]>
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 AK |
251 | /* |
252 | * When CONFIG_KASAN_SW_TAGS is enabled, ptr_addr might be tagged. | |
253 | * Normally, this doesn't cause any issues, as both set_freepointer() | |
254 | * and get_freepointer() are called with a pointer with the same tag. | |
255 | * However, there are some issues with CONFIG_SLUB_DEBUG code. For | |
256 | * example, when __free_slub() iterates over objects in a cache, it | |
257 | * passes untagged pointers to check_object(). check_object() in turns | |
258 | * calls get_freepointer() with an untagged pointer, which causes the | |
259 | * freepointer to be restored incorrectly. | |
260 | */ | |
261 | return (void *)((unsigned long)ptr ^ s->random ^ | |
1ad53d9f | 262 | swab((unsigned long)kasan_reset_tag((void *)ptr_addr))); |
2482ddec KC |
263 | #else |
264 | return ptr; | |
265 | #endif | |
266 | } | |
267 | ||
268 | /* Returns the freelist pointer recorded at location ptr_addr. */ | |
269 | static inline void *freelist_dereference(const struct kmem_cache *s, | |
270 | void *ptr_addr) | |
271 | { | |
272 | return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr), | |
273 | (unsigned long)ptr_addr); | |
274 | } | |
275 | ||
7656c72b CL |
276 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
277 | { | |
2482ddec | 278 | return freelist_dereference(s, object + s->offset); |
7656c72b CL |
279 | } |
280 | ||
0ad9500e ED |
281 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
282 | { | |
0882ff91 | 283 | prefetch(object + s->offset); |
0ad9500e ED |
284 | } |
285 | ||
1393d9a1 CL |
286 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
287 | { | |
2482ddec | 288 | unsigned long freepointer_addr; |
1393d9a1 CL |
289 | void *p; |
290 | ||
8e57f8ac | 291 | if (!debug_pagealloc_enabled_static()) |
922d566c JK |
292 | return get_freepointer(s, object); |
293 | ||
2482ddec | 294 | freepointer_addr = (unsigned long)object + s->offset; |
fe557319 | 295 | copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p)); |
2482ddec | 296 | return freelist_ptr(s, p, freepointer_addr); |
1393d9a1 CL |
297 | } |
298 | ||
7656c72b CL |
299 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
300 | { | |
2482ddec KC |
301 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
302 | ||
ce6fa91b AP |
303 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
304 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
305 | #endif | |
306 | ||
2482ddec | 307 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); |
7656c72b CL |
308 | } |
309 | ||
310 | /* Loop over all objects in a slab */ | |
224a88be | 311 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
312 | for (__p = fixup_red_left(__s, __addr); \ |
313 | __p < (__addr) + (__objects) * (__s)->size; \ | |
314 | __p += (__s)->size) | |
7656c72b | 315 | |
9736d2a9 | 316 | static inline unsigned int order_objects(unsigned int order, unsigned int size) |
ab9a0f19 | 317 | { |
9736d2a9 | 318 | return ((unsigned int)PAGE_SIZE << order) / size; |
ab9a0f19 LJ |
319 | } |
320 | ||
19af27af | 321 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, |
9736d2a9 | 322 | unsigned int size) |
834f3d11 CL |
323 | { |
324 | struct kmem_cache_order_objects x = { | |
9736d2a9 | 325 | (order << OO_SHIFT) + order_objects(order, size) |
834f3d11 CL |
326 | }; |
327 | ||
328 | return x; | |
329 | } | |
330 | ||
19af27af | 331 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) |
834f3d11 | 332 | { |
210b5c06 | 333 | return x.x >> OO_SHIFT; |
834f3d11 CL |
334 | } |
335 | ||
19af27af | 336 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) |
834f3d11 | 337 | { |
210b5c06 | 338 | return x.x & OO_MASK; |
834f3d11 CL |
339 | } |
340 | ||
881db7fb CL |
341 | /* |
342 | * Per slab locking using the pagelock | |
343 | */ | |
344 | static __always_inline void slab_lock(struct page *page) | |
345 | { | |
48c935ad | 346 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
347 | bit_spin_lock(PG_locked, &page->flags); |
348 | } | |
349 | ||
350 | static __always_inline void slab_unlock(struct page *page) | |
351 | { | |
48c935ad | 352 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
353 | __bit_spin_unlock(PG_locked, &page->flags); |
354 | } | |
355 | ||
1d07171c CL |
356 | /* Interrupts must be disabled (for the fallback code to work right) */ |
357 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
358 | void *freelist_old, unsigned long counters_old, | |
359 | void *freelist_new, unsigned long counters_new, | |
360 | const char *n) | |
361 | { | |
362 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
363 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
364 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 365 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 366 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
367 | freelist_old, counters_old, |
368 | freelist_new, counters_new)) | |
6f6528a1 | 369 | return true; |
1d07171c CL |
370 | } else |
371 | #endif | |
372 | { | |
373 | slab_lock(page); | |
d0e0ac97 CG |
374 | if (page->freelist == freelist_old && |
375 | page->counters == counters_old) { | |
1d07171c | 376 | page->freelist = freelist_new; |
7d27a04b | 377 | page->counters = counters_new; |
1d07171c | 378 | slab_unlock(page); |
6f6528a1 | 379 | return true; |
1d07171c CL |
380 | } |
381 | slab_unlock(page); | |
382 | } | |
383 | ||
384 | cpu_relax(); | |
385 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
386 | ||
387 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 388 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
389 | #endif |
390 | ||
6f6528a1 | 391 | return false; |
1d07171c CL |
392 | } |
393 | ||
b789ef51 CL |
394 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
395 | void *freelist_old, unsigned long counters_old, | |
396 | void *freelist_new, unsigned long counters_new, | |
397 | const char *n) | |
398 | { | |
2565409f HC |
399 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
400 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 401 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 402 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
403 | freelist_old, counters_old, |
404 | freelist_new, counters_new)) | |
6f6528a1 | 405 | return true; |
b789ef51 CL |
406 | } else |
407 | #endif | |
408 | { | |
1d07171c CL |
409 | unsigned long flags; |
410 | ||
411 | local_irq_save(flags); | |
881db7fb | 412 | slab_lock(page); |
d0e0ac97 CG |
413 | if (page->freelist == freelist_old && |
414 | page->counters == counters_old) { | |
b789ef51 | 415 | page->freelist = freelist_new; |
7d27a04b | 416 | page->counters = counters_new; |
881db7fb | 417 | slab_unlock(page); |
1d07171c | 418 | local_irq_restore(flags); |
6f6528a1 | 419 | return true; |
b789ef51 | 420 | } |
881db7fb | 421 | slab_unlock(page); |
1d07171c | 422 | local_irq_restore(flags); |
b789ef51 CL |
423 | } |
424 | ||
425 | cpu_relax(); | |
426 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
427 | ||
428 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 429 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
430 | #endif |
431 | ||
6f6528a1 | 432 | return false; |
b789ef51 CL |
433 | } |
434 | ||
41ecc55b | 435 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 YZ |
436 | static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)]; |
437 | static DEFINE_SPINLOCK(object_map_lock); | |
438 | ||
5f80b13a CL |
439 | /* |
440 | * Determine a map of object in use on a page. | |
441 | * | |
881db7fb | 442 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
443 | * not vanish from under us. |
444 | */ | |
90e9f6a6 | 445 | static unsigned long *get_map(struct kmem_cache *s, struct page *page) |
31364c2e | 446 | __acquires(&object_map_lock) |
5f80b13a CL |
447 | { |
448 | void *p; | |
449 | void *addr = page_address(page); | |
450 | ||
90e9f6a6 YZ |
451 | VM_BUG_ON(!irqs_disabled()); |
452 | ||
453 | spin_lock(&object_map_lock); | |
454 | ||
455 | bitmap_zero(object_map, page->objects); | |
456 | ||
5f80b13a | 457 | for (p = page->freelist; p; p = get_freepointer(s, p)) |
4138fdfc | 458 | set_bit(__obj_to_index(s, addr, p), object_map); |
90e9f6a6 YZ |
459 | |
460 | return object_map; | |
461 | } | |
462 | ||
81aba9e0 | 463 | static void put_map(unsigned long *map) __releases(&object_map_lock) |
90e9f6a6 YZ |
464 | { |
465 | VM_BUG_ON(map != object_map); | |
90e9f6a6 | 466 | spin_unlock(&object_map_lock); |
5f80b13a CL |
467 | } |
468 | ||
870b1fbb | 469 | static inline unsigned int size_from_object(struct kmem_cache *s) |
d86bd1be JK |
470 | { |
471 | if (s->flags & SLAB_RED_ZONE) | |
472 | return s->size - s->red_left_pad; | |
473 | ||
474 | return s->size; | |
475 | } | |
476 | ||
477 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
478 | { | |
479 | if (s->flags & SLAB_RED_ZONE) | |
480 | p -= s->red_left_pad; | |
481 | ||
482 | return p; | |
483 | } | |
484 | ||
41ecc55b CL |
485 | /* |
486 | * Debug settings: | |
487 | */ | |
89d3c87e | 488 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 489 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 490 | #else |
d50112ed | 491 | static slab_flags_t slub_debug; |
f0630fff | 492 | #endif |
41ecc55b | 493 | |
e17f1dfb | 494 | static char *slub_debug_string; |
fa5ec8a1 | 495 | static int disable_higher_order_debug; |
41ecc55b | 496 | |
a79316c6 AR |
497 | /* |
498 | * slub is about to manipulate internal object metadata. This memory lies | |
499 | * outside the range of the allocated object, so accessing it would normally | |
500 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
501 | * to tell kasan that these accesses are OK. | |
502 | */ | |
503 | static inline void metadata_access_enable(void) | |
504 | { | |
505 | kasan_disable_current(); | |
506 | } | |
507 | ||
508 | static inline void metadata_access_disable(void) | |
509 | { | |
510 | kasan_enable_current(); | |
511 | } | |
512 | ||
81819f0f CL |
513 | /* |
514 | * Object debugging | |
515 | */ | |
d86bd1be JK |
516 | |
517 | /* Verify that a pointer has an address that is valid within a slab page */ | |
518 | static inline int check_valid_pointer(struct kmem_cache *s, | |
519 | struct page *page, void *object) | |
520 | { | |
521 | void *base; | |
522 | ||
523 | if (!object) | |
524 | return 1; | |
525 | ||
526 | base = page_address(page); | |
338cfaad | 527 | object = kasan_reset_tag(object); |
d86bd1be JK |
528 | object = restore_red_left(s, object); |
529 | if (object < base || object >= base + page->objects * s->size || | |
530 | (object - base) % s->size) { | |
531 | return 0; | |
532 | } | |
533 | ||
534 | return 1; | |
535 | } | |
536 | ||
aa2efd5e DT |
537 | static void print_section(char *level, char *text, u8 *addr, |
538 | unsigned int length) | |
81819f0f | 539 | { |
a79316c6 | 540 | metadata_access_enable(); |
aa2efd5e | 541 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
ffc79d28 | 542 | length, 1); |
a79316c6 | 543 | metadata_access_disable(); |
81819f0f CL |
544 | } |
545 | ||
cbfc35a4 WL |
546 | /* |
547 | * See comment in calculate_sizes(). | |
548 | */ | |
549 | static inline bool freeptr_outside_object(struct kmem_cache *s) | |
550 | { | |
551 | return s->offset >= s->inuse; | |
552 | } | |
553 | ||
554 | /* | |
555 | * Return offset of the end of info block which is inuse + free pointer if | |
556 | * not overlapping with object. | |
557 | */ | |
558 | static inline unsigned int get_info_end(struct kmem_cache *s) | |
559 | { | |
560 | if (freeptr_outside_object(s)) | |
561 | return s->inuse + sizeof(void *); | |
562 | else | |
563 | return s->inuse; | |
564 | } | |
565 | ||
81819f0f CL |
566 | static struct track *get_track(struct kmem_cache *s, void *object, |
567 | enum track_item alloc) | |
568 | { | |
569 | struct track *p; | |
570 | ||
cbfc35a4 | 571 | p = object + get_info_end(s); |
81819f0f CL |
572 | |
573 | return p + alloc; | |
574 | } | |
575 | ||
576 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 577 | enum track_item alloc, unsigned long addr) |
81819f0f | 578 | { |
1a00df4a | 579 | struct track *p = get_track(s, object, alloc); |
81819f0f | 580 | |
81819f0f | 581 | if (addr) { |
d6543e39 | 582 | #ifdef CONFIG_STACKTRACE |
79716799 | 583 | unsigned int nr_entries; |
d6543e39 | 584 | |
a79316c6 | 585 | metadata_access_enable(); |
79716799 | 586 | nr_entries = stack_trace_save(p->addrs, TRACK_ADDRS_COUNT, 3); |
a79316c6 | 587 | metadata_access_disable(); |
d6543e39 | 588 | |
79716799 TG |
589 | if (nr_entries < TRACK_ADDRS_COUNT) |
590 | p->addrs[nr_entries] = 0; | |
d6543e39 | 591 | #endif |
81819f0f CL |
592 | p->addr = addr; |
593 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 594 | p->pid = current->pid; |
81819f0f | 595 | p->when = jiffies; |
b8ca7ff7 | 596 | } else { |
81819f0f | 597 | memset(p, 0, sizeof(struct track)); |
b8ca7ff7 | 598 | } |
81819f0f CL |
599 | } |
600 | ||
81819f0f CL |
601 | static void init_tracking(struct kmem_cache *s, void *object) |
602 | { | |
24922684 CL |
603 | if (!(s->flags & SLAB_STORE_USER)) |
604 | return; | |
605 | ||
ce71e27c EGM |
606 | set_track(s, object, TRACK_FREE, 0UL); |
607 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
608 | } |
609 | ||
86609d33 | 610 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f CL |
611 | { |
612 | if (!t->addr) | |
613 | return; | |
614 | ||
f9f58285 | 615 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 616 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
d6543e39 BG |
617 | #ifdef CONFIG_STACKTRACE |
618 | { | |
619 | int i; | |
620 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
621 | if (t->addrs[i]) | |
f9f58285 | 622 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
623 | else |
624 | break; | |
625 | } | |
626 | #endif | |
24922684 CL |
627 | } |
628 | ||
e42f174e | 629 | void print_tracking(struct kmem_cache *s, void *object) |
24922684 | 630 | { |
86609d33 | 631 | unsigned long pr_time = jiffies; |
24922684 CL |
632 | if (!(s->flags & SLAB_STORE_USER)) |
633 | return; | |
634 | ||
86609d33 CP |
635 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
636 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
637 | } |
638 | ||
639 | static void print_page_info(struct page *page) | |
640 | { | |
f9f58285 | 641 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 642 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
643 | |
644 | } | |
645 | ||
646 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
647 | { | |
ecc42fbe | 648 | struct va_format vaf; |
24922684 | 649 | va_list args; |
24922684 CL |
650 | |
651 | va_start(args, fmt); | |
ecc42fbe FF |
652 | vaf.fmt = fmt; |
653 | vaf.va = &args; | |
f9f58285 | 654 | pr_err("=============================================================================\n"); |
ecc42fbe | 655 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 656 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 657 | |
373d4d09 | 658 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 659 | va_end(args); |
81819f0f CL |
660 | } |
661 | ||
24922684 CL |
662 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
663 | { | |
ecc42fbe | 664 | struct va_format vaf; |
24922684 | 665 | va_list args; |
24922684 CL |
666 | |
667 | va_start(args, fmt); | |
ecc42fbe FF |
668 | vaf.fmt = fmt; |
669 | vaf.va = &args; | |
670 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 671 | va_end(args); |
24922684 CL |
672 | } |
673 | ||
52f23478 | 674 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 675 | void **freelist, void *nextfree) |
52f23478 DZ |
676 | { |
677 | if ((s->flags & SLAB_CONSISTENCY_CHECKS) && | |
dc07a728 ER |
678 | !check_valid_pointer(s, page, nextfree) && freelist) { |
679 | object_err(s, page, *freelist, "Freechain corrupt"); | |
680 | *freelist = NULL; | |
52f23478 DZ |
681 | slab_fix(s, "Isolate corrupted freechain"); |
682 | return true; | |
683 | } | |
684 | ||
685 | return false; | |
686 | } | |
687 | ||
24922684 | 688 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) |
81819f0f CL |
689 | { |
690 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 691 | u8 *addr = page_address(page); |
24922684 CL |
692 | |
693 | print_tracking(s, p); | |
694 | ||
695 | print_page_info(page); | |
696 | ||
f9f58285 FF |
697 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
698 | p, p - addr, get_freepointer(s, p)); | |
24922684 | 699 | |
d86bd1be | 700 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e DT |
701 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
702 | s->red_left_pad); | |
d86bd1be | 703 | else if (p > addr + 16) |
aa2efd5e | 704 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 705 | |
aa2efd5e | 706 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 707 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 708 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e | 709 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 710 | s->inuse - s->object_size); |
81819f0f | 711 | |
cbfc35a4 | 712 | off = get_info_end(s); |
81819f0f | 713 | |
24922684 | 714 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 715 | off += 2 * sizeof(struct track); |
81819f0f | 716 | |
80a9201a AP |
717 | off += kasan_metadata_size(s); |
718 | ||
d86bd1be | 719 | if (off != size_from_object(s)) |
81819f0f | 720 | /* Beginning of the filler is the free pointer */ |
aa2efd5e DT |
721 | print_section(KERN_ERR, "Padding ", p + off, |
722 | size_from_object(s) - off); | |
24922684 CL |
723 | |
724 | dump_stack(); | |
81819f0f CL |
725 | } |
726 | ||
75c66def | 727 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
728 | u8 *object, char *reason) |
729 | { | |
3dc50637 | 730 | slab_bug(s, "%s", reason); |
24922684 | 731 | print_trailer(s, page, object); |
81819f0f CL |
732 | } |
733 | ||
a38965bf | 734 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page, |
d0e0ac97 | 735 | const char *fmt, ...) |
81819f0f CL |
736 | { |
737 | va_list args; | |
738 | char buf[100]; | |
739 | ||
24922684 CL |
740 | va_start(args, fmt); |
741 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 742 | va_end(args); |
3dc50637 | 743 | slab_bug(s, "%s", buf); |
24922684 | 744 | print_page_info(page); |
81819f0f CL |
745 | dump_stack(); |
746 | } | |
747 | ||
f7cb1933 | 748 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
749 | { |
750 | u8 *p = object; | |
751 | ||
d86bd1be JK |
752 | if (s->flags & SLAB_RED_ZONE) |
753 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
754 | ||
81819f0f | 755 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
756 | memset(p, POISON_FREE, s->object_size - 1); |
757 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
758 | } |
759 | ||
760 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 761 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
762 | } |
763 | ||
24922684 CL |
764 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
765 | void *from, void *to) | |
766 | { | |
767 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
768 | memset(from, data, to - from); | |
769 | } | |
770 | ||
771 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
772 | u8 *object, char *what, | |
06428780 | 773 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
774 | { |
775 | u8 *fault; | |
776 | u8 *end; | |
e1b70dd1 | 777 | u8 *addr = page_address(page); |
24922684 | 778 | |
a79316c6 | 779 | metadata_access_enable(); |
79824820 | 780 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 781 | metadata_access_disable(); |
24922684 CL |
782 | if (!fault) |
783 | return 1; | |
784 | ||
785 | end = start + bytes; | |
786 | while (end > fault && end[-1] == value) | |
787 | end--; | |
788 | ||
789 | slab_bug(s, "%s overwritten", what); | |
e1b70dd1 MC |
790 | pr_err("INFO: 0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n", |
791 | fault, end - 1, fault - addr, | |
792 | fault[0], value); | |
24922684 CL |
793 | print_trailer(s, page, object); |
794 | ||
795 | restore_bytes(s, what, value, fault, end); | |
796 | return 0; | |
81819f0f CL |
797 | } |
798 | ||
81819f0f CL |
799 | /* |
800 | * Object layout: | |
801 | * | |
802 | * object address | |
803 | * Bytes of the object to be managed. | |
804 | * If the freepointer may overlay the object then the free | |
cbfc35a4 | 805 | * pointer is at the middle of the object. |
672bba3a | 806 | * |
81819f0f CL |
807 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
808 | * 0xa5 (POISON_END) | |
809 | * | |
3b0efdfa | 810 | * object + s->object_size |
81819f0f | 811 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 812 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 813 | * object_size == inuse. |
672bba3a | 814 | * |
81819f0f CL |
815 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
816 | * 0xcc (RED_ACTIVE) for objects in use. | |
817 | * | |
818 | * object + s->inuse | |
672bba3a CL |
819 | * Meta data starts here. |
820 | * | |
81819f0f CL |
821 | * A. Free pointer (if we cannot overwrite object on free) |
822 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 823 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 824 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
825 | * before the word boundary. |
826 | * | |
827 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
828 | * |
829 | * object + s->size | |
672bba3a | 830 | * Nothing is used beyond s->size. |
81819f0f | 831 | * |
3b0efdfa | 832 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 833 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
834 | * may be used with merged slabcaches. |
835 | */ | |
836 | ||
81819f0f CL |
837 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
838 | { | |
cbfc35a4 | 839 | unsigned long off = get_info_end(s); /* The end of info */ |
81819f0f CL |
840 | |
841 | if (s->flags & SLAB_STORE_USER) | |
842 | /* We also have user information there */ | |
843 | off += 2 * sizeof(struct track); | |
844 | ||
80a9201a AP |
845 | off += kasan_metadata_size(s); |
846 | ||
d86bd1be | 847 | if (size_from_object(s) == off) |
81819f0f CL |
848 | return 1; |
849 | ||
24922684 | 850 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 851 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
852 | } |
853 | ||
39b26464 | 854 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
855 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
856 | { | |
24922684 CL |
857 | u8 *start; |
858 | u8 *fault; | |
859 | u8 *end; | |
5d682681 | 860 | u8 *pad; |
24922684 CL |
861 | int length; |
862 | int remainder; | |
81819f0f CL |
863 | |
864 | if (!(s->flags & SLAB_POISON)) | |
865 | return 1; | |
866 | ||
a973e9dd | 867 | start = page_address(page); |
a50b854e | 868 | length = page_size(page); |
39b26464 CL |
869 | end = start + length; |
870 | remainder = length % s->size; | |
81819f0f CL |
871 | if (!remainder) |
872 | return 1; | |
873 | ||
5d682681 | 874 | pad = end - remainder; |
a79316c6 | 875 | metadata_access_enable(); |
5d682681 | 876 | fault = memchr_inv(pad, POISON_INUSE, remainder); |
a79316c6 | 877 | metadata_access_disable(); |
24922684 CL |
878 | if (!fault) |
879 | return 1; | |
880 | while (end > fault && end[-1] == POISON_INUSE) | |
881 | end--; | |
882 | ||
e1b70dd1 MC |
883 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p @offset=%tu", |
884 | fault, end - 1, fault - start); | |
5d682681 | 885 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 886 | |
5d682681 | 887 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
24922684 | 888 | return 0; |
81819f0f CL |
889 | } |
890 | ||
891 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 892 | void *object, u8 val) |
81819f0f CL |
893 | { |
894 | u8 *p = object; | |
3b0efdfa | 895 | u8 *endobject = object + s->object_size; |
81819f0f CL |
896 | |
897 | if (s->flags & SLAB_RED_ZONE) { | |
d86bd1be JK |
898 | if (!check_bytes_and_report(s, page, object, "Redzone", |
899 | object - s->red_left_pad, val, s->red_left_pad)) | |
900 | return 0; | |
901 | ||
24922684 | 902 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 903 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 904 | return 0; |
81819f0f | 905 | } else { |
3b0efdfa | 906 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 907 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
908 | endobject, POISON_INUSE, |
909 | s->inuse - s->object_size); | |
3adbefee | 910 | } |
81819f0f CL |
911 | } |
912 | ||
913 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 914 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 915 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 916 | POISON_FREE, s->object_size - 1) || |
24922684 | 917 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 918 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 919 | return 0; |
81819f0f CL |
920 | /* |
921 | * check_pad_bytes cleans up on its own. | |
922 | */ | |
923 | check_pad_bytes(s, page, p); | |
924 | } | |
925 | ||
cbfc35a4 | 926 | if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE) |
81819f0f CL |
927 | /* |
928 | * Object and freepointer overlap. Cannot check | |
929 | * freepointer while object is allocated. | |
930 | */ | |
931 | return 1; | |
932 | ||
933 | /* Check free pointer validity */ | |
934 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
935 | object_err(s, page, p, "Freepointer corrupt"); | |
936 | /* | |
9f6c708e | 937 | * No choice but to zap it and thus lose the remainder |
81819f0f | 938 | * of the free objects in this slab. May cause |
672bba3a | 939 | * another error because the object count is now wrong. |
81819f0f | 940 | */ |
a973e9dd | 941 | set_freepointer(s, p, NULL); |
81819f0f CL |
942 | return 0; |
943 | } | |
944 | return 1; | |
945 | } | |
946 | ||
947 | static int check_slab(struct kmem_cache *s, struct page *page) | |
948 | { | |
39b26464 CL |
949 | int maxobj; |
950 | ||
81819f0f CL |
951 | VM_BUG_ON(!irqs_disabled()); |
952 | ||
953 | if (!PageSlab(page)) { | |
24922684 | 954 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
955 | return 0; |
956 | } | |
39b26464 | 957 | |
9736d2a9 | 958 | maxobj = order_objects(compound_order(page), s->size); |
39b26464 CL |
959 | if (page->objects > maxobj) { |
960 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 961 | page->objects, maxobj); |
39b26464 CL |
962 | return 0; |
963 | } | |
964 | if (page->inuse > page->objects) { | |
24922684 | 965 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 966 | page->inuse, page->objects); |
81819f0f CL |
967 | return 0; |
968 | } | |
969 | /* Slab_pad_check fixes things up after itself */ | |
970 | slab_pad_check(s, page); | |
971 | return 1; | |
972 | } | |
973 | ||
974 | /* | |
672bba3a CL |
975 | * Determine if a certain object on a page is on the freelist. Must hold the |
976 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
977 | */ |
978 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
979 | { | |
980 | int nr = 0; | |
881db7fb | 981 | void *fp; |
81819f0f | 982 | void *object = NULL; |
f6edde9c | 983 | int max_objects; |
81819f0f | 984 | |
881db7fb | 985 | fp = page->freelist; |
39b26464 | 986 | while (fp && nr <= page->objects) { |
81819f0f CL |
987 | if (fp == search) |
988 | return 1; | |
989 | if (!check_valid_pointer(s, page, fp)) { | |
990 | if (object) { | |
991 | object_err(s, page, object, | |
992 | "Freechain corrupt"); | |
a973e9dd | 993 | set_freepointer(s, object, NULL); |
81819f0f | 994 | } else { |
24922684 | 995 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 996 | page->freelist = NULL; |
39b26464 | 997 | page->inuse = page->objects; |
24922684 | 998 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
999 | return 0; |
1000 | } | |
1001 | break; | |
1002 | } | |
1003 | object = fp; | |
1004 | fp = get_freepointer(s, object); | |
1005 | nr++; | |
1006 | } | |
1007 | ||
9736d2a9 | 1008 | max_objects = order_objects(compound_order(page), s->size); |
210b5c06 CG |
1009 | if (max_objects > MAX_OBJS_PER_PAGE) |
1010 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
1011 | |
1012 | if (page->objects != max_objects) { | |
756a025f JP |
1013 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
1014 | page->objects, max_objects); | |
224a88be CL |
1015 | page->objects = max_objects; |
1016 | slab_fix(s, "Number of objects adjusted."); | |
1017 | } | |
39b26464 | 1018 | if (page->inuse != page->objects - nr) { |
756a025f JP |
1019 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
1020 | page->inuse, page->objects - nr); | |
39b26464 | 1021 | page->inuse = page->objects - nr; |
24922684 | 1022 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
1023 | } |
1024 | return search == NULL; | |
1025 | } | |
1026 | ||
0121c619 CL |
1027 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
1028 | int alloc) | |
3ec09742 CL |
1029 | { |
1030 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 1031 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
1032 | s->name, |
1033 | alloc ? "alloc" : "free", | |
1034 | object, page->inuse, | |
1035 | page->freelist); | |
1036 | ||
1037 | if (!alloc) | |
aa2efd5e | 1038 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1039 | s->object_size); |
3ec09742 CL |
1040 | |
1041 | dump_stack(); | |
1042 | } | |
1043 | } | |
1044 | ||
643b1138 | 1045 | /* |
672bba3a | 1046 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1047 | */ |
5cc6eee8 CL |
1048 | static void add_full(struct kmem_cache *s, |
1049 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1050 | { |
5cc6eee8 CL |
1051 | if (!(s->flags & SLAB_STORE_USER)) |
1052 | return; | |
1053 | ||
255d0884 | 1054 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1055 | list_add(&page->slab_list, &n->full); |
643b1138 CL |
1056 | } |
1057 | ||
c65c1877 | 1058 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1059 | { |
643b1138 CL |
1060 | if (!(s->flags & SLAB_STORE_USER)) |
1061 | return; | |
1062 | ||
255d0884 | 1063 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1064 | list_del(&page->slab_list); |
643b1138 CL |
1065 | } |
1066 | ||
0f389ec6 CL |
1067 | /* Tracking of the number of slabs for debugging purposes */ |
1068 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1069 | { | |
1070 | struct kmem_cache_node *n = get_node(s, node); | |
1071 | ||
1072 | return atomic_long_read(&n->nr_slabs); | |
1073 | } | |
1074 | ||
26c02cf0 AB |
1075 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1076 | { | |
1077 | return atomic_long_read(&n->nr_slabs); | |
1078 | } | |
1079 | ||
205ab99d | 1080 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1081 | { |
1082 | struct kmem_cache_node *n = get_node(s, node); | |
1083 | ||
1084 | /* | |
1085 | * May be called early in order to allocate a slab for the | |
1086 | * kmem_cache_node structure. Solve the chicken-egg | |
1087 | * dilemma by deferring the increment of the count during | |
1088 | * bootstrap (see early_kmem_cache_node_alloc). | |
1089 | */ | |
338b2642 | 1090 | if (likely(n)) { |
0f389ec6 | 1091 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1092 | atomic_long_add(objects, &n->total_objects); |
1093 | } | |
0f389ec6 | 1094 | } |
205ab99d | 1095 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1096 | { |
1097 | struct kmem_cache_node *n = get_node(s, node); | |
1098 | ||
1099 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1100 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1101 | } |
1102 | ||
1103 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1104 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1105 | void *object) | |
1106 | { | |
8fc8d666 | 1107 | if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)) |
3ec09742 CL |
1108 | return; |
1109 | ||
f7cb1933 | 1110 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1111 | init_tracking(s, object); |
1112 | } | |
1113 | ||
a50b854e MWO |
1114 | static |
1115 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) | |
a7101224 | 1116 | { |
8fc8d666 | 1117 | if (!kmem_cache_debug_flags(s, SLAB_POISON)) |
a7101224 AK |
1118 | return; |
1119 | ||
1120 | metadata_access_enable(); | |
a50b854e | 1121 | memset(addr, POISON_INUSE, page_size(page)); |
a7101224 AK |
1122 | metadata_access_disable(); |
1123 | } | |
1124 | ||
becfda68 | 1125 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
278d7756 | 1126 | struct page *page, void *object) |
81819f0f CL |
1127 | { |
1128 | if (!check_slab(s, page)) | |
becfda68 | 1129 | return 0; |
81819f0f | 1130 | |
81819f0f CL |
1131 | if (!check_valid_pointer(s, page, object)) { |
1132 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1133 | return 0; |
81819f0f CL |
1134 | } |
1135 | ||
f7cb1933 | 1136 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1137 | return 0; |
1138 | ||
1139 | return 1; | |
1140 | } | |
1141 | ||
1142 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1143 | struct page *page, | |
1144 | void *object, unsigned long addr) | |
1145 | { | |
1146 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
278d7756 | 1147 | if (!alloc_consistency_checks(s, page, object)) |
becfda68 LA |
1148 | goto bad; |
1149 | } | |
81819f0f | 1150 | |
3ec09742 CL |
1151 | /* Success perform special debug activities for allocs */ |
1152 | if (s->flags & SLAB_STORE_USER) | |
1153 | set_track(s, object, TRACK_ALLOC, addr); | |
1154 | trace(s, page, object, 1); | |
f7cb1933 | 1155 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1156 | return 1; |
3ec09742 | 1157 | |
81819f0f CL |
1158 | bad: |
1159 | if (PageSlab(page)) { | |
1160 | /* | |
1161 | * If this is a slab page then lets do the best we can | |
1162 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1163 | * as used avoids touching the remaining objects. |
81819f0f | 1164 | */ |
24922684 | 1165 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1166 | page->inuse = page->objects; |
a973e9dd | 1167 | page->freelist = NULL; |
81819f0f CL |
1168 | } |
1169 | return 0; | |
1170 | } | |
1171 | ||
becfda68 LA |
1172 | static inline int free_consistency_checks(struct kmem_cache *s, |
1173 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1174 | { |
81819f0f | 1175 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1176 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1177 | return 0; |
81819f0f CL |
1178 | } |
1179 | ||
1180 | if (on_freelist(s, page, object)) { | |
24922684 | 1181 | object_err(s, page, object, "Object already free"); |
becfda68 | 1182 | return 0; |
81819f0f CL |
1183 | } |
1184 | ||
f7cb1933 | 1185 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1186 | return 0; |
81819f0f | 1187 | |
1b4f59e3 | 1188 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1189 | if (!PageSlab(page)) { |
756a025f JP |
1190 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1191 | object); | |
1b4f59e3 | 1192 | } else if (!page->slab_cache) { |
f9f58285 FF |
1193 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1194 | object); | |
70d71228 | 1195 | dump_stack(); |
06428780 | 1196 | } else |
24922684 CL |
1197 | object_err(s, page, object, |
1198 | "page slab pointer corrupt."); | |
becfda68 LA |
1199 | return 0; |
1200 | } | |
1201 | return 1; | |
1202 | } | |
1203 | ||
1204 | /* Supports checking bulk free of a constructed freelist */ | |
1205 | static noinline int free_debug_processing( | |
1206 | struct kmem_cache *s, struct page *page, | |
1207 | void *head, void *tail, int bulk_cnt, | |
1208 | unsigned long addr) | |
1209 | { | |
1210 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1211 | void *object = head; | |
1212 | int cnt = 0; | |
3f649ab7 | 1213 | unsigned long flags; |
becfda68 LA |
1214 | int ret = 0; |
1215 | ||
1216 | spin_lock_irqsave(&n->list_lock, flags); | |
1217 | slab_lock(page); | |
1218 | ||
1219 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1220 | if (!check_slab(s, page)) | |
1221 | goto out; | |
1222 | } | |
1223 | ||
1224 | next_object: | |
1225 | cnt++; | |
1226 | ||
1227 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1228 | if (!free_consistency_checks(s, page, object, addr)) | |
1229 | goto out; | |
81819f0f | 1230 | } |
3ec09742 | 1231 | |
3ec09742 CL |
1232 | if (s->flags & SLAB_STORE_USER) |
1233 | set_track(s, object, TRACK_FREE, addr); | |
1234 | trace(s, page, object, 0); | |
81084651 | 1235 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1236 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1237 | |
1238 | /* Reached end of constructed freelist yet? */ | |
1239 | if (object != tail) { | |
1240 | object = get_freepointer(s, object); | |
1241 | goto next_object; | |
1242 | } | |
804aa132 LA |
1243 | ret = 1; |
1244 | ||
5c2e4bbb | 1245 | out: |
81084651 JDB |
1246 | if (cnt != bulk_cnt) |
1247 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1248 | bulk_cnt, cnt); | |
1249 | ||
881db7fb | 1250 | slab_unlock(page); |
282acb43 | 1251 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1252 | if (!ret) |
1253 | slab_fix(s, "Object at 0x%p not freed", object); | |
1254 | return ret; | |
81819f0f CL |
1255 | } |
1256 | ||
e17f1dfb VB |
1257 | /* |
1258 | * Parse a block of slub_debug options. Blocks are delimited by ';' | |
1259 | * | |
1260 | * @str: start of block | |
1261 | * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified | |
1262 | * @slabs: return start of list of slabs, or NULL when there's no list | |
1263 | * @init: assume this is initial parsing and not per-kmem-create parsing | |
1264 | * | |
1265 | * returns the start of next block if there's any, or NULL | |
1266 | */ | |
1267 | static char * | |
1268 | parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init) | |
41ecc55b | 1269 | { |
e17f1dfb | 1270 | bool higher_order_disable = false; |
f0630fff | 1271 | |
e17f1dfb VB |
1272 | /* Skip any completely empty blocks */ |
1273 | while (*str && *str == ';') | |
1274 | str++; | |
1275 | ||
1276 | if (*str == ',') { | |
f0630fff CL |
1277 | /* |
1278 | * No options but restriction on slabs. This means full | |
1279 | * debugging for slabs matching a pattern. | |
1280 | */ | |
e17f1dfb | 1281 | *flags = DEBUG_DEFAULT_FLAGS; |
f0630fff | 1282 | goto check_slabs; |
e17f1dfb VB |
1283 | } |
1284 | *flags = 0; | |
f0630fff | 1285 | |
e17f1dfb VB |
1286 | /* Determine which debug features should be switched on */ |
1287 | for (; *str && *str != ',' && *str != ';'; str++) { | |
f0630fff | 1288 | switch (tolower(*str)) { |
e17f1dfb VB |
1289 | case '-': |
1290 | *flags = 0; | |
1291 | break; | |
f0630fff | 1292 | case 'f': |
e17f1dfb | 1293 | *flags |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1294 | break; |
1295 | case 'z': | |
e17f1dfb | 1296 | *flags |= SLAB_RED_ZONE; |
f0630fff CL |
1297 | break; |
1298 | case 'p': | |
e17f1dfb | 1299 | *flags |= SLAB_POISON; |
f0630fff CL |
1300 | break; |
1301 | case 'u': | |
e17f1dfb | 1302 | *flags |= SLAB_STORE_USER; |
f0630fff CL |
1303 | break; |
1304 | case 't': | |
e17f1dfb | 1305 | *flags |= SLAB_TRACE; |
f0630fff | 1306 | break; |
4c13dd3b | 1307 | case 'a': |
e17f1dfb | 1308 | *flags |= SLAB_FAILSLAB; |
4c13dd3b | 1309 | break; |
08303a73 CA |
1310 | case 'o': |
1311 | /* | |
1312 | * Avoid enabling debugging on caches if its minimum | |
1313 | * order would increase as a result. | |
1314 | */ | |
e17f1dfb | 1315 | higher_order_disable = true; |
08303a73 | 1316 | break; |
f0630fff | 1317 | default: |
e17f1dfb VB |
1318 | if (init) |
1319 | pr_err("slub_debug option '%c' unknown. skipped\n", *str); | |
f0630fff | 1320 | } |
41ecc55b | 1321 | } |
f0630fff | 1322 | check_slabs: |
41ecc55b | 1323 | if (*str == ',') |
e17f1dfb VB |
1324 | *slabs = ++str; |
1325 | else | |
1326 | *slabs = NULL; | |
1327 | ||
1328 | /* Skip over the slab list */ | |
1329 | while (*str && *str != ';') | |
1330 | str++; | |
1331 | ||
1332 | /* Skip any completely empty blocks */ | |
1333 | while (*str && *str == ';') | |
1334 | str++; | |
1335 | ||
1336 | if (init && higher_order_disable) | |
1337 | disable_higher_order_debug = 1; | |
1338 | ||
1339 | if (*str) | |
1340 | return str; | |
1341 | else | |
1342 | return NULL; | |
1343 | } | |
1344 | ||
1345 | static int __init setup_slub_debug(char *str) | |
1346 | { | |
1347 | slab_flags_t flags; | |
1348 | char *saved_str; | |
1349 | char *slab_list; | |
1350 | bool global_slub_debug_changed = false; | |
1351 | bool slab_list_specified = false; | |
1352 | ||
1353 | slub_debug = DEBUG_DEFAULT_FLAGS; | |
1354 | if (*str++ != '=' || !*str) | |
1355 | /* | |
1356 | * No options specified. Switch on full debugging. | |
1357 | */ | |
1358 | goto out; | |
1359 | ||
1360 | saved_str = str; | |
1361 | while (str) { | |
1362 | str = parse_slub_debug_flags(str, &flags, &slab_list, true); | |
1363 | ||
1364 | if (!slab_list) { | |
1365 | slub_debug = flags; | |
1366 | global_slub_debug_changed = true; | |
1367 | } else { | |
1368 | slab_list_specified = true; | |
1369 | } | |
1370 | } | |
1371 | ||
1372 | /* | |
1373 | * For backwards compatibility, a single list of flags with list of | |
1374 | * slabs means debugging is only enabled for those slabs, so the global | |
1375 | * slub_debug should be 0. We can extended that to multiple lists as | |
1376 | * long as there is no option specifying flags without a slab list. | |
1377 | */ | |
1378 | if (slab_list_specified) { | |
1379 | if (!global_slub_debug_changed) | |
1380 | slub_debug = 0; | |
1381 | slub_debug_string = saved_str; | |
1382 | } | |
f0630fff | 1383 | out: |
ca0cab65 VB |
1384 | if (slub_debug != 0 || slub_debug_string) |
1385 | static_branch_enable(&slub_debug_enabled); | |
6471384a AP |
1386 | if ((static_branch_unlikely(&init_on_alloc) || |
1387 | static_branch_unlikely(&init_on_free)) && | |
1388 | (slub_debug & SLAB_POISON)) | |
1389 | pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n"); | |
41ecc55b CL |
1390 | return 1; |
1391 | } | |
1392 | ||
1393 | __setup("slub_debug", setup_slub_debug); | |
1394 | ||
c5fd3ca0 AT |
1395 | /* |
1396 | * kmem_cache_flags - apply debugging options to the cache | |
1397 | * @object_size: the size of an object without meta data | |
1398 | * @flags: flags to set | |
1399 | * @name: name of the cache | |
1400 | * @ctor: constructor function | |
1401 | * | |
1402 | * Debug option(s) are applied to @flags. In addition to the debug | |
1403 | * option(s), if a slab name (or multiple) is specified i.e. | |
1404 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1405 | * then only the select slabs will receive the debug option(s). | |
1406 | */ | |
0293d1fd | 1407 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1408 | slab_flags_t flags, const char *name, |
51cc5068 | 1409 | void (*ctor)(void *)) |
41ecc55b | 1410 | { |
c5fd3ca0 AT |
1411 | char *iter; |
1412 | size_t len; | |
e17f1dfb VB |
1413 | char *next_block; |
1414 | slab_flags_t block_flags; | |
c5fd3ca0 | 1415 | |
c5fd3ca0 | 1416 | len = strlen(name); |
e17f1dfb VB |
1417 | next_block = slub_debug_string; |
1418 | /* Go through all blocks of debug options, see if any matches our slab's name */ | |
1419 | while (next_block) { | |
1420 | next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false); | |
1421 | if (!iter) | |
1422 | continue; | |
1423 | /* Found a block that has a slab list, search it */ | |
1424 | while (*iter) { | |
1425 | char *end, *glob; | |
1426 | size_t cmplen; | |
1427 | ||
1428 | end = strchrnul(iter, ','); | |
1429 | if (next_block && next_block < end) | |
1430 | end = next_block - 1; | |
1431 | ||
1432 | glob = strnchr(iter, end - iter, '*'); | |
1433 | if (glob) | |
1434 | cmplen = glob - iter; | |
1435 | else | |
1436 | cmplen = max_t(size_t, len, (end - iter)); | |
c5fd3ca0 | 1437 | |
e17f1dfb VB |
1438 | if (!strncmp(name, iter, cmplen)) { |
1439 | flags |= block_flags; | |
1440 | return flags; | |
1441 | } | |
c5fd3ca0 | 1442 | |
e17f1dfb VB |
1443 | if (!*end || *end == ';') |
1444 | break; | |
1445 | iter = end + 1; | |
c5fd3ca0 | 1446 | } |
c5fd3ca0 | 1447 | } |
ba0268a8 | 1448 | |
484cfaca | 1449 | return flags | slub_debug; |
41ecc55b | 1450 | } |
b4a64718 | 1451 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1452 | static inline void setup_object_debug(struct kmem_cache *s, |
1453 | struct page *page, void *object) {} | |
a50b854e MWO |
1454 | static inline |
1455 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {} | |
41ecc55b | 1456 | |
3ec09742 | 1457 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1458 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1459 | |
282acb43 | 1460 | static inline int free_debug_processing( |
81084651 JDB |
1461 | struct kmem_cache *s, struct page *page, |
1462 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1463 | unsigned long addr) { return 0; } |
41ecc55b | 1464 | |
41ecc55b CL |
1465 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1466 | { return 1; } | |
1467 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1468 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1469 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1470 | struct page *page) {} | |
c65c1877 PZ |
1471 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1472 | struct page *page) {} | |
0293d1fd | 1473 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1474 | slab_flags_t flags, const char *name, |
51cc5068 | 1475 | void (*ctor)(void *)) |
ba0268a8 CL |
1476 | { |
1477 | return flags; | |
1478 | } | |
41ecc55b | 1479 | #define slub_debug 0 |
0f389ec6 | 1480 | |
fdaa45e9 IM |
1481 | #define disable_higher_order_debug 0 |
1482 | ||
0f389ec6 CL |
1483 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1484 | { return 0; } | |
26c02cf0 AB |
1485 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1486 | { return 0; } | |
205ab99d CL |
1487 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1488 | int objects) {} | |
1489 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1490 | int objects) {} | |
7d550c56 | 1491 | |
52f23478 | 1492 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 1493 | void **freelist, void *nextfree) |
52f23478 DZ |
1494 | { |
1495 | return false; | |
1496 | } | |
02e72cc6 AR |
1497 | #endif /* CONFIG_SLUB_DEBUG */ |
1498 | ||
1499 | /* | |
1500 | * Hooks for other subsystems that check memory allocations. In a typical | |
1501 | * production configuration these hooks all should produce no code at all. | |
1502 | */ | |
0116523c | 1503 | static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
d56791b3 | 1504 | { |
53128245 | 1505 | ptr = kasan_kmalloc_large(ptr, size, flags); |
a2f77575 | 1506 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ |
d56791b3 | 1507 | kmemleak_alloc(ptr, size, 1, flags); |
53128245 | 1508 | return ptr; |
d56791b3 RB |
1509 | } |
1510 | ||
ee3ce779 | 1511 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1512 | { |
1513 | kmemleak_free(x); | |
ee3ce779 | 1514 | kasan_kfree_large(x, _RET_IP_); |
d56791b3 RB |
1515 | } |
1516 | ||
c3895391 | 1517 | static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x) |
d56791b3 RB |
1518 | { |
1519 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1520 | |
02e72cc6 AR |
1521 | /* |
1522 | * Trouble is that we may no longer disable interrupts in the fast path | |
1523 | * So in order to make the debug calls that expect irqs to be | |
1524 | * disabled we need to disable interrupts temporarily. | |
1525 | */ | |
4675ff05 | 1526 | #ifdef CONFIG_LOCKDEP |
02e72cc6 AR |
1527 | { |
1528 | unsigned long flags; | |
1529 | ||
1530 | local_irq_save(flags); | |
02e72cc6 AR |
1531 | debug_check_no_locks_freed(x, s->object_size); |
1532 | local_irq_restore(flags); | |
1533 | } | |
1534 | #endif | |
1535 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1536 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1537 | |
cfbe1636 ME |
1538 | /* Use KCSAN to help debug racy use-after-free. */ |
1539 | if (!(s->flags & SLAB_TYPESAFE_BY_RCU)) | |
1540 | __kcsan_check_access(x, s->object_size, | |
1541 | KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT); | |
1542 | ||
c3895391 AK |
1543 | /* KASAN might put x into memory quarantine, delaying its reuse */ |
1544 | return kasan_slab_free(s, x, _RET_IP_); | |
02e72cc6 | 1545 | } |
205ab99d | 1546 | |
c3895391 AK |
1547 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
1548 | void **head, void **tail) | |
81084651 | 1549 | { |
6471384a AP |
1550 | |
1551 | void *object; | |
1552 | void *next = *head; | |
1553 | void *old_tail = *tail ? *tail : *head; | |
1554 | int rsize; | |
1555 | ||
aea4df4c LA |
1556 | /* Head and tail of the reconstructed freelist */ |
1557 | *head = NULL; | |
1558 | *tail = NULL; | |
1b7e816f | 1559 | |
aea4df4c LA |
1560 | do { |
1561 | object = next; | |
1562 | next = get_freepointer(s, object); | |
1563 | ||
1564 | if (slab_want_init_on_free(s)) { | |
6471384a AP |
1565 | /* |
1566 | * Clear the object and the metadata, but don't touch | |
1567 | * the redzone. | |
1568 | */ | |
1569 | memset(object, 0, s->object_size); | |
1570 | rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad | |
1571 | : 0; | |
1572 | memset((char *)object + s->inuse, 0, | |
1573 | s->size - s->inuse - rsize); | |
81084651 | 1574 | |
aea4df4c | 1575 | } |
c3895391 AK |
1576 | /* If object's reuse doesn't have to be delayed */ |
1577 | if (!slab_free_hook(s, object)) { | |
1578 | /* Move object to the new freelist */ | |
1579 | set_freepointer(s, object, *head); | |
1580 | *head = object; | |
1581 | if (!*tail) | |
1582 | *tail = object; | |
1583 | } | |
1584 | } while (object != old_tail); | |
1585 | ||
1586 | if (*head == *tail) | |
1587 | *tail = NULL; | |
1588 | ||
1589 | return *head != NULL; | |
81084651 JDB |
1590 | } |
1591 | ||
4d176711 | 1592 | static void *setup_object(struct kmem_cache *s, struct page *page, |
588f8ba9 TG |
1593 | void *object) |
1594 | { | |
1595 | setup_object_debug(s, page, object); | |
4d176711 | 1596 | object = kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1597 | if (unlikely(s->ctor)) { |
1598 | kasan_unpoison_object_data(s, object); | |
1599 | s->ctor(object); | |
1600 | kasan_poison_object_data(s, object); | |
1601 | } | |
4d176711 | 1602 | return object; |
588f8ba9 TG |
1603 | } |
1604 | ||
81819f0f CL |
1605 | /* |
1606 | * Slab allocation and freeing | |
1607 | */ | |
5dfb4175 VD |
1608 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1609 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1610 | { |
5dfb4175 | 1611 | struct page *page; |
19af27af | 1612 | unsigned int order = oo_order(oo); |
65c3376a | 1613 | |
2154a336 | 1614 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1615 | page = alloc_pages(flags, order); |
65c3376a | 1616 | else |
96db800f | 1617 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1618 | |
10befea9 | 1619 | if (page) |
74d555be | 1620 | account_slab_page(page, order, s); |
5dfb4175 VD |
1621 | |
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 | |
1b4f59e3 | 1774 | page->slab_cache = s; |
c03f94cc | 1775 | __SetPageSlab(page); |
2f064f34 | 1776 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1777 | SetPageSlabPfmemalloc(page); |
81819f0f | 1778 | |
a7101224 | 1779 | kasan_poison_slab(page); |
81819f0f | 1780 | |
a7101224 | 1781 | start = page_address(page); |
81819f0f | 1782 | |
a50b854e | 1783 | setup_page_debug(s, page, start); |
0316bec2 | 1784 | |
210e7a43 TG |
1785 | shuffle = shuffle_freelist(s, page); |
1786 | ||
1787 | if (!shuffle) { | |
4d176711 AK |
1788 | start = fixup_red_left(s, start); |
1789 | start = setup_object(s, page, start); | |
1790 | page->freelist = start; | |
18e50661 AK |
1791 | for (idx = 0, p = start; idx < page->objects - 1; idx++) { |
1792 | next = p + s->size; | |
1793 | next = setup_object(s, page, next); | |
1794 | set_freepointer(s, p, next); | |
1795 | p = next; | |
1796 | } | |
1797 | set_freepointer(s, p, NULL); | |
81819f0f | 1798 | } |
81819f0f | 1799 | |
e6e82ea1 | 1800 | page->inuse = page->objects; |
8cb0a506 | 1801 | page->frozen = 1; |
588f8ba9 | 1802 | |
81819f0f | 1803 | out: |
d0164adc | 1804 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1805 | local_irq_disable(); |
1806 | if (!page) | |
1807 | return NULL; | |
1808 | ||
588f8ba9 TG |
1809 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1810 | ||
81819f0f CL |
1811 | return page; |
1812 | } | |
1813 | ||
588f8ba9 TG |
1814 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1815 | { | |
44405099 LL |
1816 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
1817 | flags = kmalloc_fix_flags(flags); | |
588f8ba9 TG |
1818 | |
1819 | return allocate_slab(s, | |
1820 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1821 | } | |
1822 | ||
81819f0f CL |
1823 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1824 | { | |
834f3d11 CL |
1825 | int order = compound_order(page); |
1826 | int pages = 1 << order; | |
81819f0f | 1827 | |
8fc8d666 | 1828 | if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) { |
81819f0f CL |
1829 | void *p; |
1830 | ||
1831 | slab_pad_check(s, page); | |
224a88be CL |
1832 | for_each_object(p, s, page_address(page), |
1833 | page->objects) | |
f7cb1933 | 1834 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1835 | } |
1836 | ||
072bb0aa | 1837 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1838 | __ClearPageSlab(page); |
0c06dd75 VB |
1839 | /* In union with page->mapping where page allocator expects NULL */ |
1840 | page->slab_cache = NULL; | |
1eb5ac64 NP |
1841 | if (current->reclaim_state) |
1842 | current->reclaim_state->reclaimed_slab += pages; | |
74d555be | 1843 | unaccount_slab_page(page, order, s); |
27ee57c9 | 1844 | __free_pages(page, order); |
81819f0f CL |
1845 | } |
1846 | ||
1847 | static void rcu_free_slab(struct rcu_head *h) | |
1848 | { | |
bf68c214 | 1849 | struct page *page = container_of(h, struct page, rcu_head); |
da9a638c | 1850 | |
1b4f59e3 | 1851 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1852 | } |
1853 | ||
1854 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1855 | { | |
5f0d5a3a | 1856 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bf68c214 | 1857 | call_rcu(&page->rcu_head, rcu_free_slab); |
81819f0f CL |
1858 | } else |
1859 | __free_slab(s, page); | |
1860 | } | |
1861 | ||
1862 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1863 | { | |
205ab99d | 1864 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1865 | free_slab(s, page); |
1866 | } | |
1867 | ||
1868 | /* | |
5cc6eee8 | 1869 | * Management of partially allocated slabs. |
81819f0f | 1870 | */ |
1e4dd946 SR |
1871 | static inline void |
1872 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1873 | { |
e95eed57 | 1874 | n->nr_partial++; |
136333d1 | 1875 | if (tail == DEACTIVATE_TO_TAIL) |
916ac052 | 1876 | list_add_tail(&page->slab_list, &n->partial); |
7c2e132c | 1877 | else |
916ac052 | 1878 | list_add(&page->slab_list, &n->partial); |
81819f0f CL |
1879 | } |
1880 | ||
1e4dd946 SR |
1881 | static inline void add_partial(struct kmem_cache_node *n, |
1882 | struct page *page, int tail) | |
62e346a8 | 1883 | { |
c65c1877 | 1884 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1885 | __add_partial(n, page, tail); |
1886 | } | |
c65c1877 | 1887 | |
1e4dd946 SR |
1888 | static inline void remove_partial(struct kmem_cache_node *n, |
1889 | struct page *page) | |
1890 | { | |
1891 | lockdep_assert_held(&n->list_lock); | |
916ac052 | 1892 | list_del(&page->slab_list); |
52b4b950 | 1893 | n->nr_partial--; |
1e4dd946 SR |
1894 | } |
1895 | ||
81819f0f | 1896 | /* |
7ced3719 CL |
1897 | * Remove slab from the partial list, freeze it and |
1898 | * return the pointer to the freelist. | |
81819f0f | 1899 | * |
497b66f2 | 1900 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1901 | */ |
497b66f2 | 1902 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1903 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1904 | int mode, int *objects) |
81819f0f | 1905 | { |
2cfb7455 CL |
1906 | void *freelist; |
1907 | unsigned long counters; | |
1908 | struct page new; | |
1909 | ||
c65c1877 PZ |
1910 | lockdep_assert_held(&n->list_lock); |
1911 | ||
2cfb7455 CL |
1912 | /* |
1913 | * Zap the freelist and set the frozen bit. | |
1914 | * The old freelist is the list of objects for the | |
1915 | * per cpu allocation list. | |
1916 | */ | |
7ced3719 CL |
1917 | freelist = page->freelist; |
1918 | counters = page->counters; | |
1919 | new.counters = counters; | |
633b0764 | 1920 | *objects = new.objects - new.inuse; |
23910c50 | 1921 | if (mode) { |
7ced3719 | 1922 | new.inuse = page->objects; |
23910c50 PE |
1923 | new.freelist = NULL; |
1924 | } else { | |
1925 | new.freelist = freelist; | |
1926 | } | |
2cfb7455 | 1927 | |
a0132ac0 | 1928 | VM_BUG_ON(new.frozen); |
7ced3719 | 1929 | new.frozen = 1; |
2cfb7455 | 1930 | |
7ced3719 | 1931 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1932 | freelist, counters, |
02d7633f | 1933 | new.freelist, new.counters, |
7ced3719 | 1934 | "acquire_slab")) |
7ced3719 | 1935 | return NULL; |
2cfb7455 CL |
1936 | |
1937 | remove_partial(n, page); | |
7ced3719 | 1938 | WARN_ON(!freelist); |
49e22585 | 1939 | return freelist; |
81819f0f CL |
1940 | } |
1941 | ||
633b0764 | 1942 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1943 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1944 | |
81819f0f | 1945 | /* |
672bba3a | 1946 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1947 | */ |
8ba00bb6 JK |
1948 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1949 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1950 | { |
49e22585 CL |
1951 | struct page *page, *page2; |
1952 | void *object = NULL; | |
e5d9998f | 1953 | unsigned int available = 0; |
633b0764 | 1954 | int objects; |
81819f0f CL |
1955 | |
1956 | /* | |
1957 | * Racy check. If we mistakenly see no partial slabs then we | |
1958 | * just allocate an empty slab. If we mistakenly try to get a | |
70b6d25e | 1959 | * partial slab and there is none available then get_partial() |
672bba3a | 1960 | * will return NULL. |
81819f0f CL |
1961 | */ |
1962 | if (!n || !n->nr_partial) | |
1963 | return NULL; | |
1964 | ||
1965 | spin_lock(&n->list_lock); | |
916ac052 | 1966 | list_for_each_entry_safe(page, page2, &n->partial, slab_list) { |
8ba00bb6 | 1967 | void *t; |
49e22585 | 1968 | |
8ba00bb6 JK |
1969 | if (!pfmemalloc_match(page, flags)) |
1970 | continue; | |
1971 | ||
633b0764 | 1972 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1973 | if (!t) |
1974 | break; | |
1975 | ||
633b0764 | 1976 | available += objects; |
12d79634 | 1977 | if (!object) { |
49e22585 | 1978 | c->page = page; |
49e22585 | 1979 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1980 | object = t; |
49e22585 | 1981 | } else { |
633b0764 | 1982 | put_cpu_partial(s, page, 0); |
8028dcea | 1983 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1984 | } |
345c905d | 1985 | if (!kmem_cache_has_cpu_partial(s) |
e6d0e1dc | 1986 | || available > slub_cpu_partial(s) / 2) |
49e22585 CL |
1987 | break; |
1988 | ||
497b66f2 | 1989 | } |
81819f0f | 1990 | spin_unlock(&n->list_lock); |
497b66f2 | 1991 | return object; |
81819f0f CL |
1992 | } |
1993 | ||
1994 | /* | |
672bba3a | 1995 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1996 | */ |
de3ec035 | 1997 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1998 | struct kmem_cache_cpu *c) |
81819f0f CL |
1999 | { |
2000 | #ifdef CONFIG_NUMA | |
2001 | struct zonelist *zonelist; | |
dd1a239f | 2002 | struct zoneref *z; |
54a6eb5c | 2003 | struct zone *zone; |
97a225e6 | 2004 | enum zone_type highest_zoneidx = gfp_zone(flags); |
497b66f2 | 2005 | void *object; |
cc9a6c87 | 2006 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
2007 | |
2008 | /* | |
672bba3a CL |
2009 | * The defrag ratio allows a configuration of the tradeoffs between |
2010 | * inter node defragmentation and node local allocations. A lower | |
2011 | * defrag_ratio increases the tendency to do local allocations | |
2012 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 2013 | * |
672bba3a CL |
2014 | * If the defrag_ratio is set to 0 then kmalloc() always |
2015 | * returns node local objects. If the ratio is higher then kmalloc() | |
2016 | * may return off node objects because partial slabs are obtained | |
2017 | * from other nodes and filled up. | |
81819f0f | 2018 | * |
43efd3ea LP |
2019 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
2020 | * (which makes defrag_ratio = 1000) then every (well almost) | |
2021 | * allocation will first attempt to defrag slab caches on other nodes. | |
2022 | * This means scanning over all nodes to look for partial slabs which | |
2023 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 2024 | * with available objects. |
81819f0f | 2025 | */ |
9824601e CL |
2026 | if (!s->remote_node_defrag_ratio || |
2027 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
2028 | return NULL; |
2029 | ||
cc9a6c87 | 2030 | do { |
d26914d1 | 2031 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 2032 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
97a225e6 | 2033 | for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) { |
cc9a6c87 MG |
2034 | struct kmem_cache_node *n; |
2035 | ||
2036 | n = get_node(s, zone_to_nid(zone)); | |
2037 | ||
dee2f8aa | 2038 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 2039 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 2040 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
2041 | if (object) { |
2042 | /* | |
d26914d1 MG |
2043 | * Don't check read_mems_allowed_retry() |
2044 | * here - if mems_allowed was updated in | |
2045 | * parallel, that was a harmless race | |
2046 | * between allocation and the cpuset | |
2047 | * update | |
cc9a6c87 | 2048 | */ |
cc9a6c87 MG |
2049 | return object; |
2050 | } | |
c0ff7453 | 2051 | } |
81819f0f | 2052 | } |
d26914d1 | 2053 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
6dfd1b65 | 2054 | #endif /* CONFIG_NUMA */ |
81819f0f CL |
2055 | return NULL; |
2056 | } | |
2057 | ||
2058 | /* | |
2059 | * Get a partial page, lock it and return it. | |
2060 | */ | |
497b66f2 | 2061 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 2062 | struct kmem_cache_cpu *c) |
81819f0f | 2063 | { |
497b66f2 | 2064 | void *object; |
a561ce00 JK |
2065 | int searchnode = node; |
2066 | ||
2067 | if (node == NUMA_NO_NODE) | |
2068 | searchnode = numa_mem_id(); | |
81819f0f | 2069 | |
8ba00bb6 | 2070 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
2071 | if (object || node != NUMA_NO_NODE) |
2072 | return object; | |
81819f0f | 2073 | |
acd19fd1 | 2074 | return get_any_partial(s, flags, c); |
81819f0f CL |
2075 | } |
2076 | ||
923717cb | 2077 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2078 | /* |
0d645ed1 | 2079 | * Calculate the next globally unique transaction for disambiguation |
8a5ec0ba CL |
2080 | * during cmpxchg. The transactions start with the cpu number and are then |
2081 | * incremented by CONFIG_NR_CPUS. | |
2082 | */ | |
2083 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
2084 | #else | |
2085 | /* | |
2086 | * No preemption supported therefore also no need to check for | |
2087 | * different cpus. | |
2088 | */ | |
2089 | #define TID_STEP 1 | |
2090 | #endif | |
2091 | ||
2092 | static inline unsigned long next_tid(unsigned long tid) | |
2093 | { | |
2094 | return tid + TID_STEP; | |
2095 | } | |
2096 | ||
9d5f0be0 | 2097 | #ifdef SLUB_DEBUG_CMPXCHG |
8a5ec0ba CL |
2098 | static inline unsigned int tid_to_cpu(unsigned long tid) |
2099 | { | |
2100 | return tid % TID_STEP; | |
2101 | } | |
2102 | ||
2103 | static inline unsigned long tid_to_event(unsigned long tid) | |
2104 | { | |
2105 | return tid / TID_STEP; | |
2106 | } | |
9d5f0be0 | 2107 | #endif |
8a5ec0ba CL |
2108 | |
2109 | static inline unsigned int init_tid(int cpu) | |
2110 | { | |
2111 | return cpu; | |
2112 | } | |
2113 | ||
2114 | static inline void note_cmpxchg_failure(const char *n, | |
2115 | const struct kmem_cache *s, unsigned long tid) | |
2116 | { | |
2117 | #ifdef SLUB_DEBUG_CMPXCHG | |
2118 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2119 | ||
f9f58285 | 2120 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba | 2121 | |
923717cb | 2122 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2123 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) |
f9f58285 | 2124 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2125 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2126 | else | |
2127 | #endif | |
2128 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2129 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2130 | tid_to_event(tid), tid_to_event(actual_tid)); |
2131 | else | |
f9f58285 | 2132 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2133 | actual_tid, tid, next_tid(tid)); |
2134 | #endif | |
4fdccdfb | 2135 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2136 | } |
2137 | ||
788e1aad | 2138 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2139 | { |
8a5ec0ba CL |
2140 | int cpu; |
2141 | ||
2142 | for_each_possible_cpu(cpu) | |
2143 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 2144 | } |
2cfb7455 | 2145 | |
81819f0f CL |
2146 | /* |
2147 | * Remove the cpu slab | |
2148 | */ | |
d0e0ac97 | 2149 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
d4ff6d35 | 2150 | void *freelist, struct kmem_cache_cpu *c) |
81819f0f | 2151 | { |
2cfb7455 | 2152 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
2153 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
2154 | int lock = 0; | |
2155 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 2156 | void *nextfree; |
136333d1 | 2157 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2158 | struct page new; |
2159 | struct page old; | |
2160 | ||
2161 | if (page->freelist) { | |
84e554e6 | 2162 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2163 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2164 | } |
2165 | ||
894b8788 | 2166 | /* |
2cfb7455 CL |
2167 | * Stage one: Free all available per cpu objects back |
2168 | * to the page freelist while it is still frozen. Leave the | |
2169 | * last one. | |
2170 | * | |
2171 | * There is no need to take the list->lock because the page | |
2172 | * is still frozen. | |
2173 | */ | |
2174 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2175 | void *prior; | |
2176 | unsigned long counters; | |
2177 | ||
52f23478 DZ |
2178 | /* |
2179 | * If 'nextfree' is invalid, it is possible that the object at | |
2180 | * 'freelist' is already corrupted. So isolate all objects | |
2181 | * starting at 'freelist'. | |
2182 | */ | |
dc07a728 | 2183 | if (freelist_corrupted(s, page, &freelist, nextfree)) |
52f23478 DZ |
2184 | break; |
2185 | ||
2cfb7455 CL |
2186 | do { |
2187 | prior = page->freelist; | |
2188 | counters = page->counters; | |
2189 | set_freepointer(s, freelist, prior); | |
2190 | new.counters = counters; | |
2191 | new.inuse--; | |
a0132ac0 | 2192 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 2193 | |
1d07171c | 2194 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2195 | prior, counters, |
2196 | freelist, new.counters, | |
2197 | "drain percpu freelist")); | |
2198 | ||
2199 | freelist = nextfree; | |
2200 | } | |
2201 | ||
894b8788 | 2202 | /* |
2cfb7455 CL |
2203 | * Stage two: Ensure that the page is unfrozen while the |
2204 | * list presence reflects the actual number of objects | |
2205 | * during unfreeze. | |
2206 | * | |
2207 | * We setup the list membership and then perform a cmpxchg | |
2208 | * with the count. If there is a mismatch then the page | |
2209 | * is not unfrozen but the page is on the wrong list. | |
2210 | * | |
2211 | * Then we restart the process which may have to remove | |
2212 | * the page from the list that we just put it on again | |
2213 | * because the number of objects in the slab may have | |
2214 | * changed. | |
894b8788 | 2215 | */ |
2cfb7455 | 2216 | redo: |
894b8788 | 2217 | |
2cfb7455 CL |
2218 | old.freelist = page->freelist; |
2219 | old.counters = page->counters; | |
a0132ac0 | 2220 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2221 | |
2cfb7455 CL |
2222 | /* Determine target state of the slab */ |
2223 | new.counters = old.counters; | |
2224 | if (freelist) { | |
2225 | new.inuse--; | |
2226 | set_freepointer(s, freelist, old.freelist); | |
2227 | new.freelist = freelist; | |
2228 | } else | |
2229 | new.freelist = old.freelist; | |
2230 | ||
2231 | new.frozen = 0; | |
2232 | ||
8a5b20ae | 2233 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2234 | m = M_FREE; |
2235 | else if (new.freelist) { | |
2236 | m = M_PARTIAL; | |
2237 | if (!lock) { | |
2238 | lock = 1; | |
2239 | /* | |
8bb4e7a2 | 2240 | * Taking the spinlock removes the possibility |
2cfb7455 CL |
2241 | * that acquire_slab() will see a slab page that |
2242 | * is frozen | |
2243 | */ | |
2244 | spin_lock(&n->list_lock); | |
2245 | } | |
2246 | } else { | |
2247 | m = M_FULL; | |
965c4848 | 2248 | if (kmem_cache_debug_flags(s, SLAB_STORE_USER) && !lock) { |
2cfb7455 CL |
2249 | lock = 1; |
2250 | /* | |
2251 | * This also ensures that the scanning of full | |
2252 | * slabs from diagnostic functions will not see | |
2253 | * any frozen slabs. | |
2254 | */ | |
2255 | spin_lock(&n->list_lock); | |
2256 | } | |
2257 | } | |
2258 | ||
2259 | if (l != m) { | |
2cfb7455 | 2260 | if (l == M_PARTIAL) |
2cfb7455 | 2261 | remove_partial(n, page); |
2cfb7455 | 2262 | else if (l == M_FULL) |
c65c1877 | 2263 | remove_full(s, n, page); |
2cfb7455 | 2264 | |
88349a28 | 2265 | if (m == M_PARTIAL) |
2cfb7455 | 2266 | add_partial(n, page, tail); |
88349a28 | 2267 | else if (m == M_FULL) |
2cfb7455 | 2268 | add_full(s, n, page); |
2cfb7455 CL |
2269 | } |
2270 | ||
2271 | l = m; | |
1d07171c | 2272 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2273 | old.freelist, old.counters, |
2274 | new.freelist, new.counters, | |
2275 | "unfreezing slab")) | |
2276 | goto redo; | |
2277 | ||
2cfb7455 CL |
2278 | if (lock) |
2279 | spin_unlock(&n->list_lock); | |
2280 | ||
88349a28 WY |
2281 | if (m == M_PARTIAL) |
2282 | stat(s, tail); | |
2283 | else if (m == M_FULL) | |
2284 | stat(s, DEACTIVATE_FULL); | |
2285 | else if (m == M_FREE) { | |
2cfb7455 CL |
2286 | stat(s, DEACTIVATE_EMPTY); |
2287 | discard_slab(s, page); | |
2288 | stat(s, FREE_SLAB); | |
894b8788 | 2289 | } |
d4ff6d35 WY |
2290 | |
2291 | c->page = NULL; | |
2292 | c->freelist = NULL; | |
81819f0f CL |
2293 | } |
2294 | ||
d24ac77f JK |
2295 | /* |
2296 | * Unfreeze all the cpu partial slabs. | |
2297 | * | |
59a09917 CL |
2298 | * This function must be called with interrupts disabled |
2299 | * for the cpu using c (or some other guarantee must be there | |
2300 | * to guarantee no concurrent accesses). | |
d24ac77f | 2301 | */ |
59a09917 CL |
2302 | static void unfreeze_partials(struct kmem_cache *s, |
2303 | struct kmem_cache_cpu *c) | |
49e22585 | 2304 | { |
345c905d | 2305 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2306 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2307 | struct page *page, *discard_page = NULL; |
49e22585 | 2308 | |
4c7ba22e | 2309 | while ((page = slub_percpu_partial(c))) { |
49e22585 CL |
2310 | struct page new; |
2311 | struct page old; | |
2312 | ||
4c7ba22e | 2313 | slub_set_percpu_partial(c, page); |
43d77867 JK |
2314 | |
2315 | n2 = get_node(s, page_to_nid(page)); | |
2316 | if (n != n2) { | |
2317 | if (n) | |
2318 | spin_unlock(&n->list_lock); | |
2319 | ||
2320 | n = n2; | |
2321 | spin_lock(&n->list_lock); | |
2322 | } | |
49e22585 CL |
2323 | |
2324 | do { | |
2325 | ||
2326 | old.freelist = page->freelist; | |
2327 | old.counters = page->counters; | |
a0132ac0 | 2328 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2329 | |
2330 | new.counters = old.counters; | |
2331 | new.freelist = old.freelist; | |
2332 | ||
2333 | new.frozen = 0; | |
2334 | ||
d24ac77f | 2335 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2336 | old.freelist, old.counters, |
2337 | new.freelist, new.counters, | |
2338 | "unfreezing slab")); | |
2339 | ||
8a5b20ae | 2340 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2341 | page->next = discard_page; |
2342 | discard_page = page; | |
43d77867 JK |
2343 | } else { |
2344 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2345 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2346 | } |
2347 | } | |
2348 | ||
2349 | if (n) | |
2350 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2351 | |
2352 | while (discard_page) { | |
2353 | page = discard_page; | |
2354 | discard_page = discard_page->next; | |
2355 | ||
2356 | stat(s, DEACTIVATE_EMPTY); | |
2357 | discard_slab(s, page); | |
2358 | stat(s, FREE_SLAB); | |
2359 | } | |
6dfd1b65 | 2360 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2361 | } |
2362 | ||
2363 | /* | |
9234bae9 WY |
2364 | * Put a page that was just frozen (in __slab_free|get_partial_node) into a |
2365 | * partial page slot if available. | |
49e22585 CL |
2366 | * |
2367 | * If we did not find a slot then simply move all the partials to the | |
2368 | * per node partial list. | |
2369 | */ | |
633b0764 | 2370 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2371 | { |
345c905d | 2372 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2373 | struct page *oldpage; |
2374 | int pages; | |
2375 | int pobjects; | |
2376 | ||
d6e0b7fa | 2377 | preempt_disable(); |
49e22585 CL |
2378 | do { |
2379 | pages = 0; | |
2380 | pobjects = 0; | |
2381 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2382 | ||
2383 | if (oldpage) { | |
2384 | pobjects = oldpage->pobjects; | |
2385 | pages = oldpage->pages; | |
bbd4e305 | 2386 | if (drain && pobjects > slub_cpu_partial(s)) { |
49e22585 CL |
2387 | unsigned long flags; |
2388 | /* | |
2389 | * partial array is full. Move the existing | |
2390 | * set to the per node partial list. | |
2391 | */ | |
2392 | local_irq_save(flags); | |
59a09917 | 2393 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2394 | local_irq_restore(flags); |
e24fc410 | 2395 | oldpage = NULL; |
49e22585 CL |
2396 | pobjects = 0; |
2397 | pages = 0; | |
8028dcea | 2398 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2399 | } |
2400 | } | |
2401 | ||
2402 | pages++; | |
2403 | pobjects += page->objects - page->inuse; | |
2404 | ||
2405 | page->pages = pages; | |
2406 | page->pobjects = pobjects; | |
2407 | page->next = oldpage; | |
2408 | ||
d0e0ac97 CG |
2409 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2410 | != oldpage); | |
bbd4e305 | 2411 | if (unlikely(!slub_cpu_partial(s))) { |
d6e0b7fa VD |
2412 | unsigned long flags; |
2413 | ||
2414 | local_irq_save(flags); | |
2415 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2416 | local_irq_restore(flags); | |
2417 | } | |
2418 | preempt_enable(); | |
6dfd1b65 | 2419 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2420 | } |
2421 | ||
dfb4f096 | 2422 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2423 | { |
84e554e6 | 2424 | stat(s, CPUSLAB_FLUSH); |
d4ff6d35 | 2425 | deactivate_slab(s, c->page, c->freelist, c); |
c17dda40 CL |
2426 | |
2427 | c->tid = next_tid(c->tid); | |
81819f0f CL |
2428 | } |
2429 | ||
2430 | /* | |
2431 | * Flush cpu slab. | |
6446faa2 | 2432 | * |
81819f0f CL |
2433 | * Called from IPI handler with interrupts disabled. |
2434 | */ | |
0c710013 | 2435 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2436 | { |
9dfc6e68 | 2437 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2438 | |
1265ef2d WY |
2439 | if (c->page) |
2440 | flush_slab(s, c); | |
49e22585 | 2441 | |
1265ef2d | 2442 | unfreeze_partials(s, c); |
81819f0f CL |
2443 | } |
2444 | ||
2445 | static void flush_cpu_slab(void *d) | |
2446 | { | |
2447 | struct kmem_cache *s = d; | |
81819f0f | 2448 | |
dfb4f096 | 2449 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2450 | } |
2451 | ||
a8364d55 GBY |
2452 | static bool has_cpu_slab(int cpu, void *info) |
2453 | { | |
2454 | struct kmem_cache *s = info; | |
2455 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2456 | ||
a93cf07b | 2457 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2458 | } |
2459 | ||
81819f0f CL |
2460 | static void flush_all(struct kmem_cache *s) |
2461 | { | |
cb923159 | 2462 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1); |
81819f0f CL |
2463 | } |
2464 | ||
a96a87bf SAS |
2465 | /* |
2466 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2467 | * necessary. | |
2468 | */ | |
2469 | static int slub_cpu_dead(unsigned int cpu) | |
2470 | { | |
2471 | struct kmem_cache *s; | |
2472 | unsigned long flags; | |
2473 | ||
2474 | mutex_lock(&slab_mutex); | |
2475 | list_for_each_entry(s, &slab_caches, list) { | |
2476 | local_irq_save(flags); | |
2477 | __flush_cpu_slab(s, cpu); | |
2478 | local_irq_restore(flags); | |
2479 | } | |
2480 | mutex_unlock(&slab_mutex); | |
2481 | return 0; | |
2482 | } | |
2483 | ||
dfb4f096 CL |
2484 | /* |
2485 | * Check if the objects in a per cpu structure fit numa | |
2486 | * locality expectations. | |
2487 | */ | |
57d437d2 | 2488 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2489 | { |
2490 | #ifdef CONFIG_NUMA | |
6159d0f5 | 2491 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2492 | return 0; |
2493 | #endif | |
2494 | return 1; | |
2495 | } | |
2496 | ||
9a02d699 | 2497 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2498 | static int count_free(struct page *page) |
2499 | { | |
2500 | return page->objects - page->inuse; | |
2501 | } | |
2502 | ||
9a02d699 DR |
2503 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2504 | { | |
2505 | return atomic_long_read(&n->total_objects); | |
2506 | } | |
2507 | #endif /* CONFIG_SLUB_DEBUG */ | |
2508 | ||
2509 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2510 | static unsigned long count_partial(struct kmem_cache_node *n, |
2511 | int (*get_count)(struct page *)) | |
2512 | { | |
2513 | unsigned long flags; | |
2514 | unsigned long x = 0; | |
2515 | struct page *page; | |
2516 | ||
2517 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 2518 | list_for_each_entry(page, &n->partial, slab_list) |
781b2ba6 PE |
2519 | x += get_count(page); |
2520 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2521 | return x; | |
2522 | } | |
9a02d699 | 2523 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2524 | |
781b2ba6 PE |
2525 | static noinline void |
2526 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2527 | { | |
9a02d699 DR |
2528 | #ifdef CONFIG_SLUB_DEBUG |
2529 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2530 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2531 | int node; |
fa45dc25 | 2532 | struct kmem_cache_node *n; |
781b2ba6 | 2533 | |
9a02d699 DR |
2534 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2535 | return; | |
2536 | ||
5b3810e5 VB |
2537 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2538 | nid, gfpflags, &gfpflags); | |
19af27af | 2539 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2540 | s->name, s->object_size, s->size, oo_order(s->oo), |
2541 | oo_order(s->min)); | |
781b2ba6 | 2542 | |
3b0efdfa | 2543 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2544 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2545 | s->name); | |
fa5ec8a1 | 2546 | |
fa45dc25 | 2547 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2548 | unsigned long nr_slabs; |
2549 | unsigned long nr_objs; | |
2550 | unsigned long nr_free; | |
2551 | ||
26c02cf0 AB |
2552 | nr_free = count_partial(n, count_free); |
2553 | nr_slabs = node_nr_slabs(n); | |
2554 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2555 | |
f9f58285 | 2556 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2557 | node, nr_slabs, nr_objs, nr_free); |
2558 | } | |
9a02d699 | 2559 | #endif |
781b2ba6 PE |
2560 | } |
2561 | ||
497b66f2 CL |
2562 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2563 | int node, struct kmem_cache_cpu **pc) | |
2564 | { | |
6faa6833 | 2565 | void *freelist; |
188fd063 CL |
2566 | struct kmem_cache_cpu *c = *pc; |
2567 | struct page *page; | |
497b66f2 | 2568 | |
128227e7 MW |
2569 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
2570 | ||
188fd063 | 2571 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2572 | |
188fd063 CL |
2573 | if (freelist) |
2574 | return freelist; | |
2575 | ||
2576 | page = new_slab(s, flags, node); | |
497b66f2 | 2577 | if (page) { |
7c8e0181 | 2578 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2579 | if (c->page) |
2580 | flush_slab(s, c); | |
2581 | ||
2582 | /* | |
2583 | * No other reference to the page yet so we can | |
2584 | * muck around with it freely without cmpxchg | |
2585 | */ | |
6faa6833 | 2586 | freelist = page->freelist; |
497b66f2 CL |
2587 | page->freelist = NULL; |
2588 | ||
2589 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2590 | c->page = page; |
2591 | *pc = c; | |
edde82b6 | 2592 | } |
497b66f2 | 2593 | |
6faa6833 | 2594 | return freelist; |
497b66f2 CL |
2595 | } |
2596 | ||
072bb0aa MG |
2597 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2598 | { | |
2599 | if (unlikely(PageSlabPfmemalloc(page))) | |
2600 | return gfp_pfmemalloc_allowed(gfpflags); | |
2601 | ||
2602 | return true; | |
2603 | } | |
2604 | ||
213eeb9f | 2605 | /* |
d0e0ac97 CG |
2606 | * Check the page->freelist of a page and either transfer the freelist to the |
2607 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2608 | * |
2609 | * The page is still frozen if the return value is not NULL. | |
2610 | * | |
2611 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2612 | * |
2613 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2614 | */ |
2615 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2616 | { | |
2617 | struct page new; | |
2618 | unsigned long counters; | |
2619 | void *freelist; | |
2620 | ||
2621 | do { | |
2622 | freelist = page->freelist; | |
2623 | counters = page->counters; | |
6faa6833 | 2624 | |
213eeb9f | 2625 | new.counters = counters; |
a0132ac0 | 2626 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2627 | |
2628 | new.inuse = page->objects; | |
2629 | new.frozen = freelist != NULL; | |
2630 | ||
d24ac77f | 2631 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2632 | freelist, counters, |
2633 | NULL, new.counters, | |
2634 | "get_freelist")); | |
2635 | ||
2636 | return freelist; | |
2637 | } | |
2638 | ||
81819f0f | 2639 | /* |
894b8788 CL |
2640 | * Slow path. The lockless freelist is empty or we need to perform |
2641 | * debugging duties. | |
2642 | * | |
894b8788 CL |
2643 | * Processing is still very fast if new objects have been freed to the |
2644 | * regular freelist. In that case we simply take over the regular freelist | |
2645 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2646 | * |
894b8788 CL |
2647 | * If that is not working then we fall back to the partial lists. We take the |
2648 | * first element of the freelist as the object to allocate now and move the | |
2649 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2650 | * |
894b8788 | 2651 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2652 | * we need to allocate a new slab. This is the slowest path since it involves |
2653 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2654 | * |
2655 | * Version of __slab_alloc to use when we know that interrupts are | |
2656 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2657 | */ |
a380a3c7 | 2658 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2659 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2660 | { |
6faa6833 | 2661 | void *freelist; |
f6e7def7 | 2662 | struct page *page; |
81819f0f | 2663 | |
9f986d99 AW |
2664 | stat(s, ALLOC_SLOWPATH); |
2665 | ||
f6e7def7 | 2666 | page = c->page; |
0715e6c5 VB |
2667 | if (!page) { |
2668 | /* | |
2669 | * if the node is not online or has no normal memory, just | |
2670 | * ignore the node constraint | |
2671 | */ | |
2672 | if (unlikely(node != NUMA_NO_NODE && | |
2673 | !node_state(node, N_NORMAL_MEMORY))) | |
2674 | node = NUMA_NO_NODE; | |
81819f0f | 2675 | goto new_slab; |
0715e6c5 | 2676 | } |
49e22585 | 2677 | redo: |
6faa6833 | 2678 | |
57d437d2 | 2679 | if (unlikely(!node_match(page, node))) { |
0715e6c5 VB |
2680 | /* |
2681 | * same as above but node_match() being false already | |
2682 | * implies node != NUMA_NO_NODE | |
2683 | */ | |
2684 | if (!node_state(node, N_NORMAL_MEMORY)) { | |
2685 | node = NUMA_NO_NODE; | |
2686 | goto redo; | |
2687 | } else { | |
a561ce00 | 2688 | stat(s, ALLOC_NODE_MISMATCH); |
d4ff6d35 | 2689 | deactivate_slab(s, page, c->freelist, c); |
a561ce00 JK |
2690 | goto new_slab; |
2691 | } | |
fc59c053 | 2692 | } |
6446faa2 | 2693 | |
072bb0aa MG |
2694 | /* |
2695 | * By rights, we should be searching for a slab page that was | |
2696 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2697 | * information when the page leaves the per-cpu allocator | |
2698 | */ | |
2699 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
d4ff6d35 | 2700 | deactivate_slab(s, page, c->freelist, c); |
072bb0aa MG |
2701 | goto new_slab; |
2702 | } | |
2703 | ||
73736e03 | 2704 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2705 | freelist = c->freelist; |
2706 | if (freelist) | |
73736e03 | 2707 | goto load_freelist; |
03e404af | 2708 | |
f6e7def7 | 2709 | freelist = get_freelist(s, page); |
6446faa2 | 2710 | |
6faa6833 | 2711 | if (!freelist) { |
03e404af CL |
2712 | c->page = NULL; |
2713 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2714 | goto new_slab; |
03e404af | 2715 | } |
6446faa2 | 2716 | |
84e554e6 | 2717 | stat(s, ALLOC_REFILL); |
6446faa2 | 2718 | |
894b8788 | 2719 | load_freelist: |
507effea CL |
2720 | /* |
2721 | * freelist is pointing to the list of objects to be used. | |
2722 | * page is pointing to the page from which the objects are obtained. | |
2723 | * That page must be frozen for per cpu allocations to work. | |
2724 | */ | |
a0132ac0 | 2725 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2726 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2727 | c->tid = next_tid(c->tid); |
6faa6833 | 2728 | return freelist; |
81819f0f | 2729 | |
81819f0f | 2730 | new_slab: |
2cfb7455 | 2731 | |
a93cf07b WY |
2732 | if (slub_percpu_partial(c)) { |
2733 | page = c->page = slub_percpu_partial(c); | |
2734 | slub_set_percpu_partial(c, page); | |
49e22585 | 2735 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2736 | goto redo; |
81819f0f CL |
2737 | } |
2738 | ||
188fd063 | 2739 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2740 | |
f4697436 | 2741 | if (unlikely(!freelist)) { |
9a02d699 | 2742 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2743 | return NULL; |
81819f0f | 2744 | } |
2cfb7455 | 2745 | |
f6e7def7 | 2746 | page = c->page; |
5091b74a | 2747 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2748 | goto load_freelist; |
2cfb7455 | 2749 | |
497b66f2 | 2750 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2751 | if (kmem_cache_debug(s) && |
2752 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2753 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2754 | |
d4ff6d35 | 2755 | deactivate_slab(s, page, get_freepointer(s, freelist), c); |
6faa6833 | 2756 | return freelist; |
894b8788 CL |
2757 | } |
2758 | ||
a380a3c7 CL |
2759 | /* |
2760 | * Another one that disabled interrupt and compensates for possible | |
2761 | * cpu changes by refetching the per cpu area pointer. | |
2762 | */ | |
2763 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2764 | unsigned long addr, struct kmem_cache_cpu *c) | |
2765 | { | |
2766 | void *p; | |
2767 | unsigned long flags; | |
2768 | ||
2769 | local_irq_save(flags); | |
923717cb | 2770 | #ifdef CONFIG_PREEMPTION |
a380a3c7 CL |
2771 | /* |
2772 | * We may have been preempted and rescheduled on a different | |
2773 | * cpu before disabling interrupts. Need to reload cpu area | |
2774 | * pointer. | |
2775 | */ | |
2776 | c = this_cpu_ptr(s->cpu_slab); | |
2777 | #endif | |
2778 | ||
2779 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2780 | local_irq_restore(flags); | |
2781 | return p; | |
2782 | } | |
2783 | ||
0f181f9f AP |
2784 | /* |
2785 | * If the object has been wiped upon free, make sure it's fully initialized by | |
2786 | * zeroing out freelist pointer. | |
2787 | */ | |
2788 | static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s, | |
2789 | void *obj) | |
2790 | { | |
2791 | if (unlikely(slab_want_init_on_free(s)) && obj) | |
2792 | memset((void *)((char *)obj + s->offset), 0, sizeof(void *)); | |
2793 | } | |
2794 | ||
894b8788 CL |
2795 | /* |
2796 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2797 | * have the fastpath folded into their functions. So no function call | |
2798 | * overhead for requests that can be satisfied on the fastpath. | |
2799 | * | |
2800 | * The fastpath works by first checking if the lockless freelist can be used. | |
2801 | * If not then __slab_alloc is called for slow processing. | |
2802 | * | |
2803 | * Otherwise we can simply pick the next object from the lockless free list. | |
2804 | */ | |
2b847c3c | 2805 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2806 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2807 | { |
03ec0ed5 | 2808 | void *object; |
dfb4f096 | 2809 | struct kmem_cache_cpu *c; |
57d437d2 | 2810 | struct page *page; |
8a5ec0ba | 2811 | unsigned long tid; |
964d4bd3 | 2812 | struct obj_cgroup *objcg = NULL; |
1f84260c | 2813 | |
964d4bd3 | 2814 | s = slab_pre_alloc_hook(s, &objcg, 1, gfpflags); |
8135be5a | 2815 | if (!s) |
773ff60e | 2816 | return NULL; |
8a5ec0ba | 2817 | redo: |
8a5ec0ba CL |
2818 | /* |
2819 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2820 | * enabled. We may switch back and forth between cpus while | |
2821 | * reading from one cpu area. That does not matter as long | |
2822 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2823 | * |
9aabf810 | 2824 | * We should guarantee that tid and kmem_cache are retrieved on |
923717cb | 2825 | * the same cpu. It could be different if CONFIG_PREEMPTION so we need |
9aabf810 | 2826 | * to check if it is matched or not. |
8a5ec0ba | 2827 | */ |
9aabf810 JK |
2828 | do { |
2829 | tid = this_cpu_read(s->cpu_slab->tid); | |
2830 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 2831 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 2832 | unlikely(tid != READ_ONCE(c->tid))); |
9aabf810 JK |
2833 | |
2834 | /* | |
2835 | * Irqless object alloc/free algorithm used here depends on sequence | |
2836 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2837 | * on c to guarantee that object and page associated with previous tid | |
2838 | * won't be used with current tid. If we fetch tid first, object and | |
2839 | * page could be one associated with next tid and our alloc/free | |
2840 | * request will be failed. In this case, we will retry. So, no problem. | |
2841 | */ | |
2842 | barrier(); | |
8a5ec0ba | 2843 | |
8a5ec0ba CL |
2844 | /* |
2845 | * The transaction ids are globally unique per cpu and per operation on | |
2846 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2847 | * occurs on the right processor and that there was no operation on the | |
2848 | * linked list in between. | |
2849 | */ | |
8a5ec0ba | 2850 | |
9dfc6e68 | 2851 | object = c->freelist; |
57d437d2 | 2852 | page = c->page; |
22e4663e | 2853 | if (unlikely(!object || !page || !node_match(page, node))) { |
dfb4f096 | 2854 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 | 2855 | } else { |
0ad9500e ED |
2856 | void *next_object = get_freepointer_safe(s, object); |
2857 | ||
8a5ec0ba | 2858 | /* |
25985edc | 2859 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2860 | * operation and if we are on the right processor. |
2861 | * | |
d0e0ac97 CG |
2862 | * The cmpxchg does the following atomically (without lock |
2863 | * semantics!) | |
8a5ec0ba CL |
2864 | * 1. Relocate first pointer to the current per cpu area. |
2865 | * 2. Verify that tid and freelist have not been changed | |
2866 | * 3. If they were not changed replace tid and freelist | |
2867 | * | |
d0e0ac97 CG |
2868 | * Since this is without lock semantics the protection is only |
2869 | * against code executing on this cpu *not* from access by | |
2870 | * other cpus. | |
8a5ec0ba | 2871 | */ |
933393f5 | 2872 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2873 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2874 | object, tid, | |
0ad9500e | 2875 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2876 | |
2877 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2878 | goto redo; | |
2879 | } | |
0ad9500e | 2880 | prefetch_freepointer(s, next_object); |
84e554e6 | 2881 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2882 | } |
0f181f9f AP |
2883 | |
2884 | maybe_wipe_obj_freeptr(s, object); | |
8a5ec0ba | 2885 | |
6471384a | 2886 | if (unlikely(slab_want_init_on_alloc(gfpflags, s)) && object) |
3b0efdfa | 2887 | memset(object, 0, s->object_size); |
d07dbea4 | 2888 | |
964d4bd3 | 2889 | slab_post_alloc_hook(s, objcg, gfpflags, 1, &object); |
5a896d9e | 2890 | |
894b8788 | 2891 | return object; |
81819f0f CL |
2892 | } |
2893 | ||
2b847c3c EG |
2894 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2895 | gfp_t gfpflags, unsigned long addr) | |
2896 | { | |
2897 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2898 | } | |
2899 | ||
81819f0f CL |
2900 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2901 | { | |
2b847c3c | 2902 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2903 | |
d0e0ac97 CG |
2904 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2905 | s->size, gfpflags); | |
5b882be4 EGM |
2906 | |
2907 | return ret; | |
81819f0f CL |
2908 | } |
2909 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2910 | ||
0f24f128 | 2911 | #ifdef CONFIG_TRACING |
4a92379b RK |
2912 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2913 | { | |
2b847c3c | 2914 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2915 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0116523c | 2916 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2917 | return ret; |
2918 | } | |
2919 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2920 | #endif |
2921 | ||
81819f0f CL |
2922 | #ifdef CONFIG_NUMA |
2923 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2924 | { | |
2b847c3c | 2925 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2926 | |
ca2b84cb | 2927 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2928 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2929 | |
2930 | return ret; | |
81819f0f CL |
2931 | } |
2932 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2933 | |
0f24f128 | 2934 | #ifdef CONFIG_TRACING |
4a92379b | 2935 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2936 | gfp_t gfpflags, |
4a92379b | 2937 | int node, size_t size) |
5b882be4 | 2938 | { |
2b847c3c | 2939 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2940 | |
2941 | trace_kmalloc_node(_RET_IP_, ret, | |
2942 | size, s->size, gfpflags, node); | |
0316bec2 | 2943 | |
0116523c | 2944 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2945 | return ret; |
5b882be4 | 2946 | } |
4a92379b | 2947 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2948 | #endif |
6dfd1b65 | 2949 | #endif /* CONFIG_NUMA */ |
5b882be4 | 2950 | |
81819f0f | 2951 | /* |
94e4d712 | 2952 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2953 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2954 | * |
894b8788 CL |
2955 | * So we still attempt to reduce cache line usage. Just take the slab |
2956 | * lock and free the item. If there is no additional partial page | |
2957 | * handling required then we can return immediately. | |
81819f0f | 2958 | */ |
894b8788 | 2959 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2960 | void *head, void *tail, int cnt, |
2961 | unsigned long addr) | |
2962 | ||
81819f0f CL |
2963 | { |
2964 | void *prior; | |
2cfb7455 | 2965 | int was_frozen; |
2cfb7455 CL |
2966 | struct page new; |
2967 | unsigned long counters; | |
2968 | struct kmem_cache_node *n = NULL; | |
3f649ab7 | 2969 | unsigned long flags; |
81819f0f | 2970 | |
8a5ec0ba | 2971 | stat(s, FREE_SLOWPATH); |
81819f0f | 2972 | |
19c7ff9e | 2973 | if (kmem_cache_debug(s) && |
282acb43 | 2974 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2975 | return; |
6446faa2 | 2976 | |
2cfb7455 | 2977 | do { |
837d678d JK |
2978 | if (unlikely(n)) { |
2979 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2980 | n = NULL; | |
2981 | } | |
2cfb7455 CL |
2982 | prior = page->freelist; |
2983 | counters = page->counters; | |
81084651 | 2984 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2985 | new.counters = counters; |
2986 | was_frozen = new.frozen; | |
81084651 | 2987 | new.inuse -= cnt; |
837d678d | 2988 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2989 | |
c65c1877 | 2990 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2991 | |
2992 | /* | |
d0e0ac97 CG |
2993 | * Slab was on no list before and will be |
2994 | * partially empty | |
2995 | * We can defer the list move and instead | |
2996 | * freeze it. | |
49e22585 CL |
2997 | */ |
2998 | new.frozen = 1; | |
2999 | ||
c65c1877 | 3000 | } else { /* Needs to be taken off a list */ |
49e22585 | 3001 | |
b455def2 | 3002 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
3003 | /* |
3004 | * Speculatively acquire the list_lock. | |
3005 | * If the cmpxchg does not succeed then we may | |
3006 | * drop the list_lock without any processing. | |
3007 | * | |
3008 | * Otherwise the list_lock will synchronize with | |
3009 | * other processors updating the list of slabs. | |
3010 | */ | |
3011 | spin_lock_irqsave(&n->list_lock, flags); | |
3012 | ||
3013 | } | |
2cfb7455 | 3014 | } |
81819f0f | 3015 | |
2cfb7455 CL |
3016 | } while (!cmpxchg_double_slab(s, page, |
3017 | prior, counters, | |
81084651 | 3018 | head, new.counters, |
2cfb7455 | 3019 | "__slab_free")); |
81819f0f | 3020 | |
2cfb7455 | 3021 | if (likely(!n)) { |
49e22585 | 3022 | |
c270cf30 AW |
3023 | if (likely(was_frozen)) { |
3024 | /* | |
3025 | * The list lock was not taken therefore no list | |
3026 | * activity can be necessary. | |
3027 | */ | |
3028 | stat(s, FREE_FROZEN); | |
3029 | } else if (new.frozen) { | |
3030 | /* | |
3031 | * If we just froze the page then put it onto the | |
3032 | * per cpu partial list. | |
3033 | */ | |
49e22585 | 3034 | put_cpu_partial(s, page, 1); |
8028dcea AS |
3035 | stat(s, CPU_PARTIAL_FREE); |
3036 | } | |
c270cf30 | 3037 | |
b455def2 L |
3038 | return; |
3039 | } | |
81819f0f | 3040 | |
8a5b20ae | 3041 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
3042 | goto slab_empty; |
3043 | ||
81819f0f | 3044 | /* |
837d678d JK |
3045 | * Objects left in the slab. If it was not on the partial list before |
3046 | * then add it. | |
81819f0f | 3047 | */ |
345c905d | 3048 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
a4d3f891 | 3049 | remove_full(s, n, page); |
837d678d JK |
3050 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
3051 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 3052 | } |
80f08c19 | 3053 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
3054 | return; |
3055 | ||
3056 | slab_empty: | |
a973e9dd | 3057 | if (prior) { |
81819f0f | 3058 | /* |
6fbabb20 | 3059 | * Slab on the partial list. |
81819f0f | 3060 | */ |
5cc6eee8 | 3061 | remove_partial(n, page); |
84e554e6 | 3062 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 3063 | } else { |
6fbabb20 | 3064 | /* Slab must be on the full list */ |
c65c1877 PZ |
3065 | remove_full(s, n, page); |
3066 | } | |
2cfb7455 | 3067 | |
80f08c19 | 3068 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 3069 | stat(s, FREE_SLAB); |
81819f0f | 3070 | discard_slab(s, page); |
81819f0f CL |
3071 | } |
3072 | ||
894b8788 CL |
3073 | /* |
3074 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
3075 | * can perform fastpath freeing without additional function calls. | |
3076 | * | |
3077 | * The fastpath is only possible if we are freeing to the current cpu slab | |
3078 | * of this processor. This typically the case if we have just allocated | |
3079 | * the item before. | |
3080 | * | |
3081 | * If fastpath is not possible then fall back to __slab_free where we deal | |
3082 | * with all sorts of special processing. | |
81084651 JDB |
3083 | * |
3084 | * Bulk free of a freelist with several objects (all pointing to the | |
3085 | * same page) possible by specifying head and tail ptr, plus objects | |
3086 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 3087 | */ |
80a9201a AP |
3088 | static __always_inline void do_slab_free(struct kmem_cache *s, |
3089 | struct page *page, void *head, void *tail, | |
3090 | int cnt, unsigned long addr) | |
894b8788 | 3091 | { |
81084651 | 3092 | void *tail_obj = tail ? : head; |
dfb4f096 | 3093 | struct kmem_cache_cpu *c; |
8a5ec0ba | 3094 | unsigned long tid; |
964d4bd3 | 3095 | |
d1b2cf6c | 3096 | memcg_slab_free_hook(s, &head, 1); |
8a5ec0ba CL |
3097 | redo: |
3098 | /* | |
3099 | * Determine the currently cpus per cpu slab. | |
3100 | * The cpu may change afterward. However that does not matter since | |
3101 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 3102 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 3103 | */ |
9aabf810 JK |
3104 | do { |
3105 | tid = this_cpu_read(s->cpu_slab->tid); | |
3106 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 3107 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 3108 | unlikely(tid != READ_ONCE(c->tid))); |
c016b0bd | 3109 | |
9aabf810 JK |
3110 | /* Same with comment on barrier() in slab_alloc_node() */ |
3111 | barrier(); | |
c016b0bd | 3112 | |
442b06bc | 3113 | if (likely(page == c->page)) { |
5076190d LT |
3114 | void **freelist = READ_ONCE(c->freelist); |
3115 | ||
3116 | set_freepointer(s, tail_obj, freelist); | |
8a5ec0ba | 3117 | |
933393f5 | 3118 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba | 3119 | s->cpu_slab->freelist, s->cpu_slab->tid, |
5076190d | 3120 | freelist, tid, |
81084651 | 3121 | head, next_tid(tid)))) { |
8a5ec0ba CL |
3122 | |
3123 | note_cmpxchg_failure("slab_free", s, tid); | |
3124 | goto redo; | |
3125 | } | |
84e554e6 | 3126 | stat(s, FREE_FASTPATH); |
894b8788 | 3127 | } else |
81084651 | 3128 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 3129 | |
894b8788 CL |
3130 | } |
3131 | ||
80a9201a AP |
3132 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
3133 | void *head, void *tail, int cnt, | |
3134 | unsigned long addr) | |
3135 | { | |
80a9201a | 3136 | /* |
c3895391 AK |
3137 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
3138 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 3139 | */ |
c3895391 AK |
3140 | if (slab_free_freelist_hook(s, &head, &tail)) |
3141 | do_slab_free(s, page, head, tail, cnt, addr); | |
80a9201a AP |
3142 | } |
3143 | ||
2bd926b4 | 3144 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
3145 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
3146 | { | |
3147 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
3148 | } | |
3149 | #endif | |
3150 | ||
81819f0f CL |
3151 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3152 | { | |
b9ce5ef4 GC |
3153 | s = cache_from_obj(s, x); |
3154 | if (!s) | |
79576102 | 3155 | return; |
81084651 | 3156 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 3157 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
3158 | } |
3159 | EXPORT_SYMBOL(kmem_cache_free); | |
3160 | ||
d0ecd894 | 3161 | struct detached_freelist { |
fbd02630 | 3162 | struct page *page; |
d0ecd894 JDB |
3163 | void *tail; |
3164 | void *freelist; | |
3165 | int cnt; | |
376bf125 | 3166 | struct kmem_cache *s; |
d0ecd894 | 3167 | }; |
fbd02630 | 3168 | |
d0ecd894 JDB |
3169 | /* |
3170 | * This function progressively scans the array with free objects (with | |
3171 | * a limited look ahead) and extract objects belonging to the same | |
3172 | * page. It builds a detached freelist directly within the given | |
3173 | * page/objects. This can happen without any need for | |
3174 | * synchronization, because the objects are owned by running process. | |
3175 | * The freelist is build up as a single linked list in the objects. | |
3176 | * The idea is, that this detached freelist can then be bulk | |
3177 | * transferred to the real freelist(s), but only requiring a single | |
3178 | * synchronization primitive. Look ahead in the array is limited due | |
3179 | * to performance reasons. | |
3180 | */ | |
376bf125 JDB |
3181 | static inline |
3182 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3183 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3184 | { |
3185 | size_t first_skipped_index = 0; | |
3186 | int lookahead = 3; | |
3187 | void *object; | |
ca257195 | 3188 | struct page *page; |
fbd02630 | 3189 | |
d0ecd894 JDB |
3190 | /* Always re-init detached_freelist */ |
3191 | df->page = NULL; | |
fbd02630 | 3192 | |
d0ecd894 JDB |
3193 | do { |
3194 | object = p[--size]; | |
ca257195 | 3195 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3196 | } while (!object && size); |
3eed034d | 3197 | |
d0ecd894 JDB |
3198 | if (!object) |
3199 | return 0; | |
fbd02630 | 3200 | |
ca257195 JDB |
3201 | page = virt_to_head_page(object); |
3202 | if (!s) { | |
3203 | /* Handle kalloc'ed objects */ | |
3204 | if (unlikely(!PageSlab(page))) { | |
3205 | BUG_ON(!PageCompound(page)); | |
3206 | kfree_hook(object); | |
4949148a | 3207 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3208 | p[size] = NULL; /* mark object processed */ |
3209 | return size; | |
3210 | } | |
3211 | /* Derive kmem_cache from object */ | |
3212 | df->s = page->slab_cache; | |
3213 | } else { | |
3214 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3215 | } | |
376bf125 | 3216 | |
d0ecd894 | 3217 | /* Start new detached freelist */ |
ca257195 | 3218 | df->page = page; |
376bf125 | 3219 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3220 | df->tail = object; |
3221 | df->freelist = object; | |
3222 | p[size] = NULL; /* mark object processed */ | |
3223 | df->cnt = 1; | |
3224 | ||
3225 | while (size) { | |
3226 | object = p[--size]; | |
3227 | if (!object) | |
3228 | continue; /* Skip processed objects */ | |
3229 | ||
3230 | /* df->page is always set at this point */ | |
3231 | if (df->page == virt_to_head_page(object)) { | |
3232 | /* Opportunity build freelist */ | |
376bf125 | 3233 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3234 | df->freelist = object; |
3235 | df->cnt++; | |
3236 | p[size] = NULL; /* mark object processed */ | |
3237 | ||
3238 | continue; | |
fbd02630 | 3239 | } |
d0ecd894 JDB |
3240 | |
3241 | /* Limit look ahead search */ | |
3242 | if (!--lookahead) | |
3243 | break; | |
3244 | ||
3245 | if (!first_skipped_index) | |
3246 | first_skipped_index = size + 1; | |
fbd02630 | 3247 | } |
d0ecd894 JDB |
3248 | |
3249 | return first_skipped_index; | |
3250 | } | |
3251 | ||
d0ecd894 | 3252 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3253 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3254 | { |
3255 | if (WARN_ON(!size)) | |
3256 | return; | |
3257 | ||
d1b2cf6c | 3258 | memcg_slab_free_hook(s, p, size); |
d0ecd894 JDB |
3259 | do { |
3260 | struct detached_freelist df; | |
3261 | ||
3262 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3263 | if (!df.page) |
d0ecd894 JDB |
3264 | continue; |
3265 | ||
376bf125 | 3266 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 3267 | } while (likely(size)); |
484748f0 CL |
3268 | } |
3269 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3270 | ||
994eb764 | 3271 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3272 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3273 | void **p) | |
484748f0 | 3274 | { |
994eb764 JDB |
3275 | struct kmem_cache_cpu *c; |
3276 | int i; | |
964d4bd3 | 3277 | struct obj_cgroup *objcg = NULL; |
994eb764 | 3278 | |
03ec0ed5 | 3279 | /* memcg and kmem_cache debug support */ |
964d4bd3 | 3280 | s = slab_pre_alloc_hook(s, &objcg, size, flags); |
03ec0ed5 JDB |
3281 | if (unlikely(!s)) |
3282 | return false; | |
994eb764 JDB |
3283 | /* |
3284 | * Drain objects in the per cpu slab, while disabling local | |
3285 | * IRQs, which protects against PREEMPT and interrupts | |
3286 | * handlers invoking normal fastpath. | |
3287 | */ | |
3288 | local_irq_disable(); | |
3289 | c = this_cpu_ptr(s->cpu_slab); | |
3290 | ||
3291 | for (i = 0; i < size; i++) { | |
3292 | void *object = c->freelist; | |
3293 | ||
ebe909e0 | 3294 | if (unlikely(!object)) { |
fd4d9c7d JH |
3295 | /* |
3296 | * We may have removed an object from c->freelist using | |
3297 | * the fastpath in the previous iteration; in that case, | |
3298 | * c->tid has not been bumped yet. | |
3299 | * Since ___slab_alloc() may reenable interrupts while | |
3300 | * allocating memory, we should bump c->tid now. | |
3301 | */ | |
3302 | c->tid = next_tid(c->tid); | |
3303 | ||
ebe909e0 JDB |
3304 | /* |
3305 | * Invoking slow path likely have side-effect | |
3306 | * of re-populating per CPU c->freelist | |
3307 | */ | |
87098373 | 3308 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3309 | _RET_IP_, c); |
87098373 CL |
3310 | if (unlikely(!p[i])) |
3311 | goto error; | |
3312 | ||
ebe909e0 | 3313 | c = this_cpu_ptr(s->cpu_slab); |
0f181f9f AP |
3314 | maybe_wipe_obj_freeptr(s, p[i]); |
3315 | ||
ebe909e0 JDB |
3316 | continue; /* goto for-loop */ |
3317 | } | |
994eb764 JDB |
3318 | c->freelist = get_freepointer(s, object); |
3319 | p[i] = object; | |
0f181f9f | 3320 | maybe_wipe_obj_freeptr(s, p[i]); |
994eb764 JDB |
3321 | } |
3322 | c->tid = next_tid(c->tid); | |
3323 | local_irq_enable(); | |
3324 | ||
3325 | /* Clear memory outside IRQ disabled fastpath loop */ | |
6471384a | 3326 | if (unlikely(slab_want_init_on_alloc(flags, s))) { |
994eb764 JDB |
3327 | int j; |
3328 | ||
3329 | for (j = 0; j < i; j++) | |
3330 | memset(p[j], 0, s->object_size); | |
3331 | } | |
3332 | ||
03ec0ed5 | 3333 | /* memcg and kmem_cache debug support */ |
964d4bd3 | 3334 | slab_post_alloc_hook(s, objcg, flags, size, p); |
865762a8 | 3335 | return i; |
87098373 | 3336 | error: |
87098373 | 3337 | local_irq_enable(); |
964d4bd3 | 3338 | slab_post_alloc_hook(s, objcg, flags, i, p); |
03ec0ed5 | 3339 | __kmem_cache_free_bulk(s, i, p); |
865762a8 | 3340 | return 0; |
484748f0 CL |
3341 | } |
3342 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3343 | ||
3344 | ||
81819f0f | 3345 | /* |
672bba3a CL |
3346 | * Object placement in a slab is made very easy because we always start at |
3347 | * offset 0. If we tune the size of the object to the alignment then we can | |
3348 | * get the required alignment by putting one properly sized object after | |
3349 | * another. | |
81819f0f CL |
3350 | * |
3351 | * Notice that the allocation order determines the sizes of the per cpu | |
3352 | * caches. Each processor has always one slab available for allocations. | |
3353 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3354 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3355 | * locking overhead. |
81819f0f CL |
3356 | */ |
3357 | ||
3358 | /* | |
3359 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3360 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3361 | * and increases the number of allocations possible without having to | |
3362 | * take the list_lock. | |
3363 | */ | |
19af27af AD |
3364 | static unsigned int slub_min_order; |
3365 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3366 | static unsigned int slub_min_objects; | |
81819f0f | 3367 | |
81819f0f CL |
3368 | /* |
3369 | * Calculate the order of allocation given an slab object size. | |
3370 | * | |
672bba3a CL |
3371 | * The order of allocation has significant impact on performance and other |
3372 | * system components. Generally order 0 allocations should be preferred since | |
3373 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3374 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3375 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3376 | * would be wasted. |
3377 | * | |
3378 | * In order to reach satisfactory performance we must ensure that a minimum | |
3379 | * number of objects is in one slab. Otherwise we may generate too much | |
3380 | * activity on the partial lists which requires taking the list_lock. This is | |
3381 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3382 | * |
672bba3a CL |
3383 | * slub_max_order specifies the order where we begin to stop considering the |
3384 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3385 | * we try to keep the page order as low as possible. So we accept more waste | |
3386 | * of space in favor of a small page order. | |
81819f0f | 3387 | * |
672bba3a CL |
3388 | * Higher order allocations also allow the placement of more objects in a |
3389 | * slab and thereby reduce object handling overhead. If the user has | |
3390 | * requested a higher mininum order then we start with that one instead of | |
3391 | * the smallest order which will fit the object. | |
81819f0f | 3392 | */ |
19af27af AD |
3393 | static inline unsigned int slab_order(unsigned int size, |
3394 | unsigned int min_objects, unsigned int max_order, | |
9736d2a9 | 3395 | unsigned int fract_leftover) |
81819f0f | 3396 | { |
19af27af AD |
3397 | unsigned int min_order = slub_min_order; |
3398 | unsigned int order; | |
81819f0f | 3399 | |
9736d2a9 | 3400 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3401 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3402 | |
9736d2a9 | 3403 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3404 | order <= max_order; order++) { |
81819f0f | 3405 | |
19af27af AD |
3406 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3407 | unsigned int rem; | |
81819f0f | 3408 | |
9736d2a9 | 3409 | rem = slab_size % size; |
81819f0f | 3410 | |
5e6d444e | 3411 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3412 | break; |
81819f0f | 3413 | } |
672bba3a | 3414 | |
81819f0f CL |
3415 | return order; |
3416 | } | |
3417 | ||
9736d2a9 | 3418 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3419 | { |
19af27af AD |
3420 | unsigned int order; |
3421 | unsigned int min_objects; | |
3422 | unsigned int max_objects; | |
5e6d444e CL |
3423 | |
3424 | /* | |
3425 | * Attempt to find best configuration for a slab. This | |
3426 | * works by first attempting to generate a layout with | |
3427 | * the best configuration and backing off gradually. | |
3428 | * | |
422ff4d7 | 3429 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3430 | * we reduce the minimum objects required in a slab. |
3431 | */ | |
3432 | min_objects = slub_min_objects; | |
9b2cd506 | 3433 | if (!min_objects) |
045ab8c9 | 3434 | min_objects = 4 * (fls(num_online_cpus()) + 1); |
9736d2a9 | 3435 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3436 | min_objects = min(min_objects, max_objects); |
3437 | ||
5e6d444e | 3438 | while (min_objects > 1) { |
19af27af AD |
3439 | unsigned int fraction; |
3440 | ||
c124f5b5 | 3441 | fraction = 16; |
5e6d444e CL |
3442 | while (fraction >= 4) { |
3443 | order = slab_order(size, min_objects, | |
9736d2a9 | 3444 | slub_max_order, fraction); |
5e6d444e CL |
3445 | if (order <= slub_max_order) |
3446 | return order; | |
3447 | fraction /= 2; | |
3448 | } | |
5086c389 | 3449 | min_objects--; |
5e6d444e CL |
3450 | } |
3451 | ||
3452 | /* | |
3453 | * We were unable to place multiple objects in a slab. Now | |
3454 | * lets see if we can place a single object there. | |
3455 | */ | |
9736d2a9 | 3456 | order = slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3457 | if (order <= slub_max_order) |
3458 | return order; | |
3459 | ||
3460 | /* | |
3461 | * Doh this slab cannot be placed using slub_max_order. | |
3462 | */ | |
9736d2a9 | 3463 | order = slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3464 | if (order < MAX_ORDER) |
5e6d444e CL |
3465 | return order; |
3466 | return -ENOSYS; | |
3467 | } | |
3468 | ||
5595cffc | 3469 | static void |
4053497d | 3470 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3471 | { |
3472 | n->nr_partial = 0; | |
81819f0f CL |
3473 | spin_lock_init(&n->list_lock); |
3474 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3475 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3476 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3477 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3478 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3479 | #endif |
81819f0f CL |
3480 | } |
3481 | ||
55136592 | 3482 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3483 | { |
6c182dc0 | 3484 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3485 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3486 | |
8a5ec0ba | 3487 | /* |
d4d84fef CM |
3488 | * Must align to double word boundary for the double cmpxchg |
3489 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3490 | */ |
d4d84fef CM |
3491 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3492 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3493 | |
3494 | if (!s->cpu_slab) | |
3495 | return 0; | |
3496 | ||
3497 | init_kmem_cache_cpus(s); | |
4c93c355 | 3498 | |
8a5ec0ba | 3499 | return 1; |
4c93c355 | 3500 | } |
4c93c355 | 3501 | |
51df1142 CL |
3502 | static struct kmem_cache *kmem_cache_node; |
3503 | ||
81819f0f CL |
3504 | /* |
3505 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3506 | * slab on the node for this slabcache. There are no concurrent accesses | |
3507 | * possible. | |
3508 | * | |
721ae22a ZYW |
3509 | * Note that this function only works on the kmem_cache_node |
3510 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3511 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3512 | */ |
55136592 | 3513 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3514 | { |
3515 | struct page *page; | |
3516 | struct kmem_cache_node *n; | |
3517 | ||
51df1142 | 3518 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3519 | |
51df1142 | 3520 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3521 | |
3522 | BUG_ON(!page); | |
a2f92ee7 | 3523 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3524 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3525 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3526 | } |
3527 | ||
81819f0f CL |
3528 | n = page->freelist; |
3529 | BUG_ON(!n); | |
8ab1372f | 3530 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3531 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3532 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3533 | #endif |
12b22386 | 3534 | n = kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
505f5dcb | 3535 | GFP_KERNEL); |
12b22386 AK |
3536 | page->freelist = get_freepointer(kmem_cache_node, n); |
3537 | page->inuse = 1; | |
3538 | page->frozen = 0; | |
3539 | kmem_cache_node->node[node] = n; | |
4053497d | 3540 | init_kmem_cache_node(n); |
51df1142 | 3541 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3542 | |
67b6c900 | 3543 | /* |
1e4dd946 SR |
3544 | * No locks need to be taken here as it has just been |
3545 | * initialized and there is no concurrent access. | |
67b6c900 | 3546 | */ |
1e4dd946 | 3547 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3548 | } |
3549 | ||
3550 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3551 | { | |
3552 | int node; | |
fa45dc25 | 3553 | struct kmem_cache_node *n; |
81819f0f | 3554 | |
fa45dc25 | 3555 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3556 | s->node[node] = NULL; |
ea37df54 | 3557 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3558 | } |
3559 | } | |
3560 | ||
52b4b950 DS |
3561 | void __kmem_cache_release(struct kmem_cache *s) |
3562 | { | |
210e7a43 | 3563 | cache_random_seq_destroy(s); |
52b4b950 DS |
3564 | free_percpu(s->cpu_slab); |
3565 | free_kmem_cache_nodes(s); | |
3566 | } | |
3567 | ||
55136592 | 3568 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3569 | { |
3570 | int node; | |
81819f0f | 3571 | |
f64dc58c | 3572 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3573 | struct kmem_cache_node *n; |
3574 | ||
73367bd8 | 3575 | if (slab_state == DOWN) { |
55136592 | 3576 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3577 | continue; |
3578 | } | |
51df1142 | 3579 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3580 | GFP_KERNEL, node); |
81819f0f | 3581 | |
73367bd8 AD |
3582 | if (!n) { |
3583 | free_kmem_cache_nodes(s); | |
3584 | return 0; | |
81819f0f | 3585 | } |
73367bd8 | 3586 | |
4053497d | 3587 | init_kmem_cache_node(n); |
ea37df54 | 3588 | s->node[node] = n; |
81819f0f CL |
3589 | } |
3590 | return 1; | |
3591 | } | |
81819f0f | 3592 | |
c0bdb232 | 3593 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3594 | { |
3595 | if (min < MIN_PARTIAL) | |
3596 | min = MIN_PARTIAL; | |
3597 | else if (min > MAX_PARTIAL) | |
3598 | min = MAX_PARTIAL; | |
3599 | s->min_partial = min; | |
3600 | } | |
3601 | ||
e6d0e1dc WY |
3602 | static void set_cpu_partial(struct kmem_cache *s) |
3603 | { | |
3604 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3605 | /* | |
3606 | * cpu_partial determined the maximum number of objects kept in the | |
3607 | * per cpu partial lists of a processor. | |
3608 | * | |
3609 | * Per cpu partial lists mainly contain slabs that just have one | |
3610 | * object freed. If they are used for allocation then they can be | |
3611 | * filled up again with minimal effort. The slab will never hit the | |
3612 | * per node partial lists and therefore no locking will be required. | |
3613 | * | |
3614 | * This setting also determines | |
3615 | * | |
3616 | * A) The number of objects from per cpu partial slabs dumped to the | |
3617 | * per node list when we reach the limit. | |
3618 | * B) The number of objects in cpu partial slabs to extract from the | |
3619 | * per node list when we run out of per cpu objects. We only fetch | |
3620 | * 50% to keep some capacity around for frees. | |
3621 | */ | |
3622 | if (!kmem_cache_has_cpu_partial(s)) | |
bbd4e305 | 3623 | slub_set_cpu_partial(s, 0); |
e6d0e1dc | 3624 | else if (s->size >= PAGE_SIZE) |
bbd4e305 | 3625 | slub_set_cpu_partial(s, 2); |
e6d0e1dc | 3626 | else if (s->size >= 1024) |
bbd4e305 | 3627 | slub_set_cpu_partial(s, 6); |
e6d0e1dc | 3628 | else if (s->size >= 256) |
bbd4e305 | 3629 | slub_set_cpu_partial(s, 13); |
e6d0e1dc | 3630 | else |
bbd4e305 | 3631 | slub_set_cpu_partial(s, 30); |
e6d0e1dc WY |
3632 | #endif |
3633 | } | |
3634 | ||
81819f0f CL |
3635 | /* |
3636 | * calculate_sizes() determines the order and the distribution of data within | |
3637 | * a slab object. | |
3638 | */ | |
06b285dc | 3639 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 3640 | { |
d50112ed | 3641 | slab_flags_t flags = s->flags; |
be4a7988 | 3642 | unsigned int size = s->object_size; |
89b83f28 | 3643 | unsigned int freepointer_area; |
19af27af | 3644 | unsigned int order; |
81819f0f | 3645 | |
d8b42bf5 CL |
3646 | /* |
3647 | * Round up object size to the next word boundary. We can only | |
3648 | * place the free pointer at word boundaries and this determines | |
3649 | * the possible location of the free pointer. | |
3650 | */ | |
3651 | size = ALIGN(size, sizeof(void *)); | |
89b83f28 KC |
3652 | /* |
3653 | * This is the area of the object where a freepointer can be | |
3654 | * safely written. If redzoning adds more to the inuse size, we | |
3655 | * can't use that portion for writing the freepointer, so | |
3656 | * s->offset must be limited within this for the general case. | |
3657 | */ | |
3658 | freepointer_area = size; | |
d8b42bf5 CL |
3659 | |
3660 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3661 | /* |
3662 | * Determine if we can poison the object itself. If the user of | |
3663 | * the slab may touch the object after free or before allocation | |
3664 | * then we should never poison the object itself. | |
3665 | */ | |
5f0d5a3a | 3666 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3667 | !s->ctor) |
81819f0f CL |
3668 | s->flags |= __OBJECT_POISON; |
3669 | else | |
3670 | s->flags &= ~__OBJECT_POISON; | |
3671 | ||
81819f0f CL |
3672 | |
3673 | /* | |
672bba3a | 3674 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3675 | * end of the object and the free pointer. If not then add an |
672bba3a | 3676 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3677 | */ |
3b0efdfa | 3678 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3679 | size += sizeof(void *); |
41ecc55b | 3680 | #endif |
81819f0f CL |
3681 | |
3682 | /* | |
672bba3a CL |
3683 | * With that we have determined the number of bytes in actual use |
3684 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3685 | */ |
3686 | s->inuse = size; | |
3687 | ||
5f0d5a3a | 3688 | if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
c59def9f | 3689 | s->ctor)) { |
81819f0f CL |
3690 | /* |
3691 | * Relocate free pointer after the object if it is not | |
3692 | * permitted to overwrite the first word of the object on | |
3693 | * kmem_cache_free. | |
3694 | * | |
3695 | * This is the case if we do RCU, have a constructor or | |
3696 | * destructor or are poisoning the objects. | |
cbfc35a4 WL |
3697 | * |
3698 | * The assumption that s->offset >= s->inuse means free | |
3699 | * pointer is outside of the object is used in the | |
3700 | * freeptr_outside_object() function. If that is no | |
3701 | * longer true, the function needs to be modified. | |
81819f0f CL |
3702 | */ |
3703 | s->offset = size; | |
3704 | size += sizeof(void *); | |
89b83f28 | 3705 | } else if (freepointer_area > sizeof(void *)) { |
3202fa62 KC |
3706 | /* |
3707 | * Store freelist pointer near middle of object to keep | |
3708 | * it away from the edges of the object to avoid small | |
3709 | * sized over/underflows from neighboring allocations. | |
3710 | */ | |
89b83f28 | 3711 | s->offset = ALIGN(freepointer_area / 2, sizeof(void *)); |
81819f0f CL |
3712 | } |
3713 | ||
c12b3c62 | 3714 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3715 | if (flags & SLAB_STORE_USER) |
3716 | /* | |
3717 | * Need to store information about allocs and frees after | |
3718 | * the object. | |
3719 | */ | |
3720 | size += 2 * sizeof(struct track); | |
80a9201a | 3721 | #endif |
81819f0f | 3722 | |
80a9201a AP |
3723 | kasan_cache_create(s, &size, &s->flags); |
3724 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3725 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3726 | /* |
3727 | * Add some empty padding so that we can catch | |
3728 | * overwrites from earlier objects rather than let | |
3729 | * tracking information or the free pointer be | |
0211a9c8 | 3730 | * corrupted if a user writes before the start |
81819f0f CL |
3731 | * of the object. |
3732 | */ | |
3733 | size += sizeof(void *); | |
d86bd1be JK |
3734 | |
3735 | s->red_left_pad = sizeof(void *); | |
3736 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3737 | size += s->red_left_pad; | |
3738 | } | |
41ecc55b | 3739 | #endif |
672bba3a | 3740 | |
81819f0f CL |
3741 | /* |
3742 | * SLUB stores one object immediately after another beginning from | |
3743 | * offset 0. In order to align the objects we have to simply size | |
3744 | * each object to conform to the alignment. | |
3745 | */ | |
45906855 | 3746 | size = ALIGN(size, s->align); |
81819f0f | 3747 | s->size = size; |
4138fdfc | 3748 | s->reciprocal_size = reciprocal_value(size); |
06b285dc CL |
3749 | if (forced_order >= 0) |
3750 | order = forced_order; | |
3751 | else | |
9736d2a9 | 3752 | order = calculate_order(size); |
81819f0f | 3753 | |
19af27af | 3754 | if ((int)order < 0) |
81819f0f CL |
3755 | return 0; |
3756 | ||
b7a49f0d | 3757 | s->allocflags = 0; |
834f3d11 | 3758 | if (order) |
b7a49f0d CL |
3759 | s->allocflags |= __GFP_COMP; |
3760 | ||
3761 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3762 | s->allocflags |= GFP_DMA; |
b7a49f0d | 3763 | |
6d6ea1e9 NB |
3764 | if (s->flags & SLAB_CACHE_DMA32) |
3765 | s->allocflags |= GFP_DMA32; | |
3766 | ||
b7a49f0d CL |
3767 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
3768 | s->allocflags |= __GFP_RECLAIMABLE; | |
3769 | ||
81819f0f CL |
3770 | /* |
3771 | * Determine the number of objects per slab | |
3772 | */ | |
9736d2a9 MW |
3773 | s->oo = oo_make(order, size); |
3774 | s->min = oo_make(get_order(size), size); | |
205ab99d CL |
3775 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3776 | s->max = s->oo; | |
81819f0f | 3777 | |
834f3d11 | 3778 | return !!oo_objects(s->oo); |
81819f0f CL |
3779 | } |
3780 | ||
d50112ed | 3781 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 3782 | { |
8a13a4cc | 3783 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
2482ddec KC |
3784 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
3785 | s->random = get_random_long(); | |
3786 | #endif | |
81819f0f | 3787 | |
06b285dc | 3788 | if (!calculate_sizes(s, -1)) |
81819f0f | 3789 | goto error; |
3de47213 DR |
3790 | if (disable_higher_order_debug) { |
3791 | /* | |
3792 | * Disable debugging flags that store metadata if the min slab | |
3793 | * order increased. | |
3794 | */ | |
3b0efdfa | 3795 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3796 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3797 | s->offset = 0; | |
3798 | if (!calculate_sizes(s, -1)) | |
3799 | goto error; | |
3800 | } | |
3801 | } | |
81819f0f | 3802 | |
2565409f HC |
3803 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3804 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3805 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3806 | /* Enable fast mode */ |
3807 | s->flags |= __CMPXCHG_DOUBLE; | |
3808 | #endif | |
3809 | ||
3b89d7d8 DR |
3810 | /* |
3811 | * The larger the object size is, the more pages we want on the partial | |
3812 | * list to avoid pounding the page allocator excessively. | |
3813 | */ | |
49e22585 CL |
3814 | set_min_partial(s, ilog2(s->size) / 2); |
3815 | ||
e6d0e1dc | 3816 | set_cpu_partial(s); |
49e22585 | 3817 | |
81819f0f | 3818 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3819 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3820 | #endif |
210e7a43 TG |
3821 | |
3822 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3823 | if (slab_state >= UP) { | |
3824 | if (init_cache_random_seq(s)) | |
3825 | goto error; | |
3826 | } | |
3827 | ||
55136592 | 3828 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3829 | goto error; |
81819f0f | 3830 | |
55136592 | 3831 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3832 | return 0; |
ff12059e | 3833 | |
4c93c355 | 3834 | free_kmem_cache_nodes(s); |
81819f0f | 3835 | error: |
278b1bb1 | 3836 | return -EINVAL; |
81819f0f | 3837 | } |
81819f0f | 3838 | |
33b12c38 | 3839 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
55860d96 | 3840 | const char *text) |
33b12c38 CL |
3841 | { |
3842 | #ifdef CONFIG_SLUB_DEBUG | |
3843 | void *addr = page_address(page); | |
55860d96 | 3844 | unsigned long *map; |
33b12c38 | 3845 | void *p; |
aa456c7a | 3846 | |
945cf2b6 | 3847 | slab_err(s, page, text, s->name); |
33b12c38 | 3848 | slab_lock(page); |
33b12c38 | 3849 | |
90e9f6a6 | 3850 | map = get_map(s, page); |
33b12c38 CL |
3851 | for_each_object(p, s, addr, page->objects) { |
3852 | ||
4138fdfc | 3853 | if (!test_bit(__obj_to_index(s, addr, p), map)) { |
f9f58285 | 3854 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3855 | print_tracking(s, p); |
3856 | } | |
3857 | } | |
55860d96 | 3858 | put_map(map); |
33b12c38 CL |
3859 | slab_unlock(page); |
3860 | #endif | |
3861 | } | |
3862 | ||
81819f0f | 3863 | /* |
599870b1 | 3864 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3865 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3866 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3867 | */ |
599870b1 | 3868 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3869 | { |
60398923 | 3870 | LIST_HEAD(discard); |
81819f0f CL |
3871 | struct page *page, *h; |
3872 | ||
52b4b950 DS |
3873 | BUG_ON(irqs_disabled()); |
3874 | spin_lock_irq(&n->list_lock); | |
916ac052 | 3875 | list_for_each_entry_safe(page, h, &n->partial, slab_list) { |
81819f0f | 3876 | if (!page->inuse) { |
52b4b950 | 3877 | remove_partial(n, page); |
916ac052 | 3878 | list_add(&page->slab_list, &discard); |
33b12c38 CL |
3879 | } else { |
3880 | list_slab_objects(s, page, | |
55860d96 | 3881 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3882 | } |
33b12c38 | 3883 | } |
52b4b950 | 3884 | spin_unlock_irq(&n->list_lock); |
60398923 | 3885 | |
916ac052 | 3886 | list_for_each_entry_safe(page, h, &discard, slab_list) |
60398923 | 3887 | discard_slab(s, page); |
81819f0f CL |
3888 | } |
3889 | ||
f9e13c0a SB |
3890 | bool __kmem_cache_empty(struct kmem_cache *s) |
3891 | { | |
3892 | int node; | |
3893 | struct kmem_cache_node *n; | |
3894 | ||
3895 | for_each_kmem_cache_node(s, node, n) | |
3896 | if (n->nr_partial || slabs_node(s, node)) | |
3897 | return false; | |
3898 | return true; | |
3899 | } | |
3900 | ||
81819f0f | 3901 | /* |
672bba3a | 3902 | * Release all resources used by a slab cache. |
81819f0f | 3903 | */ |
52b4b950 | 3904 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3905 | { |
3906 | int node; | |
fa45dc25 | 3907 | struct kmem_cache_node *n; |
81819f0f CL |
3908 | |
3909 | flush_all(s); | |
81819f0f | 3910 | /* Attempt to free all objects */ |
fa45dc25 | 3911 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3912 | free_partial(s, n); |
3913 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3914 | return 1; |
3915 | } | |
81819f0f CL |
3916 | return 0; |
3917 | } | |
3918 | ||
81819f0f CL |
3919 | /******************************************************************** |
3920 | * Kmalloc subsystem | |
3921 | *******************************************************************/ | |
3922 | ||
81819f0f CL |
3923 | static int __init setup_slub_min_order(char *str) |
3924 | { | |
19af27af | 3925 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
3926 | |
3927 | return 1; | |
3928 | } | |
3929 | ||
3930 | __setup("slub_min_order=", setup_slub_min_order); | |
3931 | ||
3932 | static int __init setup_slub_max_order(char *str) | |
3933 | { | |
19af27af AD |
3934 | get_option(&str, (int *)&slub_max_order); |
3935 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
3936 | |
3937 | return 1; | |
3938 | } | |
3939 | ||
3940 | __setup("slub_max_order=", setup_slub_max_order); | |
3941 | ||
3942 | static int __init setup_slub_min_objects(char *str) | |
3943 | { | |
19af27af | 3944 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
3945 | |
3946 | return 1; | |
3947 | } | |
3948 | ||
3949 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3950 | ||
81819f0f CL |
3951 | void *__kmalloc(size_t size, gfp_t flags) |
3952 | { | |
aadb4bc4 | 3953 | struct kmem_cache *s; |
5b882be4 | 3954 | void *ret; |
81819f0f | 3955 | |
95a05b42 | 3956 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3957 | return kmalloc_large(size, flags); |
aadb4bc4 | 3958 | |
2c59dd65 | 3959 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3960 | |
3961 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3962 | return s; |
3963 | ||
2b847c3c | 3964 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3965 | |
ca2b84cb | 3966 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3967 | |
0116523c | 3968 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3969 | |
5b882be4 | 3970 | return ret; |
81819f0f CL |
3971 | } |
3972 | EXPORT_SYMBOL(__kmalloc); | |
3973 | ||
5d1f57e4 | 3974 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3975 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3976 | { | |
b1eeab67 | 3977 | struct page *page; |
e4f7c0b4 | 3978 | void *ptr = NULL; |
6a486c0a | 3979 | unsigned int order = get_order(size); |
f619cfe1 | 3980 | |
75f296d9 | 3981 | flags |= __GFP_COMP; |
6a486c0a VB |
3982 | page = alloc_pages_node(node, flags, order); |
3983 | if (page) { | |
e4f7c0b4 | 3984 | ptr = page_address(page); |
d42f3245 RG |
3985 | mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE_B, |
3986 | PAGE_SIZE << order); | |
6a486c0a | 3987 | } |
e4f7c0b4 | 3988 | |
0116523c | 3989 | return kmalloc_large_node_hook(ptr, size, flags); |
f619cfe1 CL |
3990 | } |
3991 | ||
81819f0f CL |
3992 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3993 | { | |
aadb4bc4 | 3994 | struct kmem_cache *s; |
5b882be4 | 3995 | void *ret; |
81819f0f | 3996 | |
95a05b42 | 3997 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3998 | ret = kmalloc_large_node(size, flags, node); |
3999 | ||
ca2b84cb EGM |
4000 | trace_kmalloc_node(_RET_IP_, ret, |
4001 | size, PAGE_SIZE << get_order(size), | |
4002 | flags, node); | |
5b882be4 EGM |
4003 | |
4004 | return ret; | |
4005 | } | |
aadb4bc4 | 4006 | |
2c59dd65 | 4007 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4008 | |
4009 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4010 | return s; |
4011 | ||
2b847c3c | 4012 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 4013 | |
ca2b84cb | 4014 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 4015 | |
0116523c | 4016 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4017 | |
5b882be4 | 4018 | return ret; |
81819f0f CL |
4019 | } |
4020 | EXPORT_SYMBOL(__kmalloc_node); | |
6dfd1b65 | 4021 | #endif /* CONFIG_NUMA */ |
81819f0f | 4022 | |
ed18adc1 KC |
4023 | #ifdef CONFIG_HARDENED_USERCOPY |
4024 | /* | |
afcc90f8 KC |
4025 | * Rejects incorrectly sized objects and objects that are to be copied |
4026 | * to/from userspace but do not fall entirely within the containing slab | |
4027 | * cache's usercopy region. | |
ed18adc1 KC |
4028 | * |
4029 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4030 | * to indicate an error. | |
4031 | */ | |
f4e6e289 KC |
4032 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
4033 | bool to_user) | |
ed18adc1 KC |
4034 | { |
4035 | struct kmem_cache *s; | |
44065b2e | 4036 | unsigned int offset; |
ed18adc1 KC |
4037 | size_t object_size; |
4038 | ||
96fedce2 AK |
4039 | ptr = kasan_reset_tag(ptr); |
4040 | ||
ed18adc1 KC |
4041 | /* Find object and usable object size. */ |
4042 | s = page->slab_cache; | |
ed18adc1 KC |
4043 | |
4044 | /* Reject impossible pointers. */ | |
4045 | if (ptr < page_address(page)) | |
f4e6e289 KC |
4046 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
4047 | to_user, 0, n); | |
ed18adc1 KC |
4048 | |
4049 | /* Find offset within object. */ | |
4050 | offset = (ptr - page_address(page)) % s->size; | |
4051 | ||
4052 | /* Adjust for redzone and reject if within the redzone. */ | |
59052e89 | 4053 | if (kmem_cache_debug_flags(s, SLAB_RED_ZONE)) { |
ed18adc1 | 4054 | if (offset < s->red_left_pad) |
f4e6e289 KC |
4055 | usercopy_abort("SLUB object in left red zone", |
4056 | s->name, to_user, offset, n); | |
ed18adc1 KC |
4057 | offset -= s->red_left_pad; |
4058 | } | |
4059 | ||
afcc90f8 KC |
4060 | /* Allow address range falling entirely within usercopy region. */ |
4061 | if (offset >= s->useroffset && | |
4062 | offset - s->useroffset <= s->usersize && | |
4063 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 4064 | return; |
ed18adc1 | 4065 | |
afcc90f8 KC |
4066 | /* |
4067 | * If the copy is still within the allocated object, produce | |
4068 | * a warning instead of rejecting the copy. This is intended | |
4069 | * to be a temporary method to find any missing usercopy | |
4070 | * whitelists. | |
4071 | */ | |
4072 | object_size = slab_ksize(s); | |
2d891fbc KC |
4073 | if (usercopy_fallback && |
4074 | offset <= object_size && n <= object_size - offset) { | |
afcc90f8 KC |
4075 | usercopy_warn("SLUB object", s->name, to_user, offset, n); |
4076 | return; | |
4077 | } | |
ed18adc1 | 4078 | |
f4e6e289 | 4079 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
4080 | } |
4081 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4082 | ||
10d1f8cb | 4083 | size_t __ksize(const void *object) |
81819f0f | 4084 | { |
272c1d21 | 4085 | struct page *page; |
81819f0f | 4086 | |
ef8b4520 | 4087 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
4088 | return 0; |
4089 | ||
294a80a8 | 4090 | page = virt_to_head_page(object); |
294a80a8 | 4091 | |
76994412 PE |
4092 | if (unlikely(!PageSlab(page))) { |
4093 | WARN_ON(!PageCompound(page)); | |
a50b854e | 4094 | return page_size(page); |
76994412 | 4095 | } |
81819f0f | 4096 | |
1b4f59e3 | 4097 | return slab_ksize(page->slab_cache); |
81819f0f | 4098 | } |
10d1f8cb | 4099 | EXPORT_SYMBOL(__ksize); |
81819f0f CL |
4100 | |
4101 | void kfree(const void *x) | |
4102 | { | |
81819f0f | 4103 | struct page *page; |
5bb983b0 | 4104 | void *object = (void *)x; |
81819f0f | 4105 | |
2121db74 PE |
4106 | trace_kfree(_RET_IP_, x); |
4107 | ||
2408c550 | 4108 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
4109 | return; |
4110 | ||
b49af68f | 4111 | page = virt_to_head_page(x); |
aadb4bc4 | 4112 | if (unlikely(!PageSlab(page))) { |
6a486c0a VB |
4113 | unsigned int order = compound_order(page); |
4114 | ||
0937502a | 4115 | BUG_ON(!PageCompound(page)); |
47adccce | 4116 | kfree_hook(object); |
d42f3245 RG |
4117 | mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE_B, |
4118 | -(PAGE_SIZE << order)); | |
6a486c0a | 4119 | __free_pages(page, order); |
aadb4bc4 CL |
4120 | return; |
4121 | } | |
81084651 | 4122 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
4123 | } |
4124 | EXPORT_SYMBOL(kfree); | |
4125 | ||
832f37f5 VD |
4126 | #define SHRINK_PROMOTE_MAX 32 |
4127 | ||
2086d26a | 4128 | /* |
832f37f5 VD |
4129 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
4130 | * up most to the head of the partial lists. New allocations will then | |
4131 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
4132 | * |
4133 | * The slabs with the least items are placed last. This results in them | |
4134 | * being allocated from last increasing the chance that the last objects | |
4135 | * are freed in them. | |
2086d26a | 4136 | */ |
c9fc5864 | 4137 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
4138 | { |
4139 | int node; | |
4140 | int i; | |
4141 | struct kmem_cache_node *n; | |
4142 | struct page *page; | |
4143 | struct page *t; | |
832f37f5 VD |
4144 | struct list_head discard; |
4145 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 4146 | unsigned long flags; |
ce3712d7 | 4147 | int ret = 0; |
2086d26a | 4148 | |
2086d26a | 4149 | flush_all(s); |
fa45dc25 | 4150 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
4151 | INIT_LIST_HEAD(&discard); |
4152 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
4153 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
4154 | |
4155 | spin_lock_irqsave(&n->list_lock, flags); | |
4156 | ||
4157 | /* | |
832f37f5 | 4158 | * Build lists of slabs to discard or promote. |
2086d26a | 4159 | * |
672bba3a CL |
4160 | * Note that concurrent frees may occur while we hold the |
4161 | * list_lock. page->inuse here is the upper limit. | |
2086d26a | 4162 | */ |
916ac052 | 4163 | list_for_each_entry_safe(page, t, &n->partial, slab_list) { |
832f37f5 VD |
4164 | int free = page->objects - page->inuse; |
4165 | ||
4166 | /* Do not reread page->inuse */ | |
4167 | barrier(); | |
4168 | ||
4169 | /* We do not keep full slabs on the list */ | |
4170 | BUG_ON(free <= 0); | |
4171 | ||
4172 | if (free == page->objects) { | |
916ac052 | 4173 | list_move(&page->slab_list, &discard); |
69cb8e6b | 4174 | n->nr_partial--; |
832f37f5 | 4175 | } else if (free <= SHRINK_PROMOTE_MAX) |
916ac052 | 4176 | list_move(&page->slab_list, promote + free - 1); |
2086d26a CL |
4177 | } |
4178 | ||
2086d26a | 4179 | /* |
832f37f5 VD |
4180 | * Promote the slabs filled up most to the head of the |
4181 | * partial list. | |
2086d26a | 4182 | */ |
832f37f5 VD |
4183 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
4184 | list_splice(promote + i, &n->partial); | |
2086d26a | 4185 | |
2086d26a | 4186 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
4187 | |
4188 | /* Release empty slabs */ | |
916ac052 | 4189 | list_for_each_entry_safe(page, t, &discard, slab_list) |
69cb8e6b | 4190 | discard_slab(s, page); |
ce3712d7 VD |
4191 | |
4192 | if (slabs_node(s, node)) | |
4193 | ret = 1; | |
2086d26a CL |
4194 | } |
4195 | ||
ce3712d7 | 4196 | return ret; |
2086d26a | 4197 | } |
2086d26a | 4198 | |
b9049e23 YG |
4199 | static int slab_mem_going_offline_callback(void *arg) |
4200 | { | |
4201 | struct kmem_cache *s; | |
4202 | ||
18004c5d | 4203 | mutex_lock(&slab_mutex); |
b9049e23 | 4204 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4205 | __kmem_cache_shrink(s); |
18004c5d | 4206 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4207 | |
4208 | return 0; | |
4209 | } | |
4210 | ||
4211 | static void slab_mem_offline_callback(void *arg) | |
4212 | { | |
4213 | struct kmem_cache_node *n; | |
4214 | struct kmem_cache *s; | |
4215 | struct memory_notify *marg = arg; | |
4216 | int offline_node; | |
4217 | ||
b9d5ab25 | 4218 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4219 | |
4220 | /* | |
4221 | * If the node still has available memory. we need kmem_cache_node | |
4222 | * for it yet. | |
4223 | */ | |
4224 | if (offline_node < 0) | |
4225 | return; | |
4226 | ||
18004c5d | 4227 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4228 | list_for_each_entry(s, &slab_caches, list) { |
4229 | n = get_node(s, offline_node); | |
4230 | if (n) { | |
4231 | /* | |
4232 | * if n->nr_slabs > 0, slabs still exist on the node | |
4233 | * that is going down. We were unable to free them, | |
c9404c9c | 4234 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
4235 | * callback. So, we must fail. |
4236 | */ | |
0f389ec6 | 4237 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
4238 | |
4239 | s->node[offline_node] = NULL; | |
8de66a0c | 4240 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
4241 | } |
4242 | } | |
18004c5d | 4243 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4244 | } |
4245 | ||
4246 | static int slab_mem_going_online_callback(void *arg) | |
4247 | { | |
4248 | struct kmem_cache_node *n; | |
4249 | struct kmem_cache *s; | |
4250 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4251 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4252 | int ret = 0; |
4253 | ||
4254 | /* | |
4255 | * If the node's memory is already available, then kmem_cache_node is | |
4256 | * already created. Nothing to do. | |
4257 | */ | |
4258 | if (nid < 0) | |
4259 | return 0; | |
4260 | ||
4261 | /* | |
0121c619 | 4262 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4263 | * allocate a kmem_cache_node structure in order to bring the node |
4264 | * online. | |
4265 | */ | |
18004c5d | 4266 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4267 | list_for_each_entry(s, &slab_caches, list) { |
4268 | /* | |
4269 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4270 | * since memory is not yet available from the node that | |
4271 | * is brought up. | |
4272 | */ | |
8de66a0c | 4273 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4274 | if (!n) { |
4275 | ret = -ENOMEM; | |
4276 | goto out; | |
4277 | } | |
4053497d | 4278 | init_kmem_cache_node(n); |
b9049e23 YG |
4279 | s->node[nid] = n; |
4280 | } | |
4281 | out: | |
18004c5d | 4282 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4283 | return ret; |
4284 | } | |
4285 | ||
4286 | static int slab_memory_callback(struct notifier_block *self, | |
4287 | unsigned long action, void *arg) | |
4288 | { | |
4289 | int ret = 0; | |
4290 | ||
4291 | switch (action) { | |
4292 | case MEM_GOING_ONLINE: | |
4293 | ret = slab_mem_going_online_callback(arg); | |
4294 | break; | |
4295 | case MEM_GOING_OFFLINE: | |
4296 | ret = slab_mem_going_offline_callback(arg); | |
4297 | break; | |
4298 | case MEM_OFFLINE: | |
4299 | case MEM_CANCEL_ONLINE: | |
4300 | slab_mem_offline_callback(arg); | |
4301 | break; | |
4302 | case MEM_ONLINE: | |
4303 | case MEM_CANCEL_OFFLINE: | |
4304 | break; | |
4305 | } | |
dc19f9db KH |
4306 | if (ret) |
4307 | ret = notifier_from_errno(ret); | |
4308 | else | |
4309 | ret = NOTIFY_OK; | |
b9049e23 YG |
4310 | return ret; |
4311 | } | |
4312 | ||
3ac38faa AM |
4313 | static struct notifier_block slab_memory_callback_nb = { |
4314 | .notifier_call = slab_memory_callback, | |
4315 | .priority = SLAB_CALLBACK_PRI, | |
4316 | }; | |
b9049e23 | 4317 | |
81819f0f CL |
4318 | /******************************************************************** |
4319 | * Basic setup of slabs | |
4320 | *******************************************************************/ | |
4321 | ||
51df1142 CL |
4322 | /* |
4323 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4324 | * the page allocator. Allocate them properly then fix up the pointers |
4325 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4326 | */ |
4327 | ||
dffb4d60 | 4328 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4329 | { |
4330 | int node; | |
dffb4d60 | 4331 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4332 | struct kmem_cache_node *n; |
51df1142 | 4333 | |
dffb4d60 | 4334 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4335 | |
7d557b3c GC |
4336 | /* |
4337 | * This runs very early, and only the boot processor is supposed to be | |
4338 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4339 | * IPIs around. | |
4340 | */ | |
4341 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4342 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4343 | struct page *p; |
4344 | ||
916ac052 | 4345 | list_for_each_entry(p, &n->partial, slab_list) |
fa45dc25 | 4346 | p->slab_cache = s; |
51df1142 | 4347 | |
607bf324 | 4348 | #ifdef CONFIG_SLUB_DEBUG |
916ac052 | 4349 | list_for_each_entry(p, &n->full, slab_list) |
fa45dc25 | 4350 | p->slab_cache = s; |
51df1142 | 4351 | #endif |
51df1142 | 4352 | } |
dffb4d60 CL |
4353 | list_add(&s->list, &slab_caches); |
4354 | return s; | |
51df1142 CL |
4355 | } |
4356 | ||
81819f0f CL |
4357 | void __init kmem_cache_init(void) |
4358 | { | |
dffb4d60 CL |
4359 | static __initdata struct kmem_cache boot_kmem_cache, |
4360 | boot_kmem_cache_node; | |
51df1142 | 4361 | |
fc8d8620 SG |
4362 | if (debug_guardpage_minorder()) |
4363 | slub_max_order = 0; | |
4364 | ||
dffb4d60 CL |
4365 | kmem_cache_node = &boot_kmem_cache_node; |
4366 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4367 | |
dffb4d60 | 4368 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4369 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4370 | |
3ac38faa | 4371 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4372 | |
4373 | /* Able to allocate the per node structures */ | |
4374 | slab_state = PARTIAL; | |
4375 | ||
dffb4d60 CL |
4376 | create_boot_cache(kmem_cache, "kmem_cache", |
4377 | offsetof(struct kmem_cache, node) + | |
4378 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4379 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4380 | |
dffb4d60 | 4381 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4382 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4383 | |
4384 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4385 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4386 | create_kmalloc_caches(0); |
81819f0f | 4387 | |
210e7a43 TG |
4388 | /* Setup random freelists for each cache */ |
4389 | init_freelist_randomization(); | |
4390 | ||
a96a87bf SAS |
4391 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4392 | slub_cpu_dead); | |
81819f0f | 4393 | |
b9726c26 | 4394 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", |
f97d5f63 | 4395 | cache_line_size(), |
81819f0f CL |
4396 | slub_min_order, slub_max_order, slub_min_objects, |
4397 | nr_cpu_ids, nr_node_ids); | |
4398 | } | |
4399 | ||
7e85ee0c PE |
4400 | void __init kmem_cache_init_late(void) |
4401 | { | |
7e85ee0c PE |
4402 | } |
4403 | ||
2633d7a0 | 4404 | struct kmem_cache * |
f4957d5b | 4405 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4406 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4407 | { |
10befea9 | 4408 | struct kmem_cache *s; |
81819f0f | 4409 | |
a44cb944 | 4410 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4411 | if (s) { |
4412 | s->refcount++; | |
84d0ddd6 | 4413 | |
81819f0f CL |
4414 | /* |
4415 | * Adjust the object sizes so that we clear | |
4416 | * the complete object on kzalloc. | |
4417 | */ | |
1b473f29 | 4418 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4419 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4420 | |
7b8f3b66 | 4421 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4422 | s->refcount--; |
cbb79694 | 4423 | s = NULL; |
7b8f3b66 | 4424 | } |
a0e1d1be | 4425 | } |
6446faa2 | 4426 | |
cbb79694 CL |
4427 | return s; |
4428 | } | |
84c1cf62 | 4429 | |
d50112ed | 4430 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4431 | { |
aac3a166 PE |
4432 | int err; |
4433 | ||
4434 | err = kmem_cache_open(s, flags); | |
4435 | if (err) | |
4436 | return err; | |
20cea968 | 4437 | |
45530c44 CL |
4438 | /* Mutex is not taken during early boot */ |
4439 | if (slab_state <= UP) | |
4440 | return 0; | |
4441 | ||
aac3a166 | 4442 | err = sysfs_slab_add(s); |
aac3a166 | 4443 | if (err) |
52b4b950 | 4444 | __kmem_cache_release(s); |
20cea968 | 4445 | |
aac3a166 | 4446 | return err; |
81819f0f | 4447 | } |
81819f0f | 4448 | |
ce71e27c | 4449 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4450 | { |
aadb4bc4 | 4451 | struct kmem_cache *s; |
94b528d0 | 4452 | void *ret; |
aadb4bc4 | 4453 | |
95a05b42 | 4454 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4455 | return kmalloc_large(size, gfpflags); |
4456 | ||
2c59dd65 | 4457 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4458 | |
2408c550 | 4459 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4460 | return s; |
81819f0f | 4461 | |
2b847c3c | 4462 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4463 | |
25985edc | 4464 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4465 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4466 | |
4467 | return ret; | |
81819f0f | 4468 | } |
fd7cb575 | 4469 | EXPORT_SYMBOL(__kmalloc_track_caller); |
81819f0f | 4470 | |
5d1f57e4 | 4471 | #ifdef CONFIG_NUMA |
81819f0f | 4472 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4473 | int node, unsigned long caller) |
81819f0f | 4474 | { |
aadb4bc4 | 4475 | struct kmem_cache *s; |
94b528d0 | 4476 | void *ret; |
aadb4bc4 | 4477 | |
95a05b42 | 4478 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4479 | ret = kmalloc_large_node(size, gfpflags, node); |
4480 | ||
4481 | trace_kmalloc_node(caller, ret, | |
4482 | size, PAGE_SIZE << get_order(size), | |
4483 | gfpflags, node); | |
4484 | ||
4485 | return ret; | |
4486 | } | |
eada35ef | 4487 | |
2c59dd65 | 4488 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4489 | |
2408c550 | 4490 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4491 | return s; |
81819f0f | 4492 | |
2b847c3c | 4493 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4494 | |
25985edc | 4495 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4496 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4497 | |
4498 | return ret; | |
81819f0f | 4499 | } |
fd7cb575 | 4500 | EXPORT_SYMBOL(__kmalloc_node_track_caller); |
5d1f57e4 | 4501 | #endif |
81819f0f | 4502 | |
ab4d5ed5 | 4503 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4504 | static int count_inuse(struct page *page) |
4505 | { | |
4506 | return page->inuse; | |
4507 | } | |
4508 | ||
4509 | static int count_total(struct page *page) | |
4510 | { | |
4511 | return page->objects; | |
4512 | } | |
ab4d5ed5 | 4513 | #endif |
205ab99d | 4514 | |
ab4d5ed5 | 4515 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 | 4516 | static void validate_slab(struct kmem_cache *s, struct page *page) |
53e15af0 CL |
4517 | { |
4518 | void *p; | |
a973e9dd | 4519 | void *addr = page_address(page); |
90e9f6a6 YZ |
4520 | unsigned long *map; |
4521 | ||
4522 | slab_lock(page); | |
53e15af0 | 4523 | |
dd98afd4 | 4524 | if (!check_slab(s, page) || !on_freelist(s, page, NULL)) |
90e9f6a6 | 4525 | goto unlock; |
53e15af0 CL |
4526 | |
4527 | /* Now we know that a valid freelist exists */ | |
90e9f6a6 | 4528 | map = get_map(s, page); |
5f80b13a | 4529 | for_each_object(p, s, addr, page->objects) { |
4138fdfc | 4530 | u8 val = test_bit(__obj_to_index(s, addr, p), map) ? |
dd98afd4 | 4531 | SLUB_RED_INACTIVE : SLUB_RED_ACTIVE; |
53e15af0 | 4532 | |
dd98afd4 YZ |
4533 | if (!check_object(s, page, p, val)) |
4534 | break; | |
4535 | } | |
90e9f6a6 YZ |
4536 | put_map(map); |
4537 | unlock: | |
881db7fb | 4538 | slab_unlock(page); |
53e15af0 CL |
4539 | } |
4540 | ||
434e245d | 4541 | static int validate_slab_node(struct kmem_cache *s, |
90e9f6a6 | 4542 | struct kmem_cache_node *n) |
53e15af0 CL |
4543 | { |
4544 | unsigned long count = 0; | |
4545 | struct page *page; | |
4546 | unsigned long flags; | |
4547 | ||
4548 | spin_lock_irqsave(&n->list_lock, flags); | |
4549 | ||
916ac052 | 4550 | list_for_each_entry(page, &n->partial, slab_list) { |
90e9f6a6 | 4551 | validate_slab(s, page); |
53e15af0 CL |
4552 | count++; |
4553 | } | |
4554 | if (count != n->nr_partial) | |
f9f58285 FF |
4555 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4556 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4557 | |
4558 | if (!(s->flags & SLAB_STORE_USER)) | |
4559 | goto out; | |
4560 | ||
916ac052 | 4561 | list_for_each_entry(page, &n->full, slab_list) { |
90e9f6a6 | 4562 | validate_slab(s, page); |
53e15af0 CL |
4563 | count++; |
4564 | } | |
4565 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4566 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4567 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4568 | |
4569 | out: | |
4570 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4571 | return count; | |
4572 | } | |
4573 | ||
434e245d | 4574 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4575 | { |
4576 | int node; | |
4577 | unsigned long count = 0; | |
fa45dc25 | 4578 | struct kmem_cache_node *n; |
53e15af0 CL |
4579 | |
4580 | flush_all(s); | |
fa45dc25 | 4581 | for_each_kmem_cache_node(s, node, n) |
90e9f6a6 YZ |
4582 | count += validate_slab_node(s, n); |
4583 | ||
53e15af0 CL |
4584 | return count; |
4585 | } | |
88a420e4 | 4586 | /* |
672bba3a | 4587 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4588 | * and freed. |
4589 | */ | |
4590 | ||
4591 | struct location { | |
4592 | unsigned long count; | |
ce71e27c | 4593 | unsigned long addr; |
45edfa58 CL |
4594 | long long sum_time; |
4595 | long min_time; | |
4596 | long max_time; | |
4597 | long min_pid; | |
4598 | long max_pid; | |
174596a0 | 4599 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4600 | nodemask_t nodes; |
88a420e4 CL |
4601 | }; |
4602 | ||
4603 | struct loc_track { | |
4604 | unsigned long max; | |
4605 | unsigned long count; | |
4606 | struct location *loc; | |
4607 | }; | |
4608 | ||
4609 | static void free_loc_track(struct loc_track *t) | |
4610 | { | |
4611 | if (t->max) | |
4612 | free_pages((unsigned long)t->loc, | |
4613 | get_order(sizeof(struct location) * t->max)); | |
4614 | } | |
4615 | ||
68dff6a9 | 4616 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4617 | { |
4618 | struct location *l; | |
4619 | int order; | |
4620 | ||
88a420e4 CL |
4621 | order = get_order(sizeof(struct location) * max); |
4622 | ||
68dff6a9 | 4623 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4624 | if (!l) |
4625 | return 0; | |
4626 | ||
4627 | if (t->count) { | |
4628 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4629 | free_loc_track(t); | |
4630 | } | |
4631 | t->max = max; | |
4632 | t->loc = l; | |
4633 | return 1; | |
4634 | } | |
4635 | ||
4636 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4637 | const struct track *track) |
88a420e4 CL |
4638 | { |
4639 | long start, end, pos; | |
4640 | struct location *l; | |
ce71e27c | 4641 | unsigned long caddr; |
45edfa58 | 4642 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4643 | |
4644 | start = -1; | |
4645 | end = t->count; | |
4646 | ||
4647 | for ( ; ; ) { | |
4648 | pos = start + (end - start + 1) / 2; | |
4649 | ||
4650 | /* | |
4651 | * There is nothing at "end". If we end up there | |
4652 | * we need to add something to before end. | |
4653 | */ | |
4654 | if (pos == end) | |
4655 | break; | |
4656 | ||
4657 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4658 | if (track->addr == caddr) { |
4659 | ||
4660 | l = &t->loc[pos]; | |
4661 | l->count++; | |
4662 | if (track->when) { | |
4663 | l->sum_time += age; | |
4664 | if (age < l->min_time) | |
4665 | l->min_time = age; | |
4666 | if (age > l->max_time) | |
4667 | l->max_time = age; | |
4668 | ||
4669 | if (track->pid < l->min_pid) | |
4670 | l->min_pid = track->pid; | |
4671 | if (track->pid > l->max_pid) | |
4672 | l->max_pid = track->pid; | |
4673 | ||
174596a0 RR |
4674 | cpumask_set_cpu(track->cpu, |
4675 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4676 | } |
4677 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4678 | return 1; |
4679 | } | |
4680 | ||
45edfa58 | 4681 | if (track->addr < caddr) |
88a420e4 CL |
4682 | end = pos; |
4683 | else | |
4684 | start = pos; | |
4685 | } | |
4686 | ||
4687 | /* | |
672bba3a | 4688 | * Not found. Insert new tracking element. |
88a420e4 | 4689 | */ |
68dff6a9 | 4690 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4691 | return 0; |
4692 | ||
4693 | l = t->loc + pos; | |
4694 | if (pos < t->count) | |
4695 | memmove(l + 1, l, | |
4696 | (t->count - pos) * sizeof(struct location)); | |
4697 | t->count++; | |
4698 | l->count = 1; | |
45edfa58 CL |
4699 | l->addr = track->addr; |
4700 | l->sum_time = age; | |
4701 | l->min_time = age; | |
4702 | l->max_time = age; | |
4703 | l->min_pid = track->pid; | |
4704 | l->max_pid = track->pid; | |
174596a0 RR |
4705 | cpumask_clear(to_cpumask(l->cpus)); |
4706 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4707 | nodes_clear(l->nodes); |
4708 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4709 | return 1; |
4710 | } | |
4711 | ||
4712 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
90e9f6a6 | 4713 | struct page *page, enum track_item alloc) |
88a420e4 | 4714 | { |
a973e9dd | 4715 | void *addr = page_address(page); |
88a420e4 | 4716 | void *p; |
90e9f6a6 | 4717 | unsigned long *map; |
88a420e4 | 4718 | |
90e9f6a6 | 4719 | map = get_map(s, page); |
224a88be | 4720 | for_each_object(p, s, addr, page->objects) |
4138fdfc | 4721 | if (!test_bit(__obj_to_index(s, addr, p), map)) |
45edfa58 | 4722 | add_location(t, s, get_track(s, p, alloc)); |
90e9f6a6 | 4723 | put_map(map); |
88a420e4 CL |
4724 | } |
4725 | ||
4726 | static int list_locations(struct kmem_cache *s, char *buf, | |
4727 | enum track_item alloc) | |
4728 | { | |
e374d483 | 4729 | int len = 0; |
88a420e4 | 4730 | unsigned long i; |
68dff6a9 | 4731 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4732 | int node; |
fa45dc25 | 4733 | struct kmem_cache_node *n; |
88a420e4 | 4734 | |
90e9f6a6 YZ |
4735 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4736 | GFP_KERNEL)) { | |
68dff6a9 | 4737 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4738 | } |
88a420e4 CL |
4739 | /* Push back cpu slabs */ |
4740 | flush_all(s); | |
4741 | ||
fa45dc25 | 4742 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4743 | unsigned long flags; |
4744 | struct page *page; | |
4745 | ||
9e86943b | 4746 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4747 | continue; |
4748 | ||
4749 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 4750 | list_for_each_entry(page, &n->partial, slab_list) |
90e9f6a6 | 4751 | process_slab(&t, s, page, alloc); |
916ac052 | 4752 | list_for_each_entry(page, &n->full, slab_list) |
90e9f6a6 | 4753 | process_slab(&t, s, page, alloc); |
88a420e4 CL |
4754 | spin_unlock_irqrestore(&n->list_lock, flags); |
4755 | } | |
4756 | ||
4757 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4758 | struct location *l = &t.loc[i]; |
88a420e4 | 4759 | |
9c246247 | 4760 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4761 | break; |
e374d483 | 4762 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4763 | |
4764 | if (l->addr) | |
62c70bce | 4765 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4766 | else |
e374d483 | 4767 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4768 | |
4769 | if (l->sum_time != l->min_time) { | |
e374d483 | 4770 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4771 | l->min_time, |
4772 | (long)div_u64(l->sum_time, l->count), | |
4773 | l->max_time); | |
45edfa58 | 4774 | } else |
e374d483 | 4775 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4776 | l->min_time); |
4777 | ||
4778 | if (l->min_pid != l->max_pid) | |
e374d483 | 4779 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4780 | l->min_pid, l->max_pid); |
4781 | else | |
e374d483 | 4782 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4783 | l->min_pid); |
4784 | ||
174596a0 RR |
4785 | if (num_online_cpus() > 1 && |
4786 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4787 | len < PAGE_SIZE - 60) |
4788 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4789 | " cpus=%*pbl", | |
4790 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4791 | |
62bc62a8 | 4792 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4793 | len < PAGE_SIZE - 60) |
4794 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4795 | " nodes=%*pbl", | |
4796 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4797 | |
e374d483 | 4798 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4799 | } |
4800 | ||
4801 | free_loc_track(&t); | |
4802 | if (!t.count) | |
e374d483 HH |
4803 | len += sprintf(buf, "No data\n"); |
4804 | return len; | |
88a420e4 | 4805 | } |
6dfd1b65 | 4806 | #endif /* CONFIG_SLUB_DEBUG */ |
88a420e4 | 4807 | |
a5a84755 | 4808 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4809 | static void __init resiliency_test(void) |
a5a84755 CL |
4810 | { |
4811 | u8 *p; | |
cc252eae | 4812 | int type = KMALLOC_NORMAL; |
a5a84755 | 4813 | |
95a05b42 | 4814 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4815 | |
f9f58285 FF |
4816 | pr_err("SLUB resiliency testing\n"); |
4817 | pr_err("-----------------------\n"); | |
4818 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4819 | |
4820 | p = kzalloc(16, GFP_KERNEL); | |
4821 | p[16] = 0x12; | |
f9f58285 FF |
4822 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4823 | p + 16); | |
a5a84755 | 4824 | |
cc252eae | 4825 | validate_slab_cache(kmalloc_caches[type][4]); |
a5a84755 CL |
4826 | |
4827 | /* Hmmm... The next two are dangerous */ | |
4828 | p = kzalloc(32, GFP_KERNEL); | |
4829 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4830 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4831 | p); | |
4832 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 | 4833 | |
cc252eae | 4834 | validate_slab_cache(kmalloc_caches[type][5]); |
a5a84755 CL |
4835 | p = kzalloc(64, GFP_KERNEL); |
4836 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4837 | *p = 0x56; | |
f9f58285 FF |
4838 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4839 | p); | |
4840 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
cc252eae | 4841 | validate_slab_cache(kmalloc_caches[type][6]); |
a5a84755 | 4842 | |
f9f58285 | 4843 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4844 | p = kzalloc(128, GFP_KERNEL); |
4845 | kfree(p); | |
4846 | *p = 0x78; | |
f9f58285 | 4847 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
cc252eae | 4848 | validate_slab_cache(kmalloc_caches[type][7]); |
a5a84755 CL |
4849 | |
4850 | p = kzalloc(256, GFP_KERNEL); | |
4851 | kfree(p); | |
4852 | p[50] = 0x9a; | |
f9f58285 | 4853 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
cc252eae | 4854 | validate_slab_cache(kmalloc_caches[type][8]); |
a5a84755 CL |
4855 | |
4856 | p = kzalloc(512, GFP_KERNEL); | |
4857 | kfree(p); | |
4858 | p[512] = 0xab; | |
f9f58285 | 4859 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
cc252eae | 4860 | validate_slab_cache(kmalloc_caches[type][9]); |
a5a84755 CL |
4861 | } |
4862 | #else | |
4863 | #ifdef CONFIG_SYSFS | |
4864 | static void resiliency_test(void) {}; | |
4865 | #endif | |
6dfd1b65 | 4866 | #endif /* SLUB_RESILIENCY_TEST */ |
a5a84755 | 4867 | |
ab4d5ed5 | 4868 | #ifdef CONFIG_SYSFS |
81819f0f | 4869 | enum slab_stat_type { |
205ab99d CL |
4870 | SL_ALL, /* All slabs */ |
4871 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4872 | SL_CPU, /* Only slabs used for cpu caches */ | |
4873 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4874 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4875 | }; |
4876 | ||
205ab99d | 4877 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4878 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4879 | #define SO_CPU (1 << SL_CPU) | |
4880 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4881 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4882 | |
1663f26d TH |
4883 | #ifdef CONFIG_MEMCG |
4884 | static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON); | |
4885 | ||
4886 | static int __init setup_slub_memcg_sysfs(char *str) | |
4887 | { | |
4888 | int v; | |
4889 | ||
4890 | if (get_option(&str, &v) > 0) | |
4891 | memcg_sysfs_enabled = v; | |
4892 | ||
4893 | return 1; | |
4894 | } | |
4895 | ||
4896 | __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs); | |
4897 | #endif | |
4898 | ||
62e5c4b4 CG |
4899 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4900 | char *buf, unsigned long flags) | |
81819f0f CL |
4901 | { |
4902 | unsigned long total = 0; | |
81819f0f CL |
4903 | int node; |
4904 | int x; | |
4905 | unsigned long *nodes; | |
81819f0f | 4906 | |
6396bb22 | 4907 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
4908 | if (!nodes) |
4909 | return -ENOMEM; | |
81819f0f | 4910 | |
205ab99d CL |
4911 | if (flags & SO_CPU) { |
4912 | int cpu; | |
81819f0f | 4913 | |
205ab99d | 4914 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4915 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4916 | cpu); | |
ec3ab083 | 4917 | int node; |
49e22585 | 4918 | struct page *page; |
dfb4f096 | 4919 | |
4db0c3c2 | 4920 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4921 | if (!page) |
4922 | continue; | |
205ab99d | 4923 | |
ec3ab083 CL |
4924 | node = page_to_nid(page); |
4925 | if (flags & SO_TOTAL) | |
4926 | x = page->objects; | |
4927 | else if (flags & SO_OBJECTS) | |
4928 | x = page->inuse; | |
4929 | else | |
4930 | x = 1; | |
49e22585 | 4931 | |
ec3ab083 CL |
4932 | total += x; |
4933 | nodes[node] += x; | |
4934 | ||
a93cf07b | 4935 | page = slub_percpu_partial_read_once(c); |
49e22585 | 4936 | if (page) { |
8afb1474 LZ |
4937 | node = page_to_nid(page); |
4938 | if (flags & SO_TOTAL) | |
4939 | WARN_ON_ONCE(1); | |
4940 | else if (flags & SO_OBJECTS) | |
4941 | WARN_ON_ONCE(1); | |
4942 | else | |
4943 | x = page->pages; | |
bc6697d8 ED |
4944 | total += x; |
4945 | nodes[node] += x; | |
49e22585 | 4946 | } |
81819f0f CL |
4947 | } |
4948 | } | |
4949 | ||
e4f8e513 QC |
4950 | /* |
4951 | * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex" | |
4952 | * already held which will conflict with an existing lock order: | |
4953 | * | |
4954 | * mem_hotplug_lock->slab_mutex->kernfs_mutex | |
4955 | * | |
4956 | * We don't really need mem_hotplug_lock (to hold off | |
4957 | * slab_mem_going_offline_callback) here because slab's memory hot | |
4958 | * unplug code doesn't destroy the kmem_cache->node[] data. | |
4959 | */ | |
4960 | ||
ab4d5ed5 | 4961 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4962 | if (flags & SO_ALL) { |
fa45dc25 CL |
4963 | struct kmem_cache_node *n; |
4964 | ||
4965 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4966 | |
d0e0ac97 CG |
4967 | if (flags & SO_TOTAL) |
4968 | x = atomic_long_read(&n->total_objects); | |
4969 | else if (flags & SO_OBJECTS) | |
4970 | x = atomic_long_read(&n->total_objects) - | |
4971 | count_partial(n, count_free); | |
81819f0f | 4972 | else |
205ab99d | 4973 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4974 | total += x; |
4975 | nodes[node] += x; | |
4976 | } | |
4977 | ||
ab4d5ed5 CL |
4978 | } else |
4979 | #endif | |
4980 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4981 | struct kmem_cache_node *n; |
81819f0f | 4982 | |
fa45dc25 | 4983 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4984 | if (flags & SO_TOTAL) |
4985 | x = count_partial(n, count_total); | |
4986 | else if (flags & SO_OBJECTS) | |
4987 | x = count_partial(n, count_inuse); | |
81819f0f | 4988 | else |
205ab99d | 4989 | x = n->nr_partial; |
81819f0f CL |
4990 | total += x; |
4991 | nodes[node] += x; | |
4992 | } | |
4993 | } | |
81819f0f CL |
4994 | x = sprintf(buf, "%lu", total); |
4995 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4996 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4997 | if (nodes[node]) |
4998 | x += sprintf(buf + x, " N%d=%lu", | |
4999 | node, nodes[node]); | |
5000 | #endif | |
5001 | kfree(nodes); | |
5002 | return x + sprintf(buf + x, "\n"); | |
5003 | } | |
5004 | ||
81819f0f | 5005 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) |
497888cf | 5006 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
5007 | |
5008 | struct slab_attribute { | |
5009 | struct attribute attr; | |
5010 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
5011 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
5012 | }; | |
5013 | ||
5014 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
5015 | static struct slab_attribute _name##_attr = \ |
5016 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
5017 | |
5018 | #define SLAB_ATTR(_name) \ | |
5019 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 5020 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 5021 | |
81819f0f CL |
5022 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
5023 | { | |
44065b2e | 5024 | return sprintf(buf, "%u\n", s->size); |
81819f0f CL |
5025 | } |
5026 | SLAB_ATTR_RO(slab_size); | |
5027 | ||
5028 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
5029 | { | |
3a3791ec | 5030 | return sprintf(buf, "%u\n", s->align); |
81819f0f CL |
5031 | } |
5032 | SLAB_ATTR_RO(align); | |
5033 | ||
5034 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
5035 | { | |
1b473f29 | 5036 | return sprintf(buf, "%u\n", s->object_size); |
81819f0f CL |
5037 | } |
5038 | SLAB_ATTR_RO(object_size); | |
5039 | ||
5040 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
5041 | { | |
19af27af | 5042 | return sprintf(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
5043 | } |
5044 | SLAB_ATTR_RO(objs_per_slab); | |
5045 | ||
5046 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
5047 | { | |
19af27af | 5048 | return sprintf(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 5049 | } |
32a6f409 | 5050 | SLAB_ATTR_RO(order); |
81819f0f | 5051 | |
73d342b1 DR |
5052 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
5053 | { | |
5054 | return sprintf(buf, "%lu\n", s->min_partial); | |
5055 | } | |
5056 | ||
5057 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
5058 | size_t length) | |
5059 | { | |
5060 | unsigned long min; | |
5061 | int err; | |
5062 | ||
3dbb95f7 | 5063 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
5064 | if (err) |
5065 | return err; | |
5066 | ||
c0bdb232 | 5067 | set_min_partial(s, min); |
73d342b1 DR |
5068 | return length; |
5069 | } | |
5070 | SLAB_ATTR(min_partial); | |
5071 | ||
49e22585 CL |
5072 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
5073 | { | |
e6d0e1dc | 5074 | return sprintf(buf, "%u\n", slub_cpu_partial(s)); |
49e22585 CL |
5075 | } |
5076 | ||
5077 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
5078 | size_t length) | |
5079 | { | |
e5d9998f | 5080 | unsigned int objects; |
49e22585 CL |
5081 | int err; |
5082 | ||
e5d9998f | 5083 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
5084 | if (err) |
5085 | return err; | |
345c905d | 5086 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 5087 | return -EINVAL; |
49e22585 | 5088 | |
e6d0e1dc | 5089 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
5090 | flush_all(s); |
5091 | return length; | |
5092 | } | |
5093 | SLAB_ATTR(cpu_partial); | |
5094 | ||
81819f0f CL |
5095 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
5096 | { | |
62c70bce JP |
5097 | if (!s->ctor) |
5098 | return 0; | |
5099 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
5100 | } |
5101 | SLAB_ATTR_RO(ctor); | |
5102 | ||
81819f0f CL |
5103 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
5104 | { | |
4307c14f | 5105 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
5106 | } |
5107 | SLAB_ATTR_RO(aliases); | |
5108 | ||
81819f0f CL |
5109 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5110 | { | |
d9acf4b7 | 5111 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5112 | } |
5113 | SLAB_ATTR_RO(partial); | |
5114 | ||
5115 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5116 | { | |
d9acf4b7 | 5117 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5118 | } |
5119 | SLAB_ATTR_RO(cpu_slabs); | |
5120 | ||
5121 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5122 | { | |
205ab99d | 5123 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5124 | } |
5125 | SLAB_ATTR_RO(objects); | |
5126 | ||
205ab99d CL |
5127 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5128 | { | |
5129 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5130 | } | |
5131 | SLAB_ATTR_RO(objects_partial); | |
5132 | ||
49e22585 CL |
5133 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5134 | { | |
5135 | int objects = 0; | |
5136 | int pages = 0; | |
5137 | int cpu; | |
5138 | int len; | |
5139 | ||
5140 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5141 | struct page *page; |
5142 | ||
5143 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5144 | |
5145 | if (page) { | |
5146 | pages += page->pages; | |
5147 | objects += page->pobjects; | |
5148 | } | |
5149 | } | |
5150 | ||
5151 | len = sprintf(buf, "%d(%d)", objects, pages); | |
5152 | ||
5153 | #ifdef CONFIG_SMP | |
5154 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5155 | struct page *page; |
5156 | ||
5157 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5158 | |
5159 | if (page && len < PAGE_SIZE - 20) | |
5160 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
5161 | page->pobjects, page->pages); | |
5162 | } | |
5163 | #endif | |
5164 | return len + sprintf(buf + len, "\n"); | |
5165 | } | |
5166 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5167 | ||
a5a84755 CL |
5168 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5169 | { | |
5170 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
5171 | } | |
8f58119a | 5172 | SLAB_ATTR_RO(reclaim_account); |
a5a84755 CL |
5173 | |
5174 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5175 | { | |
5176 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
5177 | } | |
5178 | SLAB_ATTR_RO(hwcache_align); | |
5179 | ||
5180 | #ifdef CONFIG_ZONE_DMA | |
5181 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5182 | { | |
5183 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
5184 | } | |
5185 | SLAB_ATTR_RO(cache_dma); | |
5186 | #endif | |
5187 | ||
8eb8284b DW |
5188 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5189 | { | |
7bbdb81e | 5190 | return sprintf(buf, "%u\n", s->usersize); |
8eb8284b DW |
5191 | } |
5192 | SLAB_ATTR_RO(usersize); | |
5193 | ||
a5a84755 CL |
5194 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5195 | { | |
5f0d5a3a | 5196 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5197 | } |
5198 | SLAB_ATTR_RO(destroy_by_rcu); | |
5199 | ||
ab4d5ed5 | 5200 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5201 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5202 | { | |
5203 | return show_slab_objects(s, buf, SO_ALL); | |
5204 | } | |
5205 | SLAB_ATTR_RO(slabs); | |
5206 | ||
205ab99d CL |
5207 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5208 | { | |
5209 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5210 | } | |
5211 | SLAB_ATTR_RO(total_objects); | |
5212 | ||
81819f0f CL |
5213 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5214 | { | |
becfda68 | 5215 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f | 5216 | } |
060807f8 | 5217 | SLAB_ATTR_RO(sanity_checks); |
81819f0f CL |
5218 | |
5219 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5220 | { | |
5221 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
5222 | } | |
060807f8 | 5223 | SLAB_ATTR_RO(trace); |
81819f0f | 5224 | |
81819f0f CL |
5225 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5226 | { | |
5227 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
5228 | } | |
5229 | ||
ad38b5b1 | 5230 | SLAB_ATTR_RO(red_zone); |
81819f0f CL |
5231 | |
5232 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5233 | { | |
5234 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
5235 | } | |
5236 | ||
ad38b5b1 | 5237 | SLAB_ATTR_RO(poison); |
81819f0f CL |
5238 | |
5239 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5240 | { | |
5241 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
5242 | } | |
5243 | ||
ad38b5b1 | 5244 | SLAB_ATTR_RO(store_user); |
81819f0f | 5245 | |
53e15af0 CL |
5246 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5247 | { | |
5248 | return 0; | |
5249 | } | |
5250 | ||
5251 | static ssize_t validate_store(struct kmem_cache *s, | |
5252 | const char *buf, size_t length) | |
5253 | { | |
434e245d CL |
5254 | int ret = -EINVAL; |
5255 | ||
5256 | if (buf[0] == '1') { | |
5257 | ret = validate_slab_cache(s); | |
5258 | if (ret >= 0) | |
5259 | ret = length; | |
5260 | } | |
5261 | return ret; | |
53e15af0 CL |
5262 | } |
5263 | SLAB_ATTR(validate); | |
a5a84755 CL |
5264 | |
5265 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5266 | { | |
5267 | if (!(s->flags & SLAB_STORE_USER)) | |
5268 | return -ENOSYS; | |
5269 | return list_locations(s, buf, TRACK_ALLOC); | |
5270 | } | |
5271 | SLAB_ATTR_RO(alloc_calls); | |
5272 | ||
5273 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5274 | { | |
5275 | if (!(s->flags & SLAB_STORE_USER)) | |
5276 | return -ENOSYS; | |
5277 | return list_locations(s, buf, TRACK_FREE); | |
5278 | } | |
5279 | SLAB_ATTR_RO(free_calls); | |
5280 | #endif /* CONFIG_SLUB_DEBUG */ | |
5281 | ||
5282 | #ifdef CONFIG_FAILSLAB | |
5283 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5284 | { | |
5285 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
5286 | } | |
060807f8 | 5287 | SLAB_ATTR_RO(failslab); |
ab4d5ed5 | 5288 | #endif |
53e15af0 | 5289 | |
2086d26a CL |
5290 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5291 | { | |
5292 | return 0; | |
5293 | } | |
5294 | ||
5295 | static ssize_t shrink_store(struct kmem_cache *s, | |
5296 | const char *buf, size_t length) | |
5297 | { | |
832f37f5 | 5298 | if (buf[0] == '1') |
10befea9 | 5299 | kmem_cache_shrink(s); |
832f37f5 | 5300 | else |
2086d26a CL |
5301 | return -EINVAL; |
5302 | return length; | |
5303 | } | |
5304 | SLAB_ATTR(shrink); | |
5305 | ||
81819f0f | 5306 | #ifdef CONFIG_NUMA |
9824601e | 5307 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5308 | { |
eb7235eb | 5309 | return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5310 | } |
5311 | ||
9824601e | 5312 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5313 | const char *buf, size_t length) |
5314 | { | |
eb7235eb | 5315 | unsigned int ratio; |
0121c619 CL |
5316 | int err; |
5317 | ||
eb7235eb | 5318 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5319 | if (err) |
5320 | return err; | |
eb7235eb AD |
5321 | if (ratio > 100) |
5322 | return -ERANGE; | |
0121c619 | 5323 | |
eb7235eb | 5324 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5325 | |
81819f0f CL |
5326 | return length; |
5327 | } | |
9824601e | 5328 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5329 | #endif |
5330 | ||
8ff12cfc | 5331 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5332 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5333 | { | |
5334 | unsigned long sum = 0; | |
5335 | int cpu; | |
5336 | int len; | |
6da2ec56 | 5337 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5338 | |
5339 | if (!data) | |
5340 | return -ENOMEM; | |
5341 | ||
5342 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5343 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5344 | |
5345 | data[cpu] = x; | |
5346 | sum += x; | |
5347 | } | |
5348 | ||
5349 | len = sprintf(buf, "%lu", sum); | |
5350 | ||
50ef37b9 | 5351 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5352 | for_each_online_cpu(cpu) { |
5353 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5354 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5355 | } |
50ef37b9 | 5356 | #endif |
8ff12cfc CL |
5357 | kfree(data); |
5358 | return len + sprintf(buf + len, "\n"); | |
5359 | } | |
5360 | ||
78eb00cc DR |
5361 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5362 | { | |
5363 | int cpu; | |
5364 | ||
5365 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5366 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5367 | } |
5368 | ||
8ff12cfc CL |
5369 | #define STAT_ATTR(si, text) \ |
5370 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5371 | { \ | |
5372 | return show_stat(s, buf, si); \ | |
5373 | } \ | |
78eb00cc DR |
5374 | static ssize_t text##_store(struct kmem_cache *s, \ |
5375 | const char *buf, size_t length) \ | |
5376 | { \ | |
5377 | if (buf[0] != '0') \ | |
5378 | return -EINVAL; \ | |
5379 | clear_stat(s, si); \ | |
5380 | return length; \ | |
5381 | } \ | |
5382 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5383 | |
5384 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5385 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5386 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5387 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5388 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5389 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5390 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5391 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5392 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5393 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5394 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5395 | STAT_ATTR(FREE_SLAB, free_slab); |
5396 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5397 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5398 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5399 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5400 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5401 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5402 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5403 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5404 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5405 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5406 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5407 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5408 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5409 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
6dfd1b65 | 5410 | #endif /* CONFIG_SLUB_STATS */ |
8ff12cfc | 5411 | |
06428780 | 5412 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5413 | &slab_size_attr.attr, |
5414 | &object_size_attr.attr, | |
5415 | &objs_per_slab_attr.attr, | |
5416 | &order_attr.attr, | |
73d342b1 | 5417 | &min_partial_attr.attr, |
49e22585 | 5418 | &cpu_partial_attr.attr, |
81819f0f | 5419 | &objects_attr.attr, |
205ab99d | 5420 | &objects_partial_attr.attr, |
81819f0f CL |
5421 | &partial_attr.attr, |
5422 | &cpu_slabs_attr.attr, | |
5423 | &ctor_attr.attr, | |
81819f0f CL |
5424 | &aliases_attr.attr, |
5425 | &align_attr.attr, | |
81819f0f CL |
5426 | &hwcache_align_attr.attr, |
5427 | &reclaim_account_attr.attr, | |
5428 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5429 | &shrink_attr.attr, |
49e22585 | 5430 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5431 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5432 | &total_objects_attr.attr, |
5433 | &slabs_attr.attr, | |
5434 | &sanity_checks_attr.attr, | |
5435 | &trace_attr.attr, | |
81819f0f CL |
5436 | &red_zone_attr.attr, |
5437 | &poison_attr.attr, | |
5438 | &store_user_attr.attr, | |
53e15af0 | 5439 | &validate_attr.attr, |
88a420e4 CL |
5440 | &alloc_calls_attr.attr, |
5441 | &free_calls_attr.attr, | |
ab4d5ed5 | 5442 | #endif |
81819f0f CL |
5443 | #ifdef CONFIG_ZONE_DMA |
5444 | &cache_dma_attr.attr, | |
5445 | #endif | |
5446 | #ifdef CONFIG_NUMA | |
9824601e | 5447 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5448 | #endif |
5449 | #ifdef CONFIG_SLUB_STATS | |
5450 | &alloc_fastpath_attr.attr, | |
5451 | &alloc_slowpath_attr.attr, | |
5452 | &free_fastpath_attr.attr, | |
5453 | &free_slowpath_attr.attr, | |
5454 | &free_frozen_attr.attr, | |
5455 | &free_add_partial_attr.attr, | |
5456 | &free_remove_partial_attr.attr, | |
5457 | &alloc_from_partial_attr.attr, | |
5458 | &alloc_slab_attr.attr, | |
5459 | &alloc_refill_attr.attr, | |
e36a2652 | 5460 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5461 | &free_slab_attr.attr, |
5462 | &cpuslab_flush_attr.attr, | |
5463 | &deactivate_full_attr.attr, | |
5464 | &deactivate_empty_attr.attr, | |
5465 | &deactivate_to_head_attr.attr, | |
5466 | &deactivate_to_tail_attr.attr, | |
5467 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5468 | &deactivate_bypass_attr.attr, |
65c3376a | 5469 | &order_fallback_attr.attr, |
b789ef51 CL |
5470 | &cmpxchg_double_fail_attr.attr, |
5471 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5472 | &cpu_partial_alloc_attr.attr, |
5473 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5474 | &cpu_partial_node_attr.attr, |
5475 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5476 | #endif |
4c13dd3b DM |
5477 | #ifdef CONFIG_FAILSLAB |
5478 | &failslab_attr.attr, | |
5479 | #endif | |
8eb8284b | 5480 | &usersize_attr.attr, |
4c13dd3b | 5481 | |
81819f0f CL |
5482 | NULL |
5483 | }; | |
5484 | ||
1fdaaa23 | 5485 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5486 | .attrs = slab_attrs, |
5487 | }; | |
5488 | ||
5489 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5490 | struct attribute *attr, | |
5491 | char *buf) | |
5492 | { | |
5493 | struct slab_attribute *attribute; | |
5494 | struct kmem_cache *s; | |
5495 | int err; | |
5496 | ||
5497 | attribute = to_slab_attr(attr); | |
5498 | s = to_slab(kobj); | |
5499 | ||
5500 | if (!attribute->show) | |
5501 | return -EIO; | |
5502 | ||
5503 | err = attribute->show(s, buf); | |
5504 | ||
5505 | return err; | |
5506 | } | |
5507 | ||
5508 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5509 | struct attribute *attr, | |
5510 | const char *buf, size_t len) | |
5511 | { | |
5512 | struct slab_attribute *attribute; | |
5513 | struct kmem_cache *s; | |
5514 | int err; | |
5515 | ||
5516 | attribute = to_slab_attr(attr); | |
5517 | s = to_slab(kobj); | |
5518 | ||
5519 | if (!attribute->store) | |
5520 | return -EIO; | |
5521 | ||
5522 | err = attribute->store(s, buf, len); | |
81819f0f CL |
5523 | return err; |
5524 | } | |
5525 | ||
41a21285 CL |
5526 | static void kmem_cache_release(struct kobject *k) |
5527 | { | |
5528 | slab_kmem_cache_release(to_slab(k)); | |
5529 | } | |
5530 | ||
52cf25d0 | 5531 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5532 | .show = slab_attr_show, |
5533 | .store = slab_attr_store, | |
5534 | }; | |
5535 | ||
5536 | static struct kobj_type slab_ktype = { | |
5537 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5538 | .release = kmem_cache_release, |
81819f0f CL |
5539 | }; |
5540 | ||
27c3a314 | 5541 | static struct kset *slab_kset; |
81819f0f | 5542 | |
9a41707b VD |
5543 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5544 | { | |
9a41707b VD |
5545 | return slab_kset; |
5546 | } | |
5547 | ||
81819f0f CL |
5548 | #define ID_STR_LENGTH 64 |
5549 | ||
5550 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5551 | * |
5552 | * Format :[flags-]size | |
81819f0f CL |
5553 | */ |
5554 | static char *create_unique_id(struct kmem_cache *s) | |
5555 | { | |
5556 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5557 | char *p = name; | |
5558 | ||
5559 | BUG_ON(!name); | |
5560 | ||
5561 | *p++ = ':'; | |
5562 | /* | |
5563 | * First flags affecting slabcache operations. We will only | |
5564 | * get here for aliasable slabs so we do not need to support | |
5565 | * too many flags. The flags here must cover all flags that | |
5566 | * are matched during merging to guarantee that the id is | |
5567 | * unique. | |
5568 | */ | |
5569 | if (s->flags & SLAB_CACHE_DMA) | |
5570 | *p++ = 'd'; | |
6d6ea1e9 NB |
5571 | if (s->flags & SLAB_CACHE_DMA32) |
5572 | *p++ = 'D'; | |
81819f0f CL |
5573 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
5574 | *p++ = 'a'; | |
becfda68 | 5575 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5576 | *p++ = 'F'; |
230e9fc2 VD |
5577 | if (s->flags & SLAB_ACCOUNT) |
5578 | *p++ = 'A'; | |
81819f0f CL |
5579 | if (p != name + 1) |
5580 | *p++ = '-'; | |
44065b2e | 5581 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5582 | |
81819f0f CL |
5583 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5584 | return name; | |
5585 | } | |
5586 | ||
5587 | static int sysfs_slab_add(struct kmem_cache *s) | |
5588 | { | |
5589 | int err; | |
5590 | const char *name; | |
1663f26d | 5591 | struct kset *kset = cache_kset(s); |
45530c44 | 5592 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5593 | |
1663f26d TH |
5594 | if (!kset) { |
5595 | kobject_init(&s->kobj, &slab_ktype); | |
5596 | return 0; | |
5597 | } | |
5598 | ||
11066386 MC |
5599 | if (!unmergeable && disable_higher_order_debug && |
5600 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5601 | unmergeable = 1; | |
5602 | ||
81819f0f CL |
5603 | if (unmergeable) { |
5604 | /* | |
5605 | * Slabcache can never be merged so we can use the name proper. | |
5606 | * This is typically the case for debug situations. In that | |
5607 | * case we can catch duplicate names easily. | |
5608 | */ | |
27c3a314 | 5609 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5610 | name = s->name; |
5611 | } else { | |
5612 | /* | |
5613 | * Create a unique name for the slab as a target | |
5614 | * for the symlinks. | |
5615 | */ | |
5616 | name = create_unique_id(s); | |
5617 | } | |
5618 | ||
1663f26d | 5619 | s->kobj.kset = kset; |
26e4f205 | 5620 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
dde3c6b7 WH |
5621 | if (err) { |
5622 | kobject_put(&s->kobj); | |
80da026a | 5623 | goto out; |
dde3c6b7 | 5624 | } |
81819f0f CL |
5625 | |
5626 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5627 | if (err) |
5628 | goto out_del_kobj; | |
9a41707b | 5629 | |
81819f0f CL |
5630 | if (!unmergeable) { |
5631 | /* Setup first alias */ | |
5632 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5633 | } |
54b6a731 DJ |
5634 | out: |
5635 | if (!unmergeable) | |
5636 | kfree(name); | |
5637 | return err; | |
5638 | out_del_kobj: | |
5639 | kobject_del(&s->kobj); | |
54b6a731 | 5640 | goto out; |
81819f0f CL |
5641 | } |
5642 | ||
d50d82fa MP |
5643 | void sysfs_slab_unlink(struct kmem_cache *s) |
5644 | { | |
5645 | if (slab_state >= FULL) | |
5646 | kobject_del(&s->kobj); | |
5647 | } | |
5648 | ||
bf5eb3de TH |
5649 | void sysfs_slab_release(struct kmem_cache *s) |
5650 | { | |
5651 | if (slab_state >= FULL) | |
5652 | kobject_put(&s->kobj); | |
81819f0f CL |
5653 | } |
5654 | ||
5655 | /* | |
5656 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5657 | * available lest we lose that information. |
81819f0f CL |
5658 | */ |
5659 | struct saved_alias { | |
5660 | struct kmem_cache *s; | |
5661 | const char *name; | |
5662 | struct saved_alias *next; | |
5663 | }; | |
5664 | ||
5af328a5 | 5665 | static struct saved_alias *alias_list; |
81819f0f CL |
5666 | |
5667 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5668 | { | |
5669 | struct saved_alias *al; | |
5670 | ||
97d06609 | 5671 | if (slab_state == FULL) { |
81819f0f CL |
5672 | /* |
5673 | * If we have a leftover link then remove it. | |
5674 | */ | |
27c3a314 GKH |
5675 | sysfs_remove_link(&slab_kset->kobj, name); |
5676 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5677 | } |
5678 | ||
5679 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5680 | if (!al) | |
5681 | return -ENOMEM; | |
5682 | ||
5683 | al->s = s; | |
5684 | al->name = name; | |
5685 | al->next = alias_list; | |
5686 | alias_list = al; | |
5687 | return 0; | |
5688 | } | |
5689 | ||
5690 | static int __init slab_sysfs_init(void) | |
5691 | { | |
5b95a4ac | 5692 | struct kmem_cache *s; |
81819f0f CL |
5693 | int err; |
5694 | ||
18004c5d | 5695 | mutex_lock(&slab_mutex); |
2bce6485 | 5696 | |
d7660ce5 | 5697 | slab_kset = kset_create_and_add("slab", NULL, kernel_kobj); |
27c3a314 | 5698 | if (!slab_kset) { |
18004c5d | 5699 | mutex_unlock(&slab_mutex); |
f9f58285 | 5700 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5701 | return -ENOSYS; |
5702 | } | |
5703 | ||
97d06609 | 5704 | slab_state = FULL; |
26a7bd03 | 5705 | |
5b95a4ac | 5706 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5707 | err = sysfs_slab_add(s); |
5d540fb7 | 5708 | if (err) |
f9f58285 FF |
5709 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5710 | s->name); | |
26a7bd03 | 5711 | } |
81819f0f CL |
5712 | |
5713 | while (alias_list) { | |
5714 | struct saved_alias *al = alias_list; | |
5715 | ||
5716 | alias_list = alias_list->next; | |
5717 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5718 | if (err) |
f9f58285 FF |
5719 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5720 | al->name); | |
81819f0f CL |
5721 | kfree(al); |
5722 | } | |
5723 | ||
18004c5d | 5724 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5725 | resiliency_test(); |
5726 | return 0; | |
5727 | } | |
5728 | ||
5729 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5730 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5731 | |
5732 | /* | |
5733 | * The /proc/slabinfo ABI | |
5734 | */ | |
5b365771 | 5735 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 5736 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5737 | { |
57ed3eda | 5738 | unsigned long nr_slabs = 0; |
205ab99d CL |
5739 | unsigned long nr_objs = 0; |
5740 | unsigned long nr_free = 0; | |
57ed3eda | 5741 | int node; |
fa45dc25 | 5742 | struct kmem_cache_node *n; |
57ed3eda | 5743 | |
fa45dc25 | 5744 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5745 | nr_slabs += node_nr_slabs(n); |
5746 | nr_objs += node_nr_objs(n); | |
205ab99d | 5747 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5748 | } |
5749 | ||
0d7561c6 GC |
5750 | sinfo->active_objs = nr_objs - nr_free; |
5751 | sinfo->num_objs = nr_objs; | |
5752 | sinfo->active_slabs = nr_slabs; | |
5753 | sinfo->num_slabs = nr_slabs; | |
5754 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5755 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5756 | } |
5757 | ||
0d7561c6 | 5758 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5759 | { |
7b3c3a50 AD |
5760 | } |
5761 | ||
b7454ad3 GC |
5762 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5763 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5764 | { |
b7454ad3 | 5765 | return -EIO; |
7b3c3a50 | 5766 | } |
5b365771 | 5767 | #endif /* CONFIG_SLUB_DEBUG */ |