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 | |
1403 | * @ctor: constructor function | |
1404 | * | |
1405 | * Debug option(s) are applied to @flags. In addition to the debug | |
1406 | * option(s), if a slab name (or multiple) is specified i.e. | |
1407 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1408 | * then only the select slabs will receive the debug option(s). | |
1409 | */ | |
0293d1fd | 1410 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1411 | slab_flags_t flags, const char *name, |
51cc5068 | 1412 | void (*ctor)(void *)) |
41ecc55b | 1413 | { |
c5fd3ca0 AT |
1414 | char *iter; |
1415 | size_t len; | |
e17f1dfb VB |
1416 | char *next_block; |
1417 | slab_flags_t block_flags; | |
c5fd3ca0 | 1418 | |
c5fd3ca0 | 1419 | len = strlen(name); |
e17f1dfb VB |
1420 | next_block = slub_debug_string; |
1421 | /* Go through all blocks of debug options, see if any matches our slab's name */ | |
1422 | while (next_block) { | |
1423 | next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false); | |
1424 | if (!iter) | |
1425 | continue; | |
1426 | /* Found a block that has a slab list, search it */ | |
1427 | while (*iter) { | |
1428 | char *end, *glob; | |
1429 | size_t cmplen; | |
1430 | ||
1431 | end = strchrnul(iter, ','); | |
1432 | if (next_block && next_block < end) | |
1433 | end = next_block - 1; | |
1434 | ||
1435 | glob = strnchr(iter, end - iter, '*'); | |
1436 | if (glob) | |
1437 | cmplen = glob - iter; | |
1438 | else | |
1439 | cmplen = max_t(size_t, len, (end - iter)); | |
c5fd3ca0 | 1440 | |
e17f1dfb VB |
1441 | if (!strncmp(name, iter, cmplen)) { |
1442 | flags |= block_flags; | |
1443 | return flags; | |
1444 | } | |
c5fd3ca0 | 1445 | |
e17f1dfb VB |
1446 | if (!*end || *end == ';') |
1447 | break; | |
1448 | iter = end + 1; | |
c5fd3ca0 | 1449 | } |
c5fd3ca0 | 1450 | } |
ba0268a8 | 1451 | |
484cfaca | 1452 | return flags | slub_debug; |
41ecc55b | 1453 | } |
b4a64718 | 1454 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1455 | static inline void setup_object_debug(struct kmem_cache *s, |
1456 | struct page *page, void *object) {} | |
a50b854e MWO |
1457 | static inline |
1458 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {} | |
41ecc55b | 1459 | |
3ec09742 | 1460 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1461 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1462 | |
282acb43 | 1463 | static inline int free_debug_processing( |
81084651 JDB |
1464 | struct kmem_cache *s, struct page *page, |
1465 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1466 | unsigned long addr) { return 0; } |
41ecc55b | 1467 | |
41ecc55b CL |
1468 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1469 | { return 1; } | |
1470 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1471 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1472 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1473 | struct page *page) {} | |
c65c1877 PZ |
1474 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1475 | struct page *page) {} | |
0293d1fd | 1476 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1477 | slab_flags_t flags, const char *name, |
51cc5068 | 1478 | void (*ctor)(void *)) |
ba0268a8 CL |
1479 | { |
1480 | return flags; | |
1481 | } | |
41ecc55b | 1482 | #define slub_debug 0 |
0f389ec6 | 1483 | |
fdaa45e9 IM |
1484 | #define disable_higher_order_debug 0 |
1485 | ||
0f389ec6 CL |
1486 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1487 | { return 0; } | |
26c02cf0 AB |
1488 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1489 | { return 0; } | |
205ab99d CL |
1490 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1491 | int objects) {} | |
1492 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1493 | int objects) {} | |
7d550c56 | 1494 | |
52f23478 | 1495 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 1496 | void **freelist, void *nextfree) |
52f23478 DZ |
1497 | { |
1498 | return false; | |
1499 | } | |
02e72cc6 AR |
1500 | #endif /* CONFIG_SLUB_DEBUG */ |
1501 | ||
1502 | /* | |
1503 | * Hooks for other subsystems that check memory allocations. In a typical | |
1504 | * production configuration these hooks all should produce no code at all. | |
1505 | */ | |
0116523c | 1506 | static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
d56791b3 | 1507 | { |
53128245 | 1508 | ptr = kasan_kmalloc_large(ptr, size, flags); |
a2f77575 | 1509 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ |
d56791b3 | 1510 | kmemleak_alloc(ptr, size, 1, flags); |
53128245 | 1511 | return ptr; |
d56791b3 RB |
1512 | } |
1513 | ||
ee3ce779 | 1514 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1515 | { |
1516 | kmemleak_free(x); | |
ee3ce779 | 1517 | kasan_kfree_large(x, _RET_IP_); |
d56791b3 RB |
1518 | } |
1519 | ||
c3895391 | 1520 | static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x) |
d56791b3 RB |
1521 | { |
1522 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1523 | |
02e72cc6 AR |
1524 | /* |
1525 | * Trouble is that we may no longer disable interrupts in the fast path | |
1526 | * So in order to make the debug calls that expect irqs to be | |
1527 | * disabled we need to disable interrupts temporarily. | |
1528 | */ | |
4675ff05 | 1529 | #ifdef CONFIG_LOCKDEP |
02e72cc6 AR |
1530 | { |
1531 | unsigned long flags; | |
1532 | ||
1533 | local_irq_save(flags); | |
02e72cc6 AR |
1534 | debug_check_no_locks_freed(x, s->object_size); |
1535 | local_irq_restore(flags); | |
1536 | } | |
1537 | #endif | |
1538 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1539 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1540 | |
cfbe1636 ME |
1541 | /* Use KCSAN to help debug racy use-after-free. */ |
1542 | if (!(s->flags & SLAB_TYPESAFE_BY_RCU)) | |
1543 | __kcsan_check_access(x, s->object_size, | |
1544 | KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT); | |
1545 | ||
c3895391 AK |
1546 | /* KASAN might put x into memory quarantine, delaying its reuse */ |
1547 | return kasan_slab_free(s, x, _RET_IP_); | |
02e72cc6 | 1548 | } |
205ab99d | 1549 | |
c3895391 AK |
1550 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
1551 | void **head, void **tail) | |
81084651 | 1552 | { |
6471384a AP |
1553 | |
1554 | void *object; | |
1555 | void *next = *head; | |
1556 | void *old_tail = *tail ? *tail : *head; | |
1557 | int rsize; | |
1558 | ||
aea4df4c LA |
1559 | /* Head and tail of the reconstructed freelist */ |
1560 | *head = NULL; | |
1561 | *tail = NULL; | |
1b7e816f | 1562 | |
aea4df4c LA |
1563 | do { |
1564 | object = next; | |
1565 | next = get_freepointer(s, object); | |
1566 | ||
1567 | if (slab_want_init_on_free(s)) { | |
6471384a AP |
1568 | /* |
1569 | * Clear the object and the metadata, but don't touch | |
1570 | * the redzone. | |
1571 | */ | |
aa1ef4d7 | 1572 | memset(kasan_reset_tag(object), 0, s->object_size); |
6471384a AP |
1573 | rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad |
1574 | : 0; | |
aa1ef4d7 | 1575 | memset((char *)kasan_reset_tag(object) + s->inuse, 0, |
6471384a | 1576 | s->size - s->inuse - rsize); |
81084651 | 1577 | |
aea4df4c | 1578 | } |
c3895391 AK |
1579 | /* If object's reuse doesn't have to be delayed */ |
1580 | if (!slab_free_hook(s, object)) { | |
1581 | /* Move object to the new freelist */ | |
1582 | set_freepointer(s, object, *head); | |
1583 | *head = object; | |
1584 | if (!*tail) | |
1585 | *tail = object; | |
1586 | } | |
1587 | } while (object != old_tail); | |
1588 | ||
1589 | if (*head == *tail) | |
1590 | *tail = NULL; | |
1591 | ||
1592 | return *head != NULL; | |
81084651 JDB |
1593 | } |
1594 | ||
4d176711 | 1595 | static void *setup_object(struct kmem_cache *s, struct page *page, |
588f8ba9 TG |
1596 | void *object) |
1597 | { | |
1598 | setup_object_debug(s, page, object); | |
4d176711 | 1599 | object = kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1600 | if (unlikely(s->ctor)) { |
1601 | kasan_unpoison_object_data(s, object); | |
1602 | s->ctor(object); | |
1603 | kasan_poison_object_data(s, object); | |
1604 | } | |
4d176711 | 1605 | return object; |
588f8ba9 TG |
1606 | } |
1607 | ||
81819f0f CL |
1608 | /* |
1609 | * Slab allocation and freeing | |
1610 | */ | |
5dfb4175 VD |
1611 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1612 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1613 | { |
5dfb4175 | 1614 | struct page *page; |
19af27af | 1615 | unsigned int order = oo_order(oo); |
65c3376a | 1616 | |
2154a336 | 1617 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1618 | page = alloc_pages(flags, order); |
65c3376a | 1619 | else |
96db800f | 1620 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1621 | |
5dfb4175 | 1622 | return page; |
65c3376a CL |
1623 | } |
1624 | ||
210e7a43 TG |
1625 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1626 | /* Pre-initialize the random sequence cache */ | |
1627 | static int init_cache_random_seq(struct kmem_cache *s) | |
1628 | { | |
19af27af | 1629 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 1630 | int err; |
210e7a43 | 1631 | |
a810007a SR |
1632 | /* Bailout if already initialised */ |
1633 | if (s->random_seq) | |
1634 | return 0; | |
1635 | ||
210e7a43 TG |
1636 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1637 | if (err) { | |
1638 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1639 | s->name); | |
1640 | return err; | |
1641 | } | |
1642 | ||
1643 | /* Transform to an offset on the set of pages */ | |
1644 | if (s->random_seq) { | |
19af27af AD |
1645 | unsigned int i; |
1646 | ||
210e7a43 TG |
1647 | for (i = 0; i < count; i++) |
1648 | s->random_seq[i] *= s->size; | |
1649 | } | |
1650 | return 0; | |
1651 | } | |
1652 | ||
1653 | /* Initialize each random sequence freelist per cache */ | |
1654 | static void __init init_freelist_randomization(void) | |
1655 | { | |
1656 | struct kmem_cache *s; | |
1657 | ||
1658 | mutex_lock(&slab_mutex); | |
1659 | ||
1660 | list_for_each_entry(s, &slab_caches, list) | |
1661 | init_cache_random_seq(s); | |
1662 | ||
1663 | mutex_unlock(&slab_mutex); | |
1664 | } | |
1665 | ||
1666 | /* Get the next entry on the pre-computed freelist randomized */ | |
1667 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1668 | unsigned long *pos, void *start, | |
1669 | unsigned long page_limit, | |
1670 | unsigned long freelist_count) | |
1671 | { | |
1672 | unsigned int idx; | |
1673 | ||
1674 | /* | |
1675 | * If the target page allocation failed, the number of objects on the | |
1676 | * page might be smaller than the usual size defined by the cache. | |
1677 | */ | |
1678 | do { | |
1679 | idx = s->random_seq[*pos]; | |
1680 | *pos += 1; | |
1681 | if (*pos >= freelist_count) | |
1682 | *pos = 0; | |
1683 | } while (unlikely(idx >= page_limit)); | |
1684 | ||
1685 | return (char *)start + idx; | |
1686 | } | |
1687 | ||
1688 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1689 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1690 | { | |
1691 | void *start; | |
1692 | void *cur; | |
1693 | void *next; | |
1694 | unsigned long idx, pos, page_limit, freelist_count; | |
1695 | ||
1696 | if (page->objects < 2 || !s->random_seq) | |
1697 | return false; | |
1698 | ||
1699 | freelist_count = oo_objects(s->oo); | |
1700 | pos = get_random_int() % freelist_count; | |
1701 | ||
1702 | page_limit = page->objects * s->size; | |
1703 | start = fixup_red_left(s, page_address(page)); | |
1704 | ||
1705 | /* First entry is used as the base of the freelist */ | |
1706 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1707 | freelist_count); | |
4d176711 | 1708 | cur = setup_object(s, page, cur); |
210e7a43 TG |
1709 | page->freelist = cur; |
1710 | ||
1711 | for (idx = 1; idx < page->objects; idx++) { | |
210e7a43 TG |
1712 | next = next_freelist_entry(s, page, &pos, start, page_limit, |
1713 | freelist_count); | |
4d176711 | 1714 | next = setup_object(s, page, next); |
210e7a43 TG |
1715 | set_freepointer(s, cur, next); |
1716 | cur = next; | |
1717 | } | |
210e7a43 TG |
1718 | set_freepointer(s, cur, NULL); |
1719 | ||
1720 | return true; | |
1721 | } | |
1722 | #else | |
1723 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1724 | { | |
1725 | return 0; | |
1726 | } | |
1727 | static inline void init_freelist_randomization(void) { } | |
1728 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1729 | { | |
1730 | return false; | |
1731 | } | |
1732 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1733 | ||
81819f0f CL |
1734 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1735 | { | |
06428780 | 1736 | struct page *page; |
834f3d11 | 1737 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1738 | gfp_t alloc_gfp; |
4d176711 | 1739 | void *start, *p, *next; |
a50b854e | 1740 | int idx; |
210e7a43 | 1741 | bool shuffle; |
81819f0f | 1742 | |
7e0528da CL |
1743 | flags &= gfp_allowed_mask; |
1744 | ||
d0164adc | 1745 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1746 | local_irq_enable(); |
1747 | ||
b7a49f0d | 1748 | flags |= s->allocflags; |
e12ba74d | 1749 | |
ba52270d PE |
1750 | /* |
1751 | * Let the initial higher-order allocation fail under memory pressure | |
1752 | * so we fall-back to the minimum order allocation. | |
1753 | */ | |
1754 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1755 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1756 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1757 | |
5dfb4175 | 1758 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1759 | if (unlikely(!page)) { |
1760 | oo = s->min; | |
80c3a998 | 1761 | alloc_gfp = flags; |
65c3376a CL |
1762 | /* |
1763 | * Allocation may have failed due to fragmentation. | |
1764 | * Try a lower order alloc if possible | |
1765 | */ | |
5dfb4175 | 1766 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1767 | if (unlikely(!page)) |
1768 | goto out; | |
1769 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1770 | } |
5a896d9e | 1771 | |
834f3d11 | 1772 | page->objects = oo_objects(oo); |
81819f0f | 1773 | |
1f3147b4 RG |
1774 | account_slab_page(page, oo_order(oo), s); |
1775 | ||
1b4f59e3 | 1776 | page->slab_cache = s; |
c03f94cc | 1777 | __SetPageSlab(page); |
2f064f34 | 1778 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1779 | SetPageSlabPfmemalloc(page); |
81819f0f | 1780 | |
a7101224 | 1781 | kasan_poison_slab(page); |
81819f0f | 1782 | |
a7101224 | 1783 | start = page_address(page); |
81819f0f | 1784 | |
a50b854e | 1785 | setup_page_debug(s, page, start); |
0316bec2 | 1786 | |
210e7a43 TG |
1787 | shuffle = shuffle_freelist(s, page); |
1788 | ||
1789 | if (!shuffle) { | |
4d176711 AK |
1790 | start = fixup_red_left(s, start); |
1791 | start = setup_object(s, page, start); | |
1792 | page->freelist = start; | |
18e50661 AK |
1793 | for (idx = 0, p = start; idx < page->objects - 1; idx++) { |
1794 | next = p + s->size; | |
1795 | next = setup_object(s, page, next); | |
1796 | set_freepointer(s, p, next); | |
1797 | p = next; | |
1798 | } | |
1799 | set_freepointer(s, p, NULL); | |
81819f0f | 1800 | } |
81819f0f | 1801 | |
e6e82ea1 | 1802 | page->inuse = page->objects; |
8cb0a506 | 1803 | page->frozen = 1; |
588f8ba9 | 1804 | |
81819f0f | 1805 | out: |
d0164adc | 1806 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1807 | local_irq_disable(); |
1808 | if (!page) | |
1809 | return NULL; | |
1810 | ||
588f8ba9 TG |
1811 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1812 | ||
81819f0f CL |
1813 | return page; |
1814 | } | |
1815 | ||
588f8ba9 TG |
1816 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1817 | { | |
44405099 LL |
1818 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
1819 | flags = kmalloc_fix_flags(flags); | |
588f8ba9 TG |
1820 | |
1821 | return allocate_slab(s, | |
1822 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1823 | } | |
1824 | ||
81819f0f CL |
1825 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1826 | { | |
834f3d11 CL |
1827 | int order = compound_order(page); |
1828 | int pages = 1 << order; | |
81819f0f | 1829 | |
8fc8d666 | 1830 | if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) { |
81819f0f CL |
1831 | void *p; |
1832 | ||
1833 | slab_pad_check(s, page); | |
224a88be CL |
1834 | for_each_object(p, s, page_address(page), |
1835 | page->objects) | |
f7cb1933 | 1836 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1837 | } |
1838 | ||
072bb0aa | 1839 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1840 | __ClearPageSlab(page); |
0c06dd75 VB |
1841 | /* In union with page->mapping where page allocator expects NULL */ |
1842 | page->slab_cache = NULL; | |
1eb5ac64 NP |
1843 | if (current->reclaim_state) |
1844 | current->reclaim_state->reclaimed_slab += pages; | |
74d555be | 1845 | unaccount_slab_page(page, order, s); |
27ee57c9 | 1846 | __free_pages(page, order); |
81819f0f CL |
1847 | } |
1848 | ||
1849 | static void rcu_free_slab(struct rcu_head *h) | |
1850 | { | |
bf68c214 | 1851 | struct page *page = container_of(h, struct page, rcu_head); |
da9a638c | 1852 | |
1b4f59e3 | 1853 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1854 | } |
1855 | ||
1856 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1857 | { | |
5f0d5a3a | 1858 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bf68c214 | 1859 | call_rcu(&page->rcu_head, rcu_free_slab); |
81819f0f CL |
1860 | } else |
1861 | __free_slab(s, page); | |
1862 | } | |
1863 | ||
1864 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1865 | { | |
205ab99d | 1866 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1867 | free_slab(s, page); |
1868 | } | |
1869 | ||
1870 | /* | |
5cc6eee8 | 1871 | * Management of partially allocated slabs. |
81819f0f | 1872 | */ |
1e4dd946 SR |
1873 | static inline void |
1874 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1875 | { |
e95eed57 | 1876 | n->nr_partial++; |
136333d1 | 1877 | if (tail == DEACTIVATE_TO_TAIL) |
916ac052 | 1878 | list_add_tail(&page->slab_list, &n->partial); |
7c2e132c | 1879 | else |
916ac052 | 1880 | list_add(&page->slab_list, &n->partial); |
81819f0f CL |
1881 | } |
1882 | ||
1e4dd946 SR |
1883 | static inline void add_partial(struct kmem_cache_node *n, |
1884 | struct page *page, int tail) | |
62e346a8 | 1885 | { |
c65c1877 | 1886 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1887 | __add_partial(n, page, tail); |
1888 | } | |
c65c1877 | 1889 | |
1e4dd946 SR |
1890 | static inline void remove_partial(struct kmem_cache_node *n, |
1891 | struct page *page) | |
1892 | { | |
1893 | lockdep_assert_held(&n->list_lock); | |
916ac052 | 1894 | list_del(&page->slab_list); |
52b4b950 | 1895 | n->nr_partial--; |
1e4dd946 SR |
1896 | } |
1897 | ||
81819f0f | 1898 | /* |
7ced3719 CL |
1899 | * Remove slab from the partial list, freeze it and |
1900 | * return the pointer to the freelist. | |
81819f0f | 1901 | * |
497b66f2 | 1902 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1903 | */ |
497b66f2 | 1904 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1905 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1906 | int mode, int *objects) |
81819f0f | 1907 | { |
2cfb7455 CL |
1908 | void *freelist; |
1909 | unsigned long counters; | |
1910 | struct page new; | |
1911 | ||
c65c1877 PZ |
1912 | lockdep_assert_held(&n->list_lock); |
1913 | ||
2cfb7455 CL |
1914 | /* |
1915 | * Zap the freelist and set the frozen bit. | |
1916 | * The old freelist is the list of objects for the | |
1917 | * per cpu allocation list. | |
1918 | */ | |
7ced3719 CL |
1919 | freelist = page->freelist; |
1920 | counters = page->counters; | |
1921 | new.counters = counters; | |
633b0764 | 1922 | *objects = new.objects - new.inuse; |
23910c50 | 1923 | if (mode) { |
7ced3719 | 1924 | new.inuse = page->objects; |
23910c50 PE |
1925 | new.freelist = NULL; |
1926 | } else { | |
1927 | new.freelist = freelist; | |
1928 | } | |
2cfb7455 | 1929 | |
a0132ac0 | 1930 | VM_BUG_ON(new.frozen); |
7ced3719 | 1931 | new.frozen = 1; |
2cfb7455 | 1932 | |
7ced3719 | 1933 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1934 | freelist, counters, |
02d7633f | 1935 | new.freelist, new.counters, |
7ced3719 | 1936 | "acquire_slab")) |
7ced3719 | 1937 | return NULL; |
2cfb7455 CL |
1938 | |
1939 | remove_partial(n, page); | |
7ced3719 | 1940 | WARN_ON(!freelist); |
49e22585 | 1941 | return freelist; |
81819f0f CL |
1942 | } |
1943 | ||
633b0764 | 1944 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1945 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1946 | |
81819f0f | 1947 | /* |
672bba3a | 1948 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1949 | */ |
8ba00bb6 JK |
1950 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1951 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1952 | { |
49e22585 CL |
1953 | struct page *page, *page2; |
1954 | void *object = NULL; | |
e5d9998f | 1955 | unsigned int available = 0; |
633b0764 | 1956 | int objects; |
81819f0f CL |
1957 | |
1958 | /* | |
1959 | * Racy check. If we mistakenly see no partial slabs then we | |
1960 | * just allocate an empty slab. If we mistakenly try to get a | |
70b6d25e | 1961 | * partial slab and there is none available then get_partial() |
672bba3a | 1962 | * will return NULL. |
81819f0f CL |
1963 | */ |
1964 | if (!n || !n->nr_partial) | |
1965 | return NULL; | |
1966 | ||
1967 | spin_lock(&n->list_lock); | |
916ac052 | 1968 | list_for_each_entry_safe(page, page2, &n->partial, slab_list) { |
8ba00bb6 | 1969 | void *t; |
49e22585 | 1970 | |
8ba00bb6 JK |
1971 | if (!pfmemalloc_match(page, flags)) |
1972 | continue; | |
1973 | ||
633b0764 | 1974 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 | 1975 | if (!t) |
8ff60eb0 | 1976 | continue; /* cmpxchg raced */ |
49e22585 | 1977 | |
633b0764 | 1978 | available += objects; |
12d79634 | 1979 | if (!object) { |
49e22585 | 1980 | c->page = page; |
49e22585 | 1981 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1982 | object = t; |
49e22585 | 1983 | } else { |
633b0764 | 1984 | put_cpu_partial(s, page, 0); |
8028dcea | 1985 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1986 | } |
345c905d | 1987 | if (!kmem_cache_has_cpu_partial(s) |
e6d0e1dc | 1988 | || available > slub_cpu_partial(s) / 2) |
49e22585 CL |
1989 | break; |
1990 | ||
497b66f2 | 1991 | } |
81819f0f | 1992 | spin_unlock(&n->list_lock); |
497b66f2 | 1993 | return object; |
81819f0f CL |
1994 | } |
1995 | ||
1996 | /* | |
672bba3a | 1997 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1998 | */ |
de3ec035 | 1999 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 2000 | struct kmem_cache_cpu *c) |
81819f0f CL |
2001 | { |
2002 | #ifdef CONFIG_NUMA | |
2003 | struct zonelist *zonelist; | |
dd1a239f | 2004 | struct zoneref *z; |
54a6eb5c | 2005 | struct zone *zone; |
97a225e6 | 2006 | enum zone_type highest_zoneidx = gfp_zone(flags); |
497b66f2 | 2007 | void *object; |
cc9a6c87 | 2008 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
2009 | |
2010 | /* | |
672bba3a CL |
2011 | * The defrag ratio allows a configuration of the tradeoffs between |
2012 | * inter node defragmentation and node local allocations. A lower | |
2013 | * defrag_ratio increases the tendency to do local allocations | |
2014 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 2015 | * |
672bba3a CL |
2016 | * If the defrag_ratio is set to 0 then kmalloc() always |
2017 | * returns node local objects. If the ratio is higher then kmalloc() | |
2018 | * may return off node objects because partial slabs are obtained | |
2019 | * from other nodes and filled up. | |
81819f0f | 2020 | * |
43efd3ea LP |
2021 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
2022 | * (which makes defrag_ratio = 1000) then every (well almost) | |
2023 | * allocation will first attempt to defrag slab caches on other nodes. | |
2024 | * This means scanning over all nodes to look for partial slabs which | |
2025 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 2026 | * with available objects. |
81819f0f | 2027 | */ |
9824601e CL |
2028 | if (!s->remote_node_defrag_ratio || |
2029 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
2030 | return NULL; |
2031 | ||
cc9a6c87 | 2032 | do { |
d26914d1 | 2033 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 2034 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
97a225e6 | 2035 | for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) { |
cc9a6c87 MG |
2036 | struct kmem_cache_node *n; |
2037 | ||
2038 | n = get_node(s, zone_to_nid(zone)); | |
2039 | ||
dee2f8aa | 2040 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 2041 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 2042 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
2043 | if (object) { |
2044 | /* | |
d26914d1 MG |
2045 | * Don't check read_mems_allowed_retry() |
2046 | * here - if mems_allowed was updated in | |
2047 | * parallel, that was a harmless race | |
2048 | * between allocation and the cpuset | |
2049 | * update | |
cc9a6c87 | 2050 | */ |
cc9a6c87 MG |
2051 | return object; |
2052 | } | |
c0ff7453 | 2053 | } |
81819f0f | 2054 | } |
d26914d1 | 2055 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
6dfd1b65 | 2056 | #endif /* CONFIG_NUMA */ |
81819f0f CL |
2057 | return NULL; |
2058 | } | |
2059 | ||
2060 | /* | |
2061 | * Get a partial page, lock it and return it. | |
2062 | */ | |
497b66f2 | 2063 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 2064 | struct kmem_cache_cpu *c) |
81819f0f | 2065 | { |
497b66f2 | 2066 | void *object; |
a561ce00 JK |
2067 | int searchnode = node; |
2068 | ||
2069 | if (node == NUMA_NO_NODE) | |
2070 | searchnode = numa_mem_id(); | |
81819f0f | 2071 | |
8ba00bb6 | 2072 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
2073 | if (object || node != NUMA_NO_NODE) |
2074 | return object; | |
81819f0f | 2075 | |
acd19fd1 | 2076 | return get_any_partial(s, flags, c); |
81819f0f CL |
2077 | } |
2078 | ||
923717cb | 2079 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2080 | /* |
0d645ed1 | 2081 | * Calculate the next globally unique transaction for disambiguation |
8a5ec0ba CL |
2082 | * during cmpxchg. The transactions start with the cpu number and are then |
2083 | * incremented by CONFIG_NR_CPUS. | |
2084 | */ | |
2085 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
2086 | #else | |
2087 | /* | |
2088 | * No preemption supported therefore also no need to check for | |
2089 | * different cpus. | |
2090 | */ | |
2091 | #define TID_STEP 1 | |
2092 | #endif | |
2093 | ||
2094 | static inline unsigned long next_tid(unsigned long tid) | |
2095 | { | |
2096 | return tid + TID_STEP; | |
2097 | } | |
2098 | ||
9d5f0be0 | 2099 | #ifdef SLUB_DEBUG_CMPXCHG |
8a5ec0ba CL |
2100 | static inline unsigned int tid_to_cpu(unsigned long tid) |
2101 | { | |
2102 | return tid % TID_STEP; | |
2103 | } | |
2104 | ||
2105 | static inline unsigned long tid_to_event(unsigned long tid) | |
2106 | { | |
2107 | return tid / TID_STEP; | |
2108 | } | |
9d5f0be0 | 2109 | #endif |
8a5ec0ba CL |
2110 | |
2111 | static inline unsigned int init_tid(int cpu) | |
2112 | { | |
2113 | return cpu; | |
2114 | } | |
2115 | ||
2116 | static inline void note_cmpxchg_failure(const char *n, | |
2117 | const struct kmem_cache *s, unsigned long tid) | |
2118 | { | |
2119 | #ifdef SLUB_DEBUG_CMPXCHG | |
2120 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2121 | ||
f9f58285 | 2122 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba | 2123 | |
923717cb | 2124 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2125 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) |
f9f58285 | 2126 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2127 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2128 | else | |
2129 | #endif | |
2130 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2131 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2132 | tid_to_event(tid), tid_to_event(actual_tid)); |
2133 | else | |
f9f58285 | 2134 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2135 | actual_tid, tid, next_tid(tid)); |
2136 | #endif | |
4fdccdfb | 2137 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2138 | } |
2139 | ||
788e1aad | 2140 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2141 | { |
8a5ec0ba CL |
2142 | int cpu; |
2143 | ||
2144 | for_each_possible_cpu(cpu) | |
2145 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 2146 | } |
2cfb7455 | 2147 | |
81819f0f CL |
2148 | /* |
2149 | * Remove the cpu slab | |
2150 | */ | |
d0e0ac97 | 2151 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
d4ff6d35 | 2152 | void *freelist, struct kmem_cache_cpu *c) |
81819f0f | 2153 | { |
2cfb7455 | 2154 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
2155 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
2156 | int lock = 0; | |
2157 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 2158 | void *nextfree; |
136333d1 | 2159 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2160 | struct page new; |
2161 | struct page old; | |
2162 | ||
2163 | if (page->freelist) { | |
84e554e6 | 2164 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2165 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2166 | } |
2167 | ||
894b8788 | 2168 | /* |
2cfb7455 CL |
2169 | * Stage one: Free all available per cpu objects back |
2170 | * to the page freelist while it is still frozen. Leave the | |
2171 | * last one. | |
2172 | * | |
2173 | * There is no need to take the list->lock because the page | |
2174 | * is still frozen. | |
2175 | */ | |
2176 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2177 | void *prior; | |
2178 | unsigned long counters; | |
2179 | ||
52f23478 DZ |
2180 | /* |
2181 | * If 'nextfree' is invalid, it is possible that the object at | |
2182 | * 'freelist' is already corrupted. So isolate all objects | |
2183 | * starting at 'freelist'. | |
2184 | */ | |
dc07a728 | 2185 | if (freelist_corrupted(s, page, &freelist, nextfree)) |
52f23478 DZ |
2186 | break; |
2187 | ||
2cfb7455 CL |
2188 | do { |
2189 | prior = page->freelist; | |
2190 | counters = page->counters; | |
2191 | set_freepointer(s, freelist, prior); | |
2192 | new.counters = counters; | |
2193 | new.inuse--; | |
a0132ac0 | 2194 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 2195 | |
1d07171c | 2196 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2197 | prior, counters, |
2198 | freelist, new.counters, | |
2199 | "drain percpu freelist")); | |
2200 | ||
2201 | freelist = nextfree; | |
2202 | } | |
2203 | ||
894b8788 | 2204 | /* |
2cfb7455 CL |
2205 | * Stage two: Ensure that the page is unfrozen while the |
2206 | * list presence reflects the actual number of objects | |
2207 | * during unfreeze. | |
2208 | * | |
2209 | * We setup the list membership and then perform a cmpxchg | |
2210 | * with the count. If there is a mismatch then the page | |
2211 | * is not unfrozen but the page is on the wrong list. | |
2212 | * | |
2213 | * Then we restart the process which may have to remove | |
2214 | * the page from the list that we just put it on again | |
2215 | * because the number of objects in the slab may have | |
2216 | * changed. | |
894b8788 | 2217 | */ |
2cfb7455 | 2218 | redo: |
894b8788 | 2219 | |
2cfb7455 CL |
2220 | old.freelist = page->freelist; |
2221 | old.counters = page->counters; | |
a0132ac0 | 2222 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2223 | |
2cfb7455 CL |
2224 | /* Determine target state of the slab */ |
2225 | new.counters = old.counters; | |
2226 | if (freelist) { | |
2227 | new.inuse--; | |
2228 | set_freepointer(s, freelist, old.freelist); | |
2229 | new.freelist = freelist; | |
2230 | } else | |
2231 | new.freelist = old.freelist; | |
2232 | ||
2233 | new.frozen = 0; | |
2234 | ||
8a5b20ae | 2235 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2236 | m = M_FREE; |
2237 | else if (new.freelist) { | |
2238 | m = M_PARTIAL; | |
2239 | if (!lock) { | |
2240 | lock = 1; | |
2241 | /* | |
8bb4e7a2 | 2242 | * Taking the spinlock removes the possibility |
2cfb7455 CL |
2243 | * that acquire_slab() will see a slab page that |
2244 | * is frozen | |
2245 | */ | |
2246 | spin_lock(&n->list_lock); | |
2247 | } | |
2248 | } else { | |
2249 | m = M_FULL; | |
965c4848 | 2250 | if (kmem_cache_debug_flags(s, SLAB_STORE_USER) && !lock) { |
2cfb7455 CL |
2251 | lock = 1; |
2252 | /* | |
2253 | * This also ensures that the scanning of full | |
2254 | * slabs from diagnostic functions will not see | |
2255 | * any frozen slabs. | |
2256 | */ | |
2257 | spin_lock(&n->list_lock); | |
2258 | } | |
2259 | } | |
2260 | ||
2261 | if (l != m) { | |
2cfb7455 | 2262 | if (l == M_PARTIAL) |
2cfb7455 | 2263 | remove_partial(n, page); |
2cfb7455 | 2264 | else if (l == M_FULL) |
c65c1877 | 2265 | remove_full(s, n, page); |
2cfb7455 | 2266 | |
88349a28 | 2267 | if (m == M_PARTIAL) |
2cfb7455 | 2268 | add_partial(n, page, tail); |
88349a28 | 2269 | else if (m == M_FULL) |
2cfb7455 | 2270 | add_full(s, n, page); |
2cfb7455 CL |
2271 | } |
2272 | ||
2273 | l = m; | |
1d07171c | 2274 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2275 | old.freelist, old.counters, |
2276 | new.freelist, new.counters, | |
2277 | "unfreezing slab")) | |
2278 | goto redo; | |
2279 | ||
2cfb7455 CL |
2280 | if (lock) |
2281 | spin_unlock(&n->list_lock); | |
2282 | ||
88349a28 WY |
2283 | if (m == M_PARTIAL) |
2284 | stat(s, tail); | |
2285 | else if (m == M_FULL) | |
2286 | stat(s, DEACTIVATE_FULL); | |
2287 | else if (m == M_FREE) { | |
2cfb7455 CL |
2288 | stat(s, DEACTIVATE_EMPTY); |
2289 | discard_slab(s, page); | |
2290 | stat(s, FREE_SLAB); | |
894b8788 | 2291 | } |
d4ff6d35 WY |
2292 | |
2293 | c->page = NULL; | |
2294 | c->freelist = NULL; | |
81819f0f CL |
2295 | } |
2296 | ||
d24ac77f JK |
2297 | /* |
2298 | * Unfreeze all the cpu partial slabs. | |
2299 | * | |
59a09917 CL |
2300 | * This function must be called with interrupts disabled |
2301 | * for the cpu using c (or some other guarantee must be there | |
2302 | * to guarantee no concurrent accesses). | |
d24ac77f | 2303 | */ |
59a09917 CL |
2304 | static void unfreeze_partials(struct kmem_cache *s, |
2305 | struct kmem_cache_cpu *c) | |
49e22585 | 2306 | { |
345c905d | 2307 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2308 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2309 | struct page *page, *discard_page = NULL; |
49e22585 | 2310 | |
4c7ba22e | 2311 | while ((page = slub_percpu_partial(c))) { |
49e22585 CL |
2312 | struct page new; |
2313 | struct page old; | |
2314 | ||
4c7ba22e | 2315 | slub_set_percpu_partial(c, page); |
43d77867 JK |
2316 | |
2317 | n2 = get_node(s, page_to_nid(page)); | |
2318 | if (n != n2) { | |
2319 | if (n) | |
2320 | spin_unlock(&n->list_lock); | |
2321 | ||
2322 | n = n2; | |
2323 | spin_lock(&n->list_lock); | |
2324 | } | |
49e22585 CL |
2325 | |
2326 | do { | |
2327 | ||
2328 | old.freelist = page->freelist; | |
2329 | old.counters = page->counters; | |
a0132ac0 | 2330 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2331 | |
2332 | new.counters = old.counters; | |
2333 | new.freelist = old.freelist; | |
2334 | ||
2335 | new.frozen = 0; | |
2336 | ||
d24ac77f | 2337 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2338 | old.freelist, old.counters, |
2339 | new.freelist, new.counters, | |
2340 | "unfreezing slab")); | |
2341 | ||
8a5b20ae | 2342 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2343 | page->next = discard_page; |
2344 | discard_page = page; | |
43d77867 JK |
2345 | } else { |
2346 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2347 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2348 | } |
2349 | } | |
2350 | ||
2351 | if (n) | |
2352 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2353 | |
2354 | while (discard_page) { | |
2355 | page = discard_page; | |
2356 | discard_page = discard_page->next; | |
2357 | ||
2358 | stat(s, DEACTIVATE_EMPTY); | |
2359 | discard_slab(s, page); | |
2360 | stat(s, FREE_SLAB); | |
2361 | } | |
6dfd1b65 | 2362 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2363 | } |
2364 | ||
2365 | /* | |
9234bae9 WY |
2366 | * Put a page that was just frozen (in __slab_free|get_partial_node) into a |
2367 | * partial page slot if available. | |
49e22585 CL |
2368 | * |
2369 | * If we did not find a slot then simply move all the partials to the | |
2370 | * per node partial list. | |
2371 | */ | |
633b0764 | 2372 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2373 | { |
345c905d | 2374 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2375 | struct page *oldpage; |
2376 | int pages; | |
2377 | int pobjects; | |
2378 | ||
d6e0b7fa | 2379 | preempt_disable(); |
49e22585 CL |
2380 | do { |
2381 | pages = 0; | |
2382 | pobjects = 0; | |
2383 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2384 | ||
2385 | if (oldpage) { | |
2386 | pobjects = oldpage->pobjects; | |
2387 | pages = oldpage->pages; | |
bbd4e305 | 2388 | if (drain && pobjects > slub_cpu_partial(s)) { |
49e22585 CL |
2389 | unsigned long flags; |
2390 | /* | |
2391 | * partial array is full. Move the existing | |
2392 | * set to the per node partial list. | |
2393 | */ | |
2394 | local_irq_save(flags); | |
59a09917 | 2395 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2396 | local_irq_restore(flags); |
e24fc410 | 2397 | oldpage = NULL; |
49e22585 CL |
2398 | pobjects = 0; |
2399 | pages = 0; | |
8028dcea | 2400 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2401 | } |
2402 | } | |
2403 | ||
2404 | pages++; | |
2405 | pobjects += page->objects - page->inuse; | |
2406 | ||
2407 | page->pages = pages; | |
2408 | page->pobjects = pobjects; | |
2409 | page->next = oldpage; | |
2410 | ||
d0e0ac97 CG |
2411 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2412 | != oldpage); | |
bbd4e305 | 2413 | if (unlikely(!slub_cpu_partial(s))) { |
d6e0b7fa VD |
2414 | unsigned long flags; |
2415 | ||
2416 | local_irq_save(flags); | |
2417 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2418 | local_irq_restore(flags); | |
2419 | } | |
2420 | preempt_enable(); | |
6dfd1b65 | 2421 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2422 | } |
2423 | ||
dfb4f096 | 2424 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2425 | { |
84e554e6 | 2426 | stat(s, CPUSLAB_FLUSH); |
d4ff6d35 | 2427 | deactivate_slab(s, c->page, c->freelist, c); |
c17dda40 CL |
2428 | |
2429 | c->tid = next_tid(c->tid); | |
81819f0f CL |
2430 | } |
2431 | ||
2432 | /* | |
2433 | * Flush cpu slab. | |
6446faa2 | 2434 | * |
81819f0f CL |
2435 | * Called from IPI handler with interrupts disabled. |
2436 | */ | |
0c710013 | 2437 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2438 | { |
9dfc6e68 | 2439 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2440 | |
1265ef2d WY |
2441 | if (c->page) |
2442 | flush_slab(s, c); | |
49e22585 | 2443 | |
1265ef2d | 2444 | unfreeze_partials(s, c); |
81819f0f CL |
2445 | } |
2446 | ||
2447 | static void flush_cpu_slab(void *d) | |
2448 | { | |
2449 | struct kmem_cache *s = d; | |
81819f0f | 2450 | |
dfb4f096 | 2451 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2452 | } |
2453 | ||
a8364d55 GBY |
2454 | static bool has_cpu_slab(int cpu, void *info) |
2455 | { | |
2456 | struct kmem_cache *s = info; | |
2457 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2458 | ||
a93cf07b | 2459 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2460 | } |
2461 | ||
81819f0f CL |
2462 | static void flush_all(struct kmem_cache *s) |
2463 | { | |
cb923159 | 2464 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1); |
81819f0f CL |
2465 | } |
2466 | ||
a96a87bf SAS |
2467 | /* |
2468 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2469 | * necessary. | |
2470 | */ | |
2471 | static int slub_cpu_dead(unsigned int cpu) | |
2472 | { | |
2473 | struct kmem_cache *s; | |
2474 | unsigned long flags; | |
2475 | ||
2476 | mutex_lock(&slab_mutex); | |
2477 | list_for_each_entry(s, &slab_caches, list) { | |
2478 | local_irq_save(flags); | |
2479 | __flush_cpu_slab(s, cpu); | |
2480 | local_irq_restore(flags); | |
2481 | } | |
2482 | mutex_unlock(&slab_mutex); | |
2483 | return 0; | |
2484 | } | |
2485 | ||
dfb4f096 CL |
2486 | /* |
2487 | * Check if the objects in a per cpu structure fit numa | |
2488 | * locality expectations. | |
2489 | */ | |
57d437d2 | 2490 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2491 | { |
2492 | #ifdef CONFIG_NUMA | |
6159d0f5 | 2493 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2494 | return 0; |
2495 | #endif | |
2496 | return 1; | |
2497 | } | |
2498 | ||
9a02d699 | 2499 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2500 | static int count_free(struct page *page) |
2501 | { | |
2502 | return page->objects - page->inuse; | |
2503 | } | |
2504 | ||
9a02d699 DR |
2505 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2506 | { | |
2507 | return atomic_long_read(&n->total_objects); | |
2508 | } | |
2509 | #endif /* CONFIG_SLUB_DEBUG */ | |
2510 | ||
2511 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2512 | static unsigned long count_partial(struct kmem_cache_node *n, |
2513 | int (*get_count)(struct page *)) | |
2514 | { | |
2515 | unsigned long flags; | |
2516 | unsigned long x = 0; | |
2517 | struct page *page; | |
2518 | ||
2519 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 2520 | list_for_each_entry(page, &n->partial, slab_list) |
781b2ba6 PE |
2521 | x += get_count(page); |
2522 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2523 | return x; | |
2524 | } | |
9a02d699 | 2525 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2526 | |
781b2ba6 PE |
2527 | static noinline void |
2528 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2529 | { | |
9a02d699 DR |
2530 | #ifdef CONFIG_SLUB_DEBUG |
2531 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2532 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2533 | int node; |
fa45dc25 | 2534 | struct kmem_cache_node *n; |
781b2ba6 | 2535 | |
9a02d699 DR |
2536 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2537 | return; | |
2538 | ||
5b3810e5 VB |
2539 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2540 | nid, gfpflags, &gfpflags); | |
19af27af | 2541 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2542 | s->name, s->object_size, s->size, oo_order(s->oo), |
2543 | oo_order(s->min)); | |
781b2ba6 | 2544 | |
3b0efdfa | 2545 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2546 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2547 | s->name); | |
fa5ec8a1 | 2548 | |
fa45dc25 | 2549 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2550 | unsigned long nr_slabs; |
2551 | unsigned long nr_objs; | |
2552 | unsigned long nr_free; | |
2553 | ||
26c02cf0 AB |
2554 | nr_free = count_partial(n, count_free); |
2555 | nr_slabs = node_nr_slabs(n); | |
2556 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2557 | |
f9f58285 | 2558 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2559 | node, nr_slabs, nr_objs, nr_free); |
2560 | } | |
9a02d699 | 2561 | #endif |
781b2ba6 PE |
2562 | } |
2563 | ||
497b66f2 CL |
2564 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2565 | int node, struct kmem_cache_cpu **pc) | |
2566 | { | |
6faa6833 | 2567 | void *freelist; |
188fd063 CL |
2568 | struct kmem_cache_cpu *c = *pc; |
2569 | struct page *page; | |
497b66f2 | 2570 | |
128227e7 MW |
2571 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
2572 | ||
188fd063 | 2573 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2574 | |
188fd063 CL |
2575 | if (freelist) |
2576 | return freelist; | |
2577 | ||
2578 | page = new_slab(s, flags, node); | |
497b66f2 | 2579 | if (page) { |
7c8e0181 | 2580 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2581 | if (c->page) |
2582 | flush_slab(s, c); | |
2583 | ||
2584 | /* | |
2585 | * No other reference to the page yet so we can | |
2586 | * muck around with it freely without cmpxchg | |
2587 | */ | |
6faa6833 | 2588 | freelist = page->freelist; |
497b66f2 CL |
2589 | page->freelist = NULL; |
2590 | ||
2591 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2592 | c->page = page; |
2593 | *pc = c; | |
edde82b6 | 2594 | } |
497b66f2 | 2595 | |
6faa6833 | 2596 | return freelist; |
497b66f2 CL |
2597 | } |
2598 | ||
072bb0aa MG |
2599 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2600 | { | |
2601 | if (unlikely(PageSlabPfmemalloc(page))) | |
2602 | return gfp_pfmemalloc_allowed(gfpflags); | |
2603 | ||
2604 | return true; | |
2605 | } | |
2606 | ||
213eeb9f | 2607 | /* |
d0e0ac97 CG |
2608 | * Check the page->freelist of a page and either transfer the freelist to the |
2609 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2610 | * |
2611 | * The page is still frozen if the return value is not NULL. | |
2612 | * | |
2613 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2614 | * |
2615 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2616 | */ |
2617 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2618 | { | |
2619 | struct page new; | |
2620 | unsigned long counters; | |
2621 | void *freelist; | |
2622 | ||
2623 | do { | |
2624 | freelist = page->freelist; | |
2625 | counters = page->counters; | |
6faa6833 | 2626 | |
213eeb9f | 2627 | new.counters = counters; |
a0132ac0 | 2628 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2629 | |
2630 | new.inuse = page->objects; | |
2631 | new.frozen = freelist != NULL; | |
2632 | ||
d24ac77f | 2633 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2634 | freelist, counters, |
2635 | NULL, new.counters, | |
2636 | "get_freelist")); | |
2637 | ||
2638 | return freelist; | |
2639 | } | |
2640 | ||
81819f0f | 2641 | /* |
894b8788 CL |
2642 | * Slow path. The lockless freelist is empty or we need to perform |
2643 | * debugging duties. | |
2644 | * | |
894b8788 CL |
2645 | * Processing is still very fast if new objects have been freed to the |
2646 | * regular freelist. In that case we simply take over the regular freelist | |
2647 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2648 | * |
894b8788 CL |
2649 | * If that is not working then we fall back to the partial lists. We take the |
2650 | * first element of the freelist as the object to allocate now and move the | |
2651 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2652 | * |
894b8788 | 2653 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2654 | * we need to allocate a new slab. This is the slowest path since it involves |
2655 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2656 | * |
2657 | * Version of __slab_alloc to use when we know that interrupts are | |
2658 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2659 | */ |
a380a3c7 | 2660 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2661 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2662 | { |
6faa6833 | 2663 | void *freelist; |
f6e7def7 | 2664 | struct page *page; |
81819f0f | 2665 | |
9f986d99 AW |
2666 | stat(s, ALLOC_SLOWPATH); |
2667 | ||
f6e7def7 | 2668 | page = c->page; |
0715e6c5 VB |
2669 | if (!page) { |
2670 | /* | |
2671 | * if the node is not online or has no normal memory, just | |
2672 | * ignore the node constraint | |
2673 | */ | |
2674 | if (unlikely(node != NUMA_NO_NODE && | |
2675 | !node_state(node, N_NORMAL_MEMORY))) | |
2676 | node = NUMA_NO_NODE; | |
81819f0f | 2677 | goto new_slab; |
0715e6c5 | 2678 | } |
49e22585 | 2679 | redo: |
6faa6833 | 2680 | |
57d437d2 | 2681 | if (unlikely(!node_match(page, node))) { |
0715e6c5 VB |
2682 | /* |
2683 | * same as above but node_match() being false already | |
2684 | * implies node != NUMA_NO_NODE | |
2685 | */ | |
2686 | if (!node_state(node, N_NORMAL_MEMORY)) { | |
2687 | node = NUMA_NO_NODE; | |
2688 | goto redo; | |
2689 | } else { | |
a561ce00 | 2690 | stat(s, ALLOC_NODE_MISMATCH); |
d4ff6d35 | 2691 | deactivate_slab(s, page, c->freelist, c); |
a561ce00 JK |
2692 | goto new_slab; |
2693 | } | |
fc59c053 | 2694 | } |
6446faa2 | 2695 | |
072bb0aa MG |
2696 | /* |
2697 | * By rights, we should be searching for a slab page that was | |
2698 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2699 | * information when the page leaves the per-cpu allocator | |
2700 | */ | |
2701 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
d4ff6d35 | 2702 | deactivate_slab(s, page, c->freelist, c); |
072bb0aa MG |
2703 | goto new_slab; |
2704 | } | |
2705 | ||
73736e03 | 2706 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2707 | freelist = c->freelist; |
2708 | if (freelist) | |
73736e03 | 2709 | goto load_freelist; |
03e404af | 2710 | |
f6e7def7 | 2711 | freelist = get_freelist(s, page); |
6446faa2 | 2712 | |
6faa6833 | 2713 | if (!freelist) { |
03e404af CL |
2714 | c->page = NULL; |
2715 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2716 | goto new_slab; |
03e404af | 2717 | } |
6446faa2 | 2718 | |
84e554e6 | 2719 | stat(s, ALLOC_REFILL); |
6446faa2 | 2720 | |
894b8788 | 2721 | load_freelist: |
507effea CL |
2722 | /* |
2723 | * freelist is pointing to the list of objects to be used. | |
2724 | * page is pointing to the page from which the objects are obtained. | |
2725 | * That page must be frozen for per cpu allocations to work. | |
2726 | */ | |
a0132ac0 | 2727 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2728 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2729 | c->tid = next_tid(c->tid); |
6faa6833 | 2730 | return freelist; |
81819f0f | 2731 | |
81819f0f | 2732 | new_slab: |
2cfb7455 | 2733 | |
a93cf07b WY |
2734 | if (slub_percpu_partial(c)) { |
2735 | page = c->page = slub_percpu_partial(c); | |
2736 | slub_set_percpu_partial(c, page); | |
49e22585 | 2737 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2738 | goto redo; |
81819f0f CL |
2739 | } |
2740 | ||
188fd063 | 2741 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2742 | |
f4697436 | 2743 | if (unlikely(!freelist)) { |
9a02d699 | 2744 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2745 | return NULL; |
81819f0f | 2746 | } |
2cfb7455 | 2747 | |
f6e7def7 | 2748 | page = c->page; |
5091b74a | 2749 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2750 | goto load_freelist; |
2cfb7455 | 2751 | |
497b66f2 | 2752 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2753 | if (kmem_cache_debug(s) && |
2754 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2755 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2756 | |
d4ff6d35 | 2757 | deactivate_slab(s, page, get_freepointer(s, freelist), c); |
6faa6833 | 2758 | return freelist; |
894b8788 CL |
2759 | } |
2760 | ||
a380a3c7 CL |
2761 | /* |
2762 | * Another one that disabled interrupt and compensates for possible | |
2763 | * cpu changes by refetching the per cpu area pointer. | |
2764 | */ | |
2765 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2766 | unsigned long addr, struct kmem_cache_cpu *c) | |
2767 | { | |
2768 | void *p; | |
2769 | unsigned long flags; | |
2770 | ||
2771 | local_irq_save(flags); | |
923717cb | 2772 | #ifdef CONFIG_PREEMPTION |
a380a3c7 CL |
2773 | /* |
2774 | * We may have been preempted and rescheduled on a different | |
2775 | * cpu before disabling interrupts. Need to reload cpu area | |
2776 | * pointer. | |
2777 | */ | |
2778 | c = this_cpu_ptr(s->cpu_slab); | |
2779 | #endif | |
2780 | ||
2781 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2782 | local_irq_restore(flags); | |
2783 | return p; | |
2784 | } | |
2785 | ||
0f181f9f AP |
2786 | /* |
2787 | * If the object has been wiped upon free, make sure it's fully initialized by | |
2788 | * zeroing out freelist pointer. | |
2789 | */ | |
2790 | static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s, | |
2791 | void *obj) | |
2792 | { | |
2793 | if (unlikely(slab_want_init_on_free(s)) && obj) | |
ce5716c6 AK |
2794 | memset((void *)((char *)kasan_reset_tag(obj) + s->offset), |
2795 | 0, sizeof(void *)); | |
0f181f9f AP |
2796 | } |
2797 | ||
894b8788 CL |
2798 | /* |
2799 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2800 | * have the fastpath folded into their functions. So no function call | |
2801 | * overhead for requests that can be satisfied on the fastpath. | |
2802 | * | |
2803 | * The fastpath works by first checking if the lockless freelist can be used. | |
2804 | * If not then __slab_alloc is called for slow processing. | |
2805 | * | |
2806 | * Otherwise we can simply pick the next object from the lockless free list. | |
2807 | */ | |
2b847c3c | 2808 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2809 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2810 | { |
03ec0ed5 | 2811 | void *object; |
dfb4f096 | 2812 | struct kmem_cache_cpu *c; |
57d437d2 | 2813 | struct page *page; |
8a5ec0ba | 2814 | unsigned long tid; |
964d4bd3 | 2815 | struct obj_cgroup *objcg = NULL; |
1f84260c | 2816 | |
964d4bd3 | 2817 | s = slab_pre_alloc_hook(s, &objcg, 1, gfpflags); |
8135be5a | 2818 | if (!s) |
773ff60e | 2819 | return NULL; |
8a5ec0ba | 2820 | redo: |
8a5ec0ba CL |
2821 | /* |
2822 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2823 | * enabled. We may switch back and forth between cpus while | |
2824 | * reading from one cpu area. That does not matter as long | |
2825 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2826 | * |
9aabf810 | 2827 | * We should guarantee that tid and kmem_cache are retrieved on |
923717cb | 2828 | * the same cpu. It could be different if CONFIG_PREEMPTION so we need |
9aabf810 | 2829 | * to check if it is matched or not. |
8a5ec0ba | 2830 | */ |
9aabf810 JK |
2831 | do { |
2832 | tid = this_cpu_read(s->cpu_slab->tid); | |
2833 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 2834 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 2835 | unlikely(tid != READ_ONCE(c->tid))); |
9aabf810 JK |
2836 | |
2837 | /* | |
2838 | * Irqless object alloc/free algorithm used here depends on sequence | |
2839 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2840 | * on c to guarantee that object and page associated with previous tid | |
2841 | * won't be used with current tid. If we fetch tid first, object and | |
2842 | * page could be one associated with next tid and our alloc/free | |
2843 | * request will be failed. In this case, we will retry. So, no problem. | |
2844 | */ | |
2845 | barrier(); | |
8a5ec0ba | 2846 | |
8a5ec0ba CL |
2847 | /* |
2848 | * The transaction ids are globally unique per cpu and per operation on | |
2849 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2850 | * occurs on the right processor and that there was no operation on the | |
2851 | * linked list in between. | |
2852 | */ | |
8a5ec0ba | 2853 | |
9dfc6e68 | 2854 | object = c->freelist; |
57d437d2 | 2855 | page = c->page; |
22e4663e | 2856 | if (unlikely(!object || !page || !node_match(page, node))) { |
dfb4f096 | 2857 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 | 2858 | } else { |
0ad9500e ED |
2859 | void *next_object = get_freepointer_safe(s, object); |
2860 | ||
8a5ec0ba | 2861 | /* |
25985edc | 2862 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2863 | * operation and if we are on the right processor. |
2864 | * | |
d0e0ac97 CG |
2865 | * The cmpxchg does the following atomically (without lock |
2866 | * semantics!) | |
8a5ec0ba CL |
2867 | * 1. Relocate first pointer to the current per cpu area. |
2868 | * 2. Verify that tid and freelist have not been changed | |
2869 | * 3. If they were not changed replace tid and freelist | |
2870 | * | |
d0e0ac97 CG |
2871 | * Since this is without lock semantics the protection is only |
2872 | * against code executing on this cpu *not* from access by | |
2873 | * other cpus. | |
8a5ec0ba | 2874 | */ |
933393f5 | 2875 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2876 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2877 | object, tid, | |
0ad9500e | 2878 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2879 | |
2880 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2881 | goto redo; | |
2882 | } | |
0ad9500e | 2883 | prefetch_freepointer(s, next_object); |
84e554e6 | 2884 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2885 | } |
0f181f9f | 2886 | |
ce5716c6 | 2887 | maybe_wipe_obj_freeptr(s, object); |
8a5ec0ba | 2888 | |
6471384a | 2889 | if (unlikely(slab_want_init_on_alloc(gfpflags, s)) && object) |
aa1ef4d7 | 2890 | memset(kasan_reset_tag(object), 0, s->object_size); |
d07dbea4 | 2891 | |
964d4bd3 | 2892 | slab_post_alloc_hook(s, objcg, gfpflags, 1, &object); |
5a896d9e | 2893 | |
894b8788 | 2894 | return object; |
81819f0f CL |
2895 | } |
2896 | ||
2b847c3c EG |
2897 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2898 | gfp_t gfpflags, unsigned long addr) | |
2899 | { | |
2900 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2901 | } | |
2902 | ||
81819f0f CL |
2903 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2904 | { | |
2b847c3c | 2905 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2906 | |
d0e0ac97 CG |
2907 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2908 | s->size, gfpflags); | |
5b882be4 EGM |
2909 | |
2910 | return ret; | |
81819f0f CL |
2911 | } |
2912 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2913 | ||
0f24f128 | 2914 | #ifdef CONFIG_TRACING |
4a92379b RK |
2915 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2916 | { | |
2b847c3c | 2917 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2918 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0116523c | 2919 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2920 | return ret; |
2921 | } | |
2922 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2923 | #endif |
2924 | ||
81819f0f CL |
2925 | #ifdef CONFIG_NUMA |
2926 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2927 | { | |
2b847c3c | 2928 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2929 | |
ca2b84cb | 2930 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2931 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2932 | |
2933 | return ret; | |
81819f0f CL |
2934 | } |
2935 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2936 | |
0f24f128 | 2937 | #ifdef CONFIG_TRACING |
4a92379b | 2938 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2939 | gfp_t gfpflags, |
4a92379b | 2940 | int node, size_t size) |
5b882be4 | 2941 | { |
2b847c3c | 2942 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2943 | |
2944 | trace_kmalloc_node(_RET_IP_, ret, | |
2945 | size, s->size, gfpflags, node); | |
0316bec2 | 2946 | |
0116523c | 2947 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2948 | return ret; |
5b882be4 | 2949 | } |
4a92379b | 2950 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2951 | #endif |
6dfd1b65 | 2952 | #endif /* CONFIG_NUMA */ |
5b882be4 | 2953 | |
81819f0f | 2954 | /* |
94e4d712 | 2955 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2956 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2957 | * |
894b8788 CL |
2958 | * So we still attempt to reduce cache line usage. Just take the slab |
2959 | * lock and free the item. If there is no additional partial page | |
2960 | * handling required then we can return immediately. | |
81819f0f | 2961 | */ |
894b8788 | 2962 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2963 | void *head, void *tail, int cnt, |
2964 | unsigned long addr) | |
2965 | ||
81819f0f CL |
2966 | { |
2967 | void *prior; | |
2cfb7455 | 2968 | int was_frozen; |
2cfb7455 CL |
2969 | struct page new; |
2970 | unsigned long counters; | |
2971 | struct kmem_cache_node *n = NULL; | |
3f649ab7 | 2972 | unsigned long flags; |
81819f0f | 2973 | |
8a5ec0ba | 2974 | stat(s, FREE_SLOWPATH); |
81819f0f | 2975 | |
19c7ff9e | 2976 | if (kmem_cache_debug(s) && |
282acb43 | 2977 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2978 | return; |
6446faa2 | 2979 | |
2cfb7455 | 2980 | do { |
837d678d JK |
2981 | if (unlikely(n)) { |
2982 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2983 | n = NULL; | |
2984 | } | |
2cfb7455 CL |
2985 | prior = page->freelist; |
2986 | counters = page->counters; | |
81084651 | 2987 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2988 | new.counters = counters; |
2989 | was_frozen = new.frozen; | |
81084651 | 2990 | new.inuse -= cnt; |
837d678d | 2991 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2992 | |
c65c1877 | 2993 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2994 | |
2995 | /* | |
d0e0ac97 CG |
2996 | * Slab was on no list before and will be |
2997 | * partially empty | |
2998 | * We can defer the list move and instead | |
2999 | * freeze it. | |
49e22585 CL |
3000 | */ |
3001 | new.frozen = 1; | |
3002 | ||
c65c1877 | 3003 | } else { /* Needs to be taken off a list */ |
49e22585 | 3004 | |
b455def2 | 3005 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
3006 | /* |
3007 | * Speculatively acquire the list_lock. | |
3008 | * If the cmpxchg does not succeed then we may | |
3009 | * drop the list_lock without any processing. | |
3010 | * | |
3011 | * Otherwise the list_lock will synchronize with | |
3012 | * other processors updating the list of slabs. | |
3013 | */ | |
3014 | spin_lock_irqsave(&n->list_lock, flags); | |
3015 | ||
3016 | } | |
2cfb7455 | 3017 | } |
81819f0f | 3018 | |
2cfb7455 CL |
3019 | } while (!cmpxchg_double_slab(s, page, |
3020 | prior, counters, | |
81084651 | 3021 | head, new.counters, |
2cfb7455 | 3022 | "__slab_free")); |
81819f0f | 3023 | |
2cfb7455 | 3024 | if (likely(!n)) { |
49e22585 | 3025 | |
c270cf30 AW |
3026 | if (likely(was_frozen)) { |
3027 | /* | |
3028 | * The list lock was not taken therefore no list | |
3029 | * activity can be necessary. | |
3030 | */ | |
3031 | stat(s, FREE_FROZEN); | |
3032 | } else if (new.frozen) { | |
3033 | /* | |
3034 | * If we just froze the page then put it onto the | |
3035 | * per cpu partial list. | |
3036 | */ | |
49e22585 | 3037 | put_cpu_partial(s, page, 1); |
8028dcea AS |
3038 | stat(s, CPU_PARTIAL_FREE); |
3039 | } | |
c270cf30 | 3040 | |
b455def2 L |
3041 | return; |
3042 | } | |
81819f0f | 3043 | |
8a5b20ae | 3044 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
3045 | goto slab_empty; |
3046 | ||
81819f0f | 3047 | /* |
837d678d JK |
3048 | * Objects left in the slab. If it was not on the partial list before |
3049 | * then add it. | |
81819f0f | 3050 | */ |
345c905d | 3051 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
a4d3f891 | 3052 | remove_full(s, n, page); |
837d678d JK |
3053 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
3054 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 3055 | } |
80f08c19 | 3056 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
3057 | return; |
3058 | ||
3059 | slab_empty: | |
a973e9dd | 3060 | if (prior) { |
81819f0f | 3061 | /* |
6fbabb20 | 3062 | * Slab on the partial list. |
81819f0f | 3063 | */ |
5cc6eee8 | 3064 | remove_partial(n, page); |
84e554e6 | 3065 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 3066 | } else { |
6fbabb20 | 3067 | /* Slab must be on the full list */ |
c65c1877 PZ |
3068 | remove_full(s, n, page); |
3069 | } | |
2cfb7455 | 3070 | |
80f08c19 | 3071 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 3072 | stat(s, FREE_SLAB); |
81819f0f | 3073 | discard_slab(s, page); |
81819f0f CL |
3074 | } |
3075 | ||
894b8788 CL |
3076 | /* |
3077 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
3078 | * can perform fastpath freeing without additional function calls. | |
3079 | * | |
3080 | * The fastpath is only possible if we are freeing to the current cpu slab | |
3081 | * of this processor. This typically the case if we have just allocated | |
3082 | * the item before. | |
3083 | * | |
3084 | * If fastpath is not possible then fall back to __slab_free where we deal | |
3085 | * with all sorts of special processing. | |
81084651 JDB |
3086 | * |
3087 | * Bulk free of a freelist with several objects (all pointing to the | |
3088 | * same page) possible by specifying head and tail ptr, plus objects | |
3089 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 3090 | */ |
80a9201a AP |
3091 | static __always_inline void do_slab_free(struct kmem_cache *s, |
3092 | struct page *page, void *head, void *tail, | |
3093 | int cnt, unsigned long addr) | |
894b8788 | 3094 | { |
81084651 | 3095 | void *tail_obj = tail ? : head; |
dfb4f096 | 3096 | struct kmem_cache_cpu *c; |
8a5ec0ba | 3097 | unsigned long tid; |
964d4bd3 | 3098 | |
d1b2cf6c | 3099 | memcg_slab_free_hook(s, &head, 1); |
8a5ec0ba CL |
3100 | redo: |
3101 | /* | |
3102 | * Determine the currently cpus per cpu slab. | |
3103 | * The cpu may change afterward. However that does not matter since | |
3104 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 3105 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 3106 | */ |
9aabf810 JK |
3107 | do { |
3108 | tid = this_cpu_read(s->cpu_slab->tid); | |
3109 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 3110 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 3111 | unlikely(tid != READ_ONCE(c->tid))); |
c016b0bd | 3112 | |
9aabf810 JK |
3113 | /* Same with comment on barrier() in slab_alloc_node() */ |
3114 | barrier(); | |
c016b0bd | 3115 | |
442b06bc | 3116 | if (likely(page == c->page)) { |
5076190d LT |
3117 | void **freelist = READ_ONCE(c->freelist); |
3118 | ||
3119 | set_freepointer(s, tail_obj, freelist); | |
8a5ec0ba | 3120 | |
933393f5 | 3121 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba | 3122 | s->cpu_slab->freelist, s->cpu_slab->tid, |
5076190d | 3123 | freelist, tid, |
81084651 | 3124 | head, next_tid(tid)))) { |
8a5ec0ba CL |
3125 | |
3126 | note_cmpxchg_failure("slab_free", s, tid); | |
3127 | goto redo; | |
3128 | } | |
84e554e6 | 3129 | stat(s, FREE_FASTPATH); |
894b8788 | 3130 | } else |
81084651 | 3131 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 3132 | |
894b8788 CL |
3133 | } |
3134 | ||
80a9201a AP |
3135 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
3136 | void *head, void *tail, int cnt, | |
3137 | unsigned long addr) | |
3138 | { | |
80a9201a | 3139 | /* |
c3895391 AK |
3140 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
3141 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 3142 | */ |
c3895391 AK |
3143 | if (slab_free_freelist_hook(s, &head, &tail)) |
3144 | do_slab_free(s, page, head, tail, cnt, addr); | |
80a9201a AP |
3145 | } |
3146 | ||
2bd926b4 | 3147 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
3148 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
3149 | { | |
3150 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
3151 | } | |
3152 | #endif | |
3153 | ||
81819f0f CL |
3154 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3155 | { | |
b9ce5ef4 GC |
3156 | s = cache_from_obj(s, x); |
3157 | if (!s) | |
79576102 | 3158 | return; |
81084651 | 3159 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 3160 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
3161 | } |
3162 | EXPORT_SYMBOL(kmem_cache_free); | |
3163 | ||
d0ecd894 | 3164 | struct detached_freelist { |
fbd02630 | 3165 | struct page *page; |
d0ecd894 JDB |
3166 | void *tail; |
3167 | void *freelist; | |
3168 | int cnt; | |
376bf125 | 3169 | struct kmem_cache *s; |
d0ecd894 | 3170 | }; |
fbd02630 | 3171 | |
d0ecd894 JDB |
3172 | /* |
3173 | * This function progressively scans the array with free objects (with | |
3174 | * a limited look ahead) and extract objects belonging to the same | |
3175 | * page. It builds a detached freelist directly within the given | |
3176 | * page/objects. This can happen without any need for | |
3177 | * synchronization, because the objects are owned by running process. | |
3178 | * The freelist is build up as a single linked list in the objects. | |
3179 | * The idea is, that this detached freelist can then be bulk | |
3180 | * transferred to the real freelist(s), but only requiring a single | |
3181 | * synchronization primitive. Look ahead in the array is limited due | |
3182 | * to performance reasons. | |
3183 | */ | |
376bf125 JDB |
3184 | static inline |
3185 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3186 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3187 | { |
3188 | size_t first_skipped_index = 0; | |
3189 | int lookahead = 3; | |
3190 | void *object; | |
ca257195 | 3191 | struct page *page; |
fbd02630 | 3192 | |
d0ecd894 JDB |
3193 | /* Always re-init detached_freelist */ |
3194 | df->page = NULL; | |
fbd02630 | 3195 | |
d0ecd894 JDB |
3196 | do { |
3197 | object = p[--size]; | |
ca257195 | 3198 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3199 | } while (!object && size); |
3eed034d | 3200 | |
d0ecd894 JDB |
3201 | if (!object) |
3202 | return 0; | |
fbd02630 | 3203 | |
ca257195 JDB |
3204 | page = virt_to_head_page(object); |
3205 | if (!s) { | |
3206 | /* Handle kalloc'ed objects */ | |
3207 | if (unlikely(!PageSlab(page))) { | |
3208 | BUG_ON(!PageCompound(page)); | |
3209 | kfree_hook(object); | |
4949148a | 3210 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3211 | p[size] = NULL; /* mark object processed */ |
3212 | return size; | |
3213 | } | |
3214 | /* Derive kmem_cache from object */ | |
3215 | df->s = page->slab_cache; | |
3216 | } else { | |
3217 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3218 | } | |
376bf125 | 3219 | |
d0ecd894 | 3220 | /* Start new detached freelist */ |
ca257195 | 3221 | df->page = page; |
376bf125 | 3222 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3223 | df->tail = object; |
3224 | df->freelist = object; | |
3225 | p[size] = NULL; /* mark object processed */ | |
3226 | df->cnt = 1; | |
3227 | ||
3228 | while (size) { | |
3229 | object = p[--size]; | |
3230 | if (!object) | |
3231 | continue; /* Skip processed objects */ | |
3232 | ||
3233 | /* df->page is always set at this point */ | |
3234 | if (df->page == virt_to_head_page(object)) { | |
3235 | /* Opportunity build freelist */ | |
376bf125 | 3236 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3237 | df->freelist = object; |
3238 | df->cnt++; | |
3239 | p[size] = NULL; /* mark object processed */ | |
3240 | ||
3241 | continue; | |
fbd02630 | 3242 | } |
d0ecd894 JDB |
3243 | |
3244 | /* Limit look ahead search */ | |
3245 | if (!--lookahead) | |
3246 | break; | |
3247 | ||
3248 | if (!first_skipped_index) | |
3249 | first_skipped_index = size + 1; | |
fbd02630 | 3250 | } |
d0ecd894 JDB |
3251 | |
3252 | return first_skipped_index; | |
3253 | } | |
3254 | ||
d0ecd894 | 3255 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3256 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3257 | { |
3258 | if (WARN_ON(!size)) | |
3259 | return; | |
3260 | ||
d1b2cf6c | 3261 | memcg_slab_free_hook(s, p, size); |
d0ecd894 JDB |
3262 | do { |
3263 | struct detached_freelist df; | |
3264 | ||
3265 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3266 | if (!df.page) |
d0ecd894 JDB |
3267 | continue; |
3268 | ||
376bf125 | 3269 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 3270 | } while (likely(size)); |
484748f0 CL |
3271 | } |
3272 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3273 | ||
994eb764 | 3274 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3275 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3276 | void **p) | |
484748f0 | 3277 | { |
994eb764 JDB |
3278 | struct kmem_cache_cpu *c; |
3279 | int i; | |
964d4bd3 | 3280 | struct obj_cgroup *objcg = NULL; |
994eb764 | 3281 | |
03ec0ed5 | 3282 | /* memcg and kmem_cache debug support */ |
964d4bd3 | 3283 | s = slab_pre_alloc_hook(s, &objcg, size, flags); |
03ec0ed5 JDB |
3284 | if (unlikely(!s)) |
3285 | return false; | |
994eb764 JDB |
3286 | /* |
3287 | * Drain objects in the per cpu slab, while disabling local | |
3288 | * IRQs, which protects against PREEMPT and interrupts | |
3289 | * handlers invoking normal fastpath. | |
3290 | */ | |
3291 | local_irq_disable(); | |
3292 | c = this_cpu_ptr(s->cpu_slab); | |
3293 | ||
3294 | for (i = 0; i < size; i++) { | |
3295 | void *object = c->freelist; | |
3296 | ||
ebe909e0 | 3297 | if (unlikely(!object)) { |
fd4d9c7d JH |
3298 | /* |
3299 | * We may have removed an object from c->freelist using | |
3300 | * the fastpath in the previous iteration; in that case, | |
3301 | * c->tid has not been bumped yet. | |
3302 | * Since ___slab_alloc() may reenable interrupts while | |
3303 | * allocating memory, we should bump c->tid now. | |
3304 | */ | |
3305 | c->tid = next_tid(c->tid); | |
3306 | ||
ebe909e0 JDB |
3307 | /* |
3308 | * Invoking slow path likely have side-effect | |
3309 | * of re-populating per CPU c->freelist | |
3310 | */ | |
87098373 | 3311 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3312 | _RET_IP_, c); |
87098373 CL |
3313 | if (unlikely(!p[i])) |
3314 | goto error; | |
3315 | ||
ebe909e0 | 3316 | c = this_cpu_ptr(s->cpu_slab); |
0f181f9f AP |
3317 | maybe_wipe_obj_freeptr(s, p[i]); |
3318 | ||
ebe909e0 JDB |
3319 | continue; /* goto for-loop */ |
3320 | } | |
994eb764 JDB |
3321 | c->freelist = get_freepointer(s, object); |
3322 | p[i] = object; | |
0f181f9f | 3323 | maybe_wipe_obj_freeptr(s, p[i]); |
994eb764 JDB |
3324 | } |
3325 | c->tid = next_tid(c->tid); | |
3326 | local_irq_enable(); | |
3327 | ||
3328 | /* Clear memory outside IRQ disabled fastpath loop */ | |
6471384a | 3329 | if (unlikely(slab_want_init_on_alloc(flags, s))) { |
994eb764 JDB |
3330 | int j; |
3331 | ||
3332 | for (j = 0; j < i; j++) | |
ce5716c6 | 3333 | memset(kasan_reset_tag(p[j]), 0, s->object_size); |
994eb764 JDB |
3334 | } |
3335 | ||
03ec0ed5 | 3336 | /* memcg and kmem_cache debug support */ |
964d4bd3 | 3337 | slab_post_alloc_hook(s, objcg, flags, size, p); |
865762a8 | 3338 | return i; |
87098373 | 3339 | error: |
87098373 | 3340 | local_irq_enable(); |
964d4bd3 | 3341 | slab_post_alloc_hook(s, objcg, flags, i, p); |
03ec0ed5 | 3342 | __kmem_cache_free_bulk(s, i, p); |
865762a8 | 3343 | return 0; |
484748f0 CL |
3344 | } |
3345 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3346 | ||
3347 | ||
81819f0f | 3348 | /* |
672bba3a CL |
3349 | * Object placement in a slab is made very easy because we always start at |
3350 | * offset 0. If we tune the size of the object to the alignment then we can | |
3351 | * get the required alignment by putting one properly sized object after | |
3352 | * another. | |
81819f0f CL |
3353 | * |
3354 | * Notice that the allocation order determines the sizes of the per cpu | |
3355 | * caches. Each processor has always one slab available for allocations. | |
3356 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3357 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3358 | * locking overhead. |
81819f0f CL |
3359 | */ |
3360 | ||
3361 | /* | |
3362 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3363 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3364 | * and increases the number of allocations possible without having to | |
3365 | * take the list_lock. | |
3366 | */ | |
19af27af AD |
3367 | static unsigned int slub_min_order; |
3368 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3369 | static unsigned int slub_min_objects; | |
81819f0f | 3370 | |
81819f0f CL |
3371 | /* |
3372 | * Calculate the order of allocation given an slab object size. | |
3373 | * | |
672bba3a CL |
3374 | * The order of allocation has significant impact on performance and other |
3375 | * system components. Generally order 0 allocations should be preferred since | |
3376 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3377 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3378 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3379 | * would be wasted. |
3380 | * | |
3381 | * In order to reach satisfactory performance we must ensure that a minimum | |
3382 | * number of objects is in one slab. Otherwise we may generate too much | |
3383 | * activity on the partial lists which requires taking the list_lock. This is | |
3384 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3385 | * |
672bba3a CL |
3386 | * slub_max_order specifies the order where we begin to stop considering the |
3387 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3388 | * we try to keep the page order as low as possible. So we accept more waste | |
3389 | * of space in favor of a small page order. | |
81819f0f | 3390 | * |
672bba3a CL |
3391 | * Higher order allocations also allow the placement of more objects in a |
3392 | * slab and thereby reduce object handling overhead. If the user has | |
3393 | * requested a higher mininum order then we start with that one instead of | |
3394 | * the smallest order which will fit the object. | |
81819f0f | 3395 | */ |
19af27af AD |
3396 | static inline unsigned int slab_order(unsigned int size, |
3397 | unsigned int min_objects, unsigned int max_order, | |
9736d2a9 | 3398 | unsigned int fract_leftover) |
81819f0f | 3399 | { |
19af27af AD |
3400 | unsigned int min_order = slub_min_order; |
3401 | unsigned int order; | |
81819f0f | 3402 | |
9736d2a9 | 3403 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3404 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3405 | |
9736d2a9 | 3406 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3407 | order <= max_order; order++) { |
81819f0f | 3408 | |
19af27af AD |
3409 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3410 | unsigned int rem; | |
81819f0f | 3411 | |
9736d2a9 | 3412 | rem = slab_size % size; |
81819f0f | 3413 | |
5e6d444e | 3414 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3415 | break; |
81819f0f | 3416 | } |
672bba3a | 3417 | |
81819f0f CL |
3418 | return order; |
3419 | } | |
3420 | ||
9736d2a9 | 3421 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3422 | { |
19af27af AD |
3423 | unsigned int order; |
3424 | unsigned int min_objects; | |
3425 | unsigned int max_objects; | |
3286222f | 3426 | unsigned int nr_cpus; |
5e6d444e CL |
3427 | |
3428 | /* | |
3429 | * Attempt to find best configuration for a slab. This | |
3430 | * works by first attempting to generate a layout with | |
3431 | * the best configuration and backing off gradually. | |
3432 | * | |
422ff4d7 | 3433 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3434 | * we reduce the minimum objects required in a slab. |
3435 | */ | |
3436 | min_objects = slub_min_objects; | |
3286222f VB |
3437 | if (!min_objects) { |
3438 | /* | |
3439 | * Some architectures will only update present cpus when | |
3440 | * onlining them, so don't trust the number if it's just 1. But | |
3441 | * we also don't want to use nr_cpu_ids always, as on some other | |
3442 | * architectures, there can be many possible cpus, but never | |
3443 | * onlined. Here we compromise between trying to avoid too high | |
3444 | * order on systems that appear larger than they are, and too | |
3445 | * low order on systems that appear smaller than they are. | |
3446 | */ | |
3447 | nr_cpus = num_present_cpus(); | |
3448 | if (nr_cpus <= 1) | |
3449 | nr_cpus = nr_cpu_ids; | |
3450 | min_objects = 4 * (fls(nr_cpus) + 1); | |
3451 | } | |
9736d2a9 | 3452 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3453 | min_objects = min(min_objects, max_objects); |
3454 | ||
5e6d444e | 3455 | while (min_objects > 1) { |
19af27af AD |
3456 | unsigned int fraction; |
3457 | ||
c124f5b5 | 3458 | fraction = 16; |
5e6d444e CL |
3459 | while (fraction >= 4) { |
3460 | order = slab_order(size, min_objects, | |
9736d2a9 | 3461 | slub_max_order, fraction); |
5e6d444e CL |
3462 | if (order <= slub_max_order) |
3463 | return order; | |
3464 | fraction /= 2; | |
3465 | } | |
5086c389 | 3466 | min_objects--; |
5e6d444e CL |
3467 | } |
3468 | ||
3469 | /* | |
3470 | * We were unable to place multiple objects in a slab. Now | |
3471 | * lets see if we can place a single object there. | |
3472 | */ | |
9736d2a9 | 3473 | order = slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3474 | if (order <= slub_max_order) |
3475 | return order; | |
3476 | ||
3477 | /* | |
3478 | * Doh this slab cannot be placed using slub_max_order. | |
3479 | */ | |
9736d2a9 | 3480 | order = slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3481 | if (order < MAX_ORDER) |
5e6d444e CL |
3482 | return order; |
3483 | return -ENOSYS; | |
3484 | } | |
3485 | ||
5595cffc | 3486 | static void |
4053497d | 3487 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3488 | { |
3489 | n->nr_partial = 0; | |
81819f0f CL |
3490 | spin_lock_init(&n->list_lock); |
3491 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3492 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3493 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3494 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3495 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3496 | #endif |
81819f0f CL |
3497 | } |
3498 | ||
55136592 | 3499 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3500 | { |
6c182dc0 | 3501 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3502 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3503 | |
8a5ec0ba | 3504 | /* |
d4d84fef CM |
3505 | * Must align to double word boundary for the double cmpxchg |
3506 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3507 | */ |
d4d84fef CM |
3508 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3509 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3510 | |
3511 | if (!s->cpu_slab) | |
3512 | return 0; | |
3513 | ||
3514 | init_kmem_cache_cpus(s); | |
4c93c355 | 3515 | |
8a5ec0ba | 3516 | return 1; |
4c93c355 | 3517 | } |
4c93c355 | 3518 | |
51df1142 CL |
3519 | static struct kmem_cache *kmem_cache_node; |
3520 | ||
81819f0f CL |
3521 | /* |
3522 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3523 | * slab on the node for this slabcache. There are no concurrent accesses | |
3524 | * possible. | |
3525 | * | |
721ae22a ZYW |
3526 | * Note that this function only works on the kmem_cache_node |
3527 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3528 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3529 | */ |
55136592 | 3530 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3531 | { |
3532 | struct page *page; | |
3533 | struct kmem_cache_node *n; | |
3534 | ||
51df1142 | 3535 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3536 | |
51df1142 | 3537 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3538 | |
3539 | BUG_ON(!page); | |
a2f92ee7 | 3540 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3541 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3542 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3543 | } |
3544 | ||
81819f0f CL |
3545 | n = page->freelist; |
3546 | BUG_ON(!n); | |
8ab1372f | 3547 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3548 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3549 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3550 | #endif |
12b22386 | 3551 | n = kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
505f5dcb | 3552 | GFP_KERNEL); |
12b22386 AK |
3553 | page->freelist = get_freepointer(kmem_cache_node, n); |
3554 | page->inuse = 1; | |
3555 | page->frozen = 0; | |
3556 | kmem_cache_node->node[node] = n; | |
4053497d | 3557 | init_kmem_cache_node(n); |
51df1142 | 3558 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3559 | |
67b6c900 | 3560 | /* |
1e4dd946 SR |
3561 | * No locks need to be taken here as it has just been |
3562 | * initialized and there is no concurrent access. | |
67b6c900 | 3563 | */ |
1e4dd946 | 3564 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3565 | } |
3566 | ||
3567 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3568 | { | |
3569 | int node; | |
fa45dc25 | 3570 | struct kmem_cache_node *n; |
81819f0f | 3571 | |
fa45dc25 | 3572 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3573 | s->node[node] = NULL; |
ea37df54 | 3574 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3575 | } |
3576 | } | |
3577 | ||
52b4b950 DS |
3578 | void __kmem_cache_release(struct kmem_cache *s) |
3579 | { | |
210e7a43 | 3580 | cache_random_seq_destroy(s); |
52b4b950 DS |
3581 | free_percpu(s->cpu_slab); |
3582 | free_kmem_cache_nodes(s); | |
3583 | } | |
3584 | ||
55136592 | 3585 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3586 | { |
3587 | int node; | |
81819f0f | 3588 | |
f64dc58c | 3589 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3590 | struct kmem_cache_node *n; |
3591 | ||
73367bd8 | 3592 | if (slab_state == DOWN) { |
55136592 | 3593 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3594 | continue; |
3595 | } | |
51df1142 | 3596 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3597 | GFP_KERNEL, node); |
81819f0f | 3598 | |
73367bd8 AD |
3599 | if (!n) { |
3600 | free_kmem_cache_nodes(s); | |
3601 | return 0; | |
81819f0f | 3602 | } |
73367bd8 | 3603 | |
4053497d | 3604 | init_kmem_cache_node(n); |
ea37df54 | 3605 | s->node[node] = n; |
81819f0f CL |
3606 | } |
3607 | return 1; | |
3608 | } | |
81819f0f | 3609 | |
c0bdb232 | 3610 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3611 | { |
3612 | if (min < MIN_PARTIAL) | |
3613 | min = MIN_PARTIAL; | |
3614 | else if (min > MAX_PARTIAL) | |
3615 | min = MAX_PARTIAL; | |
3616 | s->min_partial = min; | |
3617 | } | |
3618 | ||
e6d0e1dc WY |
3619 | static void set_cpu_partial(struct kmem_cache *s) |
3620 | { | |
3621 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3622 | /* | |
3623 | * cpu_partial determined the maximum number of objects kept in the | |
3624 | * per cpu partial lists of a processor. | |
3625 | * | |
3626 | * Per cpu partial lists mainly contain slabs that just have one | |
3627 | * object freed. If they are used for allocation then they can be | |
3628 | * filled up again with minimal effort. The slab will never hit the | |
3629 | * per node partial lists and therefore no locking will be required. | |
3630 | * | |
3631 | * This setting also determines | |
3632 | * | |
3633 | * A) The number of objects from per cpu partial slabs dumped to the | |
3634 | * per node list when we reach the limit. | |
3635 | * B) The number of objects in cpu partial slabs to extract from the | |
3636 | * per node list when we run out of per cpu objects. We only fetch | |
3637 | * 50% to keep some capacity around for frees. | |
3638 | */ | |
3639 | if (!kmem_cache_has_cpu_partial(s)) | |
bbd4e305 | 3640 | slub_set_cpu_partial(s, 0); |
e6d0e1dc | 3641 | else if (s->size >= PAGE_SIZE) |
bbd4e305 | 3642 | slub_set_cpu_partial(s, 2); |
e6d0e1dc | 3643 | else if (s->size >= 1024) |
bbd4e305 | 3644 | slub_set_cpu_partial(s, 6); |
e6d0e1dc | 3645 | else if (s->size >= 256) |
bbd4e305 | 3646 | slub_set_cpu_partial(s, 13); |
e6d0e1dc | 3647 | else |
bbd4e305 | 3648 | slub_set_cpu_partial(s, 30); |
e6d0e1dc WY |
3649 | #endif |
3650 | } | |
3651 | ||
81819f0f CL |
3652 | /* |
3653 | * calculate_sizes() determines the order and the distribution of data within | |
3654 | * a slab object. | |
3655 | */ | |
06b285dc | 3656 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 3657 | { |
d50112ed | 3658 | slab_flags_t flags = s->flags; |
be4a7988 | 3659 | unsigned int size = s->object_size; |
89b83f28 | 3660 | unsigned int freepointer_area; |
19af27af | 3661 | unsigned int order; |
81819f0f | 3662 | |
d8b42bf5 CL |
3663 | /* |
3664 | * Round up object size to the next word boundary. We can only | |
3665 | * place the free pointer at word boundaries and this determines | |
3666 | * the possible location of the free pointer. | |
3667 | */ | |
3668 | size = ALIGN(size, sizeof(void *)); | |
89b83f28 KC |
3669 | /* |
3670 | * This is the area of the object where a freepointer can be | |
3671 | * safely written. If redzoning adds more to the inuse size, we | |
3672 | * can't use that portion for writing the freepointer, so | |
3673 | * s->offset must be limited within this for the general case. | |
3674 | */ | |
3675 | freepointer_area = size; | |
d8b42bf5 CL |
3676 | |
3677 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3678 | /* |
3679 | * Determine if we can poison the object itself. If the user of | |
3680 | * the slab may touch the object after free or before allocation | |
3681 | * then we should never poison the object itself. | |
3682 | */ | |
5f0d5a3a | 3683 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3684 | !s->ctor) |
81819f0f CL |
3685 | s->flags |= __OBJECT_POISON; |
3686 | else | |
3687 | s->flags &= ~__OBJECT_POISON; | |
3688 | ||
81819f0f CL |
3689 | |
3690 | /* | |
672bba3a | 3691 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3692 | * end of the object and the free pointer. If not then add an |
672bba3a | 3693 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3694 | */ |
3b0efdfa | 3695 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3696 | size += sizeof(void *); |
41ecc55b | 3697 | #endif |
81819f0f CL |
3698 | |
3699 | /* | |
672bba3a CL |
3700 | * With that we have determined the number of bytes in actual use |
3701 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3702 | */ |
3703 | s->inuse = size; | |
3704 | ||
5f0d5a3a | 3705 | if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
c59def9f | 3706 | s->ctor)) { |
81819f0f CL |
3707 | /* |
3708 | * Relocate free pointer after the object if it is not | |
3709 | * permitted to overwrite the first word of the object on | |
3710 | * kmem_cache_free. | |
3711 | * | |
3712 | * This is the case if we do RCU, have a constructor or | |
3713 | * destructor or are poisoning the objects. | |
cbfc35a4 WL |
3714 | * |
3715 | * The assumption that s->offset >= s->inuse means free | |
3716 | * pointer is outside of the object is used in the | |
3717 | * freeptr_outside_object() function. If that is no | |
3718 | * longer true, the function needs to be modified. | |
81819f0f CL |
3719 | */ |
3720 | s->offset = size; | |
3721 | size += sizeof(void *); | |
89b83f28 | 3722 | } else if (freepointer_area > sizeof(void *)) { |
3202fa62 KC |
3723 | /* |
3724 | * Store freelist pointer near middle of object to keep | |
3725 | * it away from the edges of the object to avoid small | |
3726 | * sized over/underflows from neighboring allocations. | |
3727 | */ | |
89b83f28 | 3728 | s->offset = ALIGN(freepointer_area / 2, sizeof(void *)); |
81819f0f CL |
3729 | } |
3730 | ||
c12b3c62 | 3731 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3732 | if (flags & SLAB_STORE_USER) |
3733 | /* | |
3734 | * Need to store information about allocs and frees after | |
3735 | * the object. | |
3736 | */ | |
3737 | size += 2 * sizeof(struct track); | |
80a9201a | 3738 | #endif |
81819f0f | 3739 | |
80a9201a AP |
3740 | kasan_cache_create(s, &size, &s->flags); |
3741 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3742 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3743 | /* |
3744 | * Add some empty padding so that we can catch | |
3745 | * overwrites from earlier objects rather than let | |
3746 | * tracking information or the free pointer be | |
0211a9c8 | 3747 | * corrupted if a user writes before the start |
81819f0f CL |
3748 | * of the object. |
3749 | */ | |
3750 | size += sizeof(void *); | |
d86bd1be JK |
3751 | |
3752 | s->red_left_pad = sizeof(void *); | |
3753 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3754 | size += s->red_left_pad; | |
3755 | } | |
41ecc55b | 3756 | #endif |
672bba3a | 3757 | |
81819f0f CL |
3758 | /* |
3759 | * SLUB stores one object immediately after another beginning from | |
3760 | * offset 0. In order to align the objects we have to simply size | |
3761 | * each object to conform to the alignment. | |
3762 | */ | |
45906855 | 3763 | size = ALIGN(size, s->align); |
81819f0f | 3764 | s->size = size; |
4138fdfc | 3765 | s->reciprocal_size = reciprocal_value(size); |
06b285dc CL |
3766 | if (forced_order >= 0) |
3767 | order = forced_order; | |
3768 | else | |
9736d2a9 | 3769 | order = calculate_order(size); |
81819f0f | 3770 | |
19af27af | 3771 | if ((int)order < 0) |
81819f0f CL |
3772 | return 0; |
3773 | ||
b7a49f0d | 3774 | s->allocflags = 0; |
834f3d11 | 3775 | if (order) |
b7a49f0d CL |
3776 | s->allocflags |= __GFP_COMP; |
3777 | ||
3778 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3779 | s->allocflags |= GFP_DMA; |
b7a49f0d | 3780 | |
6d6ea1e9 NB |
3781 | if (s->flags & SLAB_CACHE_DMA32) |
3782 | s->allocflags |= GFP_DMA32; | |
3783 | ||
b7a49f0d CL |
3784 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
3785 | s->allocflags |= __GFP_RECLAIMABLE; | |
3786 | ||
81819f0f CL |
3787 | /* |
3788 | * Determine the number of objects per slab | |
3789 | */ | |
9736d2a9 MW |
3790 | s->oo = oo_make(order, size); |
3791 | s->min = oo_make(get_order(size), size); | |
205ab99d CL |
3792 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3793 | s->max = s->oo; | |
81819f0f | 3794 | |
834f3d11 | 3795 | return !!oo_objects(s->oo); |
81819f0f CL |
3796 | } |
3797 | ||
d50112ed | 3798 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 3799 | { |
8a13a4cc | 3800 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
2482ddec KC |
3801 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
3802 | s->random = get_random_long(); | |
3803 | #endif | |
81819f0f | 3804 | |
06b285dc | 3805 | if (!calculate_sizes(s, -1)) |
81819f0f | 3806 | goto error; |
3de47213 DR |
3807 | if (disable_higher_order_debug) { |
3808 | /* | |
3809 | * Disable debugging flags that store metadata if the min slab | |
3810 | * order increased. | |
3811 | */ | |
3b0efdfa | 3812 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3813 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3814 | s->offset = 0; | |
3815 | if (!calculate_sizes(s, -1)) | |
3816 | goto error; | |
3817 | } | |
3818 | } | |
81819f0f | 3819 | |
2565409f HC |
3820 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3821 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3822 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3823 | /* Enable fast mode */ |
3824 | s->flags |= __CMPXCHG_DOUBLE; | |
3825 | #endif | |
3826 | ||
3b89d7d8 DR |
3827 | /* |
3828 | * The larger the object size is, the more pages we want on the partial | |
3829 | * list to avoid pounding the page allocator excessively. | |
3830 | */ | |
49e22585 CL |
3831 | set_min_partial(s, ilog2(s->size) / 2); |
3832 | ||
e6d0e1dc | 3833 | set_cpu_partial(s); |
49e22585 | 3834 | |
81819f0f | 3835 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3836 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3837 | #endif |
210e7a43 TG |
3838 | |
3839 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3840 | if (slab_state >= UP) { | |
3841 | if (init_cache_random_seq(s)) | |
3842 | goto error; | |
3843 | } | |
3844 | ||
55136592 | 3845 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3846 | goto error; |
81819f0f | 3847 | |
55136592 | 3848 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3849 | return 0; |
ff12059e | 3850 | |
4c93c355 | 3851 | free_kmem_cache_nodes(s); |
81819f0f | 3852 | error: |
278b1bb1 | 3853 | return -EINVAL; |
81819f0f | 3854 | } |
81819f0f | 3855 | |
33b12c38 | 3856 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
55860d96 | 3857 | const char *text) |
33b12c38 CL |
3858 | { |
3859 | #ifdef CONFIG_SLUB_DEBUG | |
3860 | void *addr = page_address(page); | |
55860d96 | 3861 | unsigned long *map; |
33b12c38 | 3862 | void *p; |
aa456c7a | 3863 | |
945cf2b6 | 3864 | slab_err(s, page, text, s->name); |
33b12c38 | 3865 | slab_lock(page); |
33b12c38 | 3866 | |
90e9f6a6 | 3867 | map = get_map(s, page); |
33b12c38 CL |
3868 | for_each_object(p, s, addr, page->objects) { |
3869 | ||
4138fdfc | 3870 | if (!test_bit(__obj_to_index(s, addr, p), map)) { |
f9f58285 | 3871 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3872 | print_tracking(s, p); |
3873 | } | |
3874 | } | |
55860d96 | 3875 | put_map(map); |
33b12c38 CL |
3876 | slab_unlock(page); |
3877 | #endif | |
3878 | } | |
3879 | ||
81819f0f | 3880 | /* |
599870b1 | 3881 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3882 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3883 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3884 | */ |
599870b1 | 3885 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3886 | { |
60398923 | 3887 | LIST_HEAD(discard); |
81819f0f CL |
3888 | struct page *page, *h; |
3889 | ||
52b4b950 DS |
3890 | BUG_ON(irqs_disabled()); |
3891 | spin_lock_irq(&n->list_lock); | |
916ac052 | 3892 | list_for_each_entry_safe(page, h, &n->partial, slab_list) { |
81819f0f | 3893 | if (!page->inuse) { |
52b4b950 | 3894 | remove_partial(n, page); |
916ac052 | 3895 | list_add(&page->slab_list, &discard); |
33b12c38 CL |
3896 | } else { |
3897 | list_slab_objects(s, page, | |
55860d96 | 3898 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3899 | } |
33b12c38 | 3900 | } |
52b4b950 | 3901 | spin_unlock_irq(&n->list_lock); |
60398923 | 3902 | |
916ac052 | 3903 | list_for_each_entry_safe(page, h, &discard, slab_list) |
60398923 | 3904 | discard_slab(s, page); |
81819f0f CL |
3905 | } |
3906 | ||
f9e13c0a SB |
3907 | bool __kmem_cache_empty(struct kmem_cache *s) |
3908 | { | |
3909 | int node; | |
3910 | struct kmem_cache_node *n; | |
3911 | ||
3912 | for_each_kmem_cache_node(s, node, n) | |
3913 | if (n->nr_partial || slabs_node(s, node)) | |
3914 | return false; | |
3915 | return true; | |
3916 | } | |
3917 | ||
81819f0f | 3918 | /* |
672bba3a | 3919 | * Release all resources used by a slab cache. |
81819f0f | 3920 | */ |
52b4b950 | 3921 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3922 | { |
3923 | int node; | |
fa45dc25 | 3924 | struct kmem_cache_node *n; |
81819f0f CL |
3925 | |
3926 | flush_all(s); | |
81819f0f | 3927 | /* Attempt to free all objects */ |
fa45dc25 | 3928 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3929 | free_partial(s, n); |
3930 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3931 | return 1; |
3932 | } | |
81819f0f CL |
3933 | return 0; |
3934 | } | |
3935 | ||
81819f0f CL |
3936 | /******************************************************************** |
3937 | * Kmalloc subsystem | |
3938 | *******************************************************************/ | |
3939 | ||
81819f0f CL |
3940 | static int __init setup_slub_min_order(char *str) |
3941 | { | |
19af27af | 3942 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
3943 | |
3944 | return 1; | |
3945 | } | |
3946 | ||
3947 | __setup("slub_min_order=", setup_slub_min_order); | |
3948 | ||
3949 | static int __init setup_slub_max_order(char *str) | |
3950 | { | |
19af27af AD |
3951 | get_option(&str, (int *)&slub_max_order); |
3952 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
3953 | |
3954 | return 1; | |
3955 | } | |
3956 | ||
3957 | __setup("slub_max_order=", setup_slub_max_order); | |
3958 | ||
3959 | static int __init setup_slub_min_objects(char *str) | |
3960 | { | |
19af27af | 3961 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
3962 | |
3963 | return 1; | |
3964 | } | |
3965 | ||
3966 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3967 | ||
81819f0f CL |
3968 | void *__kmalloc(size_t size, gfp_t flags) |
3969 | { | |
aadb4bc4 | 3970 | struct kmem_cache *s; |
5b882be4 | 3971 | void *ret; |
81819f0f | 3972 | |
95a05b42 | 3973 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3974 | return kmalloc_large(size, flags); |
aadb4bc4 | 3975 | |
2c59dd65 | 3976 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3977 | |
3978 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3979 | return s; |
3980 | ||
2b847c3c | 3981 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3982 | |
ca2b84cb | 3983 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3984 | |
0116523c | 3985 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3986 | |
5b882be4 | 3987 | return ret; |
81819f0f CL |
3988 | } |
3989 | EXPORT_SYMBOL(__kmalloc); | |
3990 | ||
5d1f57e4 | 3991 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3992 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3993 | { | |
b1eeab67 | 3994 | struct page *page; |
e4f7c0b4 | 3995 | void *ptr = NULL; |
6a486c0a | 3996 | unsigned int order = get_order(size); |
f619cfe1 | 3997 | |
75f296d9 | 3998 | flags |= __GFP_COMP; |
6a486c0a VB |
3999 | page = alloc_pages_node(node, flags, order); |
4000 | if (page) { | |
e4f7c0b4 | 4001 | ptr = page_address(page); |
d42f3245 RG |
4002 | mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE_B, |
4003 | PAGE_SIZE << order); | |
6a486c0a | 4004 | } |
e4f7c0b4 | 4005 | |
0116523c | 4006 | return kmalloc_large_node_hook(ptr, size, flags); |
f619cfe1 CL |
4007 | } |
4008 | ||
81819f0f CL |
4009 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
4010 | { | |
aadb4bc4 | 4011 | struct kmem_cache *s; |
5b882be4 | 4012 | void *ret; |
81819f0f | 4013 | |
95a05b42 | 4014 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
4015 | ret = kmalloc_large_node(size, flags, node); |
4016 | ||
ca2b84cb EGM |
4017 | trace_kmalloc_node(_RET_IP_, ret, |
4018 | size, PAGE_SIZE << get_order(size), | |
4019 | flags, node); | |
5b882be4 EGM |
4020 | |
4021 | return ret; | |
4022 | } | |
aadb4bc4 | 4023 | |
2c59dd65 | 4024 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4025 | |
4026 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4027 | return s; |
4028 | ||
2b847c3c | 4029 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 4030 | |
ca2b84cb | 4031 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 4032 | |
0116523c | 4033 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4034 | |
5b882be4 | 4035 | return ret; |
81819f0f CL |
4036 | } |
4037 | EXPORT_SYMBOL(__kmalloc_node); | |
6dfd1b65 | 4038 | #endif /* CONFIG_NUMA */ |
81819f0f | 4039 | |
ed18adc1 KC |
4040 | #ifdef CONFIG_HARDENED_USERCOPY |
4041 | /* | |
afcc90f8 KC |
4042 | * Rejects incorrectly sized objects and objects that are to be copied |
4043 | * to/from userspace but do not fall entirely within the containing slab | |
4044 | * cache's usercopy region. | |
ed18adc1 KC |
4045 | * |
4046 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4047 | * to indicate an error. | |
4048 | */ | |
f4e6e289 KC |
4049 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
4050 | bool to_user) | |
ed18adc1 KC |
4051 | { |
4052 | struct kmem_cache *s; | |
44065b2e | 4053 | unsigned int offset; |
ed18adc1 KC |
4054 | size_t object_size; |
4055 | ||
96fedce2 AK |
4056 | ptr = kasan_reset_tag(ptr); |
4057 | ||
ed18adc1 KC |
4058 | /* Find object and usable object size. */ |
4059 | s = page->slab_cache; | |
ed18adc1 KC |
4060 | |
4061 | /* Reject impossible pointers. */ | |
4062 | if (ptr < page_address(page)) | |
f4e6e289 KC |
4063 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
4064 | to_user, 0, n); | |
ed18adc1 KC |
4065 | |
4066 | /* Find offset within object. */ | |
4067 | offset = (ptr - page_address(page)) % s->size; | |
4068 | ||
4069 | /* Adjust for redzone and reject if within the redzone. */ | |
59052e89 | 4070 | if (kmem_cache_debug_flags(s, SLAB_RED_ZONE)) { |
ed18adc1 | 4071 | if (offset < s->red_left_pad) |
f4e6e289 KC |
4072 | usercopy_abort("SLUB object in left red zone", |
4073 | s->name, to_user, offset, n); | |
ed18adc1 KC |
4074 | offset -= s->red_left_pad; |
4075 | } | |
4076 | ||
afcc90f8 KC |
4077 | /* Allow address range falling entirely within usercopy region. */ |
4078 | if (offset >= s->useroffset && | |
4079 | offset - s->useroffset <= s->usersize && | |
4080 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 4081 | return; |
ed18adc1 | 4082 | |
afcc90f8 KC |
4083 | /* |
4084 | * If the copy is still within the allocated object, produce | |
4085 | * a warning instead of rejecting the copy. This is intended | |
4086 | * to be a temporary method to find any missing usercopy | |
4087 | * whitelists. | |
4088 | */ | |
4089 | object_size = slab_ksize(s); | |
2d891fbc KC |
4090 | if (usercopy_fallback && |
4091 | offset <= object_size && n <= object_size - offset) { | |
afcc90f8 KC |
4092 | usercopy_warn("SLUB object", s->name, to_user, offset, n); |
4093 | return; | |
4094 | } | |
ed18adc1 | 4095 | |
f4e6e289 | 4096 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
4097 | } |
4098 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4099 | ||
10d1f8cb | 4100 | size_t __ksize(const void *object) |
81819f0f | 4101 | { |
272c1d21 | 4102 | struct page *page; |
81819f0f | 4103 | |
ef8b4520 | 4104 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
4105 | return 0; |
4106 | ||
294a80a8 | 4107 | page = virt_to_head_page(object); |
294a80a8 | 4108 | |
76994412 PE |
4109 | if (unlikely(!PageSlab(page))) { |
4110 | WARN_ON(!PageCompound(page)); | |
a50b854e | 4111 | return page_size(page); |
76994412 | 4112 | } |
81819f0f | 4113 | |
1b4f59e3 | 4114 | return slab_ksize(page->slab_cache); |
81819f0f | 4115 | } |
10d1f8cb | 4116 | EXPORT_SYMBOL(__ksize); |
81819f0f CL |
4117 | |
4118 | void kfree(const void *x) | |
4119 | { | |
81819f0f | 4120 | struct page *page; |
5bb983b0 | 4121 | void *object = (void *)x; |
81819f0f | 4122 | |
2121db74 PE |
4123 | trace_kfree(_RET_IP_, x); |
4124 | ||
2408c550 | 4125 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
4126 | return; |
4127 | ||
b49af68f | 4128 | page = virt_to_head_page(x); |
aadb4bc4 | 4129 | if (unlikely(!PageSlab(page))) { |
6a486c0a VB |
4130 | unsigned int order = compound_order(page); |
4131 | ||
0937502a | 4132 | BUG_ON(!PageCompound(page)); |
47adccce | 4133 | kfree_hook(object); |
d42f3245 RG |
4134 | mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE_B, |
4135 | -(PAGE_SIZE << order)); | |
6a486c0a | 4136 | __free_pages(page, order); |
aadb4bc4 CL |
4137 | return; |
4138 | } | |
81084651 | 4139 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
4140 | } |
4141 | EXPORT_SYMBOL(kfree); | |
4142 | ||
832f37f5 VD |
4143 | #define SHRINK_PROMOTE_MAX 32 |
4144 | ||
2086d26a | 4145 | /* |
832f37f5 VD |
4146 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
4147 | * up most to the head of the partial lists. New allocations will then | |
4148 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
4149 | * |
4150 | * The slabs with the least items are placed last. This results in them | |
4151 | * being allocated from last increasing the chance that the last objects | |
4152 | * are freed in them. | |
2086d26a | 4153 | */ |
c9fc5864 | 4154 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
4155 | { |
4156 | int node; | |
4157 | int i; | |
4158 | struct kmem_cache_node *n; | |
4159 | struct page *page; | |
4160 | struct page *t; | |
832f37f5 VD |
4161 | struct list_head discard; |
4162 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 4163 | unsigned long flags; |
ce3712d7 | 4164 | int ret = 0; |
2086d26a | 4165 | |
2086d26a | 4166 | flush_all(s); |
fa45dc25 | 4167 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
4168 | INIT_LIST_HEAD(&discard); |
4169 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
4170 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
4171 | |
4172 | spin_lock_irqsave(&n->list_lock, flags); | |
4173 | ||
4174 | /* | |
832f37f5 | 4175 | * Build lists of slabs to discard or promote. |
2086d26a | 4176 | * |
672bba3a CL |
4177 | * Note that concurrent frees may occur while we hold the |
4178 | * list_lock. page->inuse here is the upper limit. | |
2086d26a | 4179 | */ |
916ac052 | 4180 | list_for_each_entry_safe(page, t, &n->partial, slab_list) { |
832f37f5 VD |
4181 | int free = page->objects - page->inuse; |
4182 | ||
4183 | /* Do not reread page->inuse */ | |
4184 | barrier(); | |
4185 | ||
4186 | /* We do not keep full slabs on the list */ | |
4187 | BUG_ON(free <= 0); | |
4188 | ||
4189 | if (free == page->objects) { | |
916ac052 | 4190 | list_move(&page->slab_list, &discard); |
69cb8e6b | 4191 | n->nr_partial--; |
832f37f5 | 4192 | } else if (free <= SHRINK_PROMOTE_MAX) |
916ac052 | 4193 | list_move(&page->slab_list, promote + free - 1); |
2086d26a CL |
4194 | } |
4195 | ||
2086d26a | 4196 | /* |
832f37f5 VD |
4197 | * Promote the slabs filled up most to the head of the |
4198 | * partial list. | |
2086d26a | 4199 | */ |
832f37f5 VD |
4200 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
4201 | list_splice(promote + i, &n->partial); | |
2086d26a | 4202 | |
2086d26a | 4203 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
4204 | |
4205 | /* Release empty slabs */ | |
916ac052 | 4206 | list_for_each_entry_safe(page, t, &discard, slab_list) |
69cb8e6b | 4207 | discard_slab(s, page); |
ce3712d7 VD |
4208 | |
4209 | if (slabs_node(s, node)) | |
4210 | ret = 1; | |
2086d26a CL |
4211 | } |
4212 | ||
ce3712d7 | 4213 | return ret; |
2086d26a | 4214 | } |
2086d26a | 4215 | |
b9049e23 YG |
4216 | static int slab_mem_going_offline_callback(void *arg) |
4217 | { | |
4218 | struct kmem_cache *s; | |
4219 | ||
18004c5d | 4220 | mutex_lock(&slab_mutex); |
b9049e23 | 4221 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4222 | __kmem_cache_shrink(s); |
18004c5d | 4223 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4224 | |
4225 | return 0; | |
4226 | } | |
4227 | ||
4228 | static void slab_mem_offline_callback(void *arg) | |
4229 | { | |
4230 | struct kmem_cache_node *n; | |
4231 | struct kmem_cache *s; | |
4232 | struct memory_notify *marg = arg; | |
4233 | int offline_node; | |
4234 | ||
b9d5ab25 | 4235 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4236 | |
4237 | /* | |
4238 | * If the node still has available memory. we need kmem_cache_node | |
4239 | * for it yet. | |
4240 | */ | |
4241 | if (offline_node < 0) | |
4242 | return; | |
4243 | ||
18004c5d | 4244 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4245 | list_for_each_entry(s, &slab_caches, list) { |
4246 | n = get_node(s, offline_node); | |
4247 | if (n) { | |
4248 | /* | |
4249 | * if n->nr_slabs > 0, slabs still exist on the node | |
4250 | * that is going down. We were unable to free them, | |
c9404c9c | 4251 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
4252 | * callback. So, we must fail. |
4253 | */ | |
0f389ec6 | 4254 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
4255 | |
4256 | s->node[offline_node] = NULL; | |
8de66a0c | 4257 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
4258 | } |
4259 | } | |
18004c5d | 4260 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4261 | } |
4262 | ||
4263 | static int slab_mem_going_online_callback(void *arg) | |
4264 | { | |
4265 | struct kmem_cache_node *n; | |
4266 | struct kmem_cache *s; | |
4267 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4268 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4269 | int ret = 0; |
4270 | ||
4271 | /* | |
4272 | * If the node's memory is already available, then kmem_cache_node is | |
4273 | * already created. Nothing to do. | |
4274 | */ | |
4275 | if (nid < 0) | |
4276 | return 0; | |
4277 | ||
4278 | /* | |
0121c619 | 4279 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4280 | * allocate a kmem_cache_node structure in order to bring the node |
4281 | * online. | |
4282 | */ | |
18004c5d | 4283 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4284 | list_for_each_entry(s, &slab_caches, list) { |
4285 | /* | |
4286 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4287 | * since memory is not yet available from the node that | |
4288 | * is brought up. | |
4289 | */ | |
8de66a0c | 4290 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4291 | if (!n) { |
4292 | ret = -ENOMEM; | |
4293 | goto out; | |
4294 | } | |
4053497d | 4295 | init_kmem_cache_node(n); |
b9049e23 YG |
4296 | s->node[nid] = n; |
4297 | } | |
4298 | out: | |
18004c5d | 4299 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4300 | return ret; |
4301 | } | |
4302 | ||
4303 | static int slab_memory_callback(struct notifier_block *self, | |
4304 | unsigned long action, void *arg) | |
4305 | { | |
4306 | int ret = 0; | |
4307 | ||
4308 | switch (action) { | |
4309 | case MEM_GOING_ONLINE: | |
4310 | ret = slab_mem_going_online_callback(arg); | |
4311 | break; | |
4312 | case MEM_GOING_OFFLINE: | |
4313 | ret = slab_mem_going_offline_callback(arg); | |
4314 | break; | |
4315 | case MEM_OFFLINE: | |
4316 | case MEM_CANCEL_ONLINE: | |
4317 | slab_mem_offline_callback(arg); | |
4318 | break; | |
4319 | case MEM_ONLINE: | |
4320 | case MEM_CANCEL_OFFLINE: | |
4321 | break; | |
4322 | } | |
dc19f9db KH |
4323 | if (ret) |
4324 | ret = notifier_from_errno(ret); | |
4325 | else | |
4326 | ret = NOTIFY_OK; | |
b9049e23 YG |
4327 | return ret; |
4328 | } | |
4329 | ||
3ac38faa AM |
4330 | static struct notifier_block slab_memory_callback_nb = { |
4331 | .notifier_call = slab_memory_callback, | |
4332 | .priority = SLAB_CALLBACK_PRI, | |
4333 | }; | |
b9049e23 | 4334 | |
81819f0f CL |
4335 | /******************************************************************** |
4336 | * Basic setup of slabs | |
4337 | *******************************************************************/ | |
4338 | ||
51df1142 CL |
4339 | /* |
4340 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4341 | * the page allocator. Allocate them properly then fix up the pointers |
4342 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4343 | */ |
4344 | ||
dffb4d60 | 4345 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4346 | { |
4347 | int node; | |
dffb4d60 | 4348 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4349 | struct kmem_cache_node *n; |
51df1142 | 4350 | |
dffb4d60 | 4351 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4352 | |
7d557b3c GC |
4353 | /* |
4354 | * This runs very early, and only the boot processor is supposed to be | |
4355 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4356 | * IPIs around. | |
4357 | */ | |
4358 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4359 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4360 | struct page *p; |
4361 | ||
916ac052 | 4362 | list_for_each_entry(p, &n->partial, slab_list) |
fa45dc25 | 4363 | p->slab_cache = s; |
51df1142 | 4364 | |
607bf324 | 4365 | #ifdef CONFIG_SLUB_DEBUG |
916ac052 | 4366 | list_for_each_entry(p, &n->full, slab_list) |
fa45dc25 | 4367 | p->slab_cache = s; |
51df1142 | 4368 | #endif |
51df1142 | 4369 | } |
dffb4d60 CL |
4370 | list_add(&s->list, &slab_caches); |
4371 | return s; | |
51df1142 CL |
4372 | } |
4373 | ||
81819f0f CL |
4374 | void __init kmem_cache_init(void) |
4375 | { | |
dffb4d60 CL |
4376 | static __initdata struct kmem_cache boot_kmem_cache, |
4377 | boot_kmem_cache_node; | |
51df1142 | 4378 | |
fc8d8620 SG |
4379 | if (debug_guardpage_minorder()) |
4380 | slub_max_order = 0; | |
4381 | ||
dffb4d60 CL |
4382 | kmem_cache_node = &boot_kmem_cache_node; |
4383 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4384 | |
dffb4d60 | 4385 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4386 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4387 | |
3ac38faa | 4388 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4389 | |
4390 | /* Able to allocate the per node structures */ | |
4391 | slab_state = PARTIAL; | |
4392 | ||
dffb4d60 CL |
4393 | create_boot_cache(kmem_cache, "kmem_cache", |
4394 | offsetof(struct kmem_cache, node) + | |
4395 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4396 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4397 | |
dffb4d60 | 4398 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4399 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4400 | |
4401 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4402 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4403 | create_kmalloc_caches(0); |
81819f0f | 4404 | |
210e7a43 TG |
4405 | /* Setup random freelists for each cache */ |
4406 | init_freelist_randomization(); | |
4407 | ||
a96a87bf SAS |
4408 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4409 | slub_cpu_dead); | |
81819f0f | 4410 | |
b9726c26 | 4411 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", |
f97d5f63 | 4412 | cache_line_size(), |
81819f0f CL |
4413 | slub_min_order, slub_max_order, slub_min_objects, |
4414 | nr_cpu_ids, nr_node_ids); | |
4415 | } | |
4416 | ||
7e85ee0c PE |
4417 | void __init kmem_cache_init_late(void) |
4418 | { | |
7e85ee0c PE |
4419 | } |
4420 | ||
2633d7a0 | 4421 | struct kmem_cache * |
f4957d5b | 4422 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4423 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4424 | { |
10befea9 | 4425 | struct kmem_cache *s; |
81819f0f | 4426 | |
a44cb944 | 4427 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4428 | if (s) { |
4429 | s->refcount++; | |
84d0ddd6 | 4430 | |
81819f0f CL |
4431 | /* |
4432 | * Adjust the object sizes so that we clear | |
4433 | * the complete object on kzalloc. | |
4434 | */ | |
1b473f29 | 4435 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4436 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4437 | |
7b8f3b66 | 4438 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4439 | s->refcount--; |
cbb79694 | 4440 | s = NULL; |
7b8f3b66 | 4441 | } |
a0e1d1be | 4442 | } |
6446faa2 | 4443 | |
cbb79694 CL |
4444 | return s; |
4445 | } | |
84c1cf62 | 4446 | |
d50112ed | 4447 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4448 | { |
aac3a166 PE |
4449 | int err; |
4450 | ||
4451 | err = kmem_cache_open(s, flags); | |
4452 | if (err) | |
4453 | return err; | |
20cea968 | 4454 | |
45530c44 CL |
4455 | /* Mutex is not taken during early boot */ |
4456 | if (slab_state <= UP) | |
4457 | return 0; | |
4458 | ||
aac3a166 | 4459 | err = sysfs_slab_add(s); |
aac3a166 | 4460 | if (err) |
52b4b950 | 4461 | __kmem_cache_release(s); |
20cea968 | 4462 | |
aac3a166 | 4463 | return err; |
81819f0f | 4464 | } |
81819f0f | 4465 | |
ce71e27c | 4466 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4467 | { |
aadb4bc4 | 4468 | struct kmem_cache *s; |
94b528d0 | 4469 | void *ret; |
aadb4bc4 | 4470 | |
95a05b42 | 4471 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4472 | return kmalloc_large(size, gfpflags); |
4473 | ||
2c59dd65 | 4474 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4475 | |
2408c550 | 4476 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4477 | return s; |
81819f0f | 4478 | |
2b847c3c | 4479 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4480 | |
25985edc | 4481 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4482 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4483 | |
4484 | return ret; | |
81819f0f | 4485 | } |
fd7cb575 | 4486 | EXPORT_SYMBOL(__kmalloc_track_caller); |
81819f0f | 4487 | |
5d1f57e4 | 4488 | #ifdef CONFIG_NUMA |
81819f0f | 4489 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4490 | int node, unsigned long caller) |
81819f0f | 4491 | { |
aadb4bc4 | 4492 | struct kmem_cache *s; |
94b528d0 | 4493 | void *ret; |
aadb4bc4 | 4494 | |
95a05b42 | 4495 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4496 | ret = kmalloc_large_node(size, gfpflags, node); |
4497 | ||
4498 | trace_kmalloc_node(caller, ret, | |
4499 | size, PAGE_SIZE << get_order(size), | |
4500 | gfpflags, node); | |
4501 | ||
4502 | return ret; | |
4503 | } | |
eada35ef | 4504 | |
2c59dd65 | 4505 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4506 | |
2408c550 | 4507 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4508 | return s; |
81819f0f | 4509 | |
2b847c3c | 4510 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4511 | |
25985edc | 4512 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4513 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4514 | |
4515 | return ret; | |
81819f0f | 4516 | } |
fd7cb575 | 4517 | EXPORT_SYMBOL(__kmalloc_node_track_caller); |
5d1f57e4 | 4518 | #endif |
81819f0f | 4519 | |
ab4d5ed5 | 4520 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4521 | static int count_inuse(struct page *page) |
4522 | { | |
4523 | return page->inuse; | |
4524 | } | |
4525 | ||
4526 | static int count_total(struct page *page) | |
4527 | { | |
4528 | return page->objects; | |
4529 | } | |
ab4d5ed5 | 4530 | #endif |
205ab99d | 4531 | |
ab4d5ed5 | 4532 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 | 4533 | static void validate_slab(struct kmem_cache *s, struct page *page) |
53e15af0 CL |
4534 | { |
4535 | void *p; | |
a973e9dd | 4536 | void *addr = page_address(page); |
90e9f6a6 YZ |
4537 | unsigned long *map; |
4538 | ||
4539 | slab_lock(page); | |
53e15af0 | 4540 | |
dd98afd4 | 4541 | if (!check_slab(s, page) || !on_freelist(s, page, NULL)) |
90e9f6a6 | 4542 | goto unlock; |
53e15af0 CL |
4543 | |
4544 | /* Now we know that a valid freelist exists */ | |
90e9f6a6 | 4545 | map = get_map(s, page); |
5f80b13a | 4546 | for_each_object(p, s, addr, page->objects) { |
4138fdfc | 4547 | u8 val = test_bit(__obj_to_index(s, addr, p), map) ? |
dd98afd4 | 4548 | SLUB_RED_INACTIVE : SLUB_RED_ACTIVE; |
53e15af0 | 4549 | |
dd98afd4 YZ |
4550 | if (!check_object(s, page, p, val)) |
4551 | break; | |
4552 | } | |
90e9f6a6 YZ |
4553 | put_map(map); |
4554 | unlock: | |
881db7fb | 4555 | slab_unlock(page); |
53e15af0 CL |
4556 | } |
4557 | ||
434e245d | 4558 | static int validate_slab_node(struct kmem_cache *s, |
90e9f6a6 | 4559 | struct kmem_cache_node *n) |
53e15af0 CL |
4560 | { |
4561 | unsigned long count = 0; | |
4562 | struct page *page; | |
4563 | unsigned long flags; | |
4564 | ||
4565 | spin_lock_irqsave(&n->list_lock, flags); | |
4566 | ||
916ac052 | 4567 | list_for_each_entry(page, &n->partial, slab_list) { |
90e9f6a6 | 4568 | validate_slab(s, page); |
53e15af0 CL |
4569 | count++; |
4570 | } | |
4571 | if (count != n->nr_partial) | |
f9f58285 FF |
4572 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4573 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4574 | |
4575 | if (!(s->flags & SLAB_STORE_USER)) | |
4576 | goto out; | |
4577 | ||
916ac052 | 4578 | list_for_each_entry(page, &n->full, slab_list) { |
90e9f6a6 | 4579 | validate_slab(s, page); |
53e15af0 CL |
4580 | count++; |
4581 | } | |
4582 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4583 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4584 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4585 | |
4586 | out: | |
4587 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4588 | return count; | |
4589 | } | |
4590 | ||
434e245d | 4591 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4592 | { |
4593 | int node; | |
4594 | unsigned long count = 0; | |
fa45dc25 | 4595 | struct kmem_cache_node *n; |
53e15af0 CL |
4596 | |
4597 | flush_all(s); | |
fa45dc25 | 4598 | for_each_kmem_cache_node(s, node, n) |
90e9f6a6 YZ |
4599 | count += validate_slab_node(s, n); |
4600 | ||
53e15af0 CL |
4601 | return count; |
4602 | } | |
88a420e4 | 4603 | /* |
672bba3a | 4604 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4605 | * and freed. |
4606 | */ | |
4607 | ||
4608 | struct location { | |
4609 | unsigned long count; | |
ce71e27c | 4610 | unsigned long addr; |
45edfa58 CL |
4611 | long long sum_time; |
4612 | long min_time; | |
4613 | long max_time; | |
4614 | long min_pid; | |
4615 | long max_pid; | |
174596a0 | 4616 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4617 | nodemask_t nodes; |
88a420e4 CL |
4618 | }; |
4619 | ||
4620 | struct loc_track { | |
4621 | unsigned long max; | |
4622 | unsigned long count; | |
4623 | struct location *loc; | |
4624 | }; | |
4625 | ||
4626 | static void free_loc_track(struct loc_track *t) | |
4627 | { | |
4628 | if (t->max) | |
4629 | free_pages((unsigned long)t->loc, | |
4630 | get_order(sizeof(struct location) * t->max)); | |
4631 | } | |
4632 | ||
68dff6a9 | 4633 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4634 | { |
4635 | struct location *l; | |
4636 | int order; | |
4637 | ||
88a420e4 CL |
4638 | order = get_order(sizeof(struct location) * max); |
4639 | ||
68dff6a9 | 4640 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4641 | if (!l) |
4642 | return 0; | |
4643 | ||
4644 | if (t->count) { | |
4645 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4646 | free_loc_track(t); | |
4647 | } | |
4648 | t->max = max; | |
4649 | t->loc = l; | |
4650 | return 1; | |
4651 | } | |
4652 | ||
4653 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4654 | const struct track *track) |
88a420e4 CL |
4655 | { |
4656 | long start, end, pos; | |
4657 | struct location *l; | |
ce71e27c | 4658 | unsigned long caddr; |
45edfa58 | 4659 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4660 | |
4661 | start = -1; | |
4662 | end = t->count; | |
4663 | ||
4664 | for ( ; ; ) { | |
4665 | pos = start + (end - start + 1) / 2; | |
4666 | ||
4667 | /* | |
4668 | * There is nothing at "end". If we end up there | |
4669 | * we need to add something to before end. | |
4670 | */ | |
4671 | if (pos == end) | |
4672 | break; | |
4673 | ||
4674 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4675 | if (track->addr == caddr) { |
4676 | ||
4677 | l = &t->loc[pos]; | |
4678 | l->count++; | |
4679 | if (track->when) { | |
4680 | l->sum_time += age; | |
4681 | if (age < l->min_time) | |
4682 | l->min_time = age; | |
4683 | if (age > l->max_time) | |
4684 | l->max_time = age; | |
4685 | ||
4686 | if (track->pid < l->min_pid) | |
4687 | l->min_pid = track->pid; | |
4688 | if (track->pid > l->max_pid) | |
4689 | l->max_pid = track->pid; | |
4690 | ||
174596a0 RR |
4691 | cpumask_set_cpu(track->cpu, |
4692 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4693 | } |
4694 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4695 | return 1; |
4696 | } | |
4697 | ||
45edfa58 | 4698 | if (track->addr < caddr) |
88a420e4 CL |
4699 | end = pos; |
4700 | else | |
4701 | start = pos; | |
4702 | } | |
4703 | ||
4704 | /* | |
672bba3a | 4705 | * Not found. Insert new tracking element. |
88a420e4 | 4706 | */ |
68dff6a9 | 4707 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4708 | return 0; |
4709 | ||
4710 | l = t->loc + pos; | |
4711 | if (pos < t->count) | |
4712 | memmove(l + 1, l, | |
4713 | (t->count - pos) * sizeof(struct location)); | |
4714 | t->count++; | |
4715 | l->count = 1; | |
45edfa58 CL |
4716 | l->addr = track->addr; |
4717 | l->sum_time = age; | |
4718 | l->min_time = age; | |
4719 | l->max_time = age; | |
4720 | l->min_pid = track->pid; | |
4721 | l->max_pid = track->pid; | |
174596a0 RR |
4722 | cpumask_clear(to_cpumask(l->cpus)); |
4723 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4724 | nodes_clear(l->nodes); |
4725 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4726 | return 1; |
4727 | } | |
4728 | ||
4729 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
90e9f6a6 | 4730 | struct page *page, enum track_item alloc) |
88a420e4 | 4731 | { |
a973e9dd | 4732 | void *addr = page_address(page); |
88a420e4 | 4733 | void *p; |
90e9f6a6 | 4734 | unsigned long *map; |
88a420e4 | 4735 | |
90e9f6a6 | 4736 | map = get_map(s, page); |
224a88be | 4737 | for_each_object(p, s, addr, page->objects) |
4138fdfc | 4738 | if (!test_bit(__obj_to_index(s, addr, p), map)) |
45edfa58 | 4739 | add_location(t, s, get_track(s, p, alloc)); |
90e9f6a6 | 4740 | put_map(map); |
88a420e4 CL |
4741 | } |
4742 | ||
4743 | static int list_locations(struct kmem_cache *s, char *buf, | |
bf16d19a | 4744 | enum track_item alloc) |
88a420e4 | 4745 | { |
e374d483 | 4746 | int len = 0; |
88a420e4 | 4747 | unsigned long i; |
68dff6a9 | 4748 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4749 | int node; |
fa45dc25 | 4750 | struct kmem_cache_node *n; |
88a420e4 | 4751 | |
90e9f6a6 YZ |
4752 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4753 | GFP_KERNEL)) { | |
bf16d19a | 4754 | return sysfs_emit(buf, "Out of memory\n"); |
bbd7d57b | 4755 | } |
88a420e4 CL |
4756 | /* Push back cpu slabs */ |
4757 | flush_all(s); | |
4758 | ||
fa45dc25 | 4759 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4760 | unsigned long flags; |
4761 | struct page *page; | |
4762 | ||
9e86943b | 4763 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4764 | continue; |
4765 | ||
4766 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 4767 | list_for_each_entry(page, &n->partial, slab_list) |
90e9f6a6 | 4768 | process_slab(&t, s, page, alloc); |
916ac052 | 4769 | list_for_each_entry(page, &n->full, slab_list) |
90e9f6a6 | 4770 | process_slab(&t, s, page, alloc); |
88a420e4 CL |
4771 | spin_unlock_irqrestore(&n->list_lock, flags); |
4772 | } | |
4773 | ||
4774 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4775 | struct location *l = &t.loc[i]; |
88a420e4 | 4776 | |
bf16d19a | 4777 | len += sysfs_emit_at(buf, len, "%7ld ", l->count); |
45edfa58 CL |
4778 | |
4779 | if (l->addr) | |
bf16d19a | 4780 | len += sysfs_emit_at(buf, len, "%pS", (void *)l->addr); |
88a420e4 | 4781 | else |
bf16d19a JP |
4782 | len += sysfs_emit_at(buf, len, "<not-available>"); |
4783 | ||
4784 | if (l->sum_time != l->min_time) | |
4785 | len += sysfs_emit_at(buf, len, " age=%ld/%ld/%ld", | |
4786 | l->min_time, | |
4787 | (long)div_u64(l->sum_time, | |
4788 | l->count), | |
4789 | l->max_time); | |
4790 | else | |
4791 | len += sysfs_emit_at(buf, len, " age=%ld", l->min_time); | |
45edfa58 CL |
4792 | |
4793 | if (l->min_pid != l->max_pid) | |
bf16d19a JP |
4794 | len += sysfs_emit_at(buf, len, " pid=%ld-%ld", |
4795 | l->min_pid, l->max_pid); | |
45edfa58 | 4796 | else |
bf16d19a JP |
4797 | len += sysfs_emit_at(buf, len, " pid=%ld", |
4798 | l->min_pid); | |
45edfa58 | 4799 | |
174596a0 | 4800 | if (num_online_cpus() > 1 && |
bf16d19a JP |
4801 | !cpumask_empty(to_cpumask(l->cpus))) |
4802 | len += sysfs_emit_at(buf, len, " cpus=%*pbl", | |
4803 | cpumask_pr_args(to_cpumask(l->cpus))); | |
4804 | ||
4805 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes)) | |
4806 | len += sysfs_emit_at(buf, len, " nodes=%*pbl", | |
4807 | nodemask_pr_args(&l->nodes)); | |
4808 | ||
4809 | len += sysfs_emit_at(buf, len, "\n"); | |
88a420e4 CL |
4810 | } |
4811 | ||
4812 | free_loc_track(&t); | |
4813 | if (!t.count) | |
bf16d19a JP |
4814 | len += sysfs_emit_at(buf, len, "No data\n"); |
4815 | ||
e374d483 | 4816 | return len; |
88a420e4 | 4817 | } |
6dfd1b65 | 4818 | #endif /* CONFIG_SLUB_DEBUG */ |
88a420e4 | 4819 | |
a5a84755 | 4820 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4821 | static void __init resiliency_test(void) |
a5a84755 CL |
4822 | { |
4823 | u8 *p; | |
cc252eae | 4824 | int type = KMALLOC_NORMAL; |
a5a84755 | 4825 | |
95a05b42 | 4826 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4827 | |
f9f58285 FF |
4828 | pr_err("SLUB resiliency testing\n"); |
4829 | pr_err("-----------------------\n"); | |
4830 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4831 | |
4832 | p = kzalloc(16, GFP_KERNEL); | |
4833 | p[16] = 0x12; | |
f9f58285 FF |
4834 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4835 | p + 16); | |
a5a84755 | 4836 | |
cc252eae | 4837 | validate_slab_cache(kmalloc_caches[type][4]); |
a5a84755 CL |
4838 | |
4839 | /* Hmmm... The next two are dangerous */ | |
4840 | p = kzalloc(32, GFP_KERNEL); | |
4841 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4842 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4843 | p); | |
4844 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 | 4845 | |
cc252eae | 4846 | validate_slab_cache(kmalloc_caches[type][5]); |
a5a84755 CL |
4847 | p = kzalloc(64, GFP_KERNEL); |
4848 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4849 | *p = 0x56; | |
f9f58285 FF |
4850 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4851 | p); | |
4852 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
cc252eae | 4853 | validate_slab_cache(kmalloc_caches[type][6]); |
a5a84755 | 4854 | |
f9f58285 | 4855 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4856 | p = kzalloc(128, GFP_KERNEL); |
4857 | kfree(p); | |
4858 | *p = 0x78; | |
f9f58285 | 4859 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
cc252eae | 4860 | validate_slab_cache(kmalloc_caches[type][7]); |
a5a84755 CL |
4861 | |
4862 | p = kzalloc(256, GFP_KERNEL); | |
4863 | kfree(p); | |
4864 | p[50] = 0x9a; | |
f9f58285 | 4865 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
cc252eae | 4866 | validate_slab_cache(kmalloc_caches[type][8]); |
a5a84755 CL |
4867 | |
4868 | p = kzalloc(512, GFP_KERNEL); | |
4869 | kfree(p); | |
4870 | p[512] = 0xab; | |
f9f58285 | 4871 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
cc252eae | 4872 | validate_slab_cache(kmalloc_caches[type][9]); |
a5a84755 CL |
4873 | } |
4874 | #else | |
4875 | #ifdef CONFIG_SYSFS | |
4876 | static void resiliency_test(void) {}; | |
4877 | #endif | |
6dfd1b65 | 4878 | #endif /* SLUB_RESILIENCY_TEST */ |
a5a84755 | 4879 | |
ab4d5ed5 | 4880 | #ifdef CONFIG_SYSFS |
81819f0f | 4881 | enum slab_stat_type { |
205ab99d CL |
4882 | SL_ALL, /* All slabs */ |
4883 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4884 | SL_CPU, /* Only slabs used for cpu caches */ | |
4885 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4886 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4887 | }; |
4888 | ||
205ab99d | 4889 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4890 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4891 | #define SO_CPU (1 << SL_CPU) | |
4892 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4893 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4894 | |
1663f26d TH |
4895 | #ifdef CONFIG_MEMCG |
4896 | static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON); | |
4897 | ||
4898 | static int __init setup_slub_memcg_sysfs(char *str) | |
4899 | { | |
4900 | int v; | |
4901 | ||
4902 | if (get_option(&str, &v) > 0) | |
4903 | memcg_sysfs_enabled = v; | |
4904 | ||
4905 | return 1; | |
4906 | } | |
4907 | ||
4908 | __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs); | |
4909 | #endif | |
4910 | ||
62e5c4b4 | 4911 | static ssize_t show_slab_objects(struct kmem_cache *s, |
bf16d19a | 4912 | char *buf, unsigned long flags) |
81819f0f CL |
4913 | { |
4914 | unsigned long total = 0; | |
81819f0f CL |
4915 | int node; |
4916 | int x; | |
4917 | unsigned long *nodes; | |
bf16d19a | 4918 | int len = 0; |
81819f0f | 4919 | |
6396bb22 | 4920 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
4921 | if (!nodes) |
4922 | return -ENOMEM; | |
81819f0f | 4923 | |
205ab99d CL |
4924 | if (flags & SO_CPU) { |
4925 | int cpu; | |
81819f0f | 4926 | |
205ab99d | 4927 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4928 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4929 | cpu); | |
ec3ab083 | 4930 | int node; |
49e22585 | 4931 | struct page *page; |
dfb4f096 | 4932 | |
4db0c3c2 | 4933 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4934 | if (!page) |
4935 | continue; | |
205ab99d | 4936 | |
ec3ab083 CL |
4937 | node = page_to_nid(page); |
4938 | if (flags & SO_TOTAL) | |
4939 | x = page->objects; | |
4940 | else if (flags & SO_OBJECTS) | |
4941 | x = page->inuse; | |
4942 | else | |
4943 | x = 1; | |
49e22585 | 4944 | |
ec3ab083 CL |
4945 | total += x; |
4946 | nodes[node] += x; | |
4947 | ||
a93cf07b | 4948 | page = slub_percpu_partial_read_once(c); |
49e22585 | 4949 | if (page) { |
8afb1474 LZ |
4950 | node = page_to_nid(page); |
4951 | if (flags & SO_TOTAL) | |
4952 | WARN_ON_ONCE(1); | |
4953 | else if (flags & SO_OBJECTS) | |
4954 | WARN_ON_ONCE(1); | |
4955 | else | |
4956 | x = page->pages; | |
bc6697d8 ED |
4957 | total += x; |
4958 | nodes[node] += x; | |
49e22585 | 4959 | } |
81819f0f CL |
4960 | } |
4961 | } | |
4962 | ||
e4f8e513 QC |
4963 | /* |
4964 | * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex" | |
4965 | * already held which will conflict with an existing lock order: | |
4966 | * | |
4967 | * mem_hotplug_lock->slab_mutex->kernfs_mutex | |
4968 | * | |
4969 | * We don't really need mem_hotplug_lock (to hold off | |
4970 | * slab_mem_going_offline_callback) here because slab's memory hot | |
4971 | * unplug code doesn't destroy the kmem_cache->node[] data. | |
4972 | */ | |
4973 | ||
ab4d5ed5 | 4974 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4975 | if (flags & SO_ALL) { |
fa45dc25 CL |
4976 | struct kmem_cache_node *n; |
4977 | ||
4978 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4979 | |
d0e0ac97 CG |
4980 | if (flags & SO_TOTAL) |
4981 | x = atomic_long_read(&n->total_objects); | |
4982 | else if (flags & SO_OBJECTS) | |
4983 | x = atomic_long_read(&n->total_objects) - | |
4984 | count_partial(n, count_free); | |
81819f0f | 4985 | else |
205ab99d | 4986 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4987 | total += x; |
4988 | nodes[node] += x; | |
4989 | } | |
4990 | ||
ab4d5ed5 CL |
4991 | } else |
4992 | #endif | |
4993 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4994 | struct kmem_cache_node *n; |
81819f0f | 4995 | |
fa45dc25 | 4996 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4997 | if (flags & SO_TOTAL) |
4998 | x = count_partial(n, count_total); | |
4999 | else if (flags & SO_OBJECTS) | |
5000 | x = count_partial(n, count_inuse); | |
81819f0f | 5001 | else |
205ab99d | 5002 | x = n->nr_partial; |
81819f0f CL |
5003 | total += x; |
5004 | nodes[node] += x; | |
5005 | } | |
5006 | } | |
bf16d19a JP |
5007 | |
5008 | len += sysfs_emit_at(buf, len, "%lu", total); | |
81819f0f | 5009 | #ifdef CONFIG_NUMA |
bf16d19a | 5010 | for (node = 0; node < nr_node_ids; node++) { |
81819f0f | 5011 | if (nodes[node]) |
bf16d19a JP |
5012 | len += sysfs_emit_at(buf, len, " N%d=%lu", |
5013 | node, nodes[node]); | |
5014 | } | |
81819f0f | 5015 | #endif |
bf16d19a | 5016 | len += sysfs_emit_at(buf, len, "\n"); |
81819f0f | 5017 | kfree(nodes); |
bf16d19a JP |
5018 | |
5019 | return len; | |
81819f0f CL |
5020 | } |
5021 | ||
81819f0f | 5022 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) |
497888cf | 5023 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
5024 | |
5025 | struct slab_attribute { | |
5026 | struct attribute attr; | |
5027 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
5028 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
5029 | }; | |
5030 | ||
5031 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
5032 | static struct slab_attribute _name##_attr = \ |
5033 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
5034 | |
5035 | #define SLAB_ATTR(_name) \ | |
5036 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 5037 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 5038 | |
81819f0f CL |
5039 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
5040 | { | |
bf16d19a | 5041 | return sysfs_emit(buf, "%u\n", s->size); |
81819f0f CL |
5042 | } |
5043 | SLAB_ATTR_RO(slab_size); | |
5044 | ||
5045 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
5046 | { | |
bf16d19a | 5047 | return sysfs_emit(buf, "%u\n", s->align); |
81819f0f CL |
5048 | } |
5049 | SLAB_ATTR_RO(align); | |
5050 | ||
5051 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
5052 | { | |
bf16d19a | 5053 | return sysfs_emit(buf, "%u\n", s->object_size); |
81819f0f CL |
5054 | } |
5055 | SLAB_ATTR_RO(object_size); | |
5056 | ||
5057 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
5058 | { | |
bf16d19a | 5059 | return sysfs_emit(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
5060 | } |
5061 | SLAB_ATTR_RO(objs_per_slab); | |
5062 | ||
5063 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
5064 | { | |
bf16d19a | 5065 | return sysfs_emit(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 5066 | } |
32a6f409 | 5067 | SLAB_ATTR_RO(order); |
81819f0f | 5068 | |
73d342b1 DR |
5069 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
5070 | { | |
bf16d19a | 5071 | return sysfs_emit(buf, "%lu\n", s->min_partial); |
73d342b1 DR |
5072 | } |
5073 | ||
5074 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
5075 | size_t length) | |
5076 | { | |
5077 | unsigned long min; | |
5078 | int err; | |
5079 | ||
3dbb95f7 | 5080 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
5081 | if (err) |
5082 | return err; | |
5083 | ||
c0bdb232 | 5084 | set_min_partial(s, min); |
73d342b1 DR |
5085 | return length; |
5086 | } | |
5087 | SLAB_ATTR(min_partial); | |
5088 | ||
49e22585 CL |
5089 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
5090 | { | |
bf16d19a | 5091 | return sysfs_emit(buf, "%u\n", slub_cpu_partial(s)); |
49e22585 CL |
5092 | } |
5093 | ||
5094 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
5095 | size_t length) | |
5096 | { | |
e5d9998f | 5097 | unsigned int objects; |
49e22585 CL |
5098 | int err; |
5099 | ||
e5d9998f | 5100 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
5101 | if (err) |
5102 | return err; | |
345c905d | 5103 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 5104 | return -EINVAL; |
49e22585 | 5105 | |
e6d0e1dc | 5106 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
5107 | flush_all(s); |
5108 | return length; | |
5109 | } | |
5110 | SLAB_ATTR(cpu_partial); | |
5111 | ||
81819f0f CL |
5112 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
5113 | { | |
62c70bce JP |
5114 | if (!s->ctor) |
5115 | return 0; | |
bf16d19a | 5116 | return sysfs_emit(buf, "%pS\n", s->ctor); |
81819f0f CL |
5117 | } |
5118 | SLAB_ATTR_RO(ctor); | |
5119 | ||
81819f0f CL |
5120 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
5121 | { | |
bf16d19a | 5122 | return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
5123 | } |
5124 | SLAB_ATTR_RO(aliases); | |
5125 | ||
81819f0f CL |
5126 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5127 | { | |
d9acf4b7 | 5128 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5129 | } |
5130 | SLAB_ATTR_RO(partial); | |
5131 | ||
5132 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5133 | { | |
d9acf4b7 | 5134 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5135 | } |
5136 | SLAB_ATTR_RO(cpu_slabs); | |
5137 | ||
5138 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5139 | { | |
205ab99d | 5140 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5141 | } |
5142 | SLAB_ATTR_RO(objects); | |
5143 | ||
205ab99d CL |
5144 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5145 | { | |
5146 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5147 | } | |
5148 | SLAB_ATTR_RO(objects_partial); | |
5149 | ||
49e22585 CL |
5150 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5151 | { | |
5152 | int objects = 0; | |
5153 | int pages = 0; | |
5154 | int cpu; | |
bf16d19a | 5155 | int len = 0; |
49e22585 CL |
5156 | |
5157 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5158 | struct page *page; |
5159 | ||
5160 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5161 | |
5162 | if (page) { | |
5163 | pages += page->pages; | |
5164 | objects += page->pobjects; | |
5165 | } | |
5166 | } | |
5167 | ||
bf16d19a | 5168 | len += sysfs_emit_at(buf, len, "%d(%d)", objects, pages); |
49e22585 CL |
5169 | |
5170 | #ifdef CONFIG_SMP | |
5171 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5172 | struct page *page; |
5173 | ||
5174 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
bf16d19a JP |
5175 | if (page) |
5176 | len += sysfs_emit_at(buf, len, " C%d=%d(%d)", | |
5177 | cpu, page->pobjects, page->pages); | |
49e22585 CL |
5178 | } |
5179 | #endif | |
bf16d19a JP |
5180 | len += sysfs_emit_at(buf, len, "\n"); |
5181 | ||
5182 | return len; | |
49e22585 CL |
5183 | } |
5184 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5185 | ||
a5a84755 CL |
5186 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5187 | { | |
bf16d19a | 5188 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); |
a5a84755 | 5189 | } |
8f58119a | 5190 | SLAB_ATTR_RO(reclaim_account); |
a5a84755 CL |
5191 | |
5192 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5193 | { | |
bf16d19a | 5194 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
a5a84755 CL |
5195 | } |
5196 | SLAB_ATTR_RO(hwcache_align); | |
5197 | ||
5198 | #ifdef CONFIG_ZONE_DMA | |
5199 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5200 | { | |
bf16d19a | 5201 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); |
a5a84755 CL |
5202 | } |
5203 | SLAB_ATTR_RO(cache_dma); | |
5204 | #endif | |
5205 | ||
8eb8284b DW |
5206 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5207 | { | |
bf16d19a | 5208 | return sysfs_emit(buf, "%u\n", s->usersize); |
8eb8284b DW |
5209 | } |
5210 | SLAB_ATTR_RO(usersize); | |
5211 | ||
a5a84755 CL |
5212 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5213 | { | |
bf16d19a | 5214 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5215 | } |
5216 | SLAB_ATTR_RO(destroy_by_rcu); | |
5217 | ||
ab4d5ed5 | 5218 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5219 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5220 | { | |
5221 | return show_slab_objects(s, buf, SO_ALL); | |
5222 | } | |
5223 | SLAB_ATTR_RO(slabs); | |
5224 | ||
205ab99d CL |
5225 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5226 | { | |
5227 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5228 | } | |
5229 | SLAB_ATTR_RO(total_objects); | |
5230 | ||
81819f0f CL |
5231 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5232 | { | |
bf16d19a | 5233 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f | 5234 | } |
060807f8 | 5235 | SLAB_ATTR_RO(sanity_checks); |
81819f0f CL |
5236 | |
5237 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5238 | { | |
bf16d19a | 5239 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE)); |
81819f0f | 5240 | } |
060807f8 | 5241 | SLAB_ATTR_RO(trace); |
81819f0f | 5242 | |
81819f0f CL |
5243 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5244 | { | |
bf16d19a | 5245 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); |
81819f0f CL |
5246 | } |
5247 | ||
ad38b5b1 | 5248 | SLAB_ATTR_RO(red_zone); |
81819f0f CL |
5249 | |
5250 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5251 | { | |
bf16d19a | 5252 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON)); |
81819f0f CL |
5253 | } |
5254 | ||
ad38b5b1 | 5255 | SLAB_ATTR_RO(poison); |
81819f0f CL |
5256 | |
5257 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5258 | { | |
bf16d19a | 5259 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); |
81819f0f CL |
5260 | } |
5261 | ||
ad38b5b1 | 5262 | SLAB_ATTR_RO(store_user); |
81819f0f | 5263 | |
53e15af0 CL |
5264 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5265 | { | |
5266 | return 0; | |
5267 | } | |
5268 | ||
5269 | static ssize_t validate_store(struct kmem_cache *s, | |
5270 | const char *buf, size_t length) | |
5271 | { | |
434e245d CL |
5272 | int ret = -EINVAL; |
5273 | ||
5274 | if (buf[0] == '1') { | |
5275 | ret = validate_slab_cache(s); | |
5276 | if (ret >= 0) | |
5277 | ret = length; | |
5278 | } | |
5279 | return ret; | |
53e15af0 CL |
5280 | } |
5281 | SLAB_ATTR(validate); | |
a5a84755 CL |
5282 | |
5283 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5284 | { | |
5285 | if (!(s->flags & SLAB_STORE_USER)) | |
5286 | return -ENOSYS; | |
5287 | return list_locations(s, buf, TRACK_ALLOC); | |
5288 | } | |
5289 | SLAB_ATTR_RO(alloc_calls); | |
5290 | ||
5291 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5292 | { | |
5293 | if (!(s->flags & SLAB_STORE_USER)) | |
5294 | return -ENOSYS; | |
5295 | return list_locations(s, buf, TRACK_FREE); | |
5296 | } | |
5297 | SLAB_ATTR_RO(free_calls); | |
5298 | #endif /* CONFIG_SLUB_DEBUG */ | |
5299 | ||
5300 | #ifdef CONFIG_FAILSLAB | |
5301 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5302 | { | |
bf16d19a | 5303 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); |
a5a84755 | 5304 | } |
060807f8 | 5305 | SLAB_ATTR_RO(failslab); |
ab4d5ed5 | 5306 | #endif |
53e15af0 | 5307 | |
2086d26a CL |
5308 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5309 | { | |
5310 | return 0; | |
5311 | } | |
5312 | ||
5313 | static ssize_t shrink_store(struct kmem_cache *s, | |
5314 | const char *buf, size_t length) | |
5315 | { | |
832f37f5 | 5316 | if (buf[0] == '1') |
10befea9 | 5317 | kmem_cache_shrink(s); |
832f37f5 | 5318 | else |
2086d26a CL |
5319 | return -EINVAL; |
5320 | return length; | |
5321 | } | |
5322 | SLAB_ATTR(shrink); | |
5323 | ||
81819f0f | 5324 | #ifdef CONFIG_NUMA |
9824601e | 5325 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5326 | { |
bf16d19a | 5327 | return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5328 | } |
5329 | ||
9824601e | 5330 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5331 | const char *buf, size_t length) |
5332 | { | |
eb7235eb | 5333 | unsigned int ratio; |
0121c619 CL |
5334 | int err; |
5335 | ||
eb7235eb | 5336 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5337 | if (err) |
5338 | return err; | |
eb7235eb AD |
5339 | if (ratio > 100) |
5340 | return -ERANGE; | |
0121c619 | 5341 | |
eb7235eb | 5342 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5343 | |
81819f0f CL |
5344 | return length; |
5345 | } | |
9824601e | 5346 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5347 | #endif |
5348 | ||
8ff12cfc | 5349 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5350 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5351 | { | |
5352 | unsigned long sum = 0; | |
5353 | int cpu; | |
bf16d19a | 5354 | int len = 0; |
6da2ec56 | 5355 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5356 | |
5357 | if (!data) | |
5358 | return -ENOMEM; | |
5359 | ||
5360 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5361 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5362 | |
5363 | data[cpu] = x; | |
5364 | sum += x; | |
5365 | } | |
5366 | ||
bf16d19a | 5367 | len += sysfs_emit_at(buf, len, "%lu", sum); |
8ff12cfc | 5368 | |
50ef37b9 | 5369 | #ifdef CONFIG_SMP |
8ff12cfc | 5370 | for_each_online_cpu(cpu) { |
bf16d19a JP |
5371 | if (data[cpu]) |
5372 | len += sysfs_emit_at(buf, len, " C%d=%u", | |
5373 | cpu, data[cpu]); | |
8ff12cfc | 5374 | } |
50ef37b9 | 5375 | #endif |
8ff12cfc | 5376 | kfree(data); |
bf16d19a JP |
5377 | len += sysfs_emit_at(buf, len, "\n"); |
5378 | ||
5379 | return len; | |
8ff12cfc CL |
5380 | } |
5381 | ||
78eb00cc DR |
5382 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5383 | { | |
5384 | int cpu; | |
5385 | ||
5386 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5387 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5388 | } |
5389 | ||
8ff12cfc CL |
5390 | #define STAT_ATTR(si, text) \ |
5391 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5392 | { \ | |
5393 | return show_stat(s, buf, si); \ | |
5394 | } \ | |
78eb00cc DR |
5395 | static ssize_t text##_store(struct kmem_cache *s, \ |
5396 | const char *buf, size_t length) \ | |
5397 | { \ | |
5398 | if (buf[0] != '0') \ | |
5399 | return -EINVAL; \ | |
5400 | clear_stat(s, si); \ | |
5401 | return length; \ | |
5402 | } \ | |
5403 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5404 | |
5405 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5406 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5407 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5408 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5409 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5410 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5411 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5412 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5413 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5414 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5415 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5416 | STAT_ATTR(FREE_SLAB, free_slab); |
5417 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5418 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5419 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5420 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5421 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5422 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5423 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5424 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5425 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5426 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5427 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5428 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5429 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5430 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
6dfd1b65 | 5431 | #endif /* CONFIG_SLUB_STATS */ |
8ff12cfc | 5432 | |
06428780 | 5433 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5434 | &slab_size_attr.attr, |
5435 | &object_size_attr.attr, | |
5436 | &objs_per_slab_attr.attr, | |
5437 | &order_attr.attr, | |
73d342b1 | 5438 | &min_partial_attr.attr, |
49e22585 | 5439 | &cpu_partial_attr.attr, |
81819f0f | 5440 | &objects_attr.attr, |
205ab99d | 5441 | &objects_partial_attr.attr, |
81819f0f CL |
5442 | &partial_attr.attr, |
5443 | &cpu_slabs_attr.attr, | |
5444 | &ctor_attr.attr, | |
81819f0f CL |
5445 | &aliases_attr.attr, |
5446 | &align_attr.attr, | |
81819f0f CL |
5447 | &hwcache_align_attr.attr, |
5448 | &reclaim_account_attr.attr, | |
5449 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5450 | &shrink_attr.attr, |
49e22585 | 5451 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5452 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5453 | &total_objects_attr.attr, |
5454 | &slabs_attr.attr, | |
5455 | &sanity_checks_attr.attr, | |
5456 | &trace_attr.attr, | |
81819f0f CL |
5457 | &red_zone_attr.attr, |
5458 | &poison_attr.attr, | |
5459 | &store_user_attr.attr, | |
53e15af0 | 5460 | &validate_attr.attr, |
88a420e4 CL |
5461 | &alloc_calls_attr.attr, |
5462 | &free_calls_attr.attr, | |
ab4d5ed5 | 5463 | #endif |
81819f0f CL |
5464 | #ifdef CONFIG_ZONE_DMA |
5465 | &cache_dma_attr.attr, | |
5466 | #endif | |
5467 | #ifdef CONFIG_NUMA | |
9824601e | 5468 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5469 | #endif |
5470 | #ifdef CONFIG_SLUB_STATS | |
5471 | &alloc_fastpath_attr.attr, | |
5472 | &alloc_slowpath_attr.attr, | |
5473 | &free_fastpath_attr.attr, | |
5474 | &free_slowpath_attr.attr, | |
5475 | &free_frozen_attr.attr, | |
5476 | &free_add_partial_attr.attr, | |
5477 | &free_remove_partial_attr.attr, | |
5478 | &alloc_from_partial_attr.attr, | |
5479 | &alloc_slab_attr.attr, | |
5480 | &alloc_refill_attr.attr, | |
e36a2652 | 5481 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5482 | &free_slab_attr.attr, |
5483 | &cpuslab_flush_attr.attr, | |
5484 | &deactivate_full_attr.attr, | |
5485 | &deactivate_empty_attr.attr, | |
5486 | &deactivate_to_head_attr.attr, | |
5487 | &deactivate_to_tail_attr.attr, | |
5488 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5489 | &deactivate_bypass_attr.attr, |
65c3376a | 5490 | &order_fallback_attr.attr, |
b789ef51 CL |
5491 | &cmpxchg_double_fail_attr.attr, |
5492 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5493 | &cpu_partial_alloc_attr.attr, |
5494 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5495 | &cpu_partial_node_attr.attr, |
5496 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5497 | #endif |
4c13dd3b DM |
5498 | #ifdef CONFIG_FAILSLAB |
5499 | &failslab_attr.attr, | |
5500 | #endif | |
8eb8284b | 5501 | &usersize_attr.attr, |
4c13dd3b | 5502 | |
81819f0f CL |
5503 | NULL |
5504 | }; | |
5505 | ||
1fdaaa23 | 5506 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5507 | .attrs = slab_attrs, |
5508 | }; | |
5509 | ||
5510 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5511 | struct attribute *attr, | |
5512 | char *buf) | |
5513 | { | |
5514 | struct slab_attribute *attribute; | |
5515 | struct kmem_cache *s; | |
5516 | int err; | |
5517 | ||
5518 | attribute = to_slab_attr(attr); | |
5519 | s = to_slab(kobj); | |
5520 | ||
5521 | if (!attribute->show) | |
5522 | return -EIO; | |
5523 | ||
5524 | err = attribute->show(s, buf); | |
5525 | ||
5526 | return err; | |
5527 | } | |
5528 | ||
5529 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5530 | struct attribute *attr, | |
5531 | const char *buf, size_t len) | |
5532 | { | |
5533 | struct slab_attribute *attribute; | |
5534 | struct kmem_cache *s; | |
5535 | int err; | |
5536 | ||
5537 | attribute = to_slab_attr(attr); | |
5538 | s = to_slab(kobj); | |
5539 | ||
5540 | if (!attribute->store) | |
5541 | return -EIO; | |
5542 | ||
5543 | err = attribute->store(s, buf, len); | |
81819f0f CL |
5544 | return err; |
5545 | } | |
5546 | ||
41a21285 CL |
5547 | static void kmem_cache_release(struct kobject *k) |
5548 | { | |
5549 | slab_kmem_cache_release(to_slab(k)); | |
5550 | } | |
5551 | ||
52cf25d0 | 5552 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5553 | .show = slab_attr_show, |
5554 | .store = slab_attr_store, | |
5555 | }; | |
5556 | ||
5557 | static struct kobj_type slab_ktype = { | |
5558 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5559 | .release = kmem_cache_release, |
81819f0f CL |
5560 | }; |
5561 | ||
27c3a314 | 5562 | static struct kset *slab_kset; |
81819f0f | 5563 | |
9a41707b VD |
5564 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5565 | { | |
9a41707b VD |
5566 | return slab_kset; |
5567 | } | |
5568 | ||
81819f0f CL |
5569 | #define ID_STR_LENGTH 64 |
5570 | ||
5571 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5572 | * |
5573 | * Format :[flags-]size | |
81819f0f CL |
5574 | */ |
5575 | static char *create_unique_id(struct kmem_cache *s) | |
5576 | { | |
5577 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5578 | char *p = name; | |
5579 | ||
5580 | BUG_ON(!name); | |
5581 | ||
5582 | *p++ = ':'; | |
5583 | /* | |
5584 | * First flags affecting slabcache operations. We will only | |
5585 | * get here for aliasable slabs so we do not need to support | |
5586 | * too many flags. The flags here must cover all flags that | |
5587 | * are matched during merging to guarantee that the id is | |
5588 | * unique. | |
5589 | */ | |
5590 | if (s->flags & SLAB_CACHE_DMA) | |
5591 | *p++ = 'd'; | |
6d6ea1e9 NB |
5592 | if (s->flags & SLAB_CACHE_DMA32) |
5593 | *p++ = 'D'; | |
81819f0f CL |
5594 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
5595 | *p++ = 'a'; | |
becfda68 | 5596 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5597 | *p++ = 'F'; |
230e9fc2 VD |
5598 | if (s->flags & SLAB_ACCOUNT) |
5599 | *p++ = 'A'; | |
81819f0f CL |
5600 | if (p != name + 1) |
5601 | *p++ = '-'; | |
44065b2e | 5602 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5603 | |
81819f0f CL |
5604 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5605 | return name; | |
5606 | } | |
5607 | ||
5608 | static int sysfs_slab_add(struct kmem_cache *s) | |
5609 | { | |
5610 | int err; | |
5611 | const char *name; | |
1663f26d | 5612 | struct kset *kset = cache_kset(s); |
45530c44 | 5613 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5614 | |
1663f26d TH |
5615 | if (!kset) { |
5616 | kobject_init(&s->kobj, &slab_ktype); | |
5617 | return 0; | |
5618 | } | |
5619 | ||
11066386 MC |
5620 | if (!unmergeable && disable_higher_order_debug && |
5621 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5622 | unmergeable = 1; | |
5623 | ||
81819f0f CL |
5624 | if (unmergeable) { |
5625 | /* | |
5626 | * Slabcache can never be merged so we can use the name proper. | |
5627 | * This is typically the case for debug situations. In that | |
5628 | * case we can catch duplicate names easily. | |
5629 | */ | |
27c3a314 | 5630 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5631 | name = s->name; |
5632 | } else { | |
5633 | /* | |
5634 | * Create a unique name for the slab as a target | |
5635 | * for the symlinks. | |
5636 | */ | |
5637 | name = create_unique_id(s); | |
5638 | } | |
5639 | ||
1663f26d | 5640 | s->kobj.kset = kset; |
26e4f205 | 5641 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
757fed1d | 5642 | if (err) |
80da026a | 5643 | goto out; |
81819f0f CL |
5644 | |
5645 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5646 | if (err) |
5647 | goto out_del_kobj; | |
9a41707b | 5648 | |
81819f0f CL |
5649 | if (!unmergeable) { |
5650 | /* Setup first alias */ | |
5651 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5652 | } |
54b6a731 DJ |
5653 | out: |
5654 | if (!unmergeable) | |
5655 | kfree(name); | |
5656 | return err; | |
5657 | out_del_kobj: | |
5658 | kobject_del(&s->kobj); | |
54b6a731 | 5659 | goto out; |
81819f0f CL |
5660 | } |
5661 | ||
d50d82fa MP |
5662 | void sysfs_slab_unlink(struct kmem_cache *s) |
5663 | { | |
5664 | if (slab_state >= FULL) | |
5665 | kobject_del(&s->kobj); | |
5666 | } | |
5667 | ||
bf5eb3de TH |
5668 | void sysfs_slab_release(struct kmem_cache *s) |
5669 | { | |
5670 | if (slab_state >= FULL) | |
5671 | kobject_put(&s->kobj); | |
81819f0f CL |
5672 | } |
5673 | ||
5674 | /* | |
5675 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5676 | * available lest we lose that information. |
81819f0f CL |
5677 | */ |
5678 | struct saved_alias { | |
5679 | struct kmem_cache *s; | |
5680 | const char *name; | |
5681 | struct saved_alias *next; | |
5682 | }; | |
5683 | ||
5af328a5 | 5684 | static struct saved_alias *alias_list; |
81819f0f CL |
5685 | |
5686 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5687 | { | |
5688 | struct saved_alias *al; | |
5689 | ||
97d06609 | 5690 | if (slab_state == FULL) { |
81819f0f CL |
5691 | /* |
5692 | * If we have a leftover link then remove it. | |
5693 | */ | |
27c3a314 GKH |
5694 | sysfs_remove_link(&slab_kset->kobj, name); |
5695 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5696 | } |
5697 | ||
5698 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5699 | if (!al) | |
5700 | return -ENOMEM; | |
5701 | ||
5702 | al->s = s; | |
5703 | al->name = name; | |
5704 | al->next = alias_list; | |
5705 | alias_list = al; | |
5706 | return 0; | |
5707 | } | |
5708 | ||
5709 | static int __init slab_sysfs_init(void) | |
5710 | { | |
5b95a4ac | 5711 | struct kmem_cache *s; |
81819f0f CL |
5712 | int err; |
5713 | ||
18004c5d | 5714 | mutex_lock(&slab_mutex); |
2bce6485 | 5715 | |
d7660ce5 | 5716 | slab_kset = kset_create_and_add("slab", NULL, kernel_kobj); |
27c3a314 | 5717 | if (!slab_kset) { |
18004c5d | 5718 | mutex_unlock(&slab_mutex); |
f9f58285 | 5719 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5720 | return -ENOSYS; |
5721 | } | |
5722 | ||
97d06609 | 5723 | slab_state = FULL; |
26a7bd03 | 5724 | |
5b95a4ac | 5725 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5726 | err = sysfs_slab_add(s); |
5d540fb7 | 5727 | if (err) |
f9f58285 FF |
5728 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5729 | s->name); | |
26a7bd03 | 5730 | } |
81819f0f CL |
5731 | |
5732 | while (alias_list) { | |
5733 | struct saved_alias *al = alias_list; | |
5734 | ||
5735 | alias_list = alias_list->next; | |
5736 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5737 | if (err) |
f9f58285 FF |
5738 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5739 | al->name); | |
81819f0f CL |
5740 | kfree(al); |
5741 | } | |
5742 | ||
18004c5d | 5743 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5744 | resiliency_test(); |
5745 | return 0; | |
5746 | } | |
5747 | ||
5748 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5749 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5750 | |
5751 | /* | |
5752 | * The /proc/slabinfo ABI | |
5753 | */ | |
5b365771 | 5754 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 5755 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5756 | { |
57ed3eda | 5757 | unsigned long nr_slabs = 0; |
205ab99d CL |
5758 | unsigned long nr_objs = 0; |
5759 | unsigned long nr_free = 0; | |
57ed3eda | 5760 | int node; |
fa45dc25 | 5761 | struct kmem_cache_node *n; |
57ed3eda | 5762 | |
fa45dc25 | 5763 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5764 | nr_slabs += node_nr_slabs(n); |
5765 | nr_objs += node_nr_objs(n); | |
205ab99d | 5766 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5767 | } |
5768 | ||
0d7561c6 GC |
5769 | sinfo->active_objs = nr_objs - nr_free; |
5770 | sinfo->num_objs = nr_objs; | |
5771 | sinfo->active_slabs = nr_slabs; | |
5772 | sinfo->num_slabs = nr_slabs; | |
5773 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5774 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5775 | } |
5776 | ||
0d7561c6 | 5777 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5778 | { |
7b3c3a50 AD |
5779 | } |
5780 | ||
b7454ad3 GC |
5781 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5782 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5783 | { |
b7454ad3 | 5784 | return -EIO; |
7b3c3a50 | 5785 | } |
5b365771 | 5786 | #endif /* CONFIG_SLUB_DEBUG */ |