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> |
3ac38faa | 22 | #include <linux/notifier.h> |
81819f0f | 23 | #include <linux/seq_file.h> |
a79316c6 | 24 | #include <linux/kasan.h> |
81819f0f CL |
25 | #include <linux/cpu.h> |
26 | #include <linux/cpuset.h> | |
27 | #include <linux/mempolicy.h> | |
28 | #include <linux/ctype.h> | |
3ac7fe5a | 29 | #include <linux/debugobjects.h> |
81819f0f | 30 | #include <linux/kallsyms.h> |
b9049e23 | 31 | #include <linux/memory.h> |
f8bd2258 | 32 | #include <linux/math64.h> |
773ff60e | 33 | #include <linux/fault-inject.h> |
bfa71457 | 34 | #include <linux/stacktrace.h> |
4de900b4 | 35 | #include <linux/prefetch.h> |
2633d7a0 | 36 | #include <linux/memcontrol.h> |
2482ddec | 37 | #include <linux/random.h> |
81819f0f | 38 | |
4a92379b RK |
39 | #include <trace/events/kmem.h> |
40 | ||
072bb0aa MG |
41 | #include "internal.h" |
42 | ||
81819f0f CL |
43 | /* |
44 | * Lock order: | |
18004c5d | 45 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
46 | * 2. node->list_lock |
47 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 48 | * |
18004c5d | 49 | * slab_mutex |
881db7fb | 50 | * |
18004c5d | 51 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
52 | * and to synchronize major metadata changes to slab cache structures. |
53 | * | |
54 | * The slab_lock is only used for debugging and on arches that do not | |
b7ccc7f8 | 55 | * have the ability to do a cmpxchg_double. It only protects: |
881db7fb | 56 | * A. page->freelist -> List of object free in a page |
b7ccc7f8 MW |
57 | * B. page->inuse -> Number of objects in use |
58 | * C. page->objects -> Number of objects in page | |
59 | * D. page->frozen -> frozen state | |
881db7fb CL |
60 | * |
61 | * If a slab is frozen then it is exempt from list management. It is not | |
62 | * on any list. The processor that froze the slab is the one who can | |
63 | * perform list operations on the page. Other processors may put objects | |
64 | * onto the freelist but the processor that froze the slab is the only | |
65 | * one that can retrieve the objects from the 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 | * | |
96 | * Overloading of page flags that are otherwise used for LRU management. | |
97 | * | |
4b6f0750 CL |
98 | * PageActive The slab is frozen and exempt from list processing. |
99 | * This means that the slab is dedicated to a purpose | |
100 | * such as satisfying allocations for a specific | |
101 | * processor. Objects may be freed in the slab while | |
102 | * it is frozen but slab_free will then skip the usual | |
103 | * list operations. It is up to the processor holding | |
104 | * the slab to integrate the slab into the slab lists | |
105 | * when the slab is no longer needed. | |
106 | * | |
107 | * One use of this flag is to mark slabs that are | |
108 | * used for allocations. Then such a slab becomes a cpu | |
109 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 110 | * freelist that allows lockless access to |
894b8788 CL |
111 | * free objects in addition to the regular freelist |
112 | * that requires the slab lock. | |
81819f0f CL |
113 | * |
114 | * PageError Slab requires special handling due to debug | |
115 | * options set. This moves slab handling out of | |
894b8788 | 116 | * the fast path and disables lockless freelists. |
81819f0f CL |
117 | */ |
118 | ||
af537b0a CL |
119 | static inline int kmem_cache_debug(struct kmem_cache *s) |
120 | { | |
5577bd8a | 121 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 122 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 123 | #else |
af537b0a | 124 | return 0; |
5577bd8a | 125 | #endif |
af537b0a | 126 | } |
5577bd8a | 127 | |
117d54df | 128 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be JK |
129 | { |
130 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) | |
131 | p += s->red_left_pad; | |
132 | ||
133 | return p; | |
134 | } | |
135 | ||
345c905d JK |
136 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
137 | { | |
138 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
139 | return !kmem_cache_debug(s); | |
140 | #else | |
141 | return false; | |
142 | #endif | |
143 | } | |
144 | ||
81819f0f CL |
145 | /* |
146 | * Issues still to be resolved: | |
147 | * | |
81819f0f CL |
148 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
149 | * | |
81819f0f CL |
150 | * - Variable sizing of the per node arrays |
151 | */ | |
152 | ||
153 | /* Enable to test recovery from slab corruption on boot */ | |
154 | #undef SLUB_RESILIENCY_TEST | |
155 | ||
b789ef51 CL |
156 | /* Enable to log cmpxchg failures */ |
157 | #undef SLUB_DEBUG_CMPXCHG | |
158 | ||
2086d26a CL |
159 | /* |
160 | * Mininum number of partial slabs. These will be left on the partial | |
161 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
162 | */ | |
76be8950 | 163 | #define MIN_PARTIAL 5 |
e95eed57 | 164 | |
2086d26a CL |
165 | /* |
166 | * Maximum number of desirable partial slabs. | |
167 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 168 | * sort the partial list by the number of objects in use. |
2086d26a CL |
169 | */ |
170 | #define MAX_PARTIAL 10 | |
171 | ||
becfda68 | 172 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 173 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 174 | |
149daaf3 LA |
175 | /* |
176 | * These debug flags cannot use CMPXCHG because there might be consistency | |
177 | * issues when checking or reading debug information | |
178 | */ | |
179 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
180 | SLAB_TRACE) | |
181 | ||
182 | ||
fa5ec8a1 | 183 | /* |
3de47213 DR |
184 | * Debugging flags that require metadata to be stored in the slab. These get |
185 | * disabled when slub_debug=O is used and a cache's min order increases with | |
186 | * metadata. | |
fa5ec8a1 | 187 | */ |
3de47213 | 188 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 189 | |
210b5c06 CG |
190 | #define OO_SHIFT 16 |
191 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 192 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 193 | |
81819f0f | 194 | /* Internal SLUB flags */ |
d50112ed | 195 | /* Poison object */ |
4fd0b46e | 196 | #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U) |
d50112ed | 197 | /* Use cmpxchg_double */ |
4fd0b46e | 198 | #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U) |
81819f0f | 199 | |
02cbc874 CL |
200 | /* |
201 | * Tracking user of a slab. | |
202 | */ | |
d6543e39 | 203 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 204 | struct track { |
ce71e27c | 205 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
206 | #ifdef CONFIG_STACKTRACE |
207 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
208 | #endif | |
02cbc874 CL |
209 | int cpu; /* Was running on cpu */ |
210 | int pid; /* Pid context */ | |
211 | unsigned long when; /* When did the operation occur */ | |
212 | }; | |
213 | ||
214 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
215 | ||
ab4d5ed5 | 216 | #ifdef CONFIG_SYSFS |
81819f0f CL |
217 | static int sysfs_slab_add(struct kmem_cache *); |
218 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 219 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
bf5eb3de | 220 | static void sysfs_slab_remove(struct kmem_cache *s); |
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; } | |
107dab5c | 225 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
bf5eb3de | 226 | static inline void sysfs_slab_remove(struct kmem_cache *s) { } |
81819f0f CL |
227 | #endif |
228 | ||
4fdccdfb | 229 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
230 | { |
231 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
232 | /* |
233 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
234 | * avoid this_cpu_add()'s irq-disable overhead. | |
235 | */ | |
236 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
237 | #endif |
238 | } | |
239 | ||
81819f0f CL |
240 | /******************************************************************** |
241 | * Core slab cache functions | |
242 | *******************************************************************/ | |
243 | ||
2482ddec KC |
244 | /* |
245 | * Returns freelist pointer (ptr). With hardening, this is obfuscated | |
246 | * with an XOR of the address where the pointer is held and a per-cache | |
247 | * random number. | |
248 | */ | |
249 | static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr, | |
250 | unsigned long ptr_addr) | |
251 | { | |
252 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
253 | return (void *)((unsigned long)ptr ^ s->random ^ ptr_addr); | |
254 | #else | |
255 | return ptr; | |
256 | #endif | |
257 | } | |
258 | ||
259 | /* Returns the freelist pointer recorded at location ptr_addr. */ | |
260 | static inline void *freelist_dereference(const struct kmem_cache *s, | |
261 | void *ptr_addr) | |
262 | { | |
263 | return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr), | |
264 | (unsigned long)ptr_addr); | |
265 | } | |
266 | ||
7656c72b CL |
267 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
268 | { | |
2482ddec | 269 | return freelist_dereference(s, object + s->offset); |
7656c72b CL |
270 | } |
271 | ||
0ad9500e ED |
272 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
273 | { | |
2482ddec KC |
274 | if (object) |
275 | prefetch(freelist_dereference(s, object + s->offset)); | |
0ad9500e ED |
276 | } |
277 | ||
1393d9a1 CL |
278 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
279 | { | |
2482ddec | 280 | unsigned long freepointer_addr; |
1393d9a1 CL |
281 | void *p; |
282 | ||
922d566c JK |
283 | if (!debug_pagealloc_enabled()) |
284 | return get_freepointer(s, object); | |
285 | ||
2482ddec KC |
286 | freepointer_addr = (unsigned long)object + s->offset; |
287 | probe_kernel_read(&p, (void **)freepointer_addr, sizeof(p)); | |
288 | return freelist_ptr(s, p, freepointer_addr); | |
1393d9a1 CL |
289 | } |
290 | ||
7656c72b CL |
291 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
292 | { | |
2482ddec KC |
293 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
294 | ||
ce6fa91b AP |
295 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
296 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
297 | #endif | |
298 | ||
2482ddec | 299 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); |
7656c72b CL |
300 | } |
301 | ||
302 | /* Loop over all objects in a slab */ | |
224a88be | 303 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
304 | for (__p = fixup_red_left(__s, __addr); \ |
305 | __p < (__addr) + (__objects) * (__s)->size; \ | |
306 | __p += (__s)->size) | |
7656c72b | 307 | |
54266640 | 308 | #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \ |
d86bd1be JK |
309 | for (__p = fixup_red_left(__s, __addr), __idx = 1; \ |
310 | __idx <= __objects; \ | |
311 | __p += (__s)->size, __idx++) | |
54266640 | 312 | |
7656c72b | 313 | /* Determine object index from a given position */ |
284b50dd | 314 | static inline unsigned int slab_index(void *p, struct kmem_cache *s, void *addr) |
7656c72b CL |
315 | { |
316 | return (p - addr) / s->size; | |
317 | } | |
318 | ||
9736d2a9 | 319 | static inline unsigned int order_objects(unsigned int order, unsigned int size) |
ab9a0f19 | 320 | { |
9736d2a9 | 321 | return ((unsigned int)PAGE_SIZE << order) / size; |
ab9a0f19 LJ |
322 | } |
323 | ||
19af27af | 324 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, |
9736d2a9 | 325 | unsigned int size) |
834f3d11 CL |
326 | { |
327 | struct kmem_cache_order_objects x = { | |
9736d2a9 | 328 | (order << OO_SHIFT) + order_objects(order, size) |
834f3d11 CL |
329 | }; |
330 | ||
331 | return x; | |
332 | } | |
333 | ||
19af27af | 334 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) |
834f3d11 | 335 | { |
210b5c06 | 336 | return x.x >> OO_SHIFT; |
834f3d11 CL |
337 | } |
338 | ||
19af27af | 339 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) |
834f3d11 | 340 | { |
210b5c06 | 341 | return x.x & OO_MASK; |
834f3d11 CL |
342 | } |
343 | ||
881db7fb CL |
344 | /* |
345 | * Per slab locking using the pagelock | |
346 | */ | |
347 | static __always_inline void slab_lock(struct page *page) | |
348 | { | |
48c935ad | 349 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
350 | bit_spin_lock(PG_locked, &page->flags); |
351 | } | |
352 | ||
353 | static __always_inline void slab_unlock(struct page *page) | |
354 | { | |
48c935ad | 355 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
356 | __bit_spin_unlock(PG_locked, &page->flags); |
357 | } | |
358 | ||
1d07171c CL |
359 | /* Interrupts must be disabled (for the fallback code to work right) */ |
360 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
361 | void *freelist_old, unsigned long counters_old, | |
362 | void *freelist_new, unsigned long counters_new, | |
363 | const char *n) | |
364 | { | |
365 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
366 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
367 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 368 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 369 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
370 | freelist_old, counters_old, |
371 | freelist_new, counters_new)) | |
6f6528a1 | 372 | return true; |
1d07171c CL |
373 | } else |
374 | #endif | |
375 | { | |
376 | slab_lock(page); | |
d0e0ac97 CG |
377 | if (page->freelist == freelist_old && |
378 | page->counters == counters_old) { | |
1d07171c | 379 | page->freelist = freelist_new; |
7d27a04b | 380 | page->counters = counters_new; |
1d07171c | 381 | slab_unlock(page); |
6f6528a1 | 382 | return true; |
1d07171c CL |
383 | } |
384 | slab_unlock(page); | |
385 | } | |
386 | ||
387 | cpu_relax(); | |
388 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
389 | ||
390 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 391 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
392 | #endif |
393 | ||
6f6528a1 | 394 | return false; |
1d07171c CL |
395 | } |
396 | ||
b789ef51 CL |
397 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
398 | void *freelist_old, unsigned long counters_old, | |
399 | void *freelist_new, unsigned long counters_new, | |
400 | const char *n) | |
401 | { | |
2565409f HC |
402 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
403 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 404 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 405 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
406 | freelist_old, counters_old, |
407 | freelist_new, counters_new)) | |
6f6528a1 | 408 | return true; |
b789ef51 CL |
409 | } else |
410 | #endif | |
411 | { | |
1d07171c CL |
412 | unsigned long flags; |
413 | ||
414 | local_irq_save(flags); | |
881db7fb | 415 | slab_lock(page); |
d0e0ac97 CG |
416 | if (page->freelist == freelist_old && |
417 | page->counters == counters_old) { | |
b789ef51 | 418 | page->freelist = freelist_new; |
7d27a04b | 419 | page->counters = counters_new; |
881db7fb | 420 | slab_unlock(page); |
1d07171c | 421 | local_irq_restore(flags); |
6f6528a1 | 422 | return true; |
b789ef51 | 423 | } |
881db7fb | 424 | slab_unlock(page); |
1d07171c | 425 | local_irq_restore(flags); |
b789ef51 CL |
426 | } |
427 | ||
428 | cpu_relax(); | |
429 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
430 | ||
431 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 432 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
433 | #endif |
434 | ||
6f6528a1 | 435 | return false; |
b789ef51 CL |
436 | } |
437 | ||
41ecc55b | 438 | #ifdef CONFIG_SLUB_DEBUG |
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 | */ | |
445 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
446 | { | |
447 | void *p; | |
448 | void *addr = page_address(page); | |
449 | ||
450 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
451 | set_bit(slab_index(p, s, addr), map); | |
452 | } | |
453 | ||
870b1fbb | 454 | static inline unsigned int size_from_object(struct kmem_cache *s) |
d86bd1be JK |
455 | { |
456 | if (s->flags & SLAB_RED_ZONE) | |
457 | return s->size - s->red_left_pad; | |
458 | ||
459 | return s->size; | |
460 | } | |
461 | ||
462 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
463 | { | |
464 | if (s->flags & SLAB_RED_ZONE) | |
465 | p -= s->red_left_pad; | |
466 | ||
467 | return p; | |
468 | } | |
469 | ||
41ecc55b CL |
470 | /* |
471 | * Debug settings: | |
472 | */ | |
89d3c87e | 473 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 474 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 475 | #else |
d50112ed | 476 | static slab_flags_t slub_debug; |
f0630fff | 477 | #endif |
41ecc55b CL |
478 | |
479 | static char *slub_debug_slabs; | |
fa5ec8a1 | 480 | static int disable_higher_order_debug; |
41ecc55b | 481 | |
a79316c6 AR |
482 | /* |
483 | * slub is about to manipulate internal object metadata. This memory lies | |
484 | * outside the range of the allocated object, so accessing it would normally | |
485 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
486 | * to tell kasan that these accesses are OK. | |
487 | */ | |
488 | static inline void metadata_access_enable(void) | |
489 | { | |
490 | kasan_disable_current(); | |
491 | } | |
492 | ||
493 | static inline void metadata_access_disable(void) | |
494 | { | |
495 | kasan_enable_current(); | |
496 | } | |
497 | ||
81819f0f CL |
498 | /* |
499 | * Object debugging | |
500 | */ | |
d86bd1be JK |
501 | |
502 | /* Verify that a pointer has an address that is valid within a slab page */ | |
503 | static inline int check_valid_pointer(struct kmem_cache *s, | |
504 | struct page *page, void *object) | |
505 | { | |
506 | void *base; | |
507 | ||
508 | if (!object) | |
509 | return 1; | |
510 | ||
511 | base = page_address(page); | |
512 | object = restore_red_left(s, object); | |
513 | if (object < base || object >= base + page->objects * s->size || | |
514 | (object - base) % s->size) { | |
515 | return 0; | |
516 | } | |
517 | ||
518 | return 1; | |
519 | } | |
520 | ||
aa2efd5e DT |
521 | static void print_section(char *level, char *text, u8 *addr, |
522 | unsigned int length) | |
81819f0f | 523 | { |
a79316c6 | 524 | metadata_access_enable(); |
aa2efd5e | 525 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
ffc79d28 | 526 | length, 1); |
a79316c6 | 527 | metadata_access_disable(); |
81819f0f CL |
528 | } |
529 | ||
81819f0f CL |
530 | static struct track *get_track(struct kmem_cache *s, void *object, |
531 | enum track_item alloc) | |
532 | { | |
533 | struct track *p; | |
534 | ||
535 | if (s->offset) | |
536 | p = object + s->offset + sizeof(void *); | |
537 | else | |
538 | p = object + s->inuse; | |
539 | ||
540 | return p + alloc; | |
541 | } | |
542 | ||
543 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 544 | enum track_item alloc, unsigned long addr) |
81819f0f | 545 | { |
1a00df4a | 546 | struct track *p = get_track(s, object, alloc); |
81819f0f | 547 | |
81819f0f | 548 | if (addr) { |
d6543e39 BG |
549 | #ifdef CONFIG_STACKTRACE |
550 | struct stack_trace trace; | |
551 | int i; | |
552 | ||
553 | trace.nr_entries = 0; | |
554 | trace.max_entries = TRACK_ADDRS_COUNT; | |
555 | trace.entries = p->addrs; | |
556 | trace.skip = 3; | |
a79316c6 | 557 | metadata_access_enable(); |
d6543e39 | 558 | save_stack_trace(&trace); |
a79316c6 | 559 | metadata_access_disable(); |
d6543e39 BG |
560 | |
561 | /* See rant in lockdep.c */ | |
562 | if (trace.nr_entries != 0 && | |
563 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
564 | trace.nr_entries--; | |
565 | ||
566 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
567 | p->addrs[i] = 0; | |
568 | #endif | |
81819f0f CL |
569 | p->addr = addr; |
570 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 571 | p->pid = current->pid; |
81819f0f CL |
572 | p->when = jiffies; |
573 | } else | |
574 | memset(p, 0, sizeof(struct track)); | |
575 | } | |
576 | ||
81819f0f CL |
577 | static void init_tracking(struct kmem_cache *s, void *object) |
578 | { | |
24922684 CL |
579 | if (!(s->flags & SLAB_STORE_USER)) |
580 | return; | |
581 | ||
ce71e27c EGM |
582 | set_track(s, object, TRACK_FREE, 0UL); |
583 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
584 | } |
585 | ||
86609d33 | 586 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f CL |
587 | { |
588 | if (!t->addr) | |
589 | return; | |
590 | ||
f9f58285 | 591 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 592 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
d6543e39 BG |
593 | #ifdef CONFIG_STACKTRACE |
594 | { | |
595 | int i; | |
596 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
597 | if (t->addrs[i]) | |
f9f58285 | 598 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
599 | else |
600 | break; | |
601 | } | |
602 | #endif | |
24922684 CL |
603 | } |
604 | ||
605 | static void print_tracking(struct kmem_cache *s, void *object) | |
606 | { | |
86609d33 | 607 | unsigned long pr_time = jiffies; |
24922684 CL |
608 | if (!(s->flags & SLAB_STORE_USER)) |
609 | return; | |
610 | ||
86609d33 CP |
611 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
612 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
613 | } |
614 | ||
615 | static void print_page_info(struct page *page) | |
616 | { | |
f9f58285 | 617 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 618 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
619 | |
620 | } | |
621 | ||
622 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
623 | { | |
ecc42fbe | 624 | struct va_format vaf; |
24922684 | 625 | va_list args; |
24922684 CL |
626 | |
627 | va_start(args, fmt); | |
ecc42fbe FF |
628 | vaf.fmt = fmt; |
629 | vaf.va = &args; | |
f9f58285 | 630 | pr_err("=============================================================================\n"); |
ecc42fbe | 631 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 632 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 633 | |
373d4d09 | 634 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 635 | va_end(args); |
81819f0f CL |
636 | } |
637 | ||
24922684 CL |
638 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
639 | { | |
ecc42fbe | 640 | struct va_format vaf; |
24922684 | 641 | va_list args; |
24922684 CL |
642 | |
643 | va_start(args, fmt); | |
ecc42fbe FF |
644 | vaf.fmt = fmt; |
645 | vaf.va = &args; | |
646 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 647 | va_end(args); |
24922684 CL |
648 | } |
649 | ||
650 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
651 | { |
652 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 653 | u8 *addr = page_address(page); |
24922684 CL |
654 | |
655 | print_tracking(s, p); | |
656 | ||
657 | print_page_info(page); | |
658 | ||
f9f58285 FF |
659 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
660 | p, p - addr, get_freepointer(s, p)); | |
24922684 | 661 | |
d86bd1be | 662 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e DT |
663 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
664 | s->red_left_pad); | |
d86bd1be | 665 | else if (p > addr + 16) |
aa2efd5e | 666 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 667 | |
aa2efd5e | 668 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 669 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 670 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e | 671 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 672 | s->inuse - s->object_size); |
81819f0f | 673 | |
81819f0f CL |
674 | if (s->offset) |
675 | off = s->offset + sizeof(void *); | |
676 | else | |
677 | off = s->inuse; | |
678 | ||
24922684 | 679 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 680 | off += 2 * sizeof(struct track); |
81819f0f | 681 | |
80a9201a AP |
682 | off += kasan_metadata_size(s); |
683 | ||
d86bd1be | 684 | if (off != size_from_object(s)) |
81819f0f | 685 | /* Beginning of the filler is the free pointer */ |
aa2efd5e DT |
686 | print_section(KERN_ERR, "Padding ", p + off, |
687 | size_from_object(s) - off); | |
24922684 CL |
688 | |
689 | dump_stack(); | |
81819f0f CL |
690 | } |
691 | ||
75c66def | 692 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
693 | u8 *object, char *reason) |
694 | { | |
3dc50637 | 695 | slab_bug(s, "%s", reason); |
24922684 | 696 | print_trailer(s, page, object); |
81819f0f CL |
697 | } |
698 | ||
a38965bf | 699 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page, |
d0e0ac97 | 700 | const char *fmt, ...) |
81819f0f CL |
701 | { |
702 | va_list args; | |
703 | char buf[100]; | |
704 | ||
24922684 CL |
705 | va_start(args, fmt); |
706 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 707 | va_end(args); |
3dc50637 | 708 | slab_bug(s, "%s", buf); |
24922684 | 709 | print_page_info(page); |
81819f0f CL |
710 | dump_stack(); |
711 | } | |
712 | ||
f7cb1933 | 713 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
714 | { |
715 | u8 *p = object; | |
716 | ||
d86bd1be JK |
717 | if (s->flags & SLAB_RED_ZONE) |
718 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
719 | ||
81819f0f | 720 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
721 | memset(p, POISON_FREE, s->object_size - 1); |
722 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
723 | } |
724 | ||
725 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 726 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
727 | } |
728 | ||
24922684 CL |
729 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
730 | void *from, void *to) | |
731 | { | |
732 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
733 | memset(from, data, to - from); | |
734 | } | |
735 | ||
736 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
737 | u8 *object, char *what, | |
06428780 | 738 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
739 | { |
740 | u8 *fault; | |
741 | u8 *end; | |
742 | ||
a79316c6 | 743 | metadata_access_enable(); |
79824820 | 744 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 745 | metadata_access_disable(); |
24922684 CL |
746 | if (!fault) |
747 | return 1; | |
748 | ||
749 | end = start + bytes; | |
750 | while (end > fault && end[-1] == value) | |
751 | end--; | |
752 | ||
753 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 754 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
755 | fault, end - 1, fault[0], value); |
756 | print_trailer(s, page, object); | |
757 | ||
758 | restore_bytes(s, what, value, fault, end); | |
759 | return 0; | |
81819f0f CL |
760 | } |
761 | ||
81819f0f CL |
762 | /* |
763 | * Object layout: | |
764 | * | |
765 | * object address | |
766 | * Bytes of the object to be managed. | |
767 | * If the freepointer may overlay the object then the free | |
768 | * pointer is the first word of the object. | |
672bba3a | 769 | * |
81819f0f CL |
770 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
771 | * 0xa5 (POISON_END) | |
772 | * | |
3b0efdfa | 773 | * object + s->object_size |
81819f0f | 774 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 775 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 776 | * object_size == inuse. |
672bba3a | 777 | * |
81819f0f CL |
778 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
779 | * 0xcc (RED_ACTIVE) for objects in use. | |
780 | * | |
781 | * object + s->inuse | |
672bba3a CL |
782 | * Meta data starts here. |
783 | * | |
81819f0f CL |
784 | * A. Free pointer (if we cannot overwrite object on free) |
785 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 786 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 787 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
788 | * before the word boundary. |
789 | * | |
790 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
791 | * |
792 | * object + s->size | |
672bba3a | 793 | * Nothing is used beyond s->size. |
81819f0f | 794 | * |
3b0efdfa | 795 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 796 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
797 | * may be used with merged slabcaches. |
798 | */ | |
799 | ||
81819f0f CL |
800 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
801 | { | |
802 | unsigned long off = s->inuse; /* The end of info */ | |
803 | ||
804 | if (s->offset) | |
805 | /* Freepointer is placed after the object. */ | |
806 | off += sizeof(void *); | |
807 | ||
808 | if (s->flags & SLAB_STORE_USER) | |
809 | /* We also have user information there */ | |
810 | off += 2 * sizeof(struct track); | |
811 | ||
80a9201a AP |
812 | off += kasan_metadata_size(s); |
813 | ||
d86bd1be | 814 | if (size_from_object(s) == off) |
81819f0f CL |
815 | return 1; |
816 | ||
24922684 | 817 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 818 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
819 | } |
820 | ||
39b26464 | 821 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
822 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
823 | { | |
24922684 CL |
824 | u8 *start; |
825 | u8 *fault; | |
826 | u8 *end; | |
5d682681 | 827 | u8 *pad; |
24922684 CL |
828 | int length; |
829 | int remainder; | |
81819f0f CL |
830 | |
831 | if (!(s->flags & SLAB_POISON)) | |
832 | return 1; | |
833 | ||
a973e9dd | 834 | start = page_address(page); |
9736d2a9 | 835 | length = PAGE_SIZE << compound_order(page); |
39b26464 CL |
836 | end = start + length; |
837 | remainder = length % s->size; | |
81819f0f CL |
838 | if (!remainder) |
839 | return 1; | |
840 | ||
5d682681 | 841 | pad = end - remainder; |
a79316c6 | 842 | metadata_access_enable(); |
5d682681 | 843 | fault = memchr_inv(pad, POISON_INUSE, remainder); |
a79316c6 | 844 | metadata_access_disable(); |
24922684 CL |
845 | if (!fault) |
846 | return 1; | |
847 | while (end > fault && end[-1] == POISON_INUSE) | |
848 | end--; | |
849 | ||
850 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
5d682681 | 851 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 852 | |
5d682681 | 853 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
24922684 | 854 | return 0; |
81819f0f CL |
855 | } |
856 | ||
857 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 858 | void *object, u8 val) |
81819f0f CL |
859 | { |
860 | u8 *p = object; | |
3b0efdfa | 861 | u8 *endobject = object + s->object_size; |
81819f0f CL |
862 | |
863 | if (s->flags & SLAB_RED_ZONE) { | |
d86bd1be JK |
864 | if (!check_bytes_and_report(s, page, object, "Redzone", |
865 | object - s->red_left_pad, val, s->red_left_pad)) | |
866 | return 0; | |
867 | ||
24922684 | 868 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 869 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 870 | return 0; |
81819f0f | 871 | } else { |
3b0efdfa | 872 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 873 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
874 | endobject, POISON_INUSE, |
875 | s->inuse - s->object_size); | |
3adbefee | 876 | } |
81819f0f CL |
877 | } |
878 | ||
879 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 880 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 881 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 882 | POISON_FREE, s->object_size - 1) || |
24922684 | 883 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 884 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 885 | return 0; |
81819f0f CL |
886 | /* |
887 | * check_pad_bytes cleans up on its own. | |
888 | */ | |
889 | check_pad_bytes(s, page, p); | |
890 | } | |
891 | ||
f7cb1933 | 892 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
893 | /* |
894 | * Object and freepointer overlap. Cannot check | |
895 | * freepointer while object is allocated. | |
896 | */ | |
897 | return 1; | |
898 | ||
899 | /* Check free pointer validity */ | |
900 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
901 | object_err(s, page, p, "Freepointer corrupt"); | |
902 | /* | |
9f6c708e | 903 | * No choice but to zap it and thus lose the remainder |
81819f0f | 904 | * of the free objects in this slab. May cause |
672bba3a | 905 | * another error because the object count is now wrong. |
81819f0f | 906 | */ |
a973e9dd | 907 | set_freepointer(s, p, NULL); |
81819f0f CL |
908 | return 0; |
909 | } | |
910 | return 1; | |
911 | } | |
912 | ||
913 | static int check_slab(struct kmem_cache *s, struct page *page) | |
914 | { | |
39b26464 CL |
915 | int maxobj; |
916 | ||
81819f0f CL |
917 | VM_BUG_ON(!irqs_disabled()); |
918 | ||
919 | if (!PageSlab(page)) { | |
24922684 | 920 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
921 | return 0; |
922 | } | |
39b26464 | 923 | |
9736d2a9 | 924 | maxobj = order_objects(compound_order(page), s->size); |
39b26464 CL |
925 | if (page->objects > maxobj) { |
926 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 927 | page->objects, maxobj); |
39b26464 CL |
928 | return 0; |
929 | } | |
930 | if (page->inuse > page->objects) { | |
24922684 | 931 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 932 | page->inuse, page->objects); |
81819f0f CL |
933 | return 0; |
934 | } | |
935 | /* Slab_pad_check fixes things up after itself */ | |
936 | slab_pad_check(s, page); | |
937 | return 1; | |
938 | } | |
939 | ||
940 | /* | |
672bba3a CL |
941 | * Determine if a certain object on a page is on the freelist. Must hold the |
942 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
943 | */ |
944 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
945 | { | |
946 | int nr = 0; | |
881db7fb | 947 | void *fp; |
81819f0f | 948 | void *object = NULL; |
f6edde9c | 949 | int max_objects; |
81819f0f | 950 | |
881db7fb | 951 | fp = page->freelist; |
39b26464 | 952 | while (fp && nr <= page->objects) { |
81819f0f CL |
953 | if (fp == search) |
954 | return 1; | |
955 | if (!check_valid_pointer(s, page, fp)) { | |
956 | if (object) { | |
957 | object_err(s, page, object, | |
958 | "Freechain corrupt"); | |
a973e9dd | 959 | set_freepointer(s, object, NULL); |
81819f0f | 960 | } else { |
24922684 | 961 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 962 | page->freelist = NULL; |
39b26464 | 963 | page->inuse = page->objects; |
24922684 | 964 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
965 | return 0; |
966 | } | |
967 | break; | |
968 | } | |
969 | object = fp; | |
970 | fp = get_freepointer(s, object); | |
971 | nr++; | |
972 | } | |
973 | ||
9736d2a9 | 974 | max_objects = order_objects(compound_order(page), s->size); |
210b5c06 CG |
975 | if (max_objects > MAX_OBJS_PER_PAGE) |
976 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
977 | |
978 | if (page->objects != max_objects) { | |
756a025f JP |
979 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
980 | page->objects, max_objects); | |
224a88be CL |
981 | page->objects = max_objects; |
982 | slab_fix(s, "Number of objects adjusted."); | |
983 | } | |
39b26464 | 984 | if (page->inuse != page->objects - nr) { |
756a025f JP |
985 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
986 | page->inuse, page->objects - nr); | |
39b26464 | 987 | page->inuse = page->objects - nr; |
24922684 | 988 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
989 | } |
990 | return search == NULL; | |
991 | } | |
992 | ||
0121c619 CL |
993 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
994 | int alloc) | |
3ec09742 CL |
995 | { |
996 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 997 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
998 | s->name, |
999 | alloc ? "alloc" : "free", | |
1000 | object, page->inuse, | |
1001 | page->freelist); | |
1002 | ||
1003 | if (!alloc) | |
aa2efd5e | 1004 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1005 | s->object_size); |
3ec09742 CL |
1006 | |
1007 | dump_stack(); | |
1008 | } | |
1009 | } | |
1010 | ||
643b1138 | 1011 | /* |
672bba3a | 1012 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1013 | */ |
5cc6eee8 CL |
1014 | static void add_full(struct kmem_cache *s, |
1015 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1016 | { |
5cc6eee8 CL |
1017 | if (!(s->flags & SLAB_STORE_USER)) |
1018 | return; | |
1019 | ||
255d0884 | 1020 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1021 | list_add(&page->lru, &n->full); |
643b1138 CL |
1022 | } |
1023 | ||
c65c1877 | 1024 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1025 | { |
643b1138 CL |
1026 | if (!(s->flags & SLAB_STORE_USER)) |
1027 | return; | |
1028 | ||
255d0884 | 1029 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1030 | list_del(&page->lru); |
643b1138 CL |
1031 | } |
1032 | ||
0f389ec6 CL |
1033 | /* Tracking of the number of slabs for debugging purposes */ |
1034 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1035 | { | |
1036 | struct kmem_cache_node *n = get_node(s, node); | |
1037 | ||
1038 | return atomic_long_read(&n->nr_slabs); | |
1039 | } | |
1040 | ||
26c02cf0 AB |
1041 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1042 | { | |
1043 | return atomic_long_read(&n->nr_slabs); | |
1044 | } | |
1045 | ||
205ab99d | 1046 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1047 | { |
1048 | struct kmem_cache_node *n = get_node(s, node); | |
1049 | ||
1050 | /* | |
1051 | * May be called early in order to allocate a slab for the | |
1052 | * kmem_cache_node structure. Solve the chicken-egg | |
1053 | * dilemma by deferring the increment of the count during | |
1054 | * bootstrap (see early_kmem_cache_node_alloc). | |
1055 | */ | |
338b2642 | 1056 | if (likely(n)) { |
0f389ec6 | 1057 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1058 | atomic_long_add(objects, &n->total_objects); |
1059 | } | |
0f389ec6 | 1060 | } |
205ab99d | 1061 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1062 | { |
1063 | struct kmem_cache_node *n = get_node(s, node); | |
1064 | ||
1065 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1066 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1067 | } |
1068 | ||
1069 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1070 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1071 | void *object) | |
1072 | { | |
1073 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1074 | return; | |
1075 | ||
f7cb1933 | 1076 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1077 | init_tracking(s, object); |
1078 | } | |
1079 | ||
becfda68 | 1080 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
d0e0ac97 | 1081 | struct page *page, |
ce71e27c | 1082 | void *object, unsigned long addr) |
81819f0f CL |
1083 | { |
1084 | if (!check_slab(s, page)) | |
becfda68 | 1085 | return 0; |
81819f0f | 1086 | |
81819f0f CL |
1087 | if (!check_valid_pointer(s, page, object)) { |
1088 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1089 | return 0; |
81819f0f CL |
1090 | } |
1091 | ||
f7cb1933 | 1092 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1093 | return 0; |
1094 | ||
1095 | return 1; | |
1096 | } | |
1097 | ||
1098 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1099 | struct page *page, | |
1100 | void *object, unsigned long addr) | |
1101 | { | |
1102 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1103 | if (!alloc_consistency_checks(s, page, object, addr)) | |
1104 | goto bad; | |
1105 | } | |
81819f0f | 1106 | |
3ec09742 CL |
1107 | /* Success perform special debug activities for allocs */ |
1108 | if (s->flags & SLAB_STORE_USER) | |
1109 | set_track(s, object, TRACK_ALLOC, addr); | |
1110 | trace(s, page, object, 1); | |
f7cb1933 | 1111 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1112 | return 1; |
3ec09742 | 1113 | |
81819f0f CL |
1114 | bad: |
1115 | if (PageSlab(page)) { | |
1116 | /* | |
1117 | * If this is a slab page then lets do the best we can | |
1118 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1119 | * as used avoids touching the remaining objects. |
81819f0f | 1120 | */ |
24922684 | 1121 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1122 | page->inuse = page->objects; |
a973e9dd | 1123 | page->freelist = NULL; |
81819f0f CL |
1124 | } |
1125 | return 0; | |
1126 | } | |
1127 | ||
becfda68 LA |
1128 | static inline int free_consistency_checks(struct kmem_cache *s, |
1129 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1130 | { |
81819f0f | 1131 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1132 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1133 | return 0; |
81819f0f CL |
1134 | } |
1135 | ||
1136 | if (on_freelist(s, page, object)) { | |
24922684 | 1137 | object_err(s, page, object, "Object already free"); |
becfda68 | 1138 | return 0; |
81819f0f CL |
1139 | } |
1140 | ||
f7cb1933 | 1141 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1142 | return 0; |
81819f0f | 1143 | |
1b4f59e3 | 1144 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1145 | if (!PageSlab(page)) { |
756a025f JP |
1146 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1147 | object); | |
1b4f59e3 | 1148 | } else if (!page->slab_cache) { |
f9f58285 FF |
1149 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1150 | object); | |
70d71228 | 1151 | dump_stack(); |
06428780 | 1152 | } else |
24922684 CL |
1153 | object_err(s, page, object, |
1154 | "page slab pointer corrupt."); | |
becfda68 LA |
1155 | return 0; |
1156 | } | |
1157 | return 1; | |
1158 | } | |
1159 | ||
1160 | /* Supports checking bulk free of a constructed freelist */ | |
1161 | static noinline int free_debug_processing( | |
1162 | struct kmem_cache *s, struct page *page, | |
1163 | void *head, void *tail, int bulk_cnt, | |
1164 | unsigned long addr) | |
1165 | { | |
1166 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1167 | void *object = head; | |
1168 | int cnt = 0; | |
1169 | unsigned long uninitialized_var(flags); | |
1170 | int ret = 0; | |
1171 | ||
1172 | spin_lock_irqsave(&n->list_lock, flags); | |
1173 | slab_lock(page); | |
1174 | ||
1175 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1176 | if (!check_slab(s, page)) | |
1177 | goto out; | |
1178 | } | |
1179 | ||
1180 | next_object: | |
1181 | cnt++; | |
1182 | ||
1183 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1184 | if (!free_consistency_checks(s, page, object, addr)) | |
1185 | goto out; | |
81819f0f | 1186 | } |
3ec09742 | 1187 | |
3ec09742 CL |
1188 | if (s->flags & SLAB_STORE_USER) |
1189 | set_track(s, object, TRACK_FREE, addr); | |
1190 | trace(s, page, object, 0); | |
81084651 | 1191 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1192 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1193 | |
1194 | /* Reached end of constructed freelist yet? */ | |
1195 | if (object != tail) { | |
1196 | object = get_freepointer(s, object); | |
1197 | goto next_object; | |
1198 | } | |
804aa132 LA |
1199 | ret = 1; |
1200 | ||
5c2e4bbb | 1201 | out: |
81084651 JDB |
1202 | if (cnt != bulk_cnt) |
1203 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1204 | bulk_cnt, cnt); | |
1205 | ||
881db7fb | 1206 | slab_unlock(page); |
282acb43 | 1207 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1208 | if (!ret) |
1209 | slab_fix(s, "Object at 0x%p not freed", object); | |
1210 | return ret; | |
81819f0f CL |
1211 | } |
1212 | ||
41ecc55b CL |
1213 | static int __init setup_slub_debug(char *str) |
1214 | { | |
f0630fff CL |
1215 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1216 | if (*str++ != '=' || !*str) | |
1217 | /* | |
1218 | * No options specified. Switch on full debugging. | |
1219 | */ | |
1220 | goto out; | |
1221 | ||
1222 | if (*str == ',') | |
1223 | /* | |
1224 | * No options but restriction on slabs. This means full | |
1225 | * debugging for slabs matching a pattern. | |
1226 | */ | |
1227 | goto check_slabs; | |
1228 | ||
1229 | slub_debug = 0; | |
1230 | if (*str == '-') | |
1231 | /* | |
1232 | * Switch off all debugging measures. | |
1233 | */ | |
1234 | goto out; | |
1235 | ||
1236 | /* | |
1237 | * Determine which debug features should be switched on | |
1238 | */ | |
06428780 | 1239 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1240 | switch (tolower(*str)) { |
1241 | case 'f': | |
becfda68 | 1242 | slub_debug |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1243 | break; |
1244 | case 'z': | |
1245 | slub_debug |= SLAB_RED_ZONE; | |
1246 | break; | |
1247 | case 'p': | |
1248 | slub_debug |= SLAB_POISON; | |
1249 | break; | |
1250 | case 'u': | |
1251 | slub_debug |= SLAB_STORE_USER; | |
1252 | break; | |
1253 | case 't': | |
1254 | slub_debug |= SLAB_TRACE; | |
1255 | break; | |
4c13dd3b DM |
1256 | case 'a': |
1257 | slub_debug |= SLAB_FAILSLAB; | |
1258 | break; | |
08303a73 CA |
1259 | case 'o': |
1260 | /* | |
1261 | * Avoid enabling debugging on caches if its minimum | |
1262 | * order would increase as a result. | |
1263 | */ | |
1264 | disable_higher_order_debug = 1; | |
1265 | break; | |
f0630fff | 1266 | default: |
f9f58285 FF |
1267 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1268 | *str); | |
f0630fff | 1269 | } |
41ecc55b CL |
1270 | } |
1271 | ||
f0630fff | 1272 | check_slabs: |
41ecc55b CL |
1273 | if (*str == ',') |
1274 | slub_debug_slabs = str + 1; | |
f0630fff | 1275 | out: |
41ecc55b CL |
1276 | return 1; |
1277 | } | |
1278 | ||
1279 | __setup("slub_debug", setup_slub_debug); | |
1280 | ||
0293d1fd | 1281 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1282 | slab_flags_t flags, const char *name, |
51cc5068 | 1283 | void (*ctor)(void *)) |
41ecc55b CL |
1284 | { |
1285 | /* | |
e153362a | 1286 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1287 | */ |
c6f58d9b CL |
1288 | if (slub_debug && (!slub_debug_slabs || (name && |
1289 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1290 | flags |= slub_debug; |
ba0268a8 CL |
1291 | |
1292 | return flags; | |
41ecc55b | 1293 | } |
b4a64718 | 1294 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1295 | static inline void setup_object_debug(struct kmem_cache *s, |
1296 | struct page *page, void *object) {} | |
41ecc55b | 1297 | |
3ec09742 | 1298 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1299 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1300 | |
282acb43 | 1301 | static inline int free_debug_processing( |
81084651 JDB |
1302 | struct kmem_cache *s, struct page *page, |
1303 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1304 | unsigned long addr) { return 0; } |
41ecc55b | 1305 | |
41ecc55b CL |
1306 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1307 | { return 1; } | |
1308 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1309 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1310 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1311 | struct page *page) {} | |
c65c1877 PZ |
1312 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1313 | struct page *page) {} | |
0293d1fd | 1314 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1315 | slab_flags_t flags, const char *name, |
51cc5068 | 1316 | void (*ctor)(void *)) |
ba0268a8 CL |
1317 | { |
1318 | return flags; | |
1319 | } | |
41ecc55b | 1320 | #define slub_debug 0 |
0f389ec6 | 1321 | |
fdaa45e9 IM |
1322 | #define disable_higher_order_debug 0 |
1323 | ||
0f389ec6 CL |
1324 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1325 | { return 0; } | |
26c02cf0 AB |
1326 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1327 | { return 0; } | |
205ab99d CL |
1328 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1329 | int objects) {} | |
1330 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1331 | int objects) {} | |
7d550c56 | 1332 | |
02e72cc6 AR |
1333 | #endif /* CONFIG_SLUB_DEBUG */ |
1334 | ||
1335 | /* | |
1336 | * Hooks for other subsystems that check memory allocations. In a typical | |
1337 | * production configuration these hooks all should produce no code at all. | |
1338 | */ | |
d56791b3 RB |
1339 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1340 | { | |
1341 | kmemleak_alloc(ptr, size, 1, flags); | |
505f5dcb | 1342 | kasan_kmalloc_large(ptr, size, flags); |
d56791b3 RB |
1343 | } |
1344 | ||
ee3ce779 | 1345 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1346 | { |
1347 | kmemleak_free(x); | |
ee3ce779 | 1348 | kasan_kfree_large(x, _RET_IP_); |
d56791b3 RB |
1349 | } |
1350 | ||
c3895391 | 1351 | static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x) |
d56791b3 RB |
1352 | { |
1353 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1354 | |
02e72cc6 AR |
1355 | /* |
1356 | * Trouble is that we may no longer disable interrupts in the fast path | |
1357 | * So in order to make the debug calls that expect irqs to be | |
1358 | * disabled we need to disable interrupts temporarily. | |
1359 | */ | |
4675ff05 | 1360 | #ifdef CONFIG_LOCKDEP |
02e72cc6 AR |
1361 | { |
1362 | unsigned long flags; | |
1363 | ||
1364 | local_irq_save(flags); | |
02e72cc6 AR |
1365 | debug_check_no_locks_freed(x, s->object_size); |
1366 | local_irq_restore(flags); | |
1367 | } | |
1368 | #endif | |
1369 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1370 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1371 | |
c3895391 AK |
1372 | /* KASAN might put x into memory quarantine, delaying its reuse */ |
1373 | return kasan_slab_free(s, x, _RET_IP_); | |
02e72cc6 | 1374 | } |
205ab99d | 1375 | |
c3895391 AK |
1376 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
1377 | void **head, void **tail) | |
81084651 JDB |
1378 | { |
1379 | /* | |
1380 | * Compiler cannot detect this function can be removed if slab_free_hook() | |
1381 | * evaluates to nothing. Thus, catch all relevant config debug options here. | |
1382 | */ | |
4675ff05 | 1383 | #if defined(CONFIG_LOCKDEP) || \ |
81084651 JDB |
1384 | defined(CONFIG_DEBUG_KMEMLEAK) || \ |
1385 | defined(CONFIG_DEBUG_OBJECTS_FREE) || \ | |
1386 | defined(CONFIG_KASAN) | |
1387 | ||
c3895391 AK |
1388 | void *object; |
1389 | void *next = *head; | |
1390 | void *old_tail = *tail ? *tail : *head; | |
1391 | ||
1392 | /* Head and tail of the reconstructed freelist */ | |
1393 | *head = NULL; | |
1394 | *tail = NULL; | |
81084651 JDB |
1395 | |
1396 | do { | |
c3895391 AK |
1397 | object = next; |
1398 | next = get_freepointer(s, object); | |
1399 | /* If object's reuse doesn't have to be delayed */ | |
1400 | if (!slab_free_hook(s, object)) { | |
1401 | /* Move object to the new freelist */ | |
1402 | set_freepointer(s, object, *head); | |
1403 | *head = object; | |
1404 | if (!*tail) | |
1405 | *tail = object; | |
1406 | } | |
1407 | } while (object != old_tail); | |
1408 | ||
1409 | if (*head == *tail) | |
1410 | *tail = NULL; | |
1411 | ||
1412 | return *head != NULL; | |
1413 | #else | |
1414 | return true; | |
81084651 JDB |
1415 | #endif |
1416 | } | |
1417 | ||
588f8ba9 TG |
1418 | static void setup_object(struct kmem_cache *s, struct page *page, |
1419 | void *object) | |
1420 | { | |
1421 | setup_object_debug(s, page, object); | |
b3cbd9bf | 1422 | kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1423 | if (unlikely(s->ctor)) { |
1424 | kasan_unpoison_object_data(s, object); | |
1425 | s->ctor(object); | |
1426 | kasan_poison_object_data(s, object); | |
1427 | } | |
1428 | } | |
1429 | ||
81819f0f CL |
1430 | /* |
1431 | * Slab allocation and freeing | |
1432 | */ | |
5dfb4175 VD |
1433 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1434 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1435 | { |
5dfb4175 | 1436 | struct page *page; |
19af27af | 1437 | unsigned int order = oo_order(oo); |
65c3376a | 1438 | |
2154a336 | 1439 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1440 | page = alloc_pages(flags, order); |
65c3376a | 1441 | else |
96db800f | 1442 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1443 | |
f3ccb2c4 VD |
1444 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1445 | __free_pages(page, order); | |
1446 | page = NULL; | |
1447 | } | |
5dfb4175 VD |
1448 | |
1449 | return page; | |
65c3376a CL |
1450 | } |
1451 | ||
210e7a43 TG |
1452 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1453 | /* Pre-initialize the random sequence cache */ | |
1454 | static int init_cache_random_seq(struct kmem_cache *s) | |
1455 | { | |
19af27af | 1456 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 1457 | int err; |
210e7a43 | 1458 | |
a810007a SR |
1459 | /* Bailout if already initialised */ |
1460 | if (s->random_seq) | |
1461 | return 0; | |
1462 | ||
210e7a43 TG |
1463 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1464 | if (err) { | |
1465 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1466 | s->name); | |
1467 | return err; | |
1468 | } | |
1469 | ||
1470 | /* Transform to an offset on the set of pages */ | |
1471 | if (s->random_seq) { | |
19af27af AD |
1472 | unsigned int i; |
1473 | ||
210e7a43 TG |
1474 | for (i = 0; i < count; i++) |
1475 | s->random_seq[i] *= s->size; | |
1476 | } | |
1477 | return 0; | |
1478 | } | |
1479 | ||
1480 | /* Initialize each random sequence freelist per cache */ | |
1481 | static void __init init_freelist_randomization(void) | |
1482 | { | |
1483 | struct kmem_cache *s; | |
1484 | ||
1485 | mutex_lock(&slab_mutex); | |
1486 | ||
1487 | list_for_each_entry(s, &slab_caches, list) | |
1488 | init_cache_random_seq(s); | |
1489 | ||
1490 | mutex_unlock(&slab_mutex); | |
1491 | } | |
1492 | ||
1493 | /* Get the next entry on the pre-computed freelist randomized */ | |
1494 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1495 | unsigned long *pos, void *start, | |
1496 | unsigned long page_limit, | |
1497 | unsigned long freelist_count) | |
1498 | { | |
1499 | unsigned int idx; | |
1500 | ||
1501 | /* | |
1502 | * If the target page allocation failed, the number of objects on the | |
1503 | * page might be smaller than the usual size defined by the cache. | |
1504 | */ | |
1505 | do { | |
1506 | idx = s->random_seq[*pos]; | |
1507 | *pos += 1; | |
1508 | if (*pos >= freelist_count) | |
1509 | *pos = 0; | |
1510 | } while (unlikely(idx >= page_limit)); | |
1511 | ||
1512 | return (char *)start + idx; | |
1513 | } | |
1514 | ||
1515 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1516 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1517 | { | |
1518 | void *start; | |
1519 | void *cur; | |
1520 | void *next; | |
1521 | unsigned long idx, pos, page_limit, freelist_count; | |
1522 | ||
1523 | if (page->objects < 2 || !s->random_seq) | |
1524 | return false; | |
1525 | ||
1526 | freelist_count = oo_objects(s->oo); | |
1527 | pos = get_random_int() % freelist_count; | |
1528 | ||
1529 | page_limit = page->objects * s->size; | |
1530 | start = fixup_red_left(s, page_address(page)); | |
1531 | ||
1532 | /* First entry is used as the base of the freelist */ | |
1533 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1534 | freelist_count); | |
1535 | page->freelist = cur; | |
1536 | ||
1537 | for (idx = 1; idx < page->objects; idx++) { | |
1538 | setup_object(s, page, cur); | |
1539 | next = next_freelist_entry(s, page, &pos, start, page_limit, | |
1540 | freelist_count); | |
1541 | set_freepointer(s, cur, next); | |
1542 | cur = next; | |
1543 | } | |
1544 | setup_object(s, page, cur); | |
1545 | set_freepointer(s, cur, NULL); | |
1546 | ||
1547 | return true; | |
1548 | } | |
1549 | #else | |
1550 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1551 | { | |
1552 | return 0; | |
1553 | } | |
1554 | static inline void init_freelist_randomization(void) { } | |
1555 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1556 | { | |
1557 | return false; | |
1558 | } | |
1559 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1560 | ||
81819f0f CL |
1561 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1562 | { | |
06428780 | 1563 | struct page *page; |
834f3d11 | 1564 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1565 | gfp_t alloc_gfp; |
588f8ba9 TG |
1566 | void *start, *p; |
1567 | int idx, order; | |
210e7a43 | 1568 | bool shuffle; |
81819f0f | 1569 | |
7e0528da CL |
1570 | flags &= gfp_allowed_mask; |
1571 | ||
d0164adc | 1572 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1573 | local_irq_enable(); |
1574 | ||
b7a49f0d | 1575 | flags |= s->allocflags; |
e12ba74d | 1576 | |
ba52270d PE |
1577 | /* |
1578 | * Let the initial higher-order allocation fail under memory pressure | |
1579 | * so we fall-back to the minimum order allocation. | |
1580 | */ | |
1581 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1582 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1583 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1584 | |
5dfb4175 | 1585 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1586 | if (unlikely(!page)) { |
1587 | oo = s->min; | |
80c3a998 | 1588 | alloc_gfp = flags; |
65c3376a CL |
1589 | /* |
1590 | * Allocation may have failed due to fragmentation. | |
1591 | * Try a lower order alloc if possible | |
1592 | */ | |
5dfb4175 | 1593 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1594 | if (unlikely(!page)) |
1595 | goto out; | |
1596 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1597 | } |
5a896d9e | 1598 | |
834f3d11 | 1599 | page->objects = oo_objects(oo); |
81819f0f | 1600 | |
1f458cbf | 1601 | order = compound_order(page); |
1b4f59e3 | 1602 | page->slab_cache = s; |
c03f94cc | 1603 | __SetPageSlab(page); |
2f064f34 | 1604 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1605 | SetPageSlabPfmemalloc(page); |
81819f0f CL |
1606 | |
1607 | start = page_address(page); | |
81819f0f CL |
1608 | |
1609 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1610 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1611 | |
0316bec2 AR |
1612 | kasan_poison_slab(page); |
1613 | ||
210e7a43 TG |
1614 | shuffle = shuffle_freelist(s, page); |
1615 | ||
1616 | if (!shuffle) { | |
1617 | for_each_object_idx(p, idx, s, start, page->objects) { | |
1618 | setup_object(s, page, p); | |
1619 | if (likely(idx < page->objects)) | |
1620 | set_freepointer(s, p, p + s->size); | |
1621 | else | |
1622 | set_freepointer(s, p, NULL); | |
1623 | } | |
1624 | page->freelist = fixup_red_left(s, start); | |
81819f0f | 1625 | } |
81819f0f | 1626 | |
e6e82ea1 | 1627 | page->inuse = page->objects; |
8cb0a506 | 1628 | page->frozen = 1; |
588f8ba9 | 1629 | |
81819f0f | 1630 | out: |
d0164adc | 1631 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1632 | local_irq_disable(); |
1633 | if (!page) | |
1634 | return NULL; | |
1635 | ||
7779f212 | 1636 | mod_lruvec_page_state(page, |
588f8ba9 TG |
1637 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? |
1638 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1639 | 1 << oo_order(oo)); | |
1640 | ||
1641 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1642 | ||
81819f0f CL |
1643 | return page; |
1644 | } | |
1645 | ||
588f8ba9 TG |
1646 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1647 | { | |
1648 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
bacdcb34 | 1649 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
1650 | flags &= ~GFP_SLAB_BUG_MASK; |
1651 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
1652 | invalid_mask, &invalid_mask, flags, &flags); | |
65b9de75 | 1653 | dump_stack(); |
588f8ba9 TG |
1654 | } |
1655 | ||
1656 | return allocate_slab(s, | |
1657 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1658 | } | |
1659 | ||
81819f0f CL |
1660 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1661 | { | |
834f3d11 CL |
1662 | int order = compound_order(page); |
1663 | int pages = 1 << order; | |
81819f0f | 1664 | |
becfda68 | 1665 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { |
81819f0f CL |
1666 | void *p; |
1667 | ||
1668 | slab_pad_check(s, page); | |
224a88be CL |
1669 | for_each_object(p, s, page_address(page), |
1670 | page->objects) | |
f7cb1933 | 1671 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1672 | } |
1673 | ||
7779f212 | 1674 | mod_lruvec_page_state(page, |
81819f0f CL |
1675 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? |
1676 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1677 | -pages); |
81819f0f | 1678 | |
072bb0aa | 1679 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1680 | __ClearPageSlab(page); |
1f458cbf | 1681 | |
d4fc5069 | 1682 | page->mapping = NULL; |
1eb5ac64 NP |
1683 | if (current->reclaim_state) |
1684 | current->reclaim_state->reclaimed_slab += pages; | |
27ee57c9 VD |
1685 | memcg_uncharge_slab(page, order, s); |
1686 | __free_pages(page, order); | |
81819f0f CL |
1687 | } |
1688 | ||
1689 | static void rcu_free_slab(struct rcu_head *h) | |
1690 | { | |
bf68c214 | 1691 | struct page *page = container_of(h, struct page, rcu_head); |
da9a638c | 1692 | |
1b4f59e3 | 1693 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1694 | } |
1695 | ||
1696 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1697 | { | |
5f0d5a3a | 1698 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bf68c214 | 1699 | call_rcu(&page->rcu_head, rcu_free_slab); |
81819f0f CL |
1700 | } else |
1701 | __free_slab(s, page); | |
1702 | } | |
1703 | ||
1704 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1705 | { | |
205ab99d | 1706 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1707 | free_slab(s, page); |
1708 | } | |
1709 | ||
1710 | /* | |
5cc6eee8 | 1711 | * Management of partially allocated slabs. |
81819f0f | 1712 | */ |
1e4dd946 SR |
1713 | static inline void |
1714 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1715 | { |
e95eed57 | 1716 | n->nr_partial++; |
136333d1 | 1717 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1718 | list_add_tail(&page->lru, &n->partial); |
1719 | else | |
1720 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1721 | } |
1722 | ||
1e4dd946 SR |
1723 | static inline void add_partial(struct kmem_cache_node *n, |
1724 | struct page *page, int tail) | |
62e346a8 | 1725 | { |
c65c1877 | 1726 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1727 | __add_partial(n, page, tail); |
1728 | } | |
c65c1877 | 1729 | |
1e4dd946 SR |
1730 | static inline void remove_partial(struct kmem_cache_node *n, |
1731 | struct page *page) | |
1732 | { | |
1733 | lockdep_assert_held(&n->list_lock); | |
52b4b950 DS |
1734 | list_del(&page->lru); |
1735 | n->nr_partial--; | |
1e4dd946 SR |
1736 | } |
1737 | ||
81819f0f | 1738 | /* |
7ced3719 CL |
1739 | * Remove slab from the partial list, freeze it and |
1740 | * return the pointer to the freelist. | |
81819f0f | 1741 | * |
497b66f2 | 1742 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1743 | */ |
497b66f2 | 1744 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1745 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1746 | int mode, int *objects) |
81819f0f | 1747 | { |
2cfb7455 CL |
1748 | void *freelist; |
1749 | unsigned long counters; | |
1750 | struct page new; | |
1751 | ||
c65c1877 PZ |
1752 | lockdep_assert_held(&n->list_lock); |
1753 | ||
2cfb7455 CL |
1754 | /* |
1755 | * Zap the freelist and set the frozen bit. | |
1756 | * The old freelist is the list of objects for the | |
1757 | * per cpu allocation list. | |
1758 | */ | |
7ced3719 CL |
1759 | freelist = page->freelist; |
1760 | counters = page->counters; | |
1761 | new.counters = counters; | |
633b0764 | 1762 | *objects = new.objects - new.inuse; |
23910c50 | 1763 | if (mode) { |
7ced3719 | 1764 | new.inuse = page->objects; |
23910c50 PE |
1765 | new.freelist = NULL; |
1766 | } else { | |
1767 | new.freelist = freelist; | |
1768 | } | |
2cfb7455 | 1769 | |
a0132ac0 | 1770 | VM_BUG_ON(new.frozen); |
7ced3719 | 1771 | new.frozen = 1; |
2cfb7455 | 1772 | |
7ced3719 | 1773 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1774 | freelist, counters, |
02d7633f | 1775 | new.freelist, new.counters, |
7ced3719 | 1776 | "acquire_slab")) |
7ced3719 | 1777 | return NULL; |
2cfb7455 CL |
1778 | |
1779 | remove_partial(n, page); | |
7ced3719 | 1780 | WARN_ON(!freelist); |
49e22585 | 1781 | return freelist; |
81819f0f CL |
1782 | } |
1783 | ||
633b0764 | 1784 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1785 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1786 | |
81819f0f | 1787 | /* |
672bba3a | 1788 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1789 | */ |
8ba00bb6 JK |
1790 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1791 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1792 | { |
49e22585 CL |
1793 | struct page *page, *page2; |
1794 | void *object = NULL; | |
e5d9998f | 1795 | unsigned int available = 0; |
633b0764 | 1796 | int objects; |
81819f0f CL |
1797 | |
1798 | /* | |
1799 | * Racy check. If we mistakenly see no partial slabs then we | |
1800 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1801 | * partial slab and there is none available then get_partials() |
1802 | * will return NULL. | |
81819f0f CL |
1803 | */ |
1804 | if (!n || !n->nr_partial) | |
1805 | return NULL; | |
1806 | ||
1807 | spin_lock(&n->list_lock); | |
49e22585 | 1808 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1809 | void *t; |
49e22585 | 1810 | |
8ba00bb6 JK |
1811 | if (!pfmemalloc_match(page, flags)) |
1812 | continue; | |
1813 | ||
633b0764 | 1814 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1815 | if (!t) |
1816 | break; | |
1817 | ||
633b0764 | 1818 | available += objects; |
12d79634 | 1819 | if (!object) { |
49e22585 | 1820 | c->page = page; |
49e22585 | 1821 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1822 | object = t; |
49e22585 | 1823 | } else { |
633b0764 | 1824 | put_cpu_partial(s, page, 0); |
8028dcea | 1825 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1826 | } |
345c905d | 1827 | if (!kmem_cache_has_cpu_partial(s) |
e6d0e1dc | 1828 | || available > slub_cpu_partial(s) / 2) |
49e22585 CL |
1829 | break; |
1830 | ||
497b66f2 | 1831 | } |
81819f0f | 1832 | spin_unlock(&n->list_lock); |
497b66f2 | 1833 | return object; |
81819f0f CL |
1834 | } |
1835 | ||
1836 | /* | |
672bba3a | 1837 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1838 | */ |
de3ec035 | 1839 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1840 | struct kmem_cache_cpu *c) |
81819f0f CL |
1841 | { |
1842 | #ifdef CONFIG_NUMA | |
1843 | struct zonelist *zonelist; | |
dd1a239f | 1844 | struct zoneref *z; |
54a6eb5c MG |
1845 | struct zone *zone; |
1846 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1847 | void *object; |
cc9a6c87 | 1848 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1849 | |
1850 | /* | |
672bba3a CL |
1851 | * The defrag ratio allows a configuration of the tradeoffs between |
1852 | * inter node defragmentation and node local allocations. A lower | |
1853 | * defrag_ratio increases the tendency to do local allocations | |
1854 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1855 | * |
672bba3a CL |
1856 | * If the defrag_ratio is set to 0 then kmalloc() always |
1857 | * returns node local objects. If the ratio is higher then kmalloc() | |
1858 | * may return off node objects because partial slabs are obtained | |
1859 | * from other nodes and filled up. | |
81819f0f | 1860 | * |
43efd3ea LP |
1861 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
1862 | * (which makes defrag_ratio = 1000) then every (well almost) | |
1863 | * allocation will first attempt to defrag slab caches on other nodes. | |
1864 | * This means scanning over all nodes to look for partial slabs which | |
1865 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 1866 | * with available objects. |
81819f0f | 1867 | */ |
9824601e CL |
1868 | if (!s->remote_node_defrag_ratio || |
1869 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1870 | return NULL; |
1871 | ||
cc9a6c87 | 1872 | do { |
d26914d1 | 1873 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1874 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1875 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1876 | struct kmem_cache_node *n; | |
1877 | ||
1878 | n = get_node(s, zone_to_nid(zone)); | |
1879 | ||
dee2f8aa | 1880 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1881 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1882 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1883 | if (object) { |
1884 | /* | |
d26914d1 MG |
1885 | * Don't check read_mems_allowed_retry() |
1886 | * here - if mems_allowed was updated in | |
1887 | * parallel, that was a harmless race | |
1888 | * between allocation and the cpuset | |
1889 | * update | |
cc9a6c87 | 1890 | */ |
cc9a6c87 MG |
1891 | return object; |
1892 | } | |
c0ff7453 | 1893 | } |
81819f0f | 1894 | } |
d26914d1 | 1895 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1896 | #endif |
1897 | return NULL; | |
1898 | } | |
1899 | ||
1900 | /* | |
1901 | * Get a partial page, lock it and return it. | |
1902 | */ | |
497b66f2 | 1903 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1904 | struct kmem_cache_cpu *c) |
81819f0f | 1905 | { |
497b66f2 | 1906 | void *object; |
a561ce00 JK |
1907 | int searchnode = node; |
1908 | ||
1909 | if (node == NUMA_NO_NODE) | |
1910 | searchnode = numa_mem_id(); | |
1911 | else if (!node_present_pages(node)) | |
1912 | searchnode = node_to_mem_node(node); | |
81819f0f | 1913 | |
8ba00bb6 | 1914 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1915 | if (object || node != NUMA_NO_NODE) |
1916 | return object; | |
81819f0f | 1917 | |
acd19fd1 | 1918 | return get_any_partial(s, flags, c); |
81819f0f CL |
1919 | } |
1920 | ||
8a5ec0ba CL |
1921 | #ifdef CONFIG_PREEMPT |
1922 | /* | |
1923 | * Calculate the next globally unique transaction for disambiguiation | |
1924 | * during cmpxchg. The transactions start with the cpu number and are then | |
1925 | * incremented by CONFIG_NR_CPUS. | |
1926 | */ | |
1927 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1928 | #else | |
1929 | /* | |
1930 | * No preemption supported therefore also no need to check for | |
1931 | * different cpus. | |
1932 | */ | |
1933 | #define TID_STEP 1 | |
1934 | #endif | |
1935 | ||
1936 | static inline unsigned long next_tid(unsigned long tid) | |
1937 | { | |
1938 | return tid + TID_STEP; | |
1939 | } | |
1940 | ||
1941 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1942 | { | |
1943 | return tid % TID_STEP; | |
1944 | } | |
1945 | ||
1946 | static inline unsigned long tid_to_event(unsigned long tid) | |
1947 | { | |
1948 | return tid / TID_STEP; | |
1949 | } | |
1950 | ||
1951 | static inline unsigned int init_tid(int cpu) | |
1952 | { | |
1953 | return cpu; | |
1954 | } | |
1955 | ||
1956 | static inline void note_cmpxchg_failure(const char *n, | |
1957 | const struct kmem_cache *s, unsigned long tid) | |
1958 | { | |
1959 | #ifdef SLUB_DEBUG_CMPXCHG | |
1960 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1961 | ||
f9f58285 | 1962 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1963 | |
1964 | #ifdef CONFIG_PREEMPT | |
1965 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1966 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1967 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1968 | else | |
1969 | #endif | |
1970 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1971 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1972 | tid_to_event(tid), tid_to_event(actual_tid)); |
1973 | else | |
f9f58285 | 1974 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1975 | actual_tid, tid, next_tid(tid)); |
1976 | #endif | |
4fdccdfb | 1977 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1978 | } |
1979 | ||
788e1aad | 1980 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1981 | { |
8a5ec0ba CL |
1982 | int cpu; |
1983 | ||
1984 | for_each_possible_cpu(cpu) | |
1985 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1986 | } |
2cfb7455 | 1987 | |
81819f0f CL |
1988 | /* |
1989 | * Remove the cpu slab | |
1990 | */ | |
d0e0ac97 | 1991 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
d4ff6d35 | 1992 | void *freelist, struct kmem_cache_cpu *c) |
81819f0f | 1993 | { |
2cfb7455 | 1994 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1995 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1996 | int lock = 0; | |
1997 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1998 | void *nextfree; |
136333d1 | 1999 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2000 | struct page new; |
2001 | struct page old; | |
2002 | ||
2003 | if (page->freelist) { | |
84e554e6 | 2004 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2005 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2006 | } |
2007 | ||
894b8788 | 2008 | /* |
2cfb7455 CL |
2009 | * Stage one: Free all available per cpu objects back |
2010 | * to the page freelist while it is still frozen. Leave the | |
2011 | * last one. | |
2012 | * | |
2013 | * There is no need to take the list->lock because the page | |
2014 | * is still frozen. | |
2015 | */ | |
2016 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2017 | void *prior; | |
2018 | unsigned long counters; | |
2019 | ||
2020 | do { | |
2021 | prior = page->freelist; | |
2022 | counters = page->counters; | |
2023 | set_freepointer(s, freelist, prior); | |
2024 | new.counters = counters; | |
2025 | new.inuse--; | |
a0132ac0 | 2026 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 2027 | |
1d07171c | 2028 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2029 | prior, counters, |
2030 | freelist, new.counters, | |
2031 | "drain percpu freelist")); | |
2032 | ||
2033 | freelist = nextfree; | |
2034 | } | |
2035 | ||
894b8788 | 2036 | /* |
2cfb7455 CL |
2037 | * Stage two: Ensure that the page is unfrozen while the |
2038 | * list presence reflects the actual number of objects | |
2039 | * during unfreeze. | |
2040 | * | |
2041 | * We setup the list membership and then perform a cmpxchg | |
2042 | * with the count. If there is a mismatch then the page | |
2043 | * is not unfrozen but the page is on the wrong list. | |
2044 | * | |
2045 | * Then we restart the process which may have to remove | |
2046 | * the page from the list that we just put it on again | |
2047 | * because the number of objects in the slab may have | |
2048 | * changed. | |
894b8788 | 2049 | */ |
2cfb7455 | 2050 | redo: |
894b8788 | 2051 | |
2cfb7455 CL |
2052 | old.freelist = page->freelist; |
2053 | old.counters = page->counters; | |
a0132ac0 | 2054 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2055 | |
2cfb7455 CL |
2056 | /* Determine target state of the slab */ |
2057 | new.counters = old.counters; | |
2058 | if (freelist) { | |
2059 | new.inuse--; | |
2060 | set_freepointer(s, freelist, old.freelist); | |
2061 | new.freelist = freelist; | |
2062 | } else | |
2063 | new.freelist = old.freelist; | |
2064 | ||
2065 | new.frozen = 0; | |
2066 | ||
8a5b20ae | 2067 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2068 | m = M_FREE; |
2069 | else if (new.freelist) { | |
2070 | m = M_PARTIAL; | |
2071 | if (!lock) { | |
2072 | lock = 1; | |
2073 | /* | |
2074 | * Taking the spinlock removes the possiblity | |
2075 | * that acquire_slab() will see a slab page that | |
2076 | * is frozen | |
2077 | */ | |
2078 | spin_lock(&n->list_lock); | |
2079 | } | |
2080 | } else { | |
2081 | m = M_FULL; | |
2082 | if (kmem_cache_debug(s) && !lock) { | |
2083 | lock = 1; | |
2084 | /* | |
2085 | * This also ensures that the scanning of full | |
2086 | * slabs from diagnostic functions will not see | |
2087 | * any frozen slabs. | |
2088 | */ | |
2089 | spin_lock(&n->list_lock); | |
2090 | } | |
2091 | } | |
2092 | ||
2093 | if (l != m) { | |
2094 | ||
2095 | if (l == M_PARTIAL) | |
2096 | ||
2097 | remove_partial(n, page); | |
2098 | ||
2099 | else if (l == M_FULL) | |
894b8788 | 2100 | |
c65c1877 | 2101 | remove_full(s, n, page); |
2cfb7455 CL |
2102 | |
2103 | if (m == M_PARTIAL) { | |
2104 | ||
2105 | add_partial(n, page, tail); | |
136333d1 | 2106 | stat(s, tail); |
2cfb7455 CL |
2107 | |
2108 | } else if (m == M_FULL) { | |
894b8788 | 2109 | |
2cfb7455 CL |
2110 | stat(s, DEACTIVATE_FULL); |
2111 | add_full(s, n, page); | |
2112 | ||
2113 | } | |
2114 | } | |
2115 | ||
2116 | l = m; | |
1d07171c | 2117 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2118 | old.freelist, old.counters, |
2119 | new.freelist, new.counters, | |
2120 | "unfreezing slab")) | |
2121 | goto redo; | |
2122 | ||
2cfb7455 CL |
2123 | if (lock) |
2124 | spin_unlock(&n->list_lock); | |
2125 | ||
2126 | if (m == M_FREE) { | |
2127 | stat(s, DEACTIVATE_EMPTY); | |
2128 | discard_slab(s, page); | |
2129 | stat(s, FREE_SLAB); | |
894b8788 | 2130 | } |
d4ff6d35 WY |
2131 | |
2132 | c->page = NULL; | |
2133 | c->freelist = NULL; | |
81819f0f CL |
2134 | } |
2135 | ||
d24ac77f JK |
2136 | /* |
2137 | * Unfreeze all the cpu partial slabs. | |
2138 | * | |
59a09917 CL |
2139 | * This function must be called with interrupts disabled |
2140 | * for the cpu using c (or some other guarantee must be there | |
2141 | * to guarantee no concurrent accesses). | |
d24ac77f | 2142 | */ |
59a09917 CL |
2143 | static void unfreeze_partials(struct kmem_cache *s, |
2144 | struct kmem_cache_cpu *c) | |
49e22585 | 2145 | { |
345c905d | 2146 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2147 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2148 | struct page *page, *discard_page = NULL; |
49e22585 CL |
2149 | |
2150 | while ((page = c->partial)) { | |
49e22585 CL |
2151 | struct page new; |
2152 | struct page old; | |
2153 | ||
2154 | c->partial = page->next; | |
43d77867 JK |
2155 | |
2156 | n2 = get_node(s, page_to_nid(page)); | |
2157 | if (n != n2) { | |
2158 | if (n) | |
2159 | spin_unlock(&n->list_lock); | |
2160 | ||
2161 | n = n2; | |
2162 | spin_lock(&n->list_lock); | |
2163 | } | |
49e22585 CL |
2164 | |
2165 | do { | |
2166 | ||
2167 | old.freelist = page->freelist; | |
2168 | old.counters = page->counters; | |
a0132ac0 | 2169 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2170 | |
2171 | new.counters = old.counters; | |
2172 | new.freelist = old.freelist; | |
2173 | ||
2174 | new.frozen = 0; | |
2175 | ||
d24ac77f | 2176 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2177 | old.freelist, old.counters, |
2178 | new.freelist, new.counters, | |
2179 | "unfreezing slab")); | |
2180 | ||
8a5b20ae | 2181 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2182 | page->next = discard_page; |
2183 | discard_page = page; | |
43d77867 JK |
2184 | } else { |
2185 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2186 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2187 | } |
2188 | } | |
2189 | ||
2190 | if (n) | |
2191 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2192 | |
2193 | while (discard_page) { | |
2194 | page = discard_page; | |
2195 | discard_page = discard_page->next; | |
2196 | ||
2197 | stat(s, DEACTIVATE_EMPTY); | |
2198 | discard_slab(s, page); | |
2199 | stat(s, FREE_SLAB); | |
2200 | } | |
345c905d | 2201 | #endif |
49e22585 CL |
2202 | } |
2203 | ||
2204 | /* | |
2205 | * Put a page that was just frozen (in __slab_free) into a partial page | |
0d2d5d40 | 2206 | * slot if available. |
49e22585 CL |
2207 | * |
2208 | * If we did not find a slot then simply move all the partials to the | |
2209 | * per node partial list. | |
2210 | */ | |
633b0764 | 2211 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2212 | { |
345c905d | 2213 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2214 | struct page *oldpage; |
2215 | int pages; | |
2216 | int pobjects; | |
2217 | ||
d6e0b7fa | 2218 | preempt_disable(); |
49e22585 CL |
2219 | do { |
2220 | pages = 0; | |
2221 | pobjects = 0; | |
2222 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2223 | ||
2224 | if (oldpage) { | |
2225 | pobjects = oldpage->pobjects; | |
2226 | pages = oldpage->pages; | |
2227 | if (drain && pobjects > s->cpu_partial) { | |
2228 | unsigned long flags; | |
2229 | /* | |
2230 | * partial array is full. Move the existing | |
2231 | * set to the per node partial list. | |
2232 | */ | |
2233 | local_irq_save(flags); | |
59a09917 | 2234 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2235 | local_irq_restore(flags); |
e24fc410 | 2236 | oldpage = NULL; |
49e22585 CL |
2237 | pobjects = 0; |
2238 | pages = 0; | |
8028dcea | 2239 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2240 | } |
2241 | } | |
2242 | ||
2243 | pages++; | |
2244 | pobjects += page->objects - page->inuse; | |
2245 | ||
2246 | page->pages = pages; | |
2247 | page->pobjects = pobjects; | |
2248 | page->next = oldpage; | |
2249 | ||
d0e0ac97 CG |
2250 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2251 | != oldpage); | |
d6e0b7fa VD |
2252 | if (unlikely(!s->cpu_partial)) { |
2253 | unsigned long flags; | |
2254 | ||
2255 | local_irq_save(flags); | |
2256 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2257 | local_irq_restore(flags); | |
2258 | } | |
2259 | preempt_enable(); | |
345c905d | 2260 | #endif |
49e22585 CL |
2261 | } |
2262 | ||
dfb4f096 | 2263 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2264 | { |
84e554e6 | 2265 | stat(s, CPUSLAB_FLUSH); |
d4ff6d35 | 2266 | deactivate_slab(s, c->page, c->freelist, c); |
c17dda40 CL |
2267 | |
2268 | c->tid = next_tid(c->tid); | |
81819f0f CL |
2269 | } |
2270 | ||
2271 | /* | |
2272 | * Flush cpu slab. | |
6446faa2 | 2273 | * |
81819f0f CL |
2274 | * Called from IPI handler with interrupts disabled. |
2275 | */ | |
0c710013 | 2276 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2277 | { |
9dfc6e68 | 2278 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2279 | |
49e22585 CL |
2280 | if (likely(c)) { |
2281 | if (c->page) | |
2282 | flush_slab(s, c); | |
2283 | ||
59a09917 | 2284 | unfreeze_partials(s, c); |
49e22585 | 2285 | } |
81819f0f CL |
2286 | } |
2287 | ||
2288 | static void flush_cpu_slab(void *d) | |
2289 | { | |
2290 | struct kmem_cache *s = d; | |
81819f0f | 2291 | |
dfb4f096 | 2292 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2293 | } |
2294 | ||
a8364d55 GBY |
2295 | static bool has_cpu_slab(int cpu, void *info) |
2296 | { | |
2297 | struct kmem_cache *s = info; | |
2298 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2299 | ||
a93cf07b | 2300 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2301 | } |
2302 | ||
81819f0f CL |
2303 | static void flush_all(struct kmem_cache *s) |
2304 | { | |
a8364d55 | 2305 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2306 | } |
2307 | ||
a96a87bf SAS |
2308 | /* |
2309 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2310 | * necessary. | |
2311 | */ | |
2312 | static int slub_cpu_dead(unsigned int cpu) | |
2313 | { | |
2314 | struct kmem_cache *s; | |
2315 | unsigned long flags; | |
2316 | ||
2317 | mutex_lock(&slab_mutex); | |
2318 | list_for_each_entry(s, &slab_caches, list) { | |
2319 | local_irq_save(flags); | |
2320 | __flush_cpu_slab(s, cpu); | |
2321 | local_irq_restore(flags); | |
2322 | } | |
2323 | mutex_unlock(&slab_mutex); | |
2324 | return 0; | |
2325 | } | |
2326 | ||
dfb4f096 CL |
2327 | /* |
2328 | * Check if the objects in a per cpu structure fit numa | |
2329 | * locality expectations. | |
2330 | */ | |
57d437d2 | 2331 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2332 | { |
2333 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2334 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2335 | return 0; |
2336 | #endif | |
2337 | return 1; | |
2338 | } | |
2339 | ||
9a02d699 | 2340 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2341 | static int count_free(struct page *page) |
2342 | { | |
2343 | return page->objects - page->inuse; | |
2344 | } | |
2345 | ||
9a02d699 DR |
2346 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2347 | { | |
2348 | return atomic_long_read(&n->total_objects); | |
2349 | } | |
2350 | #endif /* CONFIG_SLUB_DEBUG */ | |
2351 | ||
2352 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2353 | static unsigned long count_partial(struct kmem_cache_node *n, |
2354 | int (*get_count)(struct page *)) | |
2355 | { | |
2356 | unsigned long flags; | |
2357 | unsigned long x = 0; | |
2358 | struct page *page; | |
2359 | ||
2360 | spin_lock_irqsave(&n->list_lock, flags); | |
2361 | list_for_each_entry(page, &n->partial, lru) | |
2362 | x += get_count(page); | |
2363 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2364 | return x; | |
2365 | } | |
9a02d699 | 2366 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2367 | |
781b2ba6 PE |
2368 | static noinline void |
2369 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2370 | { | |
9a02d699 DR |
2371 | #ifdef CONFIG_SLUB_DEBUG |
2372 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2373 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2374 | int node; |
fa45dc25 | 2375 | struct kmem_cache_node *n; |
781b2ba6 | 2376 | |
9a02d699 DR |
2377 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2378 | return; | |
2379 | ||
5b3810e5 VB |
2380 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2381 | nid, gfpflags, &gfpflags); | |
19af27af | 2382 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2383 | s->name, s->object_size, s->size, oo_order(s->oo), |
2384 | oo_order(s->min)); | |
781b2ba6 | 2385 | |
3b0efdfa | 2386 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2387 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2388 | s->name); | |
fa5ec8a1 | 2389 | |
fa45dc25 | 2390 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2391 | unsigned long nr_slabs; |
2392 | unsigned long nr_objs; | |
2393 | unsigned long nr_free; | |
2394 | ||
26c02cf0 AB |
2395 | nr_free = count_partial(n, count_free); |
2396 | nr_slabs = node_nr_slabs(n); | |
2397 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2398 | |
f9f58285 | 2399 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2400 | node, nr_slabs, nr_objs, nr_free); |
2401 | } | |
9a02d699 | 2402 | #endif |
781b2ba6 PE |
2403 | } |
2404 | ||
497b66f2 CL |
2405 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2406 | int node, struct kmem_cache_cpu **pc) | |
2407 | { | |
6faa6833 | 2408 | void *freelist; |
188fd063 CL |
2409 | struct kmem_cache_cpu *c = *pc; |
2410 | struct page *page; | |
497b66f2 | 2411 | |
128227e7 MW |
2412 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
2413 | ||
188fd063 | 2414 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2415 | |
188fd063 CL |
2416 | if (freelist) |
2417 | return freelist; | |
2418 | ||
2419 | page = new_slab(s, flags, node); | |
497b66f2 | 2420 | if (page) { |
7c8e0181 | 2421 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2422 | if (c->page) |
2423 | flush_slab(s, c); | |
2424 | ||
2425 | /* | |
2426 | * No other reference to the page yet so we can | |
2427 | * muck around with it freely without cmpxchg | |
2428 | */ | |
6faa6833 | 2429 | freelist = page->freelist; |
497b66f2 CL |
2430 | page->freelist = NULL; |
2431 | ||
2432 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2433 | c->page = page; |
2434 | *pc = c; | |
2435 | } else | |
6faa6833 | 2436 | freelist = NULL; |
497b66f2 | 2437 | |
6faa6833 | 2438 | return freelist; |
497b66f2 CL |
2439 | } |
2440 | ||
072bb0aa MG |
2441 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2442 | { | |
2443 | if (unlikely(PageSlabPfmemalloc(page))) | |
2444 | return gfp_pfmemalloc_allowed(gfpflags); | |
2445 | ||
2446 | return true; | |
2447 | } | |
2448 | ||
213eeb9f | 2449 | /* |
d0e0ac97 CG |
2450 | * Check the page->freelist of a page and either transfer the freelist to the |
2451 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2452 | * |
2453 | * The page is still frozen if the return value is not NULL. | |
2454 | * | |
2455 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2456 | * |
2457 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2458 | */ |
2459 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2460 | { | |
2461 | struct page new; | |
2462 | unsigned long counters; | |
2463 | void *freelist; | |
2464 | ||
2465 | do { | |
2466 | freelist = page->freelist; | |
2467 | counters = page->counters; | |
6faa6833 | 2468 | |
213eeb9f | 2469 | new.counters = counters; |
a0132ac0 | 2470 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2471 | |
2472 | new.inuse = page->objects; | |
2473 | new.frozen = freelist != NULL; | |
2474 | ||
d24ac77f | 2475 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2476 | freelist, counters, |
2477 | NULL, new.counters, | |
2478 | "get_freelist")); | |
2479 | ||
2480 | return freelist; | |
2481 | } | |
2482 | ||
81819f0f | 2483 | /* |
894b8788 CL |
2484 | * Slow path. The lockless freelist is empty or we need to perform |
2485 | * debugging duties. | |
2486 | * | |
894b8788 CL |
2487 | * Processing is still very fast if new objects have been freed to the |
2488 | * regular freelist. In that case we simply take over the regular freelist | |
2489 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2490 | * |
894b8788 CL |
2491 | * If that is not working then we fall back to the partial lists. We take the |
2492 | * first element of the freelist as the object to allocate now and move the | |
2493 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2494 | * |
894b8788 | 2495 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2496 | * we need to allocate a new slab. This is the slowest path since it involves |
2497 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2498 | * |
2499 | * Version of __slab_alloc to use when we know that interrupts are | |
2500 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2501 | */ |
a380a3c7 | 2502 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2503 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2504 | { |
6faa6833 | 2505 | void *freelist; |
f6e7def7 | 2506 | struct page *page; |
81819f0f | 2507 | |
f6e7def7 CL |
2508 | page = c->page; |
2509 | if (!page) | |
81819f0f | 2510 | goto new_slab; |
49e22585 | 2511 | redo: |
6faa6833 | 2512 | |
57d437d2 | 2513 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2514 | int searchnode = node; |
2515 | ||
2516 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2517 | searchnode = node_to_mem_node(node); | |
2518 | ||
2519 | if (unlikely(!node_match(page, searchnode))) { | |
2520 | stat(s, ALLOC_NODE_MISMATCH); | |
d4ff6d35 | 2521 | deactivate_slab(s, page, c->freelist, c); |
a561ce00 JK |
2522 | goto new_slab; |
2523 | } | |
fc59c053 | 2524 | } |
6446faa2 | 2525 | |
072bb0aa MG |
2526 | /* |
2527 | * By rights, we should be searching for a slab page that was | |
2528 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2529 | * information when the page leaves the per-cpu allocator | |
2530 | */ | |
2531 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
d4ff6d35 | 2532 | deactivate_slab(s, page, c->freelist, c); |
072bb0aa MG |
2533 | goto new_slab; |
2534 | } | |
2535 | ||
73736e03 | 2536 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2537 | freelist = c->freelist; |
2538 | if (freelist) | |
73736e03 | 2539 | goto load_freelist; |
03e404af | 2540 | |
f6e7def7 | 2541 | freelist = get_freelist(s, page); |
6446faa2 | 2542 | |
6faa6833 | 2543 | if (!freelist) { |
03e404af CL |
2544 | c->page = NULL; |
2545 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2546 | goto new_slab; |
03e404af | 2547 | } |
6446faa2 | 2548 | |
84e554e6 | 2549 | stat(s, ALLOC_REFILL); |
6446faa2 | 2550 | |
894b8788 | 2551 | load_freelist: |
507effea CL |
2552 | /* |
2553 | * freelist is pointing to the list of objects to be used. | |
2554 | * page is pointing to the page from which the objects are obtained. | |
2555 | * That page must be frozen for per cpu allocations to work. | |
2556 | */ | |
a0132ac0 | 2557 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2558 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2559 | c->tid = next_tid(c->tid); |
6faa6833 | 2560 | return freelist; |
81819f0f | 2561 | |
81819f0f | 2562 | new_slab: |
2cfb7455 | 2563 | |
a93cf07b WY |
2564 | if (slub_percpu_partial(c)) { |
2565 | page = c->page = slub_percpu_partial(c); | |
2566 | slub_set_percpu_partial(c, page); | |
49e22585 | 2567 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2568 | goto redo; |
81819f0f CL |
2569 | } |
2570 | ||
188fd063 | 2571 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2572 | |
f4697436 | 2573 | if (unlikely(!freelist)) { |
9a02d699 | 2574 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2575 | return NULL; |
81819f0f | 2576 | } |
2cfb7455 | 2577 | |
f6e7def7 | 2578 | page = c->page; |
5091b74a | 2579 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2580 | goto load_freelist; |
2cfb7455 | 2581 | |
497b66f2 | 2582 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2583 | if (kmem_cache_debug(s) && |
2584 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2585 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2586 | |
d4ff6d35 | 2587 | deactivate_slab(s, page, get_freepointer(s, freelist), c); |
6faa6833 | 2588 | return freelist; |
894b8788 CL |
2589 | } |
2590 | ||
a380a3c7 CL |
2591 | /* |
2592 | * Another one that disabled interrupt and compensates for possible | |
2593 | * cpu changes by refetching the per cpu area pointer. | |
2594 | */ | |
2595 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2596 | unsigned long addr, struct kmem_cache_cpu *c) | |
2597 | { | |
2598 | void *p; | |
2599 | unsigned long flags; | |
2600 | ||
2601 | local_irq_save(flags); | |
2602 | #ifdef CONFIG_PREEMPT | |
2603 | /* | |
2604 | * We may have been preempted and rescheduled on a different | |
2605 | * cpu before disabling interrupts. Need to reload cpu area | |
2606 | * pointer. | |
2607 | */ | |
2608 | c = this_cpu_ptr(s->cpu_slab); | |
2609 | #endif | |
2610 | ||
2611 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2612 | local_irq_restore(flags); | |
2613 | return p; | |
2614 | } | |
2615 | ||
894b8788 CL |
2616 | /* |
2617 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2618 | * have the fastpath folded into their functions. So no function call | |
2619 | * overhead for requests that can be satisfied on the fastpath. | |
2620 | * | |
2621 | * The fastpath works by first checking if the lockless freelist can be used. | |
2622 | * If not then __slab_alloc is called for slow processing. | |
2623 | * | |
2624 | * Otherwise we can simply pick the next object from the lockless free list. | |
2625 | */ | |
2b847c3c | 2626 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2627 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2628 | { |
03ec0ed5 | 2629 | void *object; |
dfb4f096 | 2630 | struct kmem_cache_cpu *c; |
57d437d2 | 2631 | struct page *page; |
8a5ec0ba | 2632 | unsigned long tid; |
1f84260c | 2633 | |
8135be5a VD |
2634 | s = slab_pre_alloc_hook(s, gfpflags); |
2635 | if (!s) | |
773ff60e | 2636 | return NULL; |
8a5ec0ba | 2637 | redo: |
8a5ec0ba CL |
2638 | /* |
2639 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2640 | * enabled. We may switch back and forth between cpus while | |
2641 | * reading from one cpu area. That does not matter as long | |
2642 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2643 | * |
9aabf810 JK |
2644 | * We should guarantee that tid and kmem_cache are retrieved on |
2645 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2646 | * to check if it is matched or not. | |
8a5ec0ba | 2647 | */ |
9aabf810 JK |
2648 | do { |
2649 | tid = this_cpu_read(s->cpu_slab->tid); | |
2650 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2651 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2652 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2653 | |
2654 | /* | |
2655 | * Irqless object alloc/free algorithm used here depends on sequence | |
2656 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2657 | * on c to guarantee that object and page associated with previous tid | |
2658 | * won't be used with current tid. If we fetch tid first, object and | |
2659 | * page could be one associated with next tid and our alloc/free | |
2660 | * request will be failed. In this case, we will retry. So, no problem. | |
2661 | */ | |
2662 | barrier(); | |
8a5ec0ba | 2663 | |
8a5ec0ba CL |
2664 | /* |
2665 | * The transaction ids are globally unique per cpu and per operation on | |
2666 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2667 | * occurs on the right processor and that there was no operation on the | |
2668 | * linked list in between. | |
2669 | */ | |
8a5ec0ba | 2670 | |
9dfc6e68 | 2671 | object = c->freelist; |
57d437d2 | 2672 | page = c->page; |
8eae1492 | 2673 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2674 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2675 | stat(s, ALLOC_SLOWPATH); |
2676 | } else { | |
0ad9500e ED |
2677 | void *next_object = get_freepointer_safe(s, object); |
2678 | ||
8a5ec0ba | 2679 | /* |
25985edc | 2680 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2681 | * operation and if we are on the right processor. |
2682 | * | |
d0e0ac97 CG |
2683 | * The cmpxchg does the following atomically (without lock |
2684 | * semantics!) | |
8a5ec0ba CL |
2685 | * 1. Relocate first pointer to the current per cpu area. |
2686 | * 2. Verify that tid and freelist have not been changed | |
2687 | * 3. If they were not changed replace tid and freelist | |
2688 | * | |
d0e0ac97 CG |
2689 | * Since this is without lock semantics the protection is only |
2690 | * against code executing on this cpu *not* from access by | |
2691 | * other cpus. | |
8a5ec0ba | 2692 | */ |
933393f5 | 2693 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2694 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2695 | object, tid, | |
0ad9500e | 2696 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2697 | |
2698 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2699 | goto redo; | |
2700 | } | |
0ad9500e | 2701 | prefetch_freepointer(s, next_object); |
84e554e6 | 2702 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2703 | } |
8a5ec0ba | 2704 | |
74e2134f | 2705 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2706 | memset(object, 0, s->object_size); |
d07dbea4 | 2707 | |
03ec0ed5 | 2708 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2709 | |
894b8788 | 2710 | return object; |
81819f0f CL |
2711 | } |
2712 | ||
2b847c3c EG |
2713 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2714 | gfp_t gfpflags, unsigned long addr) | |
2715 | { | |
2716 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2717 | } | |
2718 | ||
81819f0f CL |
2719 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2720 | { | |
2b847c3c | 2721 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2722 | |
d0e0ac97 CG |
2723 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2724 | s->size, gfpflags); | |
5b882be4 EGM |
2725 | |
2726 | return ret; | |
81819f0f CL |
2727 | } |
2728 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2729 | ||
0f24f128 | 2730 | #ifdef CONFIG_TRACING |
4a92379b RK |
2731 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2732 | { | |
2b847c3c | 2733 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2734 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
505f5dcb | 2735 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2736 | return ret; |
2737 | } | |
2738 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2739 | #endif |
2740 | ||
81819f0f CL |
2741 | #ifdef CONFIG_NUMA |
2742 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2743 | { | |
2b847c3c | 2744 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2745 | |
ca2b84cb | 2746 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2747 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2748 | |
2749 | return ret; | |
81819f0f CL |
2750 | } |
2751 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2752 | |
0f24f128 | 2753 | #ifdef CONFIG_TRACING |
4a92379b | 2754 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2755 | gfp_t gfpflags, |
4a92379b | 2756 | int node, size_t size) |
5b882be4 | 2757 | { |
2b847c3c | 2758 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2759 | |
2760 | trace_kmalloc_node(_RET_IP_, ret, | |
2761 | size, s->size, gfpflags, node); | |
0316bec2 | 2762 | |
505f5dcb | 2763 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2764 | return ret; |
5b882be4 | 2765 | } |
4a92379b | 2766 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2767 | #endif |
5d1f57e4 | 2768 | #endif |
5b882be4 | 2769 | |
81819f0f | 2770 | /* |
94e4d712 | 2771 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2772 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2773 | * |
894b8788 CL |
2774 | * So we still attempt to reduce cache line usage. Just take the slab |
2775 | * lock and free the item. If there is no additional partial page | |
2776 | * handling required then we can return immediately. | |
81819f0f | 2777 | */ |
894b8788 | 2778 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2779 | void *head, void *tail, int cnt, |
2780 | unsigned long addr) | |
2781 | ||
81819f0f CL |
2782 | { |
2783 | void *prior; | |
2cfb7455 | 2784 | int was_frozen; |
2cfb7455 CL |
2785 | struct page new; |
2786 | unsigned long counters; | |
2787 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2788 | unsigned long uninitialized_var(flags); |
81819f0f | 2789 | |
8a5ec0ba | 2790 | stat(s, FREE_SLOWPATH); |
81819f0f | 2791 | |
19c7ff9e | 2792 | if (kmem_cache_debug(s) && |
282acb43 | 2793 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2794 | return; |
6446faa2 | 2795 | |
2cfb7455 | 2796 | do { |
837d678d JK |
2797 | if (unlikely(n)) { |
2798 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2799 | n = NULL; | |
2800 | } | |
2cfb7455 CL |
2801 | prior = page->freelist; |
2802 | counters = page->counters; | |
81084651 | 2803 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2804 | new.counters = counters; |
2805 | was_frozen = new.frozen; | |
81084651 | 2806 | new.inuse -= cnt; |
837d678d | 2807 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2808 | |
c65c1877 | 2809 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2810 | |
2811 | /* | |
d0e0ac97 CG |
2812 | * Slab was on no list before and will be |
2813 | * partially empty | |
2814 | * We can defer the list move and instead | |
2815 | * freeze it. | |
49e22585 CL |
2816 | */ |
2817 | new.frozen = 1; | |
2818 | ||
c65c1877 | 2819 | } else { /* Needs to be taken off a list */ |
49e22585 | 2820 | |
b455def2 | 2821 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2822 | /* |
2823 | * Speculatively acquire the list_lock. | |
2824 | * If the cmpxchg does not succeed then we may | |
2825 | * drop the list_lock without any processing. | |
2826 | * | |
2827 | * Otherwise the list_lock will synchronize with | |
2828 | * other processors updating the list of slabs. | |
2829 | */ | |
2830 | spin_lock_irqsave(&n->list_lock, flags); | |
2831 | ||
2832 | } | |
2cfb7455 | 2833 | } |
81819f0f | 2834 | |
2cfb7455 CL |
2835 | } while (!cmpxchg_double_slab(s, page, |
2836 | prior, counters, | |
81084651 | 2837 | head, new.counters, |
2cfb7455 | 2838 | "__slab_free")); |
81819f0f | 2839 | |
2cfb7455 | 2840 | if (likely(!n)) { |
49e22585 CL |
2841 | |
2842 | /* | |
2843 | * If we just froze the page then put it onto the | |
2844 | * per cpu partial list. | |
2845 | */ | |
8028dcea | 2846 | if (new.frozen && !was_frozen) { |
49e22585 | 2847 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2848 | stat(s, CPU_PARTIAL_FREE); |
2849 | } | |
49e22585 | 2850 | /* |
2cfb7455 CL |
2851 | * The list lock was not taken therefore no list |
2852 | * activity can be necessary. | |
2853 | */ | |
b455def2 L |
2854 | if (was_frozen) |
2855 | stat(s, FREE_FROZEN); | |
2856 | return; | |
2857 | } | |
81819f0f | 2858 | |
8a5b20ae | 2859 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2860 | goto slab_empty; |
2861 | ||
81819f0f | 2862 | /* |
837d678d JK |
2863 | * Objects left in the slab. If it was not on the partial list before |
2864 | * then add it. | |
81819f0f | 2865 | */ |
345c905d JK |
2866 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2867 | if (kmem_cache_debug(s)) | |
c65c1877 | 2868 | remove_full(s, n, page); |
837d678d JK |
2869 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2870 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2871 | } |
80f08c19 | 2872 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2873 | return; |
2874 | ||
2875 | slab_empty: | |
a973e9dd | 2876 | if (prior) { |
81819f0f | 2877 | /* |
6fbabb20 | 2878 | * Slab on the partial list. |
81819f0f | 2879 | */ |
5cc6eee8 | 2880 | remove_partial(n, page); |
84e554e6 | 2881 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2882 | } else { |
6fbabb20 | 2883 | /* Slab must be on the full list */ |
c65c1877 PZ |
2884 | remove_full(s, n, page); |
2885 | } | |
2cfb7455 | 2886 | |
80f08c19 | 2887 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2888 | stat(s, FREE_SLAB); |
81819f0f | 2889 | discard_slab(s, page); |
81819f0f CL |
2890 | } |
2891 | ||
894b8788 CL |
2892 | /* |
2893 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2894 | * can perform fastpath freeing without additional function calls. | |
2895 | * | |
2896 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2897 | * of this processor. This typically the case if we have just allocated | |
2898 | * the item before. | |
2899 | * | |
2900 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2901 | * with all sorts of special processing. | |
81084651 JDB |
2902 | * |
2903 | * Bulk free of a freelist with several objects (all pointing to the | |
2904 | * same page) possible by specifying head and tail ptr, plus objects | |
2905 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2906 | */ |
80a9201a AP |
2907 | static __always_inline void do_slab_free(struct kmem_cache *s, |
2908 | struct page *page, void *head, void *tail, | |
2909 | int cnt, unsigned long addr) | |
894b8788 | 2910 | { |
81084651 | 2911 | void *tail_obj = tail ? : head; |
dfb4f096 | 2912 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2913 | unsigned long tid; |
8a5ec0ba CL |
2914 | redo: |
2915 | /* | |
2916 | * Determine the currently cpus per cpu slab. | |
2917 | * The cpu may change afterward. However that does not matter since | |
2918 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2919 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2920 | */ |
9aabf810 JK |
2921 | do { |
2922 | tid = this_cpu_read(s->cpu_slab->tid); | |
2923 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2924 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2925 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2926 | |
9aabf810 JK |
2927 | /* Same with comment on barrier() in slab_alloc_node() */ |
2928 | barrier(); | |
c016b0bd | 2929 | |
442b06bc | 2930 | if (likely(page == c->page)) { |
81084651 | 2931 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2932 | |
933393f5 | 2933 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2934 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2935 | c->freelist, tid, | |
81084651 | 2936 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2937 | |
2938 | note_cmpxchg_failure("slab_free", s, tid); | |
2939 | goto redo; | |
2940 | } | |
84e554e6 | 2941 | stat(s, FREE_FASTPATH); |
894b8788 | 2942 | } else |
81084651 | 2943 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2944 | |
894b8788 CL |
2945 | } |
2946 | ||
80a9201a AP |
2947 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2948 | void *head, void *tail, int cnt, | |
2949 | unsigned long addr) | |
2950 | { | |
80a9201a | 2951 | /* |
c3895391 AK |
2952 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
2953 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 2954 | */ |
c3895391 AK |
2955 | if (slab_free_freelist_hook(s, &head, &tail)) |
2956 | do_slab_free(s, page, head, tail, cnt, addr); | |
80a9201a AP |
2957 | } |
2958 | ||
2959 | #ifdef CONFIG_KASAN | |
2960 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) | |
2961 | { | |
2962 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
2963 | } | |
2964 | #endif | |
2965 | ||
81819f0f CL |
2966 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2967 | { | |
b9ce5ef4 GC |
2968 | s = cache_from_obj(s, x); |
2969 | if (!s) | |
79576102 | 2970 | return; |
81084651 | 2971 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 2972 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2973 | } |
2974 | EXPORT_SYMBOL(kmem_cache_free); | |
2975 | ||
d0ecd894 | 2976 | struct detached_freelist { |
fbd02630 | 2977 | struct page *page; |
d0ecd894 JDB |
2978 | void *tail; |
2979 | void *freelist; | |
2980 | int cnt; | |
376bf125 | 2981 | struct kmem_cache *s; |
d0ecd894 | 2982 | }; |
fbd02630 | 2983 | |
d0ecd894 JDB |
2984 | /* |
2985 | * This function progressively scans the array with free objects (with | |
2986 | * a limited look ahead) and extract objects belonging to the same | |
2987 | * page. It builds a detached freelist directly within the given | |
2988 | * page/objects. This can happen without any need for | |
2989 | * synchronization, because the objects are owned by running process. | |
2990 | * The freelist is build up as a single linked list in the objects. | |
2991 | * The idea is, that this detached freelist can then be bulk | |
2992 | * transferred to the real freelist(s), but only requiring a single | |
2993 | * synchronization primitive. Look ahead in the array is limited due | |
2994 | * to performance reasons. | |
2995 | */ | |
376bf125 JDB |
2996 | static inline |
2997 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
2998 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
2999 | { |
3000 | size_t first_skipped_index = 0; | |
3001 | int lookahead = 3; | |
3002 | void *object; | |
ca257195 | 3003 | struct page *page; |
fbd02630 | 3004 | |
d0ecd894 JDB |
3005 | /* Always re-init detached_freelist */ |
3006 | df->page = NULL; | |
fbd02630 | 3007 | |
d0ecd894 JDB |
3008 | do { |
3009 | object = p[--size]; | |
ca257195 | 3010 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3011 | } while (!object && size); |
3eed034d | 3012 | |
d0ecd894 JDB |
3013 | if (!object) |
3014 | return 0; | |
fbd02630 | 3015 | |
ca257195 JDB |
3016 | page = virt_to_head_page(object); |
3017 | if (!s) { | |
3018 | /* Handle kalloc'ed objects */ | |
3019 | if (unlikely(!PageSlab(page))) { | |
3020 | BUG_ON(!PageCompound(page)); | |
3021 | kfree_hook(object); | |
4949148a | 3022 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3023 | p[size] = NULL; /* mark object processed */ |
3024 | return size; | |
3025 | } | |
3026 | /* Derive kmem_cache from object */ | |
3027 | df->s = page->slab_cache; | |
3028 | } else { | |
3029 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3030 | } | |
376bf125 | 3031 | |
d0ecd894 | 3032 | /* Start new detached freelist */ |
ca257195 | 3033 | df->page = page; |
376bf125 | 3034 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3035 | df->tail = object; |
3036 | df->freelist = object; | |
3037 | p[size] = NULL; /* mark object processed */ | |
3038 | df->cnt = 1; | |
3039 | ||
3040 | while (size) { | |
3041 | object = p[--size]; | |
3042 | if (!object) | |
3043 | continue; /* Skip processed objects */ | |
3044 | ||
3045 | /* df->page is always set at this point */ | |
3046 | if (df->page == virt_to_head_page(object)) { | |
3047 | /* Opportunity build freelist */ | |
376bf125 | 3048 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3049 | df->freelist = object; |
3050 | df->cnt++; | |
3051 | p[size] = NULL; /* mark object processed */ | |
3052 | ||
3053 | continue; | |
fbd02630 | 3054 | } |
d0ecd894 JDB |
3055 | |
3056 | /* Limit look ahead search */ | |
3057 | if (!--lookahead) | |
3058 | break; | |
3059 | ||
3060 | if (!first_skipped_index) | |
3061 | first_skipped_index = size + 1; | |
fbd02630 | 3062 | } |
d0ecd894 JDB |
3063 | |
3064 | return first_skipped_index; | |
3065 | } | |
3066 | ||
d0ecd894 | 3067 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3068 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3069 | { |
3070 | if (WARN_ON(!size)) | |
3071 | return; | |
3072 | ||
3073 | do { | |
3074 | struct detached_freelist df; | |
3075 | ||
3076 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3077 | if (!df.page) |
d0ecd894 JDB |
3078 | continue; |
3079 | ||
376bf125 | 3080 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 3081 | } while (likely(size)); |
484748f0 CL |
3082 | } |
3083 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3084 | ||
994eb764 | 3085 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3086 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3087 | void **p) | |
484748f0 | 3088 | { |
994eb764 JDB |
3089 | struct kmem_cache_cpu *c; |
3090 | int i; | |
3091 | ||
03ec0ed5 JDB |
3092 | /* memcg and kmem_cache debug support */ |
3093 | s = slab_pre_alloc_hook(s, flags); | |
3094 | if (unlikely(!s)) | |
3095 | return false; | |
994eb764 JDB |
3096 | /* |
3097 | * Drain objects in the per cpu slab, while disabling local | |
3098 | * IRQs, which protects against PREEMPT and interrupts | |
3099 | * handlers invoking normal fastpath. | |
3100 | */ | |
3101 | local_irq_disable(); | |
3102 | c = this_cpu_ptr(s->cpu_slab); | |
3103 | ||
3104 | for (i = 0; i < size; i++) { | |
3105 | void *object = c->freelist; | |
3106 | ||
ebe909e0 | 3107 | if (unlikely(!object)) { |
ebe909e0 JDB |
3108 | /* |
3109 | * Invoking slow path likely have side-effect | |
3110 | * of re-populating per CPU c->freelist | |
3111 | */ | |
87098373 | 3112 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3113 | _RET_IP_, c); |
87098373 CL |
3114 | if (unlikely(!p[i])) |
3115 | goto error; | |
3116 | ||
ebe909e0 JDB |
3117 | c = this_cpu_ptr(s->cpu_slab); |
3118 | continue; /* goto for-loop */ | |
3119 | } | |
994eb764 JDB |
3120 | c->freelist = get_freepointer(s, object); |
3121 | p[i] = object; | |
3122 | } | |
3123 | c->tid = next_tid(c->tid); | |
3124 | local_irq_enable(); | |
3125 | ||
3126 | /* Clear memory outside IRQ disabled fastpath loop */ | |
3127 | if (unlikely(flags & __GFP_ZERO)) { | |
3128 | int j; | |
3129 | ||
3130 | for (j = 0; j < i; j++) | |
3131 | memset(p[j], 0, s->object_size); | |
3132 | } | |
3133 | ||
03ec0ed5 JDB |
3134 | /* memcg and kmem_cache debug support */ |
3135 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 3136 | return i; |
87098373 | 3137 | error: |
87098373 | 3138 | local_irq_enable(); |
03ec0ed5 JDB |
3139 | slab_post_alloc_hook(s, flags, i, p); |
3140 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 3141 | return 0; |
484748f0 CL |
3142 | } |
3143 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3144 | ||
3145 | ||
81819f0f | 3146 | /* |
672bba3a CL |
3147 | * Object placement in a slab is made very easy because we always start at |
3148 | * offset 0. If we tune the size of the object to the alignment then we can | |
3149 | * get the required alignment by putting one properly sized object after | |
3150 | * another. | |
81819f0f CL |
3151 | * |
3152 | * Notice that the allocation order determines the sizes of the per cpu | |
3153 | * caches. Each processor has always one slab available for allocations. | |
3154 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3155 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3156 | * locking overhead. |
81819f0f CL |
3157 | */ |
3158 | ||
3159 | /* | |
3160 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3161 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3162 | * and increases the number of allocations possible without having to | |
3163 | * take the list_lock. | |
3164 | */ | |
19af27af AD |
3165 | static unsigned int slub_min_order; |
3166 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3167 | static unsigned int slub_min_objects; | |
81819f0f | 3168 | |
81819f0f CL |
3169 | /* |
3170 | * Calculate the order of allocation given an slab object size. | |
3171 | * | |
672bba3a CL |
3172 | * The order of allocation has significant impact on performance and other |
3173 | * system components. Generally order 0 allocations should be preferred since | |
3174 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3175 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3176 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3177 | * would be wasted. |
3178 | * | |
3179 | * In order to reach satisfactory performance we must ensure that a minimum | |
3180 | * number of objects is in one slab. Otherwise we may generate too much | |
3181 | * activity on the partial lists which requires taking the list_lock. This is | |
3182 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3183 | * |
672bba3a CL |
3184 | * slub_max_order specifies the order where we begin to stop considering the |
3185 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3186 | * we try to keep the page order as low as possible. So we accept more waste | |
3187 | * of space in favor of a small page order. | |
81819f0f | 3188 | * |
672bba3a CL |
3189 | * Higher order allocations also allow the placement of more objects in a |
3190 | * slab and thereby reduce object handling overhead. If the user has | |
3191 | * requested a higher mininum order then we start with that one instead of | |
3192 | * the smallest order which will fit the object. | |
81819f0f | 3193 | */ |
19af27af AD |
3194 | static inline unsigned int slab_order(unsigned int size, |
3195 | unsigned int min_objects, unsigned int max_order, | |
9736d2a9 | 3196 | unsigned int fract_leftover) |
81819f0f | 3197 | { |
19af27af AD |
3198 | unsigned int min_order = slub_min_order; |
3199 | unsigned int order; | |
81819f0f | 3200 | |
9736d2a9 | 3201 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3202 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3203 | |
9736d2a9 | 3204 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3205 | order <= max_order; order++) { |
81819f0f | 3206 | |
19af27af AD |
3207 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3208 | unsigned int rem; | |
81819f0f | 3209 | |
9736d2a9 | 3210 | rem = slab_size % size; |
81819f0f | 3211 | |
5e6d444e | 3212 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3213 | break; |
81819f0f | 3214 | } |
672bba3a | 3215 | |
81819f0f CL |
3216 | return order; |
3217 | } | |
3218 | ||
9736d2a9 | 3219 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3220 | { |
19af27af AD |
3221 | unsigned int order; |
3222 | unsigned int min_objects; | |
3223 | unsigned int max_objects; | |
5e6d444e CL |
3224 | |
3225 | /* | |
3226 | * Attempt to find best configuration for a slab. This | |
3227 | * works by first attempting to generate a layout with | |
3228 | * the best configuration and backing off gradually. | |
3229 | * | |
422ff4d7 | 3230 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3231 | * we reduce the minimum objects required in a slab. |
3232 | */ | |
3233 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3234 | if (!min_objects) |
3235 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
9736d2a9 | 3236 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3237 | min_objects = min(min_objects, max_objects); |
3238 | ||
5e6d444e | 3239 | while (min_objects > 1) { |
19af27af AD |
3240 | unsigned int fraction; |
3241 | ||
c124f5b5 | 3242 | fraction = 16; |
5e6d444e CL |
3243 | while (fraction >= 4) { |
3244 | order = slab_order(size, min_objects, | |
9736d2a9 | 3245 | slub_max_order, fraction); |
5e6d444e CL |
3246 | if (order <= slub_max_order) |
3247 | return order; | |
3248 | fraction /= 2; | |
3249 | } | |
5086c389 | 3250 | min_objects--; |
5e6d444e CL |
3251 | } |
3252 | ||
3253 | /* | |
3254 | * We were unable to place multiple objects in a slab. Now | |
3255 | * lets see if we can place a single object there. | |
3256 | */ | |
9736d2a9 | 3257 | order = slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3258 | if (order <= slub_max_order) |
3259 | return order; | |
3260 | ||
3261 | /* | |
3262 | * Doh this slab cannot be placed using slub_max_order. | |
3263 | */ | |
9736d2a9 | 3264 | order = slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3265 | if (order < MAX_ORDER) |
5e6d444e CL |
3266 | return order; |
3267 | return -ENOSYS; | |
3268 | } | |
3269 | ||
5595cffc | 3270 | static void |
4053497d | 3271 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3272 | { |
3273 | n->nr_partial = 0; | |
81819f0f CL |
3274 | spin_lock_init(&n->list_lock); |
3275 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3276 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3277 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3278 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3279 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3280 | #endif |
81819f0f CL |
3281 | } |
3282 | ||
55136592 | 3283 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3284 | { |
6c182dc0 | 3285 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3286 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3287 | |
8a5ec0ba | 3288 | /* |
d4d84fef CM |
3289 | * Must align to double word boundary for the double cmpxchg |
3290 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3291 | */ |
d4d84fef CM |
3292 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3293 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3294 | |
3295 | if (!s->cpu_slab) | |
3296 | return 0; | |
3297 | ||
3298 | init_kmem_cache_cpus(s); | |
4c93c355 | 3299 | |
8a5ec0ba | 3300 | return 1; |
4c93c355 | 3301 | } |
4c93c355 | 3302 | |
51df1142 CL |
3303 | static struct kmem_cache *kmem_cache_node; |
3304 | ||
81819f0f CL |
3305 | /* |
3306 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3307 | * slab on the node for this slabcache. There are no concurrent accesses | |
3308 | * possible. | |
3309 | * | |
721ae22a ZYW |
3310 | * Note that this function only works on the kmem_cache_node |
3311 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3312 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3313 | */ |
55136592 | 3314 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3315 | { |
3316 | struct page *page; | |
3317 | struct kmem_cache_node *n; | |
3318 | ||
51df1142 | 3319 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3320 | |
51df1142 | 3321 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3322 | |
3323 | BUG_ON(!page); | |
a2f92ee7 | 3324 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3325 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3326 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3327 | } |
3328 | ||
81819f0f CL |
3329 | n = page->freelist; |
3330 | BUG_ON(!n); | |
51df1142 | 3331 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 3332 | page->inuse = 1; |
8cb0a506 | 3333 | page->frozen = 0; |
51df1142 | 3334 | kmem_cache_node->node[node] = n; |
8ab1372f | 3335 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3336 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3337 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3338 | #endif |
505f5dcb AP |
3339 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
3340 | GFP_KERNEL); | |
4053497d | 3341 | init_kmem_cache_node(n); |
51df1142 | 3342 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3343 | |
67b6c900 | 3344 | /* |
1e4dd946 SR |
3345 | * No locks need to be taken here as it has just been |
3346 | * initialized and there is no concurrent access. | |
67b6c900 | 3347 | */ |
1e4dd946 | 3348 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3349 | } |
3350 | ||
3351 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3352 | { | |
3353 | int node; | |
fa45dc25 | 3354 | struct kmem_cache_node *n; |
81819f0f | 3355 | |
fa45dc25 | 3356 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3357 | s->node[node] = NULL; |
ea37df54 | 3358 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3359 | } |
3360 | } | |
3361 | ||
52b4b950 DS |
3362 | void __kmem_cache_release(struct kmem_cache *s) |
3363 | { | |
210e7a43 | 3364 | cache_random_seq_destroy(s); |
52b4b950 DS |
3365 | free_percpu(s->cpu_slab); |
3366 | free_kmem_cache_nodes(s); | |
3367 | } | |
3368 | ||
55136592 | 3369 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3370 | { |
3371 | int node; | |
81819f0f | 3372 | |
f64dc58c | 3373 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3374 | struct kmem_cache_node *n; |
3375 | ||
73367bd8 | 3376 | if (slab_state == DOWN) { |
55136592 | 3377 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3378 | continue; |
3379 | } | |
51df1142 | 3380 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3381 | GFP_KERNEL, node); |
81819f0f | 3382 | |
73367bd8 AD |
3383 | if (!n) { |
3384 | free_kmem_cache_nodes(s); | |
3385 | return 0; | |
81819f0f | 3386 | } |
73367bd8 | 3387 | |
4053497d | 3388 | init_kmem_cache_node(n); |
ea37df54 | 3389 | s->node[node] = n; |
81819f0f CL |
3390 | } |
3391 | return 1; | |
3392 | } | |
81819f0f | 3393 | |
c0bdb232 | 3394 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3395 | { |
3396 | if (min < MIN_PARTIAL) | |
3397 | min = MIN_PARTIAL; | |
3398 | else if (min > MAX_PARTIAL) | |
3399 | min = MAX_PARTIAL; | |
3400 | s->min_partial = min; | |
3401 | } | |
3402 | ||
e6d0e1dc WY |
3403 | static void set_cpu_partial(struct kmem_cache *s) |
3404 | { | |
3405 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3406 | /* | |
3407 | * cpu_partial determined the maximum number of objects kept in the | |
3408 | * per cpu partial lists of a processor. | |
3409 | * | |
3410 | * Per cpu partial lists mainly contain slabs that just have one | |
3411 | * object freed. If they are used for allocation then they can be | |
3412 | * filled up again with minimal effort. The slab will never hit the | |
3413 | * per node partial lists and therefore no locking will be required. | |
3414 | * | |
3415 | * This setting also determines | |
3416 | * | |
3417 | * A) The number of objects from per cpu partial slabs dumped to the | |
3418 | * per node list when we reach the limit. | |
3419 | * B) The number of objects in cpu partial slabs to extract from the | |
3420 | * per node list when we run out of per cpu objects. We only fetch | |
3421 | * 50% to keep some capacity around for frees. | |
3422 | */ | |
3423 | if (!kmem_cache_has_cpu_partial(s)) | |
3424 | s->cpu_partial = 0; | |
3425 | else if (s->size >= PAGE_SIZE) | |
3426 | s->cpu_partial = 2; | |
3427 | else if (s->size >= 1024) | |
3428 | s->cpu_partial = 6; | |
3429 | else if (s->size >= 256) | |
3430 | s->cpu_partial = 13; | |
3431 | else | |
3432 | s->cpu_partial = 30; | |
3433 | #endif | |
3434 | } | |
3435 | ||
81819f0f CL |
3436 | /* |
3437 | * calculate_sizes() determines the order and the distribution of data within | |
3438 | * a slab object. | |
3439 | */ | |
06b285dc | 3440 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 3441 | { |
d50112ed | 3442 | slab_flags_t flags = s->flags; |
be4a7988 | 3443 | unsigned int size = s->object_size; |
19af27af | 3444 | unsigned int order; |
81819f0f | 3445 | |
d8b42bf5 CL |
3446 | /* |
3447 | * Round up object size to the next word boundary. We can only | |
3448 | * place the free pointer at word boundaries and this determines | |
3449 | * the possible location of the free pointer. | |
3450 | */ | |
3451 | size = ALIGN(size, sizeof(void *)); | |
3452 | ||
3453 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3454 | /* |
3455 | * Determine if we can poison the object itself. If the user of | |
3456 | * the slab may touch the object after free or before allocation | |
3457 | * then we should never poison the object itself. | |
3458 | */ | |
5f0d5a3a | 3459 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3460 | !s->ctor) |
81819f0f CL |
3461 | s->flags |= __OBJECT_POISON; |
3462 | else | |
3463 | s->flags &= ~__OBJECT_POISON; | |
3464 | ||
81819f0f CL |
3465 | |
3466 | /* | |
672bba3a | 3467 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3468 | * end of the object and the free pointer. If not then add an |
672bba3a | 3469 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3470 | */ |
3b0efdfa | 3471 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3472 | size += sizeof(void *); |
41ecc55b | 3473 | #endif |
81819f0f CL |
3474 | |
3475 | /* | |
672bba3a CL |
3476 | * With that we have determined the number of bytes in actual use |
3477 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3478 | */ |
3479 | s->inuse = size; | |
3480 | ||
5f0d5a3a | 3481 | if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
c59def9f | 3482 | s->ctor)) { |
81819f0f CL |
3483 | /* |
3484 | * Relocate free pointer after the object if it is not | |
3485 | * permitted to overwrite the first word of the object on | |
3486 | * kmem_cache_free. | |
3487 | * | |
3488 | * This is the case if we do RCU, have a constructor or | |
3489 | * destructor or are poisoning the objects. | |
3490 | */ | |
3491 | s->offset = size; | |
3492 | size += sizeof(void *); | |
3493 | } | |
3494 | ||
c12b3c62 | 3495 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3496 | if (flags & SLAB_STORE_USER) |
3497 | /* | |
3498 | * Need to store information about allocs and frees after | |
3499 | * the object. | |
3500 | */ | |
3501 | size += 2 * sizeof(struct track); | |
80a9201a | 3502 | #endif |
81819f0f | 3503 | |
80a9201a AP |
3504 | kasan_cache_create(s, &size, &s->flags); |
3505 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3506 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3507 | /* |
3508 | * Add some empty padding so that we can catch | |
3509 | * overwrites from earlier objects rather than let | |
3510 | * tracking information or the free pointer be | |
0211a9c8 | 3511 | * corrupted if a user writes before the start |
81819f0f CL |
3512 | * of the object. |
3513 | */ | |
3514 | size += sizeof(void *); | |
d86bd1be JK |
3515 | |
3516 | s->red_left_pad = sizeof(void *); | |
3517 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3518 | size += s->red_left_pad; | |
3519 | } | |
41ecc55b | 3520 | #endif |
672bba3a | 3521 | |
81819f0f CL |
3522 | /* |
3523 | * SLUB stores one object immediately after another beginning from | |
3524 | * offset 0. In order to align the objects we have to simply size | |
3525 | * each object to conform to the alignment. | |
3526 | */ | |
45906855 | 3527 | size = ALIGN(size, s->align); |
81819f0f | 3528 | s->size = size; |
06b285dc CL |
3529 | if (forced_order >= 0) |
3530 | order = forced_order; | |
3531 | else | |
9736d2a9 | 3532 | order = calculate_order(size); |
81819f0f | 3533 | |
19af27af | 3534 | if ((int)order < 0) |
81819f0f CL |
3535 | return 0; |
3536 | ||
b7a49f0d | 3537 | s->allocflags = 0; |
834f3d11 | 3538 | if (order) |
b7a49f0d CL |
3539 | s->allocflags |= __GFP_COMP; |
3540 | ||
3541 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3542 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3543 | |
3544 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3545 | s->allocflags |= __GFP_RECLAIMABLE; | |
3546 | ||
81819f0f CL |
3547 | /* |
3548 | * Determine the number of objects per slab | |
3549 | */ | |
9736d2a9 MW |
3550 | s->oo = oo_make(order, size); |
3551 | s->min = oo_make(get_order(size), size); | |
205ab99d CL |
3552 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3553 | s->max = s->oo; | |
81819f0f | 3554 | |
834f3d11 | 3555 | return !!oo_objects(s->oo); |
81819f0f CL |
3556 | } |
3557 | ||
d50112ed | 3558 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 3559 | { |
8a13a4cc | 3560 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
2482ddec KC |
3561 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
3562 | s->random = get_random_long(); | |
3563 | #endif | |
81819f0f | 3564 | |
06b285dc | 3565 | if (!calculate_sizes(s, -1)) |
81819f0f | 3566 | goto error; |
3de47213 DR |
3567 | if (disable_higher_order_debug) { |
3568 | /* | |
3569 | * Disable debugging flags that store metadata if the min slab | |
3570 | * order increased. | |
3571 | */ | |
3b0efdfa | 3572 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3573 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3574 | s->offset = 0; | |
3575 | if (!calculate_sizes(s, -1)) | |
3576 | goto error; | |
3577 | } | |
3578 | } | |
81819f0f | 3579 | |
2565409f HC |
3580 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3581 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3582 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3583 | /* Enable fast mode */ |
3584 | s->flags |= __CMPXCHG_DOUBLE; | |
3585 | #endif | |
3586 | ||
3b89d7d8 DR |
3587 | /* |
3588 | * The larger the object size is, the more pages we want on the partial | |
3589 | * list to avoid pounding the page allocator excessively. | |
3590 | */ | |
49e22585 CL |
3591 | set_min_partial(s, ilog2(s->size) / 2); |
3592 | ||
e6d0e1dc | 3593 | set_cpu_partial(s); |
49e22585 | 3594 | |
81819f0f | 3595 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3596 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3597 | #endif |
210e7a43 TG |
3598 | |
3599 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3600 | if (slab_state >= UP) { | |
3601 | if (init_cache_random_seq(s)) | |
3602 | goto error; | |
3603 | } | |
3604 | ||
55136592 | 3605 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3606 | goto error; |
81819f0f | 3607 | |
55136592 | 3608 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3609 | return 0; |
ff12059e | 3610 | |
4c93c355 | 3611 | free_kmem_cache_nodes(s); |
81819f0f CL |
3612 | error: |
3613 | if (flags & SLAB_PANIC) | |
44065b2e AD |
3614 | panic("Cannot create slab %s size=%u realsize=%u order=%u offset=%u flags=%lx\n", |
3615 | s->name, s->size, s->size, | |
4fd0b46e | 3616 | oo_order(s->oo), s->offset, (unsigned long)flags); |
278b1bb1 | 3617 | return -EINVAL; |
81819f0f | 3618 | } |
81819f0f | 3619 | |
33b12c38 CL |
3620 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3621 | const char *text) | |
3622 | { | |
3623 | #ifdef CONFIG_SLUB_DEBUG | |
3624 | void *addr = page_address(page); | |
3625 | void *p; | |
6396bb22 KC |
3626 | unsigned long *map = kcalloc(BITS_TO_LONGS(page->objects), |
3627 | sizeof(long), | |
3628 | GFP_ATOMIC); | |
bbd7d57b ED |
3629 | if (!map) |
3630 | return; | |
945cf2b6 | 3631 | slab_err(s, page, text, s->name); |
33b12c38 | 3632 | slab_lock(page); |
33b12c38 | 3633 | |
5f80b13a | 3634 | get_map(s, page, map); |
33b12c38 CL |
3635 | for_each_object(p, s, addr, page->objects) { |
3636 | ||
3637 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3638 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3639 | print_tracking(s, p); |
3640 | } | |
3641 | } | |
3642 | slab_unlock(page); | |
bbd7d57b | 3643 | kfree(map); |
33b12c38 CL |
3644 | #endif |
3645 | } | |
3646 | ||
81819f0f | 3647 | /* |
599870b1 | 3648 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3649 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3650 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3651 | */ |
599870b1 | 3652 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3653 | { |
60398923 | 3654 | LIST_HEAD(discard); |
81819f0f CL |
3655 | struct page *page, *h; |
3656 | ||
52b4b950 DS |
3657 | BUG_ON(irqs_disabled()); |
3658 | spin_lock_irq(&n->list_lock); | |
33b12c38 | 3659 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3660 | if (!page->inuse) { |
52b4b950 | 3661 | remove_partial(n, page); |
60398923 | 3662 | list_add(&page->lru, &discard); |
33b12c38 CL |
3663 | } else { |
3664 | list_slab_objects(s, page, | |
52b4b950 | 3665 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3666 | } |
33b12c38 | 3667 | } |
52b4b950 | 3668 | spin_unlock_irq(&n->list_lock); |
60398923 CW |
3669 | |
3670 | list_for_each_entry_safe(page, h, &discard, lru) | |
3671 | discard_slab(s, page); | |
81819f0f CL |
3672 | } |
3673 | ||
f9e13c0a SB |
3674 | bool __kmem_cache_empty(struct kmem_cache *s) |
3675 | { | |
3676 | int node; | |
3677 | struct kmem_cache_node *n; | |
3678 | ||
3679 | for_each_kmem_cache_node(s, node, n) | |
3680 | if (n->nr_partial || slabs_node(s, node)) | |
3681 | return false; | |
3682 | return true; | |
3683 | } | |
3684 | ||
81819f0f | 3685 | /* |
672bba3a | 3686 | * Release all resources used by a slab cache. |
81819f0f | 3687 | */ |
52b4b950 | 3688 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3689 | { |
3690 | int node; | |
fa45dc25 | 3691 | struct kmem_cache_node *n; |
81819f0f CL |
3692 | |
3693 | flush_all(s); | |
81819f0f | 3694 | /* Attempt to free all objects */ |
fa45dc25 | 3695 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3696 | free_partial(s, n); |
3697 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3698 | return 1; |
3699 | } | |
bf5eb3de | 3700 | sysfs_slab_remove(s); |
81819f0f CL |
3701 | return 0; |
3702 | } | |
3703 | ||
81819f0f CL |
3704 | /******************************************************************** |
3705 | * Kmalloc subsystem | |
3706 | *******************************************************************/ | |
3707 | ||
81819f0f CL |
3708 | static int __init setup_slub_min_order(char *str) |
3709 | { | |
19af27af | 3710 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
3711 | |
3712 | return 1; | |
3713 | } | |
3714 | ||
3715 | __setup("slub_min_order=", setup_slub_min_order); | |
3716 | ||
3717 | static int __init setup_slub_max_order(char *str) | |
3718 | { | |
19af27af AD |
3719 | get_option(&str, (int *)&slub_max_order); |
3720 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
3721 | |
3722 | return 1; | |
3723 | } | |
3724 | ||
3725 | __setup("slub_max_order=", setup_slub_max_order); | |
3726 | ||
3727 | static int __init setup_slub_min_objects(char *str) | |
3728 | { | |
19af27af | 3729 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
3730 | |
3731 | return 1; | |
3732 | } | |
3733 | ||
3734 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3735 | ||
81819f0f CL |
3736 | void *__kmalloc(size_t size, gfp_t flags) |
3737 | { | |
aadb4bc4 | 3738 | struct kmem_cache *s; |
5b882be4 | 3739 | void *ret; |
81819f0f | 3740 | |
95a05b42 | 3741 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3742 | return kmalloc_large(size, flags); |
aadb4bc4 | 3743 | |
2c59dd65 | 3744 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3745 | |
3746 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3747 | return s; |
3748 | ||
2b847c3c | 3749 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3750 | |
ca2b84cb | 3751 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3752 | |
505f5dcb | 3753 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3754 | |
5b882be4 | 3755 | return ret; |
81819f0f CL |
3756 | } |
3757 | EXPORT_SYMBOL(__kmalloc); | |
3758 | ||
5d1f57e4 | 3759 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3760 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3761 | { | |
b1eeab67 | 3762 | struct page *page; |
e4f7c0b4 | 3763 | void *ptr = NULL; |
f619cfe1 | 3764 | |
75f296d9 | 3765 | flags |= __GFP_COMP; |
4949148a | 3766 | page = alloc_pages_node(node, flags, get_order(size)); |
f619cfe1 | 3767 | if (page) |
e4f7c0b4 CM |
3768 | ptr = page_address(page); |
3769 | ||
d56791b3 | 3770 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3771 | return ptr; |
f619cfe1 CL |
3772 | } |
3773 | ||
81819f0f CL |
3774 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3775 | { | |
aadb4bc4 | 3776 | struct kmem_cache *s; |
5b882be4 | 3777 | void *ret; |
81819f0f | 3778 | |
95a05b42 | 3779 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3780 | ret = kmalloc_large_node(size, flags, node); |
3781 | ||
ca2b84cb EGM |
3782 | trace_kmalloc_node(_RET_IP_, ret, |
3783 | size, PAGE_SIZE << get_order(size), | |
3784 | flags, node); | |
5b882be4 EGM |
3785 | |
3786 | return ret; | |
3787 | } | |
aadb4bc4 | 3788 | |
2c59dd65 | 3789 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3790 | |
3791 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3792 | return s; |
3793 | ||
2b847c3c | 3794 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3795 | |
ca2b84cb | 3796 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3797 | |
505f5dcb | 3798 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3799 | |
5b882be4 | 3800 | return ret; |
81819f0f CL |
3801 | } |
3802 | EXPORT_SYMBOL(__kmalloc_node); | |
3803 | #endif | |
3804 | ||
ed18adc1 KC |
3805 | #ifdef CONFIG_HARDENED_USERCOPY |
3806 | /* | |
afcc90f8 KC |
3807 | * Rejects incorrectly sized objects and objects that are to be copied |
3808 | * to/from userspace but do not fall entirely within the containing slab | |
3809 | * cache's usercopy region. | |
ed18adc1 KC |
3810 | * |
3811 | * Returns NULL if check passes, otherwise const char * to name of cache | |
3812 | * to indicate an error. | |
3813 | */ | |
f4e6e289 KC |
3814 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
3815 | bool to_user) | |
ed18adc1 KC |
3816 | { |
3817 | struct kmem_cache *s; | |
44065b2e | 3818 | unsigned int offset; |
ed18adc1 KC |
3819 | size_t object_size; |
3820 | ||
3821 | /* Find object and usable object size. */ | |
3822 | s = page->slab_cache; | |
ed18adc1 KC |
3823 | |
3824 | /* Reject impossible pointers. */ | |
3825 | if (ptr < page_address(page)) | |
f4e6e289 KC |
3826 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
3827 | to_user, 0, n); | |
ed18adc1 KC |
3828 | |
3829 | /* Find offset within object. */ | |
3830 | offset = (ptr - page_address(page)) % s->size; | |
3831 | ||
3832 | /* Adjust for redzone and reject if within the redzone. */ | |
3833 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) { | |
3834 | if (offset < s->red_left_pad) | |
f4e6e289 KC |
3835 | usercopy_abort("SLUB object in left red zone", |
3836 | s->name, to_user, offset, n); | |
ed18adc1 KC |
3837 | offset -= s->red_left_pad; |
3838 | } | |
3839 | ||
afcc90f8 KC |
3840 | /* Allow address range falling entirely within usercopy region. */ |
3841 | if (offset >= s->useroffset && | |
3842 | offset - s->useroffset <= s->usersize && | |
3843 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 3844 | return; |
ed18adc1 | 3845 | |
afcc90f8 KC |
3846 | /* |
3847 | * If the copy is still within the allocated object, produce | |
3848 | * a warning instead of rejecting the copy. This is intended | |
3849 | * to be a temporary method to find any missing usercopy | |
3850 | * whitelists. | |
3851 | */ | |
3852 | object_size = slab_ksize(s); | |
2d891fbc KC |
3853 | if (usercopy_fallback && |
3854 | offset <= object_size && n <= object_size - offset) { | |
afcc90f8 KC |
3855 | usercopy_warn("SLUB object", s->name, to_user, offset, n); |
3856 | return; | |
3857 | } | |
ed18adc1 | 3858 | |
f4e6e289 | 3859 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
3860 | } |
3861 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
3862 | ||
0316bec2 | 3863 | static size_t __ksize(const void *object) |
81819f0f | 3864 | { |
272c1d21 | 3865 | struct page *page; |
81819f0f | 3866 | |
ef8b4520 | 3867 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3868 | return 0; |
3869 | ||
294a80a8 | 3870 | page = virt_to_head_page(object); |
294a80a8 | 3871 | |
76994412 PE |
3872 | if (unlikely(!PageSlab(page))) { |
3873 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3874 | return PAGE_SIZE << compound_order(page); |
76994412 | 3875 | } |
81819f0f | 3876 | |
1b4f59e3 | 3877 | return slab_ksize(page->slab_cache); |
81819f0f | 3878 | } |
0316bec2 AR |
3879 | |
3880 | size_t ksize(const void *object) | |
3881 | { | |
3882 | size_t size = __ksize(object); | |
3883 | /* We assume that ksize callers could use whole allocated area, | |
4ebb31a4 AP |
3884 | * so we need to unpoison this area. |
3885 | */ | |
3886 | kasan_unpoison_shadow(object, size); | |
0316bec2 AR |
3887 | return size; |
3888 | } | |
b1aabecd | 3889 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3890 | |
3891 | void kfree(const void *x) | |
3892 | { | |
81819f0f | 3893 | struct page *page; |
5bb983b0 | 3894 | void *object = (void *)x; |
81819f0f | 3895 | |
2121db74 PE |
3896 | trace_kfree(_RET_IP_, x); |
3897 | ||
2408c550 | 3898 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3899 | return; |
3900 | ||
b49af68f | 3901 | page = virt_to_head_page(x); |
aadb4bc4 | 3902 | if (unlikely(!PageSlab(page))) { |
0937502a | 3903 | BUG_ON(!PageCompound(page)); |
47adccce | 3904 | kfree_hook(object); |
4949148a | 3905 | __free_pages(page, compound_order(page)); |
aadb4bc4 CL |
3906 | return; |
3907 | } | |
81084651 | 3908 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3909 | } |
3910 | EXPORT_SYMBOL(kfree); | |
3911 | ||
832f37f5 VD |
3912 | #define SHRINK_PROMOTE_MAX 32 |
3913 | ||
2086d26a | 3914 | /* |
832f37f5 VD |
3915 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3916 | * up most to the head of the partial lists. New allocations will then | |
3917 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3918 | * |
3919 | * The slabs with the least items are placed last. This results in them | |
3920 | * being allocated from last increasing the chance that the last objects | |
3921 | * are freed in them. | |
2086d26a | 3922 | */ |
c9fc5864 | 3923 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
3924 | { |
3925 | int node; | |
3926 | int i; | |
3927 | struct kmem_cache_node *n; | |
3928 | struct page *page; | |
3929 | struct page *t; | |
832f37f5 VD |
3930 | struct list_head discard; |
3931 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3932 | unsigned long flags; |
ce3712d7 | 3933 | int ret = 0; |
2086d26a | 3934 | |
2086d26a | 3935 | flush_all(s); |
fa45dc25 | 3936 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3937 | INIT_LIST_HEAD(&discard); |
3938 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3939 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3940 | |
3941 | spin_lock_irqsave(&n->list_lock, flags); | |
3942 | ||
3943 | /* | |
832f37f5 | 3944 | * Build lists of slabs to discard or promote. |
2086d26a | 3945 | * |
672bba3a CL |
3946 | * Note that concurrent frees may occur while we hold the |
3947 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3948 | */ |
3949 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3950 | int free = page->objects - page->inuse; |
3951 | ||
3952 | /* Do not reread page->inuse */ | |
3953 | barrier(); | |
3954 | ||
3955 | /* We do not keep full slabs on the list */ | |
3956 | BUG_ON(free <= 0); | |
3957 | ||
3958 | if (free == page->objects) { | |
3959 | list_move(&page->lru, &discard); | |
69cb8e6b | 3960 | n->nr_partial--; |
832f37f5 VD |
3961 | } else if (free <= SHRINK_PROMOTE_MAX) |
3962 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3963 | } |
3964 | ||
2086d26a | 3965 | /* |
832f37f5 VD |
3966 | * Promote the slabs filled up most to the head of the |
3967 | * partial list. | |
2086d26a | 3968 | */ |
832f37f5 VD |
3969 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3970 | list_splice(promote + i, &n->partial); | |
2086d26a | 3971 | |
2086d26a | 3972 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3973 | |
3974 | /* Release empty slabs */ | |
832f37f5 | 3975 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3976 | discard_slab(s, page); |
ce3712d7 VD |
3977 | |
3978 | if (slabs_node(s, node)) | |
3979 | ret = 1; | |
2086d26a CL |
3980 | } |
3981 | ||
ce3712d7 | 3982 | return ret; |
2086d26a | 3983 | } |
2086d26a | 3984 | |
c9fc5864 | 3985 | #ifdef CONFIG_MEMCG |
01fb58bc TH |
3986 | static void kmemcg_cache_deact_after_rcu(struct kmem_cache *s) |
3987 | { | |
50862ce7 TH |
3988 | /* |
3989 | * Called with all the locks held after a sched RCU grace period. | |
3990 | * Even if @s becomes empty after shrinking, we can't know that @s | |
3991 | * doesn't have allocations already in-flight and thus can't | |
3992 | * destroy @s until the associated memcg is released. | |
3993 | * | |
3994 | * However, let's remove the sysfs files for empty caches here. | |
3995 | * Each cache has a lot of interface files which aren't | |
3996 | * particularly useful for empty draining caches; otherwise, we can | |
3997 | * easily end up with millions of unnecessary sysfs files on | |
3998 | * systems which have a lot of memory and transient cgroups. | |
3999 | */ | |
4000 | if (!__kmem_cache_shrink(s)) | |
4001 | sysfs_slab_remove(s); | |
01fb58bc TH |
4002 | } |
4003 | ||
c9fc5864 TH |
4004 | void __kmemcg_cache_deactivate(struct kmem_cache *s) |
4005 | { | |
4006 | /* | |
4007 | * Disable empty slabs caching. Used to avoid pinning offline | |
4008 | * memory cgroups by kmem pages that can be freed. | |
4009 | */ | |
e6d0e1dc | 4010 | slub_set_cpu_partial(s, 0); |
c9fc5864 TH |
4011 | s->min_partial = 0; |
4012 | ||
4013 | /* | |
4014 | * s->cpu_partial is checked locklessly (see put_cpu_partial), so | |
01fb58bc | 4015 | * we have to make sure the change is visible before shrinking. |
c9fc5864 | 4016 | */ |
01fb58bc | 4017 | slab_deactivate_memcg_cache_rcu_sched(s, kmemcg_cache_deact_after_rcu); |
c9fc5864 TH |
4018 | } |
4019 | #endif | |
4020 | ||
b9049e23 YG |
4021 | static int slab_mem_going_offline_callback(void *arg) |
4022 | { | |
4023 | struct kmem_cache *s; | |
4024 | ||
18004c5d | 4025 | mutex_lock(&slab_mutex); |
b9049e23 | 4026 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4027 | __kmem_cache_shrink(s); |
18004c5d | 4028 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4029 | |
4030 | return 0; | |
4031 | } | |
4032 | ||
4033 | static void slab_mem_offline_callback(void *arg) | |
4034 | { | |
4035 | struct kmem_cache_node *n; | |
4036 | struct kmem_cache *s; | |
4037 | struct memory_notify *marg = arg; | |
4038 | int offline_node; | |
4039 | ||
b9d5ab25 | 4040 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4041 | |
4042 | /* | |
4043 | * If the node still has available memory. we need kmem_cache_node | |
4044 | * for it yet. | |
4045 | */ | |
4046 | if (offline_node < 0) | |
4047 | return; | |
4048 | ||
18004c5d | 4049 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4050 | list_for_each_entry(s, &slab_caches, list) { |
4051 | n = get_node(s, offline_node); | |
4052 | if (n) { | |
4053 | /* | |
4054 | * if n->nr_slabs > 0, slabs still exist on the node | |
4055 | * that is going down. We were unable to free them, | |
c9404c9c | 4056 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
4057 | * callback. So, we must fail. |
4058 | */ | |
0f389ec6 | 4059 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
4060 | |
4061 | s->node[offline_node] = NULL; | |
8de66a0c | 4062 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
4063 | } |
4064 | } | |
18004c5d | 4065 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4066 | } |
4067 | ||
4068 | static int slab_mem_going_online_callback(void *arg) | |
4069 | { | |
4070 | struct kmem_cache_node *n; | |
4071 | struct kmem_cache *s; | |
4072 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4073 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4074 | int ret = 0; |
4075 | ||
4076 | /* | |
4077 | * If the node's memory is already available, then kmem_cache_node is | |
4078 | * already created. Nothing to do. | |
4079 | */ | |
4080 | if (nid < 0) | |
4081 | return 0; | |
4082 | ||
4083 | /* | |
0121c619 | 4084 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4085 | * allocate a kmem_cache_node structure in order to bring the node |
4086 | * online. | |
4087 | */ | |
18004c5d | 4088 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4089 | list_for_each_entry(s, &slab_caches, list) { |
4090 | /* | |
4091 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4092 | * since memory is not yet available from the node that | |
4093 | * is brought up. | |
4094 | */ | |
8de66a0c | 4095 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4096 | if (!n) { |
4097 | ret = -ENOMEM; | |
4098 | goto out; | |
4099 | } | |
4053497d | 4100 | init_kmem_cache_node(n); |
b9049e23 YG |
4101 | s->node[nid] = n; |
4102 | } | |
4103 | out: | |
18004c5d | 4104 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4105 | return ret; |
4106 | } | |
4107 | ||
4108 | static int slab_memory_callback(struct notifier_block *self, | |
4109 | unsigned long action, void *arg) | |
4110 | { | |
4111 | int ret = 0; | |
4112 | ||
4113 | switch (action) { | |
4114 | case MEM_GOING_ONLINE: | |
4115 | ret = slab_mem_going_online_callback(arg); | |
4116 | break; | |
4117 | case MEM_GOING_OFFLINE: | |
4118 | ret = slab_mem_going_offline_callback(arg); | |
4119 | break; | |
4120 | case MEM_OFFLINE: | |
4121 | case MEM_CANCEL_ONLINE: | |
4122 | slab_mem_offline_callback(arg); | |
4123 | break; | |
4124 | case MEM_ONLINE: | |
4125 | case MEM_CANCEL_OFFLINE: | |
4126 | break; | |
4127 | } | |
dc19f9db KH |
4128 | if (ret) |
4129 | ret = notifier_from_errno(ret); | |
4130 | else | |
4131 | ret = NOTIFY_OK; | |
b9049e23 YG |
4132 | return ret; |
4133 | } | |
4134 | ||
3ac38faa AM |
4135 | static struct notifier_block slab_memory_callback_nb = { |
4136 | .notifier_call = slab_memory_callback, | |
4137 | .priority = SLAB_CALLBACK_PRI, | |
4138 | }; | |
b9049e23 | 4139 | |
81819f0f CL |
4140 | /******************************************************************** |
4141 | * Basic setup of slabs | |
4142 | *******************************************************************/ | |
4143 | ||
51df1142 CL |
4144 | /* |
4145 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4146 | * the page allocator. Allocate them properly then fix up the pointers |
4147 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4148 | */ |
4149 | ||
dffb4d60 | 4150 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4151 | { |
4152 | int node; | |
dffb4d60 | 4153 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4154 | struct kmem_cache_node *n; |
51df1142 | 4155 | |
dffb4d60 | 4156 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4157 | |
7d557b3c GC |
4158 | /* |
4159 | * This runs very early, and only the boot processor is supposed to be | |
4160 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4161 | * IPIs around. | |
4162 | */ | |
4163 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4164 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4165 | struct page *p; |
4166 | ||
fa45dc25 CL |
4167 | list_for_each_entry(p, &n->partial, lru) |
4168 | p->slab_cache = s; | |
51df1142 | 4169 | |
607bf324 | 4170 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
4171 | list_for_each_entry(p, &n->full, lru) |
4172 | p->slab_cache = s; | |
51df1142 | 4173 | #endif |
51df1142 | 4174 | } |
f7ce3190 | 4175 | slab_init_memcg_params(s); |
dffb4d60 | 4176 | list_add(&s->list, &slab_caches); |
510ded33 | 4177 | memcg_link_cache(s); |
dffb4d60 | 4178 | return s; |
51df1142 CL |
4179 | } |
4180 | ||
81819f0f CL |
4181 | void __init kmem_cache_init(void) |
4182 | { | |
dffb4d60 CL |
4183 | static __initdata struct kmem_cache boot_kmem_cache, |
4184 | boot_kmem_cache_node; | |
51df1142 | 4185 | |
fc8d8620 SG |
4186 | if (debug_guardpage_minorder()) |
4187 | slub_max_order = 0; | |
4188 | ||
dffb4d60 CL |
4189 | kmem_cache_node = &boot_kmem_cache_node; |
4190 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4191 | |
dffb4d60 | 4192 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4193 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4194 | |
3ac38faa | 4195 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4196 | |
4197 | /* Able to allocate the per node structures */ | |
4198 | slab_state = PARTIAL; | |
4199 | ||
dffb4d60 CL |
4200 | create_boot_cache(kmem_cache, "kmem_cache", |
4201 | offsetof(struct kmem_cache, node) + | |
4202 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4203 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4204 | |
dffb4d60 | 4205 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4206 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4207 | |
4208 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4209 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4210 | create_kmalloc_caches(0); |
81819f0f | 4211 | |
210e7a43 TG |
4212 | /* Setup random freelists for each cache */ |
4213 | init_freelist_randomization(); | |
4214 | ||
a96a87bf SAS |
4215 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4216 | slub_cpu_dead); | |
81819f0f | 4217 | |
19af27af | 4218 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%d\n", |
f97d5f63 | 4219 | cache_line_size(), |
81819f0f CL |
4220 | slub_min_order, slub_max_order, slub_min_objects, |
4221 | nr_cpu_ids, nr_node_ids); | |
4222 | } | |
4223 | ||
7e85ee0c PE |
4224 | void __init kmem_cache_init_late(void) |
4225 | { | |
7e85ee0c PE |
4226 | } |
4227 | ||
2633d7a0 | 4228 | struct kmem_cache * |
f4957d5b | 4229 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4230 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4231 | { |
426589f5 | 4232 | struct kmem_cache *s, *c; |
81819f0f | 4233 | |
a44cb944 | 4234 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4235 | if (s) { |
4236 | s->refcount++; | |
84d0ddd6 | 4237 | |
81819f0f CL |
4238 | /* |
4239 | * Adjust the object sizes so that we clear | |
4240 | * the complete object on kzalloc. | |
4241 | */ | |
1b473f29 | 4242 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4243 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4244 | |
426589f5 | 4245 | for_each_memcg_cache(c, s) { |
84d0ddd6 | 4246 | c->object_size = s->object_size; |
52ee6d74 | 4247 | c->inuse = max(c->inuse, ALIGN(size, sizeof(void *))); |
84d0ddd6 VD |
4248 | } |
4249 | ||
7b8f3b66 | 4250 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4251 | s->refcount--; |
cbb79694 | 4252 | s = NULL; |
7b8f3b66 | 4253 | } |
a0e1d1be | 4254 | } |
6446faa2 | 4255 | |
cbb79694 CL |
4256 | return s; |
4257 | } | |
84c1cf62 | 4258 | |
d50112ed | 4259 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4260 | { |
aac3a166 PE |
4261 | int err; |
4262 | ||
4263 | err = kmem_cache_open(s, flags); | |
4264 | if (err) | |
4265 | return err; | |
20cea968 | 4266 | |
45530c44 CL |
4267 | /* Mutex is not taken during early boot */ |
4268 | if (slab_state <= UP) | |
4269 | return 0; | |
4270 | ||
107dab5c | 4271 | memcg_propagate_slab_attrs(s); |
aac3a166 | 4272 | err = sysfs_slab_add(s); |
aac3a166 | 4273 | if (err) |
52b4b950 | 4274 | __kmem_cache_release(s); |
20cea968 | 4275 | |
aac3a166 | 4276 | return err; |
81819f0f | 4277 | } |
81819f0f | 4278 | |
ce71e27c | 4279 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4280 | { |
aadb4bc4 | 4281 | struct kmem_cache *s; |
94b528d0 | 4282 | void *ret; |
aadb4bc4 | 4283 | |
95a05b42 | 4284 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4285 | return kmalloc_large(size, gfpflags); |
4286 | ||
2c59dd65 | 4287 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4288 | |
2408c550 | 4289 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4290 | return s; |
81819f0f | 4291 | |
2b847c3c | 4292 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4293 | |
25985edc | 4294 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4295 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4296 | |
4297 | return ret; | |
81819f0f CL |
4298 | } |
4299 | ||
5d1f57e4 | 4300 | #ifdef CONFIG_NUMA |
81819f0f | 4301 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4302 | int node, unsigned long caller) |
81819f0f | 4303 | { |
aadb4bc4 | 4304 | struct kmem_cache *s; |
94b528d0 | 4305 | void *ret; |
aadb4bc4 | 4306 | |
95a05b42 | 4307 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4308 | ret = kmalloc_large_node(size, gfpflags, node); |
4309 | ||
4310 | trace_kmalloc_node(caller, ret, | |
4311 | size, PAGE_SIZE << get_order(size), | |
4312 | gfpflags, node); | |
4313 | ||
4314 | return ret; | |
4315 | } | |
eada35ef | 4316 | |
2c59dd65 | 4317 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4318 | |
2408c550 | 4319 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4320 | return s; |
81819f0f | 4321 | |
2b847c3c | 4322 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4323 | |
25985edc | 4324 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4325 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4326 | |
4327 | return ret; | |
81819f0f | 4328 | } |
5d1f57e4 | 4329 | #endif |
81819f0f | 4330 | |
ab4d5ed5 | 4331 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4332 | static int count_inuse(struct page *page) |
4333 | { | |
4334 | return page->inuse; | |
4335 | } | |
4336 | ||
4337 | static int count_total(struct page *page) | |
4338 | { | |
4339 | return page->objects; | |
4340 | } | |
ab4d5ed5 | 4341 | #endif |
205ab99d | 4342 | |
ab4d5ed5 | 4343 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4344 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4345 | unsigned long *map) | |
53e15af0 CL |
4346 | { |
4347 | void *p; | |
a973e9dd | 4348 | void *addr = page_address(page); |
53e15af0 CL |
4349 | |
4350 | if (!check_slab(s, page) || | |
4351 | !on_freelist(s, page, NULL)) | |
4352 | return 0; | |
4353 | ||
4354 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4355 | bitmap_zero(map, page->objects); |
53e15af0 | 4356 | |
5f80b13a CL |
4357 | get_map(s, page, map); |
4358 | for_each_object(p, s, addr, page->objects) { | |
4359 | if (test_bit(slab_index(p, s, addr), map)) | |
4360 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4361 | return 0; | |
53e15af0 CL |
4362 | } |
4363 | ||
224a88be | 4364 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4365 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4366 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4367 | return 0; |
4368 | return 1; | |
4369 | } | |
4370 | ||
434e245d CL |
4371 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4372 | unsigned long *map) | |
53e15af0 | 4373 | { |
881db7fb CL |
4374 | slab_lock(page); |
4375 | validate_slab(s, page, map); | |
4376 | slab_unlock(page); | |
53e15af0 CL |
4377 | } |
4378 | ||
434e245d CL |
4379 | static int validate_slab_node(struct kmem_cache *s, |
4380 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4381 | { |
4382 | unsigned long count = 0; | |
4383 | struct page *page; | |
4384 | unsigned long flags; | |
4385 | ||
4386 | spin_lock_irqsave(&n->list_lock, flags); | |
4387 | ||
4388 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4389 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4390 | count++; |
4391 | } | |
4392 | if (count != n->nr_partial) | |
f9f58285 FF |
4393 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4394 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4395 | |
4396 | if (!(s->flags & SLAB_STORE_USER)) | |
4397 | goto out; | |
4398 | ||
4399 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4400 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4401 | count++; |
4402 | } | |
4403 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4404 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4405 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4406 | |
4407 | out: | |
4408 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4409 | return count; | |
4410 | } | |
4411 | ||
434e245d | 4412 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4413 | { |
4414 | int node; | |
4415 | unsigned long count = 0; | |
6da2ec56 KC |
4416 | unsigned long *map = kmalloc_array(BITS_TO_LONGS(oo_objects(s->max)), |
4417 | sizeof(unsigned long), | |
4418 | GFP_KERNEL); | |
fa45dc25 | 4419 | struct kmem_cache_node *n; |
434e245d CL |
4420 | |
4421 | if (!map) | |
4422 | return -ENOMEM; | |
53e15af0 CL |
4423 | |
4424 | flush_all(s); | |
fa45dc25 | 4425 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4426 | count += validate_slab_node(s, n, map); |
434e245d | 4427 | kfree(map); |
53e15af0 CL |
4428 | return count; |
4429 | } | |
88a420e4 | 4430 | /* |
672bba3a | 4431 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4432 | * and freed. |
4433 | */ | |
4434 | ||
4435 | struct location { | |
4436 | unsigned long count; | |
ce71e27c | 4437 | unsigned long addr; |
45edfa58 CL |
4438 | long long sum_time; |
4439 | long min_time; | |
4440 | long max_time; | |
4441 | long min_pid; | |
4442 | long max_pid; | |
174596a0 | 4443 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4444 | nodemask_t nodes; |
88a420e4 CL |
4445 | }; |
4446 | ||
4447 | struct loc_track { | |
4448 | unsigned long max; | |
4449 | unsigned long count; | |
4450 | struct location *loc; | |
4451 | }; | |
4452 | ||
4453 | static void free_loc_track(struct loc_track *t) | |
4454 | { | |
4455 | if (t->max) | |
4456 | free_pages((unsigned long)t->loc, | |
4457 | get_order(sizeof(struct location) * t->max)); | |
4458 | } | |
4459 | ||
68dff6a9 | 4460 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4461 | { |
4462 | struct location *l; | |
4463 | int order; | |
4464 | ||
88a420e4 CL |
4465 | order = get_order(sizeof(struct location) * max); |
4466 | ||
68dff6a9 | 4467 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4468 | if (!l) |
4469 | return 0; | |
4470 | ||
4471 | if (t->count) { | |
4472 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4473 | free_loc_track(t); | |
4474 | } | |
4475 | t->max = max; | |
4476 | t->loc = l; | |
4477 | return 1; | |
4478 | } | |
4479 | ||
4480 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4481 | const struct track *track) |
88a420e4 CL |
4482 | { |
4483 | long start, end, pos; | |
4484 | struct location *l; | |
ce71e27c | 4485 | unsigned long caddr; |
45edfa58 | 4486 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4487 | |
4488 | start = -1; | |
4489 | end = t->count; | |
4490 | ||
4491 | for ( ; ; ) { | |
4492 | pos = start + (end - start + 1) / 2; | |
4493 | ||
4494 | /* | |
4495 | * There is nothing at "end". If we end up there | |
4496 | * we need to add something to before end. | |
4497 | */ | |
4498 | if (pos == end) | |
4499 | break; | |
4500 | ||
4501 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4502 | if (track->addr == caddr) { |
4503 | ||
4504 | l = &t->loc[pos]; | |
4505 | l->count++; | |
4506 | if (track->when) { | |
4507 | l->sum_time += age; | |
4508 | if (age < l->min_time) | |
4509 | l->min_time = age; | |
4510 | if (age > l->max_time) | |
4511 | l->max_time = age; | |
4512 | ||
4513 | if (track->pid < l->min_pid) | |
4514 | l->min_pid = track->pid; | |
4515 | if (track->pid > l->max_pid) | |
4516 | l->max_pid = track->pid; | |
4517 | ||
174596a0 RR |
4518 | cpumask_set_cpu(track->cpu, |
4519 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4520 | } |
4521 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4522 | return 1; |
4523 | } | |
4524 | ||
45edfa58 | 4525 | if (track->addr < caddr) |
88a420e4 CL |
4526 | end = pos; |
4527 | else | |
4528 | start = pos; | |
4529 | } | |
4530 | ||
4531 | /* | |
672bba3a | 4532 | * Not found. Insert new tracking element. |
88a420e4 | 4533 | */ |
68dff6a9 | 4534 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4535 | return 0; |
4536 | ||
4537 | l = t->loc + pos; | |
4538 | if (pos < t->count) | |
4539 | memmove(l + 1, l, | |
4540 | (t->count - pos) * sizeof(struct location)); | |
4541 | t->count++; | |
4542 | l->count = 1; | |
45edfa58 CL |
4543 | l->addr = track->addr; |
4544 | l->sum_time = age; | |
4545 | l->min_time = age; | |
4546 | l->max_time = age; | |
4547 | l->min_pid = track->pid; | |
4548 | l->max_pid = track->pid; | |
174596a0 RR |
4549 | cpumask_clear(to_cpumask(l->cpus)); |
4550 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4551 | nodes_clear(l->nodes); |
4552 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4553 | return 1; |
4554 | } | |
4555 | ||
4556 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4557 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4558 | unsigned long *map) |
88a420e4 | 4559 | { |
a973e9dd | 4560 | void *addr = page_address(page); |
88a420e4 CL |
4561 | void *p; |
4562 | ||
39b26464 | 4563 | bitmap_zero(map, page->objects); |
5f80b13a | 4564 | get_map(s, page, map); |
88a420e4 | 4565 | |
224a88be | 4566 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4567 | if (!test_bit(slab_index(p, s, addr), map)) |
4568 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4569 | } |
4570 | ||
4571 | static int list_locations(struct kmem_cache *s, char *buf, | |
4572 | enum track_item alloc) | |
4573 | { | |
e374d483 | 4574 | int len = 0; |
88a420e4 | 4575 | unsigned long i; |
68dff6a9 | 4576 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4577 | int node; |
6da2ec56 KC |
4578 | unsigned long *map = kmalloc_array(BITS_TO_LONGS(oo_objects(s->max)), |
4579 | sizeof(unsigned long), | |
4580 | GFP_KERNEL); | |
fa45dc25 | 4581 | struct kmem_cache_node *n; |
88a420e4 | 4582 | |
bbd7d57b | 4583 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
0ee931c4 | 4584 | GFP_KERNEL)) { |
bbd7d57b | 4585 | kfree(map); |
68dff6a9 | 4586 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4587 | } |
88a420e4 CL |
4588 | /* Push back cpu slabs */ |
4589 | flush_all(s); | |
4590 | ||
fa45dc25 | 4591 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4592 | unsigned long flags; |
4593 | struct page *page; | |
4594 | ||
9e86943b | 4595 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4596 | continue; |
4597 | ||
4598 | spin_lock_irqsave(&n->list_lock, flags); | |
4599 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4600 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4601 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4602 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4603 | spin_unlock_irqrestore(&n->list_lock, flags); |
4604 | } | |
4605 | ||
4606 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4607 | struct location *l = &t.loc[i]; |
88a420e4 | 4608 | |
9c246247 | 4609 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4610 | break; |
e374d483 | 4611 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4612 | |
4613 | if (l->addr) | |
62c70bce | 4614 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4615 | else |
e374d483 | 4616 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4617 | |
4618 | if (l->sum_time != l->min_time) { | |
e374d483 | 4619 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4620 | l->min_time, |
4621 | (long)div_u64(l->sum_time, l->count), | |
4622 | l->max_time); | |
45edfa58 | 4623 | } else |
e374d483 | 4624 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4625 | l->min_time); |
4626 | ||
4627 | if (l->min_pid != l->max_pid) | |
e374d483 | 4628 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4629 | l->min_pid, l->max_pid); |
4630 | else | |
e374d483 | 4631 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4632 | l->min_pid); |
4633 | ||
174596a0 RR |
4634 | if (num_online_cpus() > 1 && |
4635 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4636 | len < PAGE_SIZE - 60) |
4637 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4638 | " cpus=%*pbl", | |
4639 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4640 | |
62bc62a8 | 4641 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4642 | len < PAGE_SIZE - 60) |
4643 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4644 | " nodes=%*pbl", | |
4645 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4646 | |
e374d483 | 4647 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4648 | } |
4649 | ||
4650 | free_loc_track(&t); | |
bbd7d57b | 4651 | kfree(map); |
88a420e4 | 4652 | if (!t.count) |
e374d483 HH |
4653 | len += sprintf(buf, "No data\n"); |
4654 | return len; | |
88a420e4 | 4655 | } |
ab4d5ed5 | 4656 | #endif |
88a420e4 | 4657 | |
a5a84755 | 4658 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4659 | static void __init resiliency_test(void) |
a5a84755 CL |
4660 | { |
4661 | u8 *p; | |
4662 | ||
95a05b42 | 4663 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4664 | |
f9f58285 FF |
4665 | pr_err("SLUB resiliency testing\n"); |
4666 | pr_err("-----------------------\n"); | |
4667 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4668 | |
4669 | p = kzalloc(16, GFP_KERNEL); | |
4670 | p[16] = 0x12; | |
f9f58285 FF |
4671 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4672 | p + 16); | |
a5a84755 CL |
4673 | |
4674 | validate_slab_cache(kmalloc_caches[4]); | |
4675 | ||
4676 | /* Hmmm... The next two are dangerous */ | |
4677 | p = kzalloc(32, GFP_KERNEL); | |
4678 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4679 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4680 | p); | |
4681 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4682 | |
4683 | validate_slab_cache(kmalloc_caches[5]); | |
4684 | p = kzalloc(64, GFP_KERNEL); | |
4685 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4686 | *p = 0x56; | |
f9f58285 FF |
4687 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4688 | p); | |
4689 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4690 | validate_slab_cache(kmalloc_caches[6]); |
4691 | ||
f9f58285 | 4692 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4693 | p = kzalloc(128, GFP_KERNEL); |
4694 | kfree(p); | |
4695 | *p = 0x78; | |
f9f58285 | 4696 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4697 | validate_slab_cache(kmalloc_caches[7]); |
4698 | ||
4699 | p = kzalloc(256, GFP_KERNEL); | |
4700 | kfree(p); | |
4701 | p[50] = 0x9a; | |
f9f58285 | 4702 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4703 | validate_slab_cache(kmalloc_caches[8]); |
4704 | ||
4705 | p = kzalloc(512, GFP_KERNEL); | |
4706 | kfree(p); | |
4707 | p[512] = 0xab; | |
f9f58285 | 4708 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4709 | validate_slab_cache(kmalloc_caches[9]); |
4710 | } | |
4711 | #else | |
4712 | #ifdef CONFIG_SYSFS | |
4713 | static void resiliency_test(void) {}; | |
4714 | #endif | |
4715 | #endif | |
4716 | ||
ab4d5ed5 | 4717 | #ifdef CONFIG_SYSFS |
81819f0f | 4718 | enum slab_stat_type { |
205ab99d CL |
4719 | SL_ALL, /* All slabs */ |
4720 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4721 | SL_CPU, /* Only slabs used for cpu caches */ | |
4722 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4723 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4724 | }; |
4725 | ||
205ab99d | 4726 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4727 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4728 | #define SO_CPU (1 << SL_CPU) | |
4729 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4730 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4731 | |
1663f26d TH |
4732 | #ifdef CONFIG_MEMCG |
4733 | static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON); | |
4734 | ||
4735 | static int __init setup_slub_memcg_sysfs(char *str) | |
4736 | { | |
4737 | int v; | |
4738 | ||
4739 | if (get_option(&str, &v) > 0) | |
4740 | memcg_sysfs_enabled = v; | |
4741 | ||
4742 | return 1; | |
4743 | } | |
4744 | ||
4745 | __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs); | |
4746 | #endif | |
4747 | ||
62e5c4b4 CG |
4748 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4749 | char *buf, unsigned long flags) | |
81819f0f CL |
4750 | { |
4751 | unsigned long total = 0; | |
81819f0f CL |
4752 | int node; |
4753 | int x; | |
4754 | unsigned long *nodes; | |
81819f0f | 4755 | |
6396bb22 | 4756 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
4757 | if (!nodes) |
4758 | return -ENOMEM; | |
81819f0f | 4759 | |
205ab99d CL |
4760 | if (flags & SO_CPU) { |
4761 | int cpu; | |
81819f0f | 4762 | |
205ab99d | 4763 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4764 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4765 | cpu); | |
ec3ab083 | 4766 | int node; |
49e22585 | 4767 | struct page *page; |
dfb4f096 | 4768 | |
4db0c3c2 | 4769 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4770 | if (!page) |
4771 | continue; | |
205ab99d | 4772 | |
ec3ab083 CL |
4773 | node = page_to_nid(page); |
4774 | if (flags & SO_TOTAL) | |
4775 | x = page->objects; | |
4776 | else if (flags & SO_OBJECTS) | |
4777 | x = page->inuse; | |
4778 | else | |
4779 | x = 1; | |
49e22585 | 4780 | |
ec3ab083 CL |
4781 | total += x; |
4782 | nodes[node] += x; | |
4783 | ||
a93cf07b | 4784 | page = slub_percpu_partial_read_once(c); |
49e22585 | 4785 | if (page) { |
8afb1474 LZ |
4786 | node = page_to_nid(page); |
4787 | if (flags & SO_TOTAL) | |
4788 | WARN_ON_ONCE(1); | |
4789 | else if (flags & SO_OBJECTS) | |
4790 | WARN_ON_ONCE(1); | |
4791 | else | |
4792 | x = page->pages; | |
bc6697d8 ED |
4793 | total += x; |
4794 | nodes[node] += x; | |
49e22585 | 4795 | } |
81819f0f CL |
4796 | } |
4797 | } | |
4798 | ||
bfc8c901 | 4799 | get_online_mems(); |
ab4d5ed5 | 4800 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4801 | if (flags & SO_ALL) { |
fa45dc25 CL |
4802 | struct kmem_cache_node *n; |
4803 | ||
4804 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4805 | |
d0e0ac97 CG |
4806 | if (flags & SO_TOTAL) |
4807 | x = atomic_long_read(&n->total_objects); | |
4808 | else if (flags & SO_OBJECTS) | |
4809 | x = atomic_long_read(&n->total_objects) - | |
4810 | count_partial(n, count_free); | |
81819f0f | 4811 | else |
205ab99d | 4812 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4813 | total += x; |
4814 | nodes[node] += x; | |
4815 | } | |
4816 | ||
ab4d5ed5 CL |
4817 | } else |
4818 | #endif | |
4819 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4820 | struct kmem_cache_node *n; |
81819f0f | 4821 | |
fa45dc25 | 4822 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4823 | if (flags & SO_TOTAL) |
4824 | x = count_partial(n, count_total); | |
4825 | else if (flags & SO_OBJECTS) | |
4826 | x = count_partial(n, count_inuse); | |
81819f0f | 4827 | else |
205ab99d | 4828 | x = n->nr_partial; |
81819f0f CL |
4829 | total += x; |
4830 | nodes[node] += x; | |
4831 | } | |
4832 | } | |
81819f0f CL |
4833 | x = sprintf(buf, "%lu", total); |
4834 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4835 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4836 | if (nodes[node]) |
4837 | x += sprintf(buf + x, " N%d=%lu", | |
4838 | node, nodes[node]); | |
4839 | #endif | |
bfc8c901 | 4840 | put_online_mems(); |
81819f0f CL |
4841 | kfree(nodes); |
4842 | return x + sprintf(buf + x, "\n"); | |
4843 | } | |
4844 | ||
ab4d5ed5 | 4845 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4846 | static int any_slab_objects(struct kmem_cache *s) |
4847 | { | |
4848 | int node; | |
fa45dc25 | 4849 | struct kmem_cache_node *n; |
81819f0f | 4850 | |
fa45dc25 | 4851 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4852 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4853 | return 1; |
fa45dc25 | 4854 | |
81819f0f CL |
4855 | return 0; |
4856 | } | |
ab4d5ed5 | 4857 | #endif |
81819f0f CL |
4858 | |
4859 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4860 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4861 | |
4862 | struct slab_attribute { | |
4863 | struct attribute attr; | |
4864 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4865 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4866 | }; | |
4867 | ||
4868 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4869 | static struct slab_attribute _name##_attr = \ |
4870 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4871 | |
4872 | #define SLAB_ATTR(_name) \ | |
4873 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4874 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4875 | |
81819f0f CL |
4876 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4877 | { | |
44065b2e | 4878 | return sprintf(buf, "%u\n", s->size); |
81819f0f CL |
4879 | } |
4880 | SLAB_ATTR_RO(slab_size); | |
4881 | ||
4882 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4883 | { | |
3a3791ec | 4884 | return sprintf(buf, "%u\n", s->align); |
81819f0f CL |
4885 | } |
4886 | SLAB_ATTR_RO(align); | |
4887 | ||
4888 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4889 | { | |
1b473f29 | 4890 | return sprintf(buf, "%u\n", s->object_size); |
81819f0f CL |
4891 | } |
4892 | SLAB_ATTR_RO(object_size); | |
4893 | ||
4894 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4895 | { | |
19af27af | 4896 | return sprintf(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
4897 | } |
4898 | SLAB_ATTR_RO(objs_per_slab); | |
4899 | ||
06b285dc CL |
4900 | static ssize_t order_store(struct kmem_cache *s, |
4901 | const char *buf, size_t length) | |
4902 | { | |
19af27af | 4903 | unsigned int order; |
0121c619 CL |
4904 | int err; |
4905 | ||
19af27af | 4906 | err = kstrtouint(buf, 10, &order); |
0121c619 CL |
4907 | if (err) |
4908 | return err; | |
06b285dc CL |
4909 | |
4910 | if (order > slub_max_order || order < slub_min_order) | |
4911 | return -EINVAL; | |
4912 | ||
4913 | calculate_sizes(s, order); | |
4914 | return length; | |
4915 | } | |
4916 | ||
81819f0f CL |
4917 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4918 | { | |
19af27af | 4919 | return sprintf(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 4920 | } |
06b285dc | 4921 | SLAB_ATTR(order); |
81819f0f | 4922 | |
73d342b1 DR |
4923 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4924 | { | |
4925 | return sprintf(buf, "%lu\n", s->min_partial); | |
4926 | } | |
4927 | ||
4928 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4929 | size_t length) | |
4930 | { | |
4931 | unsigned long min; | |
4932 | int err; | |
4933 | ||
3dbb95f7 | 4934 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4935 | if (err) |
4936 | return err; | |
4937 | ||
c0bdb232 | 4938 | set_min_partial(s, min); |
73d342b1 DR |
4939 | return length; |
4940 | } | |
4941 | SLAB_ATTR(min_partial); | |
4942 | ||
49e22585 CL |
4943 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4944 | { | |
e6d0e1dc | 4945 | return sprintf(buf, "%u\n", slub_cpu_partial(s)); |
49e22585 CL |
4946 | } |
4947 | ||
4948 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4949 | size_t length) | |
4950 | { | |
e5d9998f | 4951 | unsigned int objects; |
49e22585 CL |
4952 | int err; |
4953 | ||
e5d9998f | 4954 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
4955 | if (err) |
4956 | return err; | |
345c905d | 4957 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4958 | return -EINVAL; |
49e22585 | 4959 | |
e6d0e1dc | 4960 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
4961 | flush_all(s); |
4962 | return length; | |
4963 | } | |
4964 | SLAB_ATTR(cpu_partial); | |
4965 | ||
81819f0f CL |
4966 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4967 | { | |
62c70bce JP |
4968 | if (!s->ctor) |
4969 | return 0; | |
4970 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4971 | } |
4972 | SLAB_ATTR_RO(ctor); | |
4973 | ||
81819f0f CL |
4974 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4975 | { | |
4307c14f | 4976 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4977 | } |
4978 | SLAB_ATTR_RO(aliases); | |
4979 | ||
81819f0f CL |
4980 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4981 | { | |
d9acf4b7 | 4982 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4983 | } |
4984 | SLAB_ATTR_RO(partial); | |
4985 | ||
4986 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4987 | { | |
d9acf4b7 | 4988 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4989 | } |
4990 | SLAB_ATTR_RO(cpu_slabs); | |
4991 | ||
4992 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4993 | { | |
205ab99d | 4994 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4995 | } |
4996 | SLAB_ATTR_RO(objects); | |
4997 | ||
205ab99d CL |
4998 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4999 | { | |
5000 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5001 | } | |
5002 | SLAB_ATTR_RO(objects_partial); | |
5003 | ||
49e22585 CL |
5004 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5005 | { | |
5006 | int objects = 0; | |
5007 | int pages = 0; | |
5008 | int cpu; | |
5009 | int len; | |
5010 | ||
5011 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5012 | struct page *page; |
5013 | ||
5014 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5015 | |
5016 | if (page) { | |
5017 | pages += page->pages; | |
5018 | objects += page->pobjects; | |
5019 | } | |
5020 | } | |
5021 | ||
5022 | len = sprintf(buf, "%d(%d)", objects, pages); | |
5023 | ||
5024 | #ifdef CONFIG_SMP | |
5025 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5026 | struct page *page; |
5027 | ||
5028 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5029 | |
5030 | if (page && len < PAGE_SIZE - 20) | |
5031 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
5032 | page->pobjects, page->pages); | |
5033 | } | |
5034 | #endif | |
5035 | return len + sprintf(buf + len, "\n"); | |
5036 | } | |
5037 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5038 | ||
a5a84755 CL |
5039 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5040 | { | |
5041 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
5042 | } | |
5043 | ||
5044 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
5045 | const char *buf, size_t length) | |
5046 | { | |
5047 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
5048 | if (buf[0] == '1') | |
5049 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
5050 | return length; | |
5051 | } | |
5052 | SLAB_ATTR(reclaim_account); | |
5053 | ||
5054 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5055 | { | |
5056 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
5057 | } | |
5058 | SLAB_ATTR_RO(hwcache_align); | |
5059 | ||
5060 | #ifdef CONFIG_ZONE_DMA | |
5061 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5062 | { | |
5063 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
5064 | } | |
5065 | SLAB_ATTR_RO(cache_dma); | |
5066 | #endif | |
5067 | ||
8eb8284b DW |
5068 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5069 | { | |
7bbdb81e | 5070 | return sprintf(buf, "%u\n", s->usersize); |
8eb8284b DW |
5071 | } |
5072 | SLAB_ATTR_RO(usersize); | |
5073 | ||
a5a84755 CL |
5074 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5075 | { | |
5f0d5a3a | 5076 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5077 | } |
5078 | SLAB_ATTR_RO(destroy_by_rcu); | |
5079 | ||
ab4d5ed5 | 5080 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5081 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5082 | { | |
5083 | return show_slab_objects(s, buf, SO_ALL); | |
5084 | } | |
5085 | SLAB_ATTR_RO(slabs); | |
5086 | ||
205ab99d CL |
5087 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5088 | { | |
5089 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5090 | } | |
5091 | SLAB_ATTR_RO(total_objects); | |
5092 | ||
81819f0f CL |
5093 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5094 | { | |
becfda68 | 5095 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f CL |
5096 | } |
5097 | ||
5098 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
5099 | const char *buf, size_t length) | |
5100 | { | |
becfda68 | 5101 | s->flags &= ~SLAB_CONSISTENCY_CHECKS; |
b789ef51 CL |
5102 | if (buf[0] == '1') { |
5103 | s->flags &= ~__CMPXCHG_DOUBLE; | |
becfda68 | 5104 | s->flags |= SLAB_CONSISTENCY_CHECKS; |
b789ef51 | 5105 | } |
81819f0f CL |
5106 | return length; |
5107 | } | |
5108 | SLAB_ATTR(sanity_checks); | |
5109 | ||
5110 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5111 | { | |
5112 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
5113 | } | |
5114 | ||
5115 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
5116 | size_t length) | |
5117 | { | |
c9e16131 CL |
5118 | /* |
5119 | * Tracing a merged cache is going to give confusing results | |
5120 | * as well as cause other issues like converting a mergeable | |
5121 | * cache into an umergeable one. | |
5122 | */ | |
5123 | if (s->refcount > 1) | |
5124 | return -EINVAL; | |
5125 | ||
81819f0f | 5126 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
5127 | if (buf[0] == '1') { |
5128 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5129 | s->flags |= SLAB_TRACE; |
b789ef51 | 5130 | } |
81819f0f CL |
5131 | return length; |
5132 | } | |
5133 | SLAB_ATTR(trace); | |
5134 | ||
81819f0f CL |
5135 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5136 | { | |
5137 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
5138 | } | |
5139 | ||
5140 | static ssize_t red_zone_store(struct kmem_cache *s, | |
5141 | const char *buf, size_t length) | |
5142 | { | |
5143 | if (any_slab_objects(s)) | |
5144 | return -EBUSY; | |
5145 | ||
5146 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 | 5147 | if (buf[0] == '1') { |
81819f0f | 5148 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 5149 | } |
06b285dc | 5150 | calculate_sizes(s, -1); |
81819f0f CL |
5151 | return length; |
5152 | } | |
5153 | SLAB_ATTR(red_zone); | |
5154 | ||
5155 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5156 | { | |
5157 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
5158 | } | |
5159 | ||
5160 | static ssize_t poison_store(struct kmem_cache *s, | |
5161 | const char *buf, size_t length) | |
5162 | { | |
5163 | if (any_slab_objects(s)) | |
5164 | return -EBUSY; | |
5165 | ||
5166 | s->flags &= ~SLAB_POISON; | |
b789ef51 | 5167 | if (buf[0] == '1') { |
81819f0f | 5168 | s->flags |= SLAB_POISON; |
b789ef51 | 5169 | } |
06b285dc | 5170 | calculate_sizes(s, -1); |
81819f0f CL |
5171 | return length; |
5172 | } | |
5173 | SLAB_ATTR(poison); | |
5174 | ||
5175 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5176 | { | |
5177 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
5178 | } | |
5179 | ||
5180 | static ssize_t store_user_store(struct kmem_cache *s, | |
5181 | const char *buf, size_t length) | |
5182 | { | |
5183 | if (any_slab_objects(s)) | |
5184 | return -EBUSY; | |
5185 | ||
5186 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
5187 | if (buf[0] == '1') { |
5188 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5189 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 5190 | } |
06b285dc | 5191 | calculate_sizes(s, -1); |
81819f0f CL |
5192 | return length; |
5193 | } | |
5194 | SLAB_ATTR(store_user); | |
5195 | ||
53e15af0 CL |
5196 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5197 | { | |
5198 | return 0; | |
5199 | } | |
5200 | ||
5201 | static ssize_t validate_store(struct kmem_cache *s, | |
5202 | const char *buf, size_t length) | |
5203 | { | |
434e245d CL |
5204 | int ret = -EINVAL; |
5205 | ||
5206 | if (buf[0] == '1') { | |
5207 | ret = validate_slab_cache(s); | |
5208 | if (ret >= 0) | |
5209 | ret = length; | |
5210 | } | |
5211 | return ret; | |
53e15af0 CL |
5212 | } |
5213 | SLAB_ATTR(validate); | |
a5a84755 CL |
5214 | |
5215 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5216 | { | |
5217 | if (!(s->flags & SLAB_STORE_USER)) | |
5218 | return -ENOSYS; | |
5219 | return list_locations(s, buf, TRACK_ALLOC); | |
5220 | } | |
5221 | SLAB_ATTR_RO(alloc_calls); | |
5222 | ||
5223 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5224 | { | |
5225 | if (!(s->flags & SLAB_STORE_USER)) | |
5226 | return -ENOSYS; | |
5227 | return list_locations(s, buf, TRACK_FREE); | |
5228 | } | |
5229 | SLAB_ATTR_RO(free_calls); | |
5230 | #endif /* CONFIG_SLUB_DEBUG */ | |
5231 | ||
5232 | #ifdef CONFIG_FAILSLAB | |
5233 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5234 | { | |
5235 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
5236 | } | |
5237 | ||
5238 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
5239 | size_t length) | |
5240 | { | |
c9e16131 CL |
5241 | if (s->refcount > 1) |
5242 | return -EINVAL; | |
5243 | ||
a5a84755 CL |
5244 | s->flags &= ~SLAB_FAILSLAB; |
5245 | if (buf[0] == '1') | |
5246 | s->flags |= SLAB_FAILSLAB; | |
5247 | return length; | |
5248 | } | |
5249 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 5250 | #endif |
53e15af0 | 5251 | |
2086d26a CL |
5252 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5253 | { | |
5254 | return 0; | |
5255 | } | |
5256 | ||
5257 | static ssize_t shrink_store(struct kmem_cache *s, | |
5258 | const char *buf, size_t length) | |
5259 | { | |
832f37f5 VD |
5260 | if (buf[0] == '1') |
5261 | kmem_cache_shrink(s); | |
5262 | else | |
2086d26a CL |
5263 | return -EINVAL; |
5264 | return length; | |
5265 | } | |
5266 | SLAB_ATTR(shrink); | |
5267 | ||
81819f0f | 5268 | #ifdef CONFIG_NUMA |
9824601e | 5269 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5270 | { |
eb7235eb | 5271 | return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5272 | } |
5273 | ||
9824601e | 5274 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5275 | const char *buf, size_t length) |
5276 | { | |
eb7235eb | 5277 | unsigned int ratio; |
0121c619 CL |
5278 | int err; |
5279 | ||
eb7235eb | 5280 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5281 | if (err) |
5282 | return err; | |
eb7235eb AD |
5283 | if (ratio > 100) |
5284 | return -ERANGE; | |
0121c619 | 5285 | |
eb7235eb | 5286 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5287 | |
81819f0f CL |
5288 | return length; |
5289 | } | |
9824601e | 5290 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5291 | #endif |
5292 | ||
8ff12cfc | 5293 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5294 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5295 | { | |
5296 | unsigned long sum = 0; | |
5297 | int cpu; | |
5298 | int len; | |
6da2ec56 | 5299 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5300 | |
5301 | if (!data) | |
5302 | return -ENOMEM; | |
5303 | ||
5304 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5305 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5306 | |
5307 | data[cpu] = x; | |
5308 | sum += x; | |
5309 | } | |
5310 | ||
5311 | len = sprintf(buf, "%lu", sum); | |
5312 | ||
50ef37b9 | 5313 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5314 | for_each_online_cpu(cpu) { |
5315 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5316 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5317 | } |
50ef37b9 | 5318 | #endif |
8ff12cfc CL |
5319 | kfree(data); |
5320 | return len + sprintf(buf + len, "\n"); | |
5321 | } | |
5322 | ||
78eb00cc DR |
5323 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5324 | { | |
5325 | int cpu; | |
5326 | ||
5327 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5328 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5329 | } |
5330 | ||
8ff12cfc CL |
5331 | #define STAT_ATTR(si, text) \ |
5332 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5333 | { \ | |
5334 | return show_stat(s, buf, si); \ | |
5335 | } \ | |
78eb00cc DR |
5336 | static ssize_t text##_store(struct kmem_cache *s, \ |
5337 | const char *buf, size_t length) \ | |
5338 | { \ | |
5339 | if (buf[0] != '0') \ | |
5340 | return -EINVAL; \ | |
5341 | clear_stat(s, si); \ | |
5342 | return length; \ | |
5343 | } \ | |
5344 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5345 | |
5346 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5347 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5348 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5349 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5350 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5351 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5352 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5353 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5354 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5355 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5356 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5357 | STAT_ATTR(FREE_SLAB, free_slab); |
5358 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5359 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5360 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5361 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5362 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5363 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5364 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5365 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5366 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5367 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5368 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5369 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5370 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5371 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5372 | #endif |
5373 | ||
06428780 | 5374 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5375 | &slab_size_attr.attr, |
5376 | &object_size_attr.attr, | |
5377 | &objs_per_slab_attr.attr, | |
5378 | &order_attr.attr, | |
73d342b1 | 5379 | &min_partial_attr.attr, |
49e22585 | 5380 | &cpu_partial_attr.attr, |
81819f0f | 5381 | &objects_attr.attr, |
205ab99d | 5382 | &objects_partial_attr.attr, |
81819f0f CL |
5383 | &partial_attr.attr, |
5384 | &cpu_slabs_attr.attr, | |
5385 | &ctor_attr.attr, | |
81819f0f CL |
5386 | &aliases_attr.attr, |
5387 | &align_attr.attr, | |
81819f0f CL |
5388 | &hwcache_align_attr.attr, |
5389 | &reclaim_account_attr.attr, | |
5390 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5391 | &shrink_attr.attr, |
49e22585 | 5392 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5393 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5394 | &total_objects_attr.attr, |
5395 | &slabs_attr.attr, | |
5396 | &sanity_checks_attr.attr, | |
5397 | &trace_attr.attr, | |
81819f0f CL |
5398 | &red_zone_attr.attr, |
5399 | &poison_attr.attr, | |
5400 | &store_user_attr.attr, | |
53e15af0 | 5401 | &validate_attr.attr, |
88a420e4 CL |
5402 | &alloc_calls_attr.attr, |
5403 | &free_calls_attr.attr, | |
ab4d5ed5 | 5404 | #endif |
81819f0f CL |
5405 | #ifdef CONFIG_ZONE_DMA |
5406 | &cache_dma_attr.attr, | |
5407 | #endif | |
5408 | #ifdef CONFIG_NUMA | |
9824601e | 5409 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5410 | #endif |
5411 | #ifdef CONFIG_SLUB_STATS | |
5412 | &alloc_fastpath_attr.attr, | |
5413 | &alloc_slowpath_attr.attr, | |
5414 | &free_fastpath_attr.attr, | |
5415 | &free_slowpath_attr.attr, | |
5416 | &free_frozen_attr.attr, | |
5417 | &free_add_partial_attr.attr, | |
5418 | &free_remove_partial_attr.attr, | |
5419 | &alloc_from_partial_attr.attr, | |
5420 | &alloc_slab_attr.attr, | |
5421 | &alloc_refill_attr.attr, | |
e36a2652 | 5422 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5423 | &free_slab_attr.attr, |
5424 | &cpuslab_flush_attr.attr, | |
5425 | &deactivate_full_attr.attr, | |
5426 | &deactivate_empty_attr.attr, | |
5427 | &deactivate_to_head_attr.attr, | |
5428 | &deactivate_to_tail_attr.attr, | |
5429 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5430 | &deactivate_bypass_attr.attr, |
65c3376a | 5431 | &order_fallback_attr.attr, |
b789ef51 CL |
5432 | &cmpxchg_double_fail_attr.attr, |
5433 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5434 | &cpu_partial_alloc_attr.attr, |
5435 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5436 | &cpu_partial_node_attr.attr, |
5437 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5438 | #endif |
4c13dd3b DM |
5439 | #ifdef CONFIG_FAILSLAB |
5440 | &failslab_attr.attr, | |
5441 | #endif | |
8eb8284b | 5442 | &usersize_attr.attr, |
4c13dd3b | 5443 | |
81819f0f CL |
5444 | NULL |
5445 | }; | |
5446 | ||
1fdaaa23 | 5447 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5448 | .attrs = slab_attrs, |
5449 | }; | |
5450 | ||
5451 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5452 | struct attribute *attr, | |
5453 | char *buf) | |
5454 | { | |
5455 | struct slab_attribute *attribute; | |
5456 | struct kmem_cache *s; | |
5457 | int err; | |
5458 | ||
5459 | attribute = to_slab_attr(attr); | |
5460 | s = to_slab(kobj); | |
5461 | ||
5462 | if (!attribute->show) | |
5463 | return -EIO; | |
5464 | ||
5465 | err = attribute->show(s, buf); | |
5466 | ||
5467 | return err; | |
5468 | } | |
5469 | ||
5470 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5471 | struct attribute *attr, | |
5472 | const char *buf, size_t len) | |
5473 | { | |
5474 | struct slab_attribute *attribute; | |
5475 | struct kmem_cache *s; | |
5476 | int err; | |
5477 | ||
5478 | attribute = to_slab_attr(attr); | |
5479 | s = to_slab(kobj); | |
5480 | ||
5481 | if (!attribute->store) | |
5482 | return -EIO; | |
5483 | ||
5484 | err = attribute->store(s, buf, len); | |
127424c8 | 5485 | #ifdef CONFIG_MEMCG |
107dab5c | 5486 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5487 | struct kmem_cache *c; |
81819f0f | 5488 | |
107dab5c GC |
5489 | mutex_lock(&slab_mutex); |
5490 | if (s->max_attr_size < len) | |
5491 | s->max_attr_size = len; | |
5492 | ||
ebe945c2 GC |
5493 | /* |
5494 | * This is a best effort propagation, so this function's return | |
5495 | * value will be determined by the parent cache only. This is | |
5496 | * basically because not all attributes will have a well | |
5497 | * defined semantics for rollbacks - most of the actions will | |
5498 | * have permanent effects. | |
5499 | * | |
5500 | * Returning the error value of any of the children that fail | |
5501 | * is not 100 % defined, in the sense that users seeing the | |
5502 | * error code won't be able to know anything about the state of | |
5503 | * the cache. | |
5504 | * | |
5505 | * Only returning the error code for the parent cache at least | |
5506 | * has well defined semantics. The cache being written to | |
5507 | * directly either failed or succeeded, in which case we loop | |
5508 | * through the descendants with best-effort propagation. | |
5509 | */ | |
426589f5 VD |
5510 | for_each_memcg_cache(c, s) |
5511 | attribute->store(c, buf, len); | |
107dab5c GC |
5512 | mutex_unlock(&slab_mutex); |
5513 | } | |
5514 | #endif | |
81819f0f CL |
5515 | return err; |
5516 | } | |
5517 | ||
107dab5c GC |
5518 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5519 | { | |
127424c8 | 5520 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5521 | int i; |
5522 | char *buffer = NULL; | |
93030d83 | 5523 | struct kmem_cache *root_cache; |
107dab5c | 5524 | |
93030d83 | 5525 | if (is_root_cache(s)) |
107dab5c GC |
5526 | return; |
5527 | ||
f7ce3190 | 5528 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5529 | |
107dab5c GC |
5530 | /* |
5531 | * This mean this cache had no attribute written. Therefore, no point | |
5532 | * in copying default values around | |
5533 | */ | |
93030d83 | 5534 | if (!root_cache->max_attr_size) |
107dab5c GC |
5535 | return; |
5536 | ||
5537 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5538 | char mbuf[64]; | |
5539 | char *buf; | |
5540 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
478fe303 | 5541 | ssize_t len; |
107dab5c GC |
5542 | |
5543 | if (!attr || !attr->store || !attr->show) | |
5544 | continue; | |
5545 | ||
5546 | /* | |
5547 | * It is really bad that we have to allocate here, so we will | |
5548 | * do it only as a fallback. If we actually allocate, though, | |
5549 | * we can just use the allocated buffer until the end. | |
5550 | * | |
5551 | * Most of the slub attributes will tend to be very small in | |
5552 | * size, but sysfs allows buffers up to a page, so they can | |
5553 | * theoretically happen. | |
5554 | */ | |
5555 | if (buffer) | |
5556 | buf = buffer; | |
93030d83 | 5557 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5558 | buf = mbuf; |
5559 | else { | |
5560 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5561 | if (WARN_ON(!buffer)) | |
5562 | continue; | |
5563 | buf = buffer; | |
5564 | } | |
5565 | ||
478fe303 TG |
5566 | len = attr->show(root_cache, buf); |
5567 | if (len > 0) | |
5568 | attr->store(s, buf, len); | |
107dab5c GC |
5569 | } |
5570 | ||
5571 | if (buffer) | |
5572 | free_page((unsigned long)buffer); | |
5573 | #endif | |
5574 | } | |
5575 | ||
41a21285 CL |
5576 | static void kmem_cache_release(struct kobject *k) |
5577 | { | |
5578 | slab_kmem_cache_release(to_slab(k)); | |
5579 | } | |
5580 | ||
52cf25d0 | 5581 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5582 | .show = slab_attr_show, |
5583 | .store = slab_attr_store, | |
5584 | }; | |
5585 | ||
5586 | static struct kobj_type slab_ktype = { | |
5587 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5588 | .release = kmem_cache_release, |
81819f0f CL |
5589 | }; |
5590 | ||
5591 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5592 | { | |
5593 | struct kobj_type *ktype = get_ktype(kobj); | |
5594 | ||
5595 | if (ktype == &slab_ktype) | |
5596 | return 1; | |
5597 | return 0; | |
5598 | } | |
5599 | ||
9cd43611 | 5600 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5601 | .filter = uevent_filter, |
5602 | }; | |
5603 | ||
27c3a314 | 5604 | static struct kset *slab_kset; |
81819f0f | 5605 | |
9a41707b VD |
5606 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5607 | { | |
127424c8 | 5608 | #ifdef CONFIG_MEMCG |
9a41707b | 5609 | if (!is_root_cache(s)) |
f7ce3190 | 5610 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5611 | #endif |
5612 | return slab_kset; | |
5613 | } | |
5614 | ||
81819f0f CL |
5615 | #define ID_STR_LENGTH 64 |
5616 | ||
5617 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5618 | * |
5619 | * Format :[flags-]size | |
81819f0f CL |
5620 | */ |
5621 | static char *create_unique_id(struct kmem_cache *s) | |
5622 | { | |
5623 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5624 | char *p = name; | |
5625 | ||
5626 | BUG_ON(!name); | |
5627 | ||
5628 | *p++ = ':'; | |
5629 | /* | |
5630 | * First flags affecting slabcache operations. We will only | |
5631 | * get here for aliasable slabs so we do not need to support | |
5632 | * too many flags. The flags here must cover all flags that | |
5633 | * are matched during merging to guarantee that the id is | |
5634 | * unique. | |
5635 | */ | |
5636 | if (s->flags & SLAB_CACHE_DMA) | |
5637 | *p++ = 'd'; | |
5638 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5639 | *p++ = 'a'; | |
becfda68 | 5640 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5641 | *p++ = 'F'; |
230e9fc2 VD |
5642 | if (s->flags & SLAB_ACCOUNT) |
5643 | *p++ = 'A'; | |
81819f0f CL |
5644 | if (p != name + 1) |
5645 | *p++ = '-'; | |
44065b2e | 5646 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5647 | |
81819f0f CL |
5648 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5649 | return name; | |
5650 | } | |
5651 | ||
3b7b3140 TH |
5652 | static void sysfs_slab_remove_workfn(struct work_struct *work) |
5653 | { | |
5654 | struct kmem_cache *s = | |
5655 | container_of(work, struct kmem_cache, kobj_remove_work); | |
5656 | ||
5657 | if (!s->kobj.state_in_sysfs) | |
5658 | /* | |
5659 | * For a memcg cache, this may be called during | |
5660 | * deactivation and again on shutdown. Remove only once. | |
5661 | * A cache is never shut down before deactivation is | |
5662 | * complete, so no need to worry about synchronization. | |
5663 | */ | |
f6ba4880 | 5664 | goto out; |
3b7b3140 TH |
5665 | |
5666 | #ifdef CONFIG_MEMCG | |
5667 | kset_unregister(s->memcg_kset); | |
5668 | #endif | |
5669 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
f6ba4880 | 5670 | out: |
3b7b3140 TH |
5671 | kobject_put(&s->kobj); |
5672 | } | |
5673 | ||
81819f0f CL |
5674 | static int sysfs_slab_add(struct kmem_cache *s) |
5675 | { | |
5676 | int err; | |
5677 | const char *name; | |
1663f26d | 5678 | struct kset *kset = cache_kset(s); |
45530c44 | 5679 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5680 | |
3b7b3140 TH |
5681 | INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn); |
5682 | ||
1663f26d TH |
5683 | if (!kset) { |
5684 | kobject_init(&s->kobj, &slab_ktype); | |
5685 | return 0; | |
5686 | } | |
5687 | ||
11066386 MC |
5688 | if (!unmergeable && disable_higher_order_debug && |
5689 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5690 | unmergeable = 1; | |
5691 | ||
81819f0f CL |
5692 | if (unmergeable) { |
5693 | /* | |
5694 | * Slabcache can never be merged so we can use the name proper. | |
5695 | * This is typically the case for debug situations. In that | |
5696 | * case we can catch duplicate names easily. | |
5697 | */ | |
27c3a314 | 5698 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5699 | name = s->name; |
5700 | } else { | |
5701 | /* | |
5702 | * Create a unique name for the slab as a target | |
5703 | * for the symlinks. | |
5704 | */ | |
5705 | name = create_unique_id(s); | |
5706 | } | |
5707 | ||
1663f26d | 5708 | s->kobj.kset = kset; |
26e4f205 | 5709 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5710 | if (err) |
80da026a | 5711 | goto out; |
81819f0f CL |
5712 | |
5713 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5714 | if (err) |
5715 | goto out_del_kobj; | |
9a41707b | 5716 | |
127424c8 | 5717 | #ifdef CONFIG_MEMCG |
1663f26d | 5718 | if (is_root_cache(s) && memcg_sysfs_enabled) { |
9a41707b VD |
5719 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); |
5720 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5721 | err = -ENOMEM; |
5722 | goto out_del_kobj; | |
9a41707b VD |
5723 | } |
5724 | } | |
5725 | #endif | |
5726 | ||
81819f0f CL |
5727 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5728 | if (!unmergeable) { | |
5729 | /* Setup first alias */ | |
5730 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5731 | } |
54b6a731 DJ |
5732 | out: |
5733 | if (!unmergeable) | |
5734 | kfree(name); | |
5735 | return err; | |
5736 | out_del_kobj: | |
5737 | kobject_del(&s->kobj); | |
54b6a731 | 5738 | goto out; |
81819f0f CL |
5739 | } |
5740 | ||
bf5eb3de | 5741 | static void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5742 | { |
97d06609 | 5743 | if (slab_state < FULL) |
2bce6485 CL |
5744 | /* |
5745 | * Sysfs has not been setup yet so no need to remove the | |
5746 | * cache from sysfs. | |
5747 | */ | |
5748 | return; | |
5749 | ||
3b7b3140 TH |
5750 | kobject_get(&s->kobj); |
5751 | schedule_work(&s->kobj_remove_work); | |
bf5eb3de TH |
5752 | } |
5753 | ||
d50d82fa MP |
5754 | void sysfs_slab_unlink(struct kmem_cache *s) |
5755 | { | |
5756 | if (slab_state >= FULL) | |
5757 | kobject_del(&s->kobj); | |
5758 | } | |
5759 | ||
bf5eb3de TH |
5760 | void sysfs_slab_release(struct kmem_cache *s) |
5761 | { | |
5762 | if (slab_state >= FULL) | |
5763 | kobject_put(&s->kobj); | |
81819f0f CL |
5764 | } |
5765 | ||
5766 | /* | |
5767 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5768 | * available lest we lose that information. |
81819f0f CL |
5769 | */ |
5770 | struct saved_alias { | |
5771 | struct kmem_cache *s; | |
5772 | const char *name; | |
5773 | struct saved_alias *next; | |
5774 | }; | |
5775 | ||
5af328a5 | 5776 | static struct saved_alias *alias_list; |
81819f0f CL |
5777 | |
5778 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5779 | { | |
5780 | struct saved_alias *al; | |
5781 | ||
97d06609 | 5782 | if (slab_state == FULL) { |
81819f0f CL |
5783 | /* |
5784 | * If we have a leftover link then remove it. | |
5785 | */ | |
27c3a314 GKH |
5786 | sysfs_remove_link(&slab_kset->kobj, name); |
5787 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5788 | } |
5789 | ||
5790 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5791 | if (!al) | |
5792 | return -ENOMEM; | |
5793 | ||
5794 | al->s = s; | |
5795 | al->name = name; | |
5796 | al->next = alias_list; | |
5797 | alias_list = al; | |
5798 | return 0; | |
5799 | } | |
5800 | ||
5801 | static int __init slab_sysfs_init(void) | |
5802 | { | |
5b95a4ac | 5803 | struct kmem_cache *s; |
81819f0f CL |
5804 | int err; |
5805 | ||
18004c5d | 5806 | mutex_lock(&slab_mutex); |
2bce6485 | 5807 | |
0ff21e46 | 5808 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5809 | if (!slab_kset) { |
18004c5d | 5810 | mutex_unlock(&slab_mutex); |
f9f58285 | 5811 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5812 | return -ENOSYS; |
5813 | } | |
5814 | ||
97d06609 | 5815 | slab_state = FULL; |
26a7bd03 | 5816 | |
5b95a4ac | 5817 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5818 | err = sysfs_slab_add(s); |
5d540fb7 | 5819 | if (err) |
f9f58285 FF |
5820 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5821 | s->name); | |
26a7bd03 | 5822 | } |
81819f0f CL |
5823 | |
5824 | while (alias_list) { | |
5825 | struct saved_alias *al = alias_list; | |
5826 | ||
5827 | alias_list = alias_list->next; | |
5828 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5829 | if (err) |
f9f58285 FF |
5830 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5831 | al->name); | |
81819f0f CL |
5832 | kfree(al); |
5833 | } | |
5834 | ||
18004c5d | 5835 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5836 | resiliency_test(); |
5837 | return 0; | |
5838 | } | |
5839 | ||
5840 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5841 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5842 | |
5843 | /* | |
5844 | * The /proc/slabinfo ABI | |
5845 | */ | |
5b365771 | 5846 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 5847 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5848 | { |
57ed3eda | 5849 | unsigned long nr_slabs = 0; |
205ab99d CL |
5850 | unsigned long nr_objs = 0; |
5851 | unsigned long nr_free = 0; | |
57ed3eda | 5852 | int node; |
fa45dc25 | 5853 | struct kmem_cache_node *n; |
57ed3eda | 5854 | |
fa45dc25 | 5855 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5856 | nr_slabs += node_nr_slabs(n); |
5857 | nr_objs += node_nr_objs(n); | |
205ab99d | 5858 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5859 | } |
5860 | ||
0d7561c6 GC |
5861 | sinfo->active_objs = nr_objs - nr_free; |
5862 | sinfo->num_objs = nr_objs; | |
5863 | sinfo->active_slabs = nr_slabs; | |
5864 | sinfo->num_slabs = nr_slabs; | |
5865 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5866 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5867 | } |
5868 | ||
0d7561c6 | 5869 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5870 | { |
7b3c3a50 AD |
5871 | } |
5872 | ||
b7454ad3 GC |
5873 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5874 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5875 | { |
b7454ad3 | 5876 | return -EIO; |
7b3c3a50 | 5877 | } |
5b365771 | 5878 | #endif /* CONFIG_SLUB_DEBUG */ |