0f CL |
5591 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
5592 | *p++ = 'a'; | |
becfda68 | 5593 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5594 | *p++ = 'F'; |
230e9fc2 VD |
5595 | if (s->flags & SLAB_ACCOUNT) |
5596 | *p++ = 'A'; | |
81819f0f CL |
5597 | if (p != name + 1) |
5598 | *p++ = '-'; | |
44065b2e | 5599 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5600 | |
81819f0f CL |
5601 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5602 | return name; | |
5603 | } | |
5604 | ||
5605 | static int sysfs_slab_add(struct kmem_cache *s) | |
5606 | { | |
5607 | int err; | |
5608 | const char *name; | |
1663f26d | 5609 | struct kset *kset = cache_kset(s); |
45530c44 | 5610 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5611 | |
1663f26d TH |
5612 | if (!kset) { |
5613 | kobject_init(&s->kobj, &slab_ktype); | |
5614 | return 0; | |
5615 | } | |
5616 | ||
11066386 MC |
5617 | if (!unmergeable && disable_higher_order_debug && |
5618 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5619 | unmergeable = 1; | |
5620 | ||
81819f0f CL |
5621 | if (unmergeable) { |
5622 | /* | |
5623 | * Slabcache can never be merged so we can use the name proper. | |
5624 | * This is typically the case for debug situations. In that | |
5625 | * case we can catch duplicate names easily. | |
5626 | */ | |
27c3a314 | 5627 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5628 | name = s->name; |
5629 | } else { | |
5630 | /* | |
5631 | * Create a unique name for the slab as a target | |
5632 | * for the symlinks. | |
5633 | */ | |
5634 | name = create_unique_id(s); | |
5635 | } | |
5636 | ||
1663f26d | 5637 | s->kobj.kset = kset; |
26e4f205 | 5638 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
757fed1d | 5639 | if (err) |
80da026a | 5640 | goto out; |
81819f0f CL |
5641 | |
5642 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5643 | if (err) |
5644 | goto out_del_kobj; | |
9a41707b | 5645 | |
81819f0f CL |
5646 | if (!unmergeable) { |
5647 | /* Setup first alias */ | |
5648 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5649 | } |
54b6a731 DJ |
5650 | out: |
5651 | if (!unmergeable) | |
5652 | kfree(name); | |
5653 | return err; | |
5654 | out_del_kobj: | |
5655 | kobject_del(&s->kobj); | |
54b6a731 | 5656 | goto out; |
81819f0f CL |
5657 | } |
5658 | ||
d50d82fa MP |
5659 | void sysfs_slab_unlink(struct kmem_cache *s) |
5660 | { | |
5661 | if (slab_state >= FULL) | |
5662 | kobject_del(&s->kobj); | |
5663 | } | |
5664 | ||
bf5eb3de TH |
5665 | void sysfs_slab_release(struct kmem_cache *s) |
5666 | { | |
5667 | if (slab_state >= FULL) | |
5668 | kobject_put(&s->kobj); | |
81819f0f CL |
5669 | } |
5670 | ||
5671 | /* | |
5672 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5673 | * available lest we lose that information. |
81819f0f CL |
5674 | */ |
5675 | struct saved_alias { | |
5676 | struct kmem_cache *s; | |
5677 | const char *name; | |
5678 | struct saved_alias *next; | |
5679 | }; | |
5680 | ||
5af328a5 | 5681 | static struct saved_alias *alias_list; |
81819f0f CL |
5682 | |
5683 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5684 | { | |
5685 | struct saved_alias *al; | |
5686 | ||
97d06609 | 5687 | if (slab_state == FULL) { |
81819f0f CL |
5688 | /* |
5689 | * If we have a leftover link then remove it. | |
5690 | */ | |
27c3a314 GKH |
5691 | sysfs_remove_link(&slab_kset->kobj, name); |
5692 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5693 | } |
5694 | ||
5695 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5696 | if (!al) | |
5697 | return -ENOMEM; | |
5698 | ||
5699 | al->s = s; | |
5700 | al->name = name; | |
5701 | al->next = alias_list; | |
5702 | alias_list = al; | |
5703 | return 0; | |
5704 | } | |
5705 | ||
5706 | static int __init slab_sysfs_init(void) | |
5707 | { | |
5b95a4ac | 5708 | struct kmem_cache *s; |
81819f0f CL |
5709 | int err; |
5710 | ||
18004c5d | 5711 | mutex_lock(&slab_mutex); |
2bce6485 | 5712 | |
d7660ce5 | 5713 | slab_kset = kset_create_and_add("slab", NULL, kernel_kobj); |
27c3a314 | 5714 | if (!slab_kset) { |
18004c5d | 5715 | mutex_unlock(&slab_mutex); |
f9f58285 | 5716 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5717 | return -ENOSYS; |
5718 | } | |
5719 | ||
97d06609 | 5720 | slab_state = FULL; |
26a7bd03 | 5721 | |
5b95a4ac | 5722 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5723 | err = sysfs_slab_add(s); |
5d540fb7 | 5724 | if (err) |
f9f58285 FF |
5725 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5726 | s->name); | |
26a7bd03 | 5727 | } |
81819f0f CL |
5728 | |
5729 | while (alias_list) { | |
5730 | struct saved_alias *al = alias_list; | |
5731 | ||
5732 | alias_list = alias_list->next; | |
5733 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5734 | if (err) |
f9f58285 FF |
5735 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5736 | al->name); | |
81819f0f CL |
5737 | kfree(al); |
5738 | } | |
5739 | ||
18004c5d | 5740 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5741 | resiliency_test(); |
5742 | return 0; | |
5743 | } | |
5744 | ||
5745 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5746 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5747 | |
5748 | /* | |
5749 | * The /proc/slabinfo ABI | |
5750 | */ | |
5b365771 | 5751 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 5752 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5753 | { |
57ed3eda | 5754 | unsigned long nr_slabs = 0; |
205ab99d CL |
5755 | unsigned long nr_objs = 0; |
5756 | unsigned long nr_free = 0; | |
57ed3eda | 5757 | int node; |
fa45dc25 | 5758 | struct kmem_cache_node *n; |
57ed3eda | 5759 | |
fa45dc25 | 5760 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5761 | nr_slabs += node_nr_slabs(n); |
5762 | nr_objs += node_nr_objs(n); | |
205ab99d | 5763 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5764 | } |
5765 | ||
0d7561c6 GC |
5766 | sinfo->active_objs = nr_objs - nr_free; |
5767 | sinfo->num_objs = nr_objs; | |
5768 | sinfo->active_slabs = nr_slabs; | |
5769 | sinfo->num_slabs = nr_slabs; | |
5770 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5771 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5772 | } |
5773 | ||
0d7561c6 | 5774 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5775 | { |
7b3c3a50 AD |
5776 | } |
5777 | ||
b7454ad3 GC |
5778 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5779 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5780 | { |
b7454ad3 | 5781 | return -EIO; |
7b3c3a50 | 5782 | } |
5b365771 | 5783 | #endif /* CONFIG_SLUB_DEBUG */ |