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