]>
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/seq_file.h> | |
23 | #include <linux/kasan.h> | |
24 | #include <linux/cpu.h> | |
25 | #include <linux/cpuset.h> | |
26 | #include <linux/mempolicy.h> | |
27 | #include <linux/ctype.h> | |
28 | #include <linux/debugobjects.h> | |
29 | #include <linux/kallsyms.h> | |
30 | #include <linux/memory.h> | |
31 | #include <linux/math64.h> | |
32 | #include <linux/fault-inject.h> | |
33 | #include <linux/stacktrace.h> | |
34 | #include <linux/prefetch.h> | |
35 | #include <linux/memcontrol.h> | |
36 | #include <linux/random.h> | |
37 | ||
38 | #include <trace/events/kmem.h> | |
39 | ||
40 | #include "internal.h" | |
41 | ||
42 | /* | |
43 | * Lock order: | |
44 | * 1. slab_mutex (Global Mutex) | |
45 | * 2. node->list_lock | |
46 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
47 | * | |
48 | * slab_mutex | |
49 | * | |
50 | * The role of the slab_mutex is to protect the list of all the slabs | |
51 | * and to synchronize major metadata changes to slab cache structures. | |
52 | * | |
53 | * The slab_lock is only used for debugging and on arches that do not | |
54 | * have the ability to do a cmpxchg_double. It only protects: | |
55 | * A. page->freelist -> List of object free in a page | |
56 | * B. page->inuse -> Number of objects in use | |
57 | * C. page->objects -> Number of objects in page | |
58 | * D. page->frozen -> frozen state | |
59 | * | |
60 | * If a slab is frozen then it is exempt from list management. It is not | |
61 | * on any list except per cpu partial list. The processor that froze the | |
62 | * slab is the one who can perform list operations on the page. Other | |
63 | * processors may put objects onto the freelist but the processor that | |
64 | * froze the slab is the only one that can retrieve the objects from the | |
65 | * page's freelist. | |
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 | * page->frozen The slab is frozen and exempt from list processing. | |
97 | * This means that the slab is dedicated to a purpose | |
98 | * such as satisfying allocations for a specific | |
99 | * processor. Objects may be freed in the slab while | |
100 | * it is frozen but slab_free will then skip the usual | |
101 | * list operations. It is up to the processor holding | |
102 | * the slab to integrate the slab into the slab lists | |
103 | * when the slab is no longer needed. | |
104 | * | |
105 | * One use of this flag is to mark slabs that are | |
106 | * used for allocations. Then such a slab becomes a cpu | |
107 | * slab. The cpu slab may be equipped with an additional | |
108 | * freelist that allows lockless access to | |
109 | * free objects in addition to the regular freelist | |
110 | * that requires the slab lock. | |
111 | * | |
112 | * SLAB_DEBUG_FLAGS Slab requires special handling due to debug | |
113 | * options set. This moves slab handling out of | |
114 | * the fast path and disables lockless freelists. | |
115 | */ | |
116 | ||
117 | static inline int kmem_cache_debug(struct kmem_cache *s) | |
118 | { | |
119 | #ifdef CONFIG_SLUB_DEBUG | |
120 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); | |
121 | #else | |
122 | return 0; | |
123 | #endif | |
124 | } | |
125 | ||
126 | void *fixup_red_left(struct kmem_cache *s, void *p) | |
127 | { | |
128 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) | |
129 | p += s->red_left_pad; | |
130 | ||
131 | return p; | |
132 | } | |
133 | ||
134 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) | |
135 | { | |
136 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
137 | return !kmem_cache_debug(s); | |
138 | #else | |
139 | return false; | |
140 | #endif | |
141 | } | |
142 | ||
143 | /* | |
144 | * Issues still to be resolved: | |
145 | * | |
146 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. | |
147 | * | |
148 | * - Variable sizing of the per node arrays | |
149 | */ | |
150 | ||
151 | /* Enable to test recovery from slab corruption on boot */ | |
152 | #undef SLUB_RESILIENCY_TEST | |
153 | ||
154 | /* Enable to log cmpxchg failures */ | |
155 | #undef SLUB_DEBUG_CMPXCHG | |
156 | ||
157 | /* | |
158 | * Mininum number of partial slabs. These will be left on the partial | |
159 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
160 | */ | |
161 | #define MIN_PARTIAL 5 | |
162 | ||
163 | /* | |
164 | * Maximum number of desirable partial slabs. | |
165 | * The existence of more partial slabs makes kmem_cache_shrink | |
166 | * sort the partial list by the number of objects in use. | |
167 | */ | |
168 | #define MAX_PARTIAL 10 | |
169 | ||
170 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ | |
171 | SLAB_POISON | SLAB_STORE_USER) | |
172 | ||
173 | /* | |
174 | * These debug flags cannot use CMPXCHG because there might be consistency | |
175 | * issues when checking or reading debug information | |
176 | */ | |
177 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
178 | SLAB_TRACE) | |
179 | ||
180 | ||
181 | /* | |
182 | * Debugging flags that require metadata to be stored in the slab. These get | |
183 | * disabled when slub_debug=O is used and a cache's min order increases with | |
184 | * metadata. | |
185 | */ | |
186 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) | |
187 | ||
188 | #define OO_SHIFT 16 | |
189 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
190 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ | |
191 | ||
192 | /* Internal SLUB flags */ | |
193 | /* Poison object */ | |
194 | #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U) | |
195 | /* Use cmpxchg_double */ | |
196 | #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U) | |
197 | ||
198 | /* | |
199 | * Tracking user of a slab. | |
200 | */ | |
201 | #define TRACK_ADDRS_COUNT 16 | |
202 | struct track { | |
203 | unsigned long addr; /* Called from address */ | |
204 | #ifdef CONFIG_STACKTRACE | |
205 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
206 | #endif | |
207 | int cpu; /* Was running on cpu */ | |
208 | int pid; /* Pid context */ | |
209 | unsigned long when; /* When did the operation occur */ | |
210 | }; | |
211 | ||
212 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
213 | ||
214 | #ifdef CONFIG_SYSFS | |
215 | static int sysfs_slab_add(struct kmem_cache *); | |
216 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
217 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); | |
218 | static void sysfs_slab_remove(struct kmem_cache *s); | |
219 | #else | |
220 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } | |
221 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
222 | { return 0; } | |
223 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } | |
224 | static inline void sysfs_slab_remove(struct kmem_cache *s) { } | |
225 | #endif | |
226 | ||
227 | static inline void stat(const struct kmem_cache *s, enum stat_item si) | |
228 | { | |
229 | #ifdef CONFIG_SLUB_STATS | |
230 | /* | |
231 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
232 | * avoid this_cpu_add()'s irq-disable overhead. | |
233 | */ | |
234 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
235 | #endif | |
236 | } | |
237 | ||
238 | /******************************************************************** | |
239 | * Core slab cache functions | |
240 | *******************************************************************/ | |
241 | ||
242 | /* | |
243 | * Returns freelist pointer (ptr). With hardening, this is obfuscated | |
244 | * with an XOR of the address where the pointer is held and a per-cache | |
245 | * random number. | |
246 | */ | |
247 | static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr, | |
248 | unsigned long ptr_addr) | |
249 | { | |
250 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
251 | /* | |
252 | * When CONFIG_KASAN_SW_TAGS is enabled, ptr_addr might be tagged. | |
253 | * Normally, this doesn't cause any issues, as both set_freepointer() | |
254 | * and get_freepointer() are called with a pointer with the same tag. | |
255 | * However, there are some issues with CONFIG_SLUB_DEBUG code. For | |
256 | * example, when __free_slub() iterates over objects in a cache, it | |
257 | * passes untagged pointers to check_object(). check_object() in turns | |
258 | * calls get_freepointer() with an untagged pointer, which causes the | |
259 | * freepointer to be restored incorrectly. | |
260 | */ | |
261 | return (void *)((unsigned long)ptr ^ s->random ^ | |
262 | (unsigned long)kasan_reset_tag((void *)ptr_addr)); | |
263 | #else | |
264 | return ptr; | |
265 | #endif | |
266 | } | |
267 | ||
268 | /* Returns the freelist pointer recorded at location ptr_addr. */ | |
269 | static inline void *freelist_dereference(const struct kmem_cache *s, | |
270 | void *ptr_addr) | |
271 | { | |
272 | return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr), | |
273 | (unsigned long)ptr_addr); | |
274 | } | |
275 | ||
276 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
277 | { | |
278 | return freelist_dereference(s, object + s->offset); | |
279 | } | |
280 | ||
281 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) | |
282 | { | |
283 | prefetch(object + s->offset); | |
284 | } | |
285 | ||
286 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) | |
287 | { | |
288 | unsigned long freepointer_addr; | |
289 | void *p; | |
290 | ||
291 | if (!debug_pagealloc_enabled()) | |
292 | return get_freepointer(s, object); | |
293 | ||
294 | freepointer_addr = (unsigned long)object + s->offset; | |
295 | probe_kernel_read(&p, (void **)freepointer_addr, sizeof(p)); | |
296 | return freelist_ptr(s, p, freepointer_addr); | |
297 | } | |
298 | ||
299 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
300 | { | |
301 | unsigned long freeptr_addr = (unsigned long)object + s->offset; | |
302 | ||
303 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
304 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
305 | #endif | |
306 | ||
307 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); | |
308 | } | |
309 | ||
310 | /* Loop over all objects in a slab */ | |
311 | #define for_each_object(__p, __s, __addr, __objects) \ | |
312 | for (__p = fixup_red_left(__s, __addr); \ | |
313 | __p < (__addr) + (__objects) * (__s)->size; \ | |
314 | __p += (__s)->size) | |
315 | ||
316 | /* Determine object index from a given position */ | |
317 | static inline unsigned int slab_index(void *p, struct kmem_cache *s, void *addr) | |
318 | { | |
319 | return (kasan_reset_tag(p) - addr) / s->size; | |
320 | } | |
321 | ||
322 | static inline unsigned int order_objects(unsigned int order, unsigned int size) | |
323 | { | |
324 | return ((unsigned int)PAGE_SIZE << order) / size; | |
325 | } | |
326 | ||
327 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, | |
328 | unsigned int size) | |
329 | { | |
330 | struct kmem_cache_order_objects x = { | |
331 | (order << OO_SHIFT) + order_objects(order, size) | |
332 | }; | |
333 | ||
334 | return x; | |
335 | } | |
336 | ||
337 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) | |
338 | { | |
339 | return x.x >> OO_SHIFT; | |
340 | } | |
341 | ||
342 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) | |
343 | { | |
344 | return x.x & OO_MASK; | |
345 | } | |
346 | ||
347 | /* | |
348 | * Per slab locking using the pagelock | |
349 | */ | |
350 | static __always_inline void slab_lock(struct page *page) | |
351 | { | |
352 | VM_BUG_ON_PAGE(PageTail(page), page); | |
353 | bit_spin_lock(PG_locked, &page->flags); | |
354 | } | |
355 | ||
356 | static __always_inline void slab_unlock(struct page *page) | |
357 | { | |
358 | VM_BUG_ON_PAGE(PageTail(page), page); | |
359 | __bit_spin_unlock(PG_locked, &page->flags); | |
360 | } | |
361 | ||
362 | /* Interrupts must be disabled (for the fallback code to work right) */ | |
363 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
364 | void *freelist_old, unsigned long counters_old, | |
365 | void *freelist_new, unsigned long counters_new, | |
366 | const char *n) | |
367 | { | |
368 | VM_BUG_ON(!irqs_disabled()); | |
369 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ | |
370 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
371 | if (s->flags & __CMPXCHG_DOUBLE) { | |
372 | if (cmpxchg_double(&page->freelist, &page->counters, | |
373 | freelist_old, counters_old, | |
374 | freelist_new, counters_new)) | |
375 | return true; | |
376 | } else | |
377 | #endif | |
378 | { | |
379 | slab_lock(page); | |
380 | if (page->freelist == freelist_old && | |
381 | page->counters == counters_old) { | |
382 | page->freelist = freelist_new; | |
383 | page->counters = counters_new; | |
384 | slab_unlock(page); | |
385 | return true; | |
386 | } | |
387 | slab_unlock(page); | |
388 | } | |
389 | ||
390 | cpu_relax(); | |
391 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
392 | ||
393 | #ifdef SLUB_DEBUG_CMPXCHG | |
394 | pr_info("%s %s: cmpxchg double redo ", n, s->name); | |
395 | #endif | |
396 | ||
397 | return false; | |
398 | } | |
399 | ||
400 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
401 | void *freelist_old, unsigned long counters_old, | |
402 | void *freelist_new, unsigned long counters_new, | |
403 | const char *n) | |
404 | { | |
405 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ | |
406 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
407 | if (s->flags & __CMPXCHG_DOUBLE) { | |
408 | if (cmpxchg_double(&page->freelist, &page->counters, | |
409 | freelist_old, counters_old, | |
410 | freelist_new, counters_new)) | |
411 | return true; | |
412 | } else | |
413 | #endif | |
414 | { | |
415 | unsigned long flags; | |
416 | ||
417 | local_irq_save(flags); | |
418 | slab_lock(page); | |
419 | if (page->freelist == freelist_old && | |
420 | page->counters == counters_old) { | |
421 | page->freelist = freelist_new; | |
422 | page->counters = counters_new; | |
423 | slab_unlock(page); | |
424 | local_irq_restore(flags); | |
425 | return true; | |
426 | } | |
427 | slab_unlock(page); | |
428 | local_irq_restore(flags); | |
429 | } | |
430 | ||
431 | cpu_relax(); | |
432 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
433 | ||
434 | #ifdef SLUB_DEBUG_CMPXCHG | |
435 | pr_info("%s %s: cmpxchg double redo ", n, s->name); | |
436 | #endif | |
437 | ||
438 | return false; | |
439 | } | |
440 | ||
441 | #ifdef CONFIG_SLUB_DEBUG | |
442 | /* | |
443 | * Determine a map of object in use on a page. | |
444 | * | |
445 | * Node listlock must be held to guarantee that the page does | |
446 | * not vanish from under us. | |
447 | */ | |
448 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
449 | { | |
450 | void *p; | |
451 | void *addr = page_address(page); | |
452 | ||
453 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
454 | set_bit(slab_index(p, s, addr), map); | |
455 | } | |
456 | ||
457 | static inline unsigned int size_from_object(struct kmem_cache *s) | |
458 | { | |
459 | if (s->flags & SLAB_RED_ZONE) | |
460 | return s->size - s->red_left_pad; | |
461 | ||
462 | return s->size; | |
463 | } | |
464 | ||
465 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
466 | { | |
467 | if (s->flags & SLAB_RED_ZONE) | |
468 | p -= s->red_left_pad; | |
469 | ||
470 | return p; | |
471 | } | |
472 | ||
473 | /* | |
474 | * Debug settings: | |
475 | */ | |
476 | #if defined(CONFIG_SLUB_DEBUG_ON) | |
477 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; | |
478 | #else | |
479 | static slab_flags_t slub_debug; | |
480 | #endif | |
481 | ||
482 | static char *slub_debug_slabs; | |
483 | static int disable_higher_order_debug; | |
484 | ||
485 | /* | |
486 | * slub is about to manipulate internal object metadata. This memory lies | |
487 | * outside the range of the allocated object, so accessing it would normally | |
488 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
489 | * to tell kasan that these accesses are OK. | |
490 | */ | |
491 | static inline void metadata_access_enable(void) | |
492 | { | |
493 | kasan_disable_current(); | |
494 | } | |
495 | ||
496 | static inline void metadata_access_disable(void) | |
497 | { | |
498 | kasan_enable_current(); | |
499 | } | |
500 | ||
501 | /* | |
502 | * Object debugging | |
503 | */ | |
504 | ||
505 | /* Verify that a pointer has an address that is valid within a slab page */ | |
506 | static inline int check_valid_pointer(struct kmem_cache *s, | |
507 | struct page *page, void *object) | |
508 | { | |
509 | void *base; | |
510 | ||
511 | if (!object) | |
512 | return 1; | |
513 | ||
514 | base = page_address(page); | |
515 | object = kasan_reset_tag(object); | |
516 | object = restore_red_left(s, object); | |
517 | if (object < base || object >= base + page->objects * s->size || | |
518 | (object - base) % s->size) { | |
519 | return 0; | |
520 | } | |
521 | ||
522 | return 1; | |
523 | } | |
524 | ||
525 | static void print_section(char *level, char *text, u8 *addr, | |
526 | unsigned int length) | |
527 | { | |
528 | metadata_access_enable(); | |
529 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, | |
530 | length, 1); | |
531 | metadata_access_disable(); | |
532 | } | |
533 | ||
534 | static struct track *get_track(struct kmem_cache *s, void *object, | |
535 | enum track_item alloc) | |
536 | { | |
537 | struct track *p; | |
538 | ||
539 | if (s->offset) | |
540 | p = object + s->offset + sizeof(void *); | |
541 | else | |
542 | p = object + s->inuse; | |
543 | ||
544 | return p + alloc; | |
545 | } | |
546 | ||
547 | static void set_track(struct kmem_cache *s, void *object, | |
548 | enum track_item alloc, unsigned long addr) | |
549 | { | |
550 | struct track *p = get_track(s, object, alloc); | |
551 | ||
552 | if (addr) { | |
553 | #ifdef CONFIG_STACKTRACE | |
554 | unsigned int nr_entries; | |
555 | ||
556 | metadata_access_enable(); | |
557 | nr_entries = stack_trace_save(p->addrs, TRACK_ADDRS_COUNT, 3); | |
558 | metadata_access_disable(); | |
559 | ||
560 | if (nr_entries < TRACK_ADDRS_COUNT) | |
561 | p->addrs[nr_entries] = 0; | |
562 | #endif | |
563 | p->addr = addr; | |
564 | p->cpu = smp_processor_id(); | |
565 | p->pid = current->pid; | |
566 | p->when = jiffies; | |
567 | } else { | |
568 | memset(p, 0, sizeof(struct track)); | |
569 | } | |
570 | } | |
571 | ||
572 | static void init_tracking(struct kmem_cache *s, void *object) | |
573 | { | |
574 | if (!(s->flags & SLAB_STORE_USER)) | |
575 | return; | |
576 | ||
577 | set_track(s, object, TRACK_FREE, 0UL); | |
578 | set_track(s, object, TRACK_ALLOC, 0UL); | |
579 | } | |
580 | ||
581 | static void print_track(const char *s, struct track *t, unsigned long pr_time) | |
582 | { | |
583 | if (!t->addr) | |
584 | return; | |
585 | ||
586 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", | |
587 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); | |
588 | #ifdef CONFIG_STACKTRACE | |
589 | { | |
590 | int i; | |
591 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
592 | if (t->addrs[i]) | |
593 | pr_err("\t%pS\n", (void *)t->addrs[i]); | |
594 | else | |
595 | break; | |
596 | } | |
597 | #endif | |
598 | } | |
599 | ||
600 | static void print_tracking(struct kmem_cache *s, void *object) | |
601 | { | |
602 | unsigned long pr_time = jiffies; | |
603 | if (!(s->flags & SLAB_STORE_USER)) | |
604 | return; | |
605 | ||
606 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); | |
607 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
608 | } | |
609 | ||
610 | static void print_page_info(struct page *page) | |
611 | { | |
612 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", | |
613 | page, page->objects, page->inuse, page->freelist, page->flags); | |
614 | ||
615 | } | |
616 | ||
617 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
618 | { | |
619 | struct va_format vaf; | |
620 | va_list args; | |
621 | ||
622 | va_start(args, fmt); | |
623 | vaf.fmt = fmt; | |
624 | vaf.va = &args; | |
625 | pr_err("=============================================================================\n"); | |
626 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); | |
627 | pr_err("-----------------------------------------------------------------------------\n\n"); | |
628 | ||
629 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
630 | va_end(args); | |
631 | } | |
632 | ||
633 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) | |
634 | { | |
635 | struct va_format vaf; | |
636 | va_list args; | |
637 | ||
638 | va_start(args, fmt); | |
639 | vaf.fmt = fmt; | |
640 | vaf.va = &args; | |
641 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
642 | va_end(args); | |
643 | } | |
644 | ||
645 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
646 | { | |
647 | unsigned int off; /* Offset of last byte */ | |
648 | u8 *addr = page_address(page); | |
649 | ||
650 | print_tracking(s, p); | |
651 | ||
652 | print_page_info(page); | |
653 | ||
654 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
655 | p, p - addr, get_freepointer(s, p)); | |
656 | ||
657 | if (s->flags & SLAB_RED_ZONE) | |
658 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, | |
659 | s->red_left_pad); | |
660 | else if (p > addr + 16) | |
661 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); | |
662 | ||
663 | print_section(KERN_ERR, "Object ", p, | |
664 | min_t(unsigned int, s->object_size, PAGE_SIZE)); | |
665 | if (s->flags & SLAB_RED_ZONE) | |
666 | print_section(KERN_ERR, "Redzone ", p + s->object_size, | |
667 | s->inuse - s->object_size); | |
668 | ||
669 | if (s->offset) | |
670 | off = s->offset + sizeof(void *); | |
671 | else | |
672 | off = s->inuse; | |
673 | ||
674 | if (s->flags & SLAB_STORE_USER) | |
675 | off += 2 * sizeof(struct track); | |
676 | ||
677 | off += kasan_metadata_size(s); | |
678 | ||
679 | if (off != size_from_object(s)) | |
680 | /* Beginning of the filler is the free pointer */ | |
681 | print_section(KERN_ERR, "Padding ", p + off, | |
682 | size_from_object(s) - off); | |
683 | ||
684 | dump_stack(); | |
685 | } | |
686 | ||
687 | void object_err(struct kmem_cache *s, struct page *page, | |
688 | u8 *object, char *reason) | |
689 | { | |
690 | slab_bug(s, "%s", reason); | |
691 | print_trailer(s, page, object); | |
692 | } | |
693 | ||
694 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page, | |
695 | const char *fmt, ...) | |
696 | { | |
697 | va_list args; | |
698 | char buf[100]; | |
699 | ||
700 | va_start(args, fmt); | |
701 | vsnprintf(buf, sizeof(buf), fmt, args); | |
702 | va_end(args); | |
703 | slab_bug(s, "%s", buf); | |
704 | print_page_info(page); | |
705 | dump_stack(); | |
706 | } | |
707 | ||
708 | static void init_object(struct kmem_cache *s, void *object, u8 val) | |
709 | { | |
710 | u8 *p = object; | |
711 | ||
712 | if (s->flags & SLAB_RED_ZONE) | |
713 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
714 | ||
715 | if (s->flags & __OBJECT_POISON) { | |
716 | memset(p, POISON_FREE, s->object_size - 1); | |
717 | p[s->object_size - 1] = POISON_END; | |
718 | } | |
719 | ||
720 | if (s->flags & SLAB_RED_ZONE) | |
721 | memset(p + s->object_size, val, s->inuse - s->object_size); | |
722 | } | |
723 | ||
724 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
725 | void *from, void *to) | |
726 | { | |
727 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
728 | memset(from, data, to - from); | |
729 | } | |
730 | ||
731 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
732 | u8 *object, char *what, | |
733 | u8 *start, unsigned int value, unsigned int bytes) | |
734 | { | |
735 | u8 *fault; | |
736 | u8 *end; | |
737 | u8 *addr = page_address(page); | |
738 | ||
739 | metadata_access_enable(); | |
740 | fault = memchr_inv(start, value, bytes); | |
741 | metadata_access_disable(); | |
742 | if (!fault) | |
743 | return 1; | |
744 | ||
745 | end = start + bytes; | |
746 | while (end > fault && end[-1] == value) | |
747 | end--; | |
748 | ||
749 | slab_bug(s, "%s overwritten", what); | |
750 | pr_err("INFO: 0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n", | |
751 | fault, end - 1, fault - addr, | |
752 | fault[0], value); | |
753 | print_trailer(s, page, object); | |
754 | ||
755 | restore_bytes(s, what, value, fault, end); | |
756 | return 0; | |
757 | } | |
758 | ||
759 | /* | |
760 | * Object layout: | |
761 | * | |
762 | * object address | |
763 | * Bytes of the object to be managed. | |
764 | * If the freepointer may overlay the object then the free | |
765 | * pointer is the first word of the object. | |
766 | * | |
767 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is | |
768 | * 0xa5 (POISON_END) | |
769 | * | |
770 | * object + s->object_size | |
771 | * Padding to reach word boundary. This is also used for Redzoning. | |
772 | * Padding is extended by another word if Redzoning is enabled and | |
773 | * object_size == inuse. | |
774 | * | |
775 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with | |
776 | * 0xcc (RED_ACTIVE) for objects in use. | |
777 | * | |
778 | * object + s->inuse | |
779 | * Meta data starts here. | |
780 | * | |
781 | * A. Free pointer (if we cannot overwrite object on free) | |
782 | * B. Tracking data for SLAB_STORE_USER | |
783 | * C. Padding to reach required alignment boundary or at mininum | |
784 | * one word if debugging is on to be able to detect writes | |
785 | * before the word boundary. | |
786 | * | |
787 | * Padding is done using 0x5a (POISON_INUSE) | |
788 | * | |
789 | * object + s->size | |
790 | * Nothing is used beyond s->size. | |
791 | * | |
792 | * If slabcaches are merged then the object_size and inuse boundaries are mostly | |
793 | * ignored. And therefore no slab options that rely on these boundaries | |
794 | * may be used with merged slabcaches. | |
795 | */ | |
796 | ||
797 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) | |
798 | { | |
799 | unsigned long off = s->inuse; /* The end of info */ | |
800 | ||
801 | if (s->offset) | |
802 | /* Freepointer is placed after the object. */ | |
803 | off += sizeof(void *); | |
804 | ||
805 | if (s->flags & SLAB_STORE_USER) | |
806 | /* We also have user information there */ | |
807 | off += 2 * sizeof(struct track); | |
808 | ||
809 | off += kasan_metadata_size(s); | |
810 | ||
811 | if (size_from_object(s) == off) | |
812 | return 1; | |
813 | ||
814 | return check_bytes_and_report(s, page, p, "Object padding", | |
815 | p + off, POISON_INUSE, size_from_object(s) - off); | |
816 | } | |
817 | ||
818 | /* Check the pad bytes at the end of a slab page */ | |
819 | static int slab_pad_check(struct kmem_cache *s, struct page *page) | |
820 | { | |
821 | u8 *start; | |
822 | u8 *fault; | |
823 | u8 *end; | |
824 | u8 *pad; | |
825 | int length; | |
826 | int remainder; | |
827 | ||
828 | if (!(s->flags & SLAB_POISON)) | |
829 | return 1; | |
830 | ||
831 | start = page_address(page); | |
832 | length = page_size(page); | |
833 | end = start + length; | |
834 | remainder = length % s->size; | |
835 | if (!remainder) | |
836 | return 1; | |
837 | ||
838 | pad = end - remainder; | |
839 | metadata_access_enable(); | |
840 | fault = memchr_inv(pad, POISON_INUSE, remainder); | |
841 | metadata_access_disable(); | |
842 | if (!fault) | |
843 | return 1; | |
844 | while (end > fault && end[-1] == POISON_INUSE) | |
845 | end--; | |
846 | ||
847 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p @offset=%tu", | |
848 | fault, end - 1, fault - start); | |
849 | print_section(KERN_ERR, "Padding ", pad, remainder); | |
850 | ||
851 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); | |
852 | return 0; | |
853 | } | |
854 | ||
855 | static int check_object(struct kmem_cache *s, struct page *page, | |
856 | void *object, u8 val) | |
857 | { | |
858 | u8 *p = object; | |
859 | u8 *endobject = object + s->object_size; | |
860 | ||
861 | if (s->flags & SLAB_RED_ZONE) { | |
862 | if (!check_bytes_and_report(s, page, object, "Redzone", | |
863 | object - s->red_left_pad, val, s->red_left_pad)) | |
864 | return 0; | |
865 | ||
866 | if (!check_bytes_and_report(s, page, object, "Redzone", | |
867 | endobject, val, s->inuse - s->object_size)) | |
868 | return 0; | |
869 | } else { | |
870 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { | |
871 | check_bytes_and_report(s, page, p, "Alignment padding", | |
872 | endobject, POISON_INUSE, | |
873 | s->inuse - s->object_size); | |
874 | } | |
875 | } | |
876 | ||
877 | if (s->flags & SLAB_POISON) { | |
878 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && | |
879 | (!check_bytes_and_report(s, page, p, "Poison", p, | |
880 | POISON_FREE, s->object_size - 1) || | |
881 | !check_bytes_and_report(s, page, p, "Poison", | |
882 | p + s->object_size - 1, POISON_END, 1))) | |
883 | return 0; | |
884 | /* | |
885 | * check_pad_bytes cleans up on its own. | |
886 | */ | |
887 | check_pad_bytes(s, page, p); | |
888 | } | |
889 | ||
890 | if (!s->offset && val == SLUB_RED_ACTIVE) | |
891 | /* | |
892 | * Object and freepointer overlap. Cannot check | |
893 | * freepointer while object is allocated. | |
894 | */ | |
895 | return 1; | |
896 | ||
897 | /* Check free pointer validity */ | |
898 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
899 | object_err(s, page, p, "Freepointer corrupt"); | |
900 | /* | |
901 | * No choice but to zap it and thus lose the remainder | |
902 | * of the free objects in this slab. May cause | |
903 | * another error because the object count is now wrong. | |
904 | */ | |
905 | set_freepointer(s, p, NULL); | |
906 | return 0; | |
907 | } | |
908 | return 1; | |
909 | } | |
910 | ||
911 | static int check_slab(struct kmem_cache *s, struct page *page) | |
912 | { | |
913 | int maxobj; | |
914 | ||
915 | VM_BUG_ON(!irqs_disabled()); | |
916 | ||
917 | if (!PageSlab(page)) { | |
918 | slab_err(s, page, "Not a valid slab page"); | |
919 | return 0; | |
920 | } | |
921 | ||
922 | maxobj = order_objects(compound_order(page), s->size); | |
923 | if (page->objects > maxobj) { | |
924 | slab_err(s, page, "objects %u > max %u", | |
925 | page->objects, maxobj); | |
926 | return 0; | |
927 | } | |
928 | if (page->inuse > page->objects) { | |
929 | slab_err(s, page, "inuse %u > max %u", | |
930 | page->inuse, page->objects); | |
931 | return 0; | |
932 | } | |
933 | /* Slab_pad_check fixes things up after itself */ | |
934 | slab_pad_check(s, page); | |
935 | return 1; | |
936 | } | |
937 | ||
938 | /* | |
939 | * Determine if a certain object on a page is on the freelist. Must hold the | |
940 | * slab lock to guarantee that the chains are in a consistent state. | |
941 | */ | |
942 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
943 | { | |
944 | int nr = 0; | |
945 | void *fp; | |
946 | void *object = NULL; | |
947 | int max_objects; | |
948 | ||
949 | fp = page->freelist; | |
950 | while (fp && nr <= page->objects) { | |
951 | if (fp == search) | |
952 | return 1; | |
953 | if (!check_valid_pointer(s, page, fp)) { | |
954 | if (object) { | |
955 | object_err(s, page, object, | |
956 | "Freechain corrupt"); | |
957 | set_freepointer(s, object, NULL); | |
958 | } else { | |
959 | slab_err(s, page, "Freepointer corrupt"); | |
960 | page->freelist = NULL; | |
961 | page->inuse = page->objects; | |
962 | slab_fix(s, "Freelist cleared"); | |
963 | return 0; | |
964 | } | |
965 | break; | |
966 | } | |
967 | object = fp; | |
968 | fp = get_freepointer(s, object); | |
969 | nr++; | |
970 | } | |
971 | ||
972 | max_objects = order_objects(compound_order(page), s->size); | |
973 | if (max_objects > MAX_OBJS_PER_PAGE) | |
974 | max_objects = MAX_OBJS_PER_PAGE; | |
975 | ||
976 | if (page->objects != max_objects) { | |
977 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", | |
978 | page->objects, max_objects); | |
979 | page->objects = max_objects; | |
980 | slab_fix(s, "Number of objects adjusted."); | |
981 | } | |
982 | if (page->inuse != page->objects - nr) { | |
983 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", | |
984 | page->inuse, page->objects - nr); | |
985 | page->inuse = page->objects - nr; | |
986 | slab_fix(s, "Object count adjusted."); | |
987 | } | |
988 | return search == NULL; | |
989 | } | |
990 | ||
991 | static void trace(struct kmem_cache *s, struct page *page, void *object, | |
992 | int alloc) | |
993 | { | |
994 | if (s->flags & SLAB_TRACE) { | |
995 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
996 | s->name, | |
997 | alloc ? "alloc" : "free", | |
998 | object, page->inuse, | |
999 | page->freelist); | |
1000 | ||
1001 | if (!alloc) | |
1002 | print_section(KERN_INFO, "Object ", (void *)object, | |
1003 | s->object_size); | |
1004 | ||
1005 | dump_stack(); | |
1006 | } | |
1007 | } | |
1008 | ||
1009 | /* | |
1010 | * Tracking of fully allocated slabs for debugging purposes. | |
1011 | */ | |
1012 | static void add_full(struct kmem_cache *s, | |
1013 | struct kmem_cache_node *n, struct page *page) | |
1014 | { | |
1015 | if (!(s->flags & SLAB_STORE_USER)) | |
1016 | return; | |
1017 | ||
1018 | lockdep_assert_held(&n->list_lock); | |
1019 | list_add(&page->slab_list, &n->full); | |
1020 | } | |
1021 | ||
1022 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) | |
1023 | { | |
1024 | if (!(s->flags & SLAB_STORE_USER)) | |
1025 | return; | |
1026 | ||
1027 | lockdep_assert_held(&n->list_lock); | |
1028 | list_del(&page->slab_list); | |
1029 | } | |
1030 | ||
1031 | /* Tracking of the number of slabs for debugging purposes */ | |
1032 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1033 | { | |
1034 | struct kmem_cache_node *n = get_node(s, node); | |
1035 | ||
1036 | return atomic_long_read(&n->nr_slabs); | |
1037 | } | |
1038 | ||
1039 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) | |
1040 | { | |
1041 | return atomic_long_read(&n->nr_slabs); | |
1042 | } | |
1043 | ||
1044 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) | |
1045 | { | |
1046 | struct kmem_cache_node *n = get_node(s, node); | |
1047 | ||
1048 | /* | |
1049 | * May be called early in order to allocate a slab for the | |
1050 | * kmem_cache_node structure. Solve the chicken-egg | |
1051 | * dilemma by deferring the increment of the count during | |
1052 | * bootstrap (see early_kmem_cache_node_alloc). | |
1053 | */ | |
1054 | if (likely(n)) { | |
1055 | atomic_long_inc(&n->nr_slabs); | |
1056 | atomic_long_add(objects, &n->total_objects); | |
1057 | } | |
1058 | } | |
1059 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) | |
1060 | { | |
1061 | struct kmem_cache_node *n = get_node(s, node); | |
1062 | ||
1063 | atomic_long_dec(&n->nr_slabs); | |
1064 | atomic_long_sub(objects, &n->total_objects); | |
1065 | } | |
1066 | ||
1067 | /* Object debug checks for alloc/free paths */ | |
1068 | static void setup_object_debug(struct kmem_cache *s, struct page *page, | |
1069 | void *object) | |
1070 | { | |
1071 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1072 | return; | |
1073 | ||
1074 | init_object(s, object, SLUB_RED_INACTIVE); | |
1075 | init_tracking(s, object); | |
1076 | } | |
1077 | ||
1078 | static | |
1079 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) | |
1080 | { | |
1081 | if (!(s->flags & SLAB_POISON)) | |
1082 | return; | |
1083 | ||
1084 | metadata_access_enable(); | |
1085 | memset(addr, POISON_INUSE, page_size(page)); | |
1086 | metadata_access_disable(); | |
1087 | } | |
1088 | ||
1089 | static inline int alloc_consistency_checks(struct kmem_cache *s, | |
1090 | struct page *page, void *object) | |
1091 | { | |
1092 | if (!check_slab(s, page)) | |
1093 | return 0; | |
1094 | ||
1095 | if (!check_valid_pointer(s, page, object)) { | |
1096 | object_err(s, page, object, "Freelist Pointer check fails"); | |
1097 | return 0; | |
1098 | } | |
1099 | ||
1100 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) | |
1101 | return 0; | |
1102 | ||
1103 | return 1; | |
1104 | } | |
1105 | ||
1106 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1107 | struct page *page, | |
1108 | void *object, unsigned long addr) | |
1109 | { | |
1110 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1111 | if (!alloc_consistency_checks(s, page, object)) | |
1112 | goto bad; | |
1113 | } | |
1114 | ||
1115 | /* Success perform special debug activities for allocs */ | |
1116 | if (s->flags & SLAB_STORE_USER) | |
1117 | set_track(s, object, TRACK_ALLOC, addr); | |
1118 | trace(s, page, object, 1); | |
1119 | init_object(s, object, SLUB_RED_ACTIVE); | |
1120 | return 1; | |
1121 | ||
1122 | bad: | |
1123 | if (PageSlab(page)) { | |
1124 | /* | |
1125 | * If this is a slab page then lets do the best we can | |
1126 | * to avoid issues in the future. Marking all objects | |
1127 | * as used avoids touching the remaining objects. | |
1128 | */ | |
1129 | slab_fix(s, "Marking all objects used"); | |
1130 | page->inuse = page->objects; | |
1131 | page->freelist = NULL; | |
1132 | } | |
1133 | return 0; | |
1134 | } | |
1135 | ||
1136 | static inline int free_consistency_checks(struct kmem_cache *s, | |
1137 | struct page *page, void *object, unsigned long addr) | |
1138 | { | |
1139 | if (!check_valid_pointer(s, page, object)) { | |
1140 | slab_err(s, page, "Invalid object pointer 0x%p", object); | |
1141 | return 0; | |
1142 | } | |
1143 | ||
1144 | if (on_freelist(s, page, object)) { | |
1145 | object_err(s, page, object, "Object already free"); | |
1146 | return 0; | |
1147 | } | |
1148 | ||
1149 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) | |
1150 | return 0; | |
1151 | ||
1152 | if (unlikely(s != page->slab_cache)) { | |
1153 | if (!PageSlab(page)) { | |
1154 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", | |
1155 | object); | |
1156 | } else if (!page->slab_cache) { | |
1157 | pr_err("SLUB <none>: no slab for object 0x%p.\n", | |
1158 | object); | |
1159 | dump_stack(); | |
1160 | } else | |
1161 | object_err(s, page, object, | |
1162 | "page slab pointer corrupt."); | |
1163 | return 0; | |
1164 | } | |
1165 | return 1; | |
1166 | } | |
1167 | ||
1168 | /* Supports checking bulk free of a constructed freelist */ | |
1169 | static noinline int free_debug_processing( | |
1170 | struct kmem_cache *s, struct page *page, | |
1171 | void *head, void *tail, int bulk_cnt, | |
1172 | unsigned long addr) | |
1173 | { | |
1174 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1175 | void *object = head; | |
1176 | int cnt = 0; | |
1177 | unsigned long uninitialized_var(flags); | |
1178 | int ret = 0; | |
1179 | ||
1180 | spin_lock_irqsave(&n->list_lock, flags); | |
1181 | slab_lock(page); | |
1182 | ||
1183 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1184 | if (!check_slab(s, page)) | |
1185 | goto out; | |
1186 | } | |
1187 | ||
1188 | next_object: | |
1189 | cnt++; | |
1190 | ||
1191 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1192 | if (!free_consistency_checks(s, page, object, addr)) | |
1193 | goto out; | |
1194 | } | |
1195 | ||
1196 | if (s->flags & SLAB_STORE_USER) | |
1197 | set_track(s, object, TRACK_FREE, addr); | |
1198 | trace(s, page, object, 0); | |
1199 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ | |
1200 | init_object(s, object, SLUB_RED_INACTIVE); | |
1201 | ||
1202 | /* Reached end of constructed freelist yet? */ | |
1203 | if (object != tail) { | |
1204 | object = get_freepointer(s, object); | |
1205 | goto next_object; | |
1206 | } | |
1207 | ret = 1; | |
1208 | ||
1209 | out: | |
1210 | if (cnt != bulk_cnt) | |
1211 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1212 | bulk_cnt, cnt); | |
1213 | ||
1214 | slab_unlock(page); | |
1215 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1216 | if (!ret) | |
1217 | slab_fix(s, "Object at 0x%p not freed", object); | |
1218 | return ret; | |
1219 | } | |
1220 | ||
1221 | static int __init setup_slub_debug(char *str) | |
1222 | { | |
1223 | slub_debug = DEBUG_DEFAULT_FLAGS; | |
1224 | if (*str++ != '=' || !*str) | |
1225 | /* | |
1226 | * No options specified. Switch on full debugging. | |
1227 | */ | |
1228 | goto out; | |
1229 | ||
1230 | if (*str == ',') | |
1231 | /* | |
1232 | * No options but restriction on slabs. This means full | |
1233 | * debugging for slabs matching a pattern. | |
1234 | */ | |
1235 | goto check_slabs; | |
1236 | ||
1237 | slub_debug = 0; | |
1238 | if (*str == '-') | |
1239 | /* | |
1240 | * Switch off all debugging measures. | |
1241 | */ | |
1242 | goto out; | |
1243 | ||
1244 | /* | |
1245 | * Determine which debug features should be switched on | |
1246 | */ | |
1247 | for (; *str && *str != ','; str++) { | |
1248 | switch (tolower(*str)) { | |
1249 | case 'f': | |
1250 | slub_debug |= SLAB_CONSISTENCY_CHECKS; | |
1251 | break; | |
1252 | case 'z': | |
1253 | slub_debug |= SLAB_RED_ZONE; | |
1254 | break; | |
1255 | case 'p': | |
1256 | slub_debug |= SLAB_POISON; | |
1257 | break; | |
1258 | case 'u': | |
1259 | slub_debug |= SLAB_STORE_USER; | |
1260 | break; | |
1261 | case 't': | |
1262 | slub_debug |= SLAB_TRACE; | |
1263 | break; | |
1264 | case 'a': | |
1265 | slub_debug |= SLAB_FAILSLAB; | |
1266 | break; | |
1267 | case 'o': | |
1268 | /* | |
1269 | * Avoid enabling debugging on caches if its minimum | |
1270 | * order would increase as a result. | |
1271 | */ | |
1272 | disable_higher_order_debug = 1; | |
1273 | break; | |
1274 | default: | |
1275 | pr_err("slub_debug option '%c' unknown. skipped\n", | |
1276 | *str); | |
1277 | } | |
1278 | } | |
1279 | ||
1280 | check_slabs: | |
1281 | if (*str == ',') | |
1282 | slub_debug_slabs = str + 1; | |
1283 | out: | |
1284 | if ((static_branch_unlikely(&init_on_alloc) || | |
1285 | static_branch_unlikely(&init_on_free)) && | |
1286 | (slub_debug & SLAB_POISON)) | |
1287 | pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n"); | |
1288 | return 1; | |
1289 | } | |
1290 | ||
1291 | __setup("slub_debug", setup_slub_debug); | |
1292 | ||
1293 | /* | |
1294 | * kmem_cache_flags - apply debugging options to the cache | |
1295 | * @object_size: the size of an object without meta data | |
1296 | * @flags: flags to set | |
1297 | * @name: name of the cache | |
1298 | * @ctor: constructor function | |
1299 | * | |
1300 | * Debug option(s) are applied to @flags. In addition to the debug | |
1301 | * option(s), if a slab name (or multiple) is specified i.e. | |
1302 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1303 | * then only the select slabs will receive the debug option(s). | |
1304 | */ | |
1305 | slab_flags_t kmem_cache_flags(unsigned int object_size, | |
1306 | slab_flags_t flags, const char *name, | |
1307 | void (*ctor)(void *)) | |
1308 | { | |
1309 | char *iter; | |
1310 | size_t len; | |
1311 | ||
1312 | /* If slub_debug = 0, it folds into the if conditional. */ | |
1313 | if (!slub_debug_slabs) | |
1314 | return flags | slub_debug; | |
1315 | ||
1316 | len = strlen(name); | |
1317 | iter = slub_debug_slabs; | |
1318 | while (*iter) { | |
1319 | char *end, *glob; | |
1320 | size_t cmplen; | |
1321 | ||
1322 | end = strchrnul(iter, ','); | |
1323 | ||
1324 | glob = strnchr(iter, end - iter, '*'); | |
1325 | if (glob) | |
1326 | cmplen = glob - iter; | |
1327 | else | |
1328 | cmplen = max_t(size_t, len, (end - iter)); | |
1329 | ||
1330 | if (!strncmp(name, iter, cmplen)) { | |
1331 | flags |= slub_debug; | |
1332 | break; | |
1333 | } | |
1334 | ||
1335 | if (!*end) | |
1336 | break; | |
1337 | iter = end + 1; | |
1338 | } | |
1339 | ||
1340 | return flags; | |
1341 | } | |
1342 | #else /* !CONFIG_SLUB_DEBUG */ | |
1343 | static inline void setup_object_debug(struct kmem_cache *s, | |
1344 | struct page *page, void *object) {} | |
1345 | static inline | |
1346 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {} | |
1347 | ||
1348 | static inline int alloc_debug_processing(struct kmem_cache *s, | |
1349 | struct page *page, void *object, unsigned long addr) { return 0; } | |
1350 | ||
1351 | static inline int free_debug_processing( | |
1352 | struct kmem_cache *s, struct page *page, | |
1353 | void *head, void *tail, int bulk_cnt, | |
1354 | unsigned long addr) { return 0; } | |
1355 | ||
1356 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) | |
1357 | { return 1; } | |
1358 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1359 | void *object, u8 val) { return 1; } | |
1360 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, | |
1361 | struct page *page) {} | |
1362 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, | |
1363 | struct page *page) {} | |
1364 | slab_flags_t kmem_cache_flags(unsigned int object_size, | |
1365 | slab_flags_t flags, const char *name, | |
1366 | void (*ctor)(void *)) | |
1367 | { | |
1368 | return flags; | |
1369 | } | |
1370 | #define slub_debug 0 | |
1371 | ||
1372 | #define disable_higher_order_debug 0 | |
1373 | ||
1374 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1375 | { return 0; } | |
1376 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) | |
1377 | { return 0; } | |
1378 | static inline void inc_slabs_node(struct kmem_cache *s, int node, | |
1379 | int objects) {} | |
1380 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1381 | int objects) {} | |
1382 | ||
1383 | #endif /* CONFIG_SLUB_DEBUG */ | |
1384 | ||
1385 | /* | |
1386 | * Hooks for other subsystems that check memory allocations. In a typical | |
1387 | * production configuration these hooks all should produce no code at all. | |
1388 | */ | |
1389 | static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) | |
1390 | { | |
1391 | ptr = kasan_kmalloc_large(ptr, size, flags); | |
1392 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ | |
1393 | kmemleak_alloc(ptr, size, 1, flags); | |
1394 | return ptr; | |
1395 | } | |
1396 | ||
1397 | static __always_inline void kfree_hook(void *x) | |
1398 | { | |
1399 | kmemleak_free(x); | |
1400 | kasan_kfree_large(x, _RET_IP_); | |
1401 | } | |
1402 | ||
1403 | static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x) | |
1404 | { | |
1405 | kmemleak_free_recursive(x, s->flags); | |
1406 | ||
1407 | /* | |
1408 | * Trouble is that we may no longer disable interrupts in the fast path | |
1409 | * So in order to make the debug calls that expect irqs to be | |
1410 | * disabled we need to disable interrupts temporarily. | |
1411 | */ | |
1412 | #ifdef CONFIG_LOCKDEP | |
1413 | { | |
1414 | unsigned long flags; | |
1415 | ||
1416 | local_irq_save(flags); | |
1417 | debug_check_no_locks_freed(x, s->object_size); | |
1418 | local_irq_restore(flags); | |
1419 | } | |
1420 | #endif | |
1421 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1422 | debug_check_no_obj_freed(x, s->object_size); | |
1423 | ||
1424 | /* KASAN might put x into memory quarantine, delaying its reuse */ | |
1425 | return kasan_slab_free(s, x, _RET_IP_); | |
1426 | } | |
1427 | ||
1428 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, | |
1429 | void **head, void **tail) | |
1430 | { | |
1431 | ||
1432 | void *object; | |
1433 | void *next = *head; | |
1434 | void *old_tail = *tail ? *tail : *head; | |
1435 | int rsize; | |
1436 | ||
1437 | /* Head and tail of the reconstructed freelist */ | |
1438 | *head = NULL; | |
1439 | *tail = NULL; | |
1440 | ||
1441 | do { | |
1442 | object = next; | |
1443 | next = get_freepointer(s, object); | |
1444 | ||
1445 | if (slab_want_init_on_free(s)) { | |
1446 | /* | |
1447 | * Clear the object and the metadata, but don't touch | |
1448 | * the redzone. | |
1449 | */ | |
1450 | memset(object, 0, s->object_size); | |
1451 | rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad | |
1452 | : 0; | |
1453 | memset((char *)object + s->inuse, 0, | |
1454 | s->size - s->inuse - rsize); | |
1455 | ||
1456 | } | |
1457 | /* If object's reuse doesn't have to be delayed */ | |
1458 | if (!slab_free_hook(s, object)) { | |
1459 | /* Move object to the new freelist */ | |
1460 | set_freepointer(s, object, *head); | |
1461 | *head = object; | |
1462 | if (!*tail) | |
1463 | *tail = object; | |
1464 | } | |
1465 | } while (object != old_tail); | |
1466 | ||
1467 | if (*head == *tail) | |
1468 | *tail = NULL; | |
1469 | ||
1470 | return *head != NULL; | |
1471 | } | |
1472 | ||
1473 | static void *setup_object(struct kmem_cache *s, struct page *page, | |
1474 | void *object) | |
1475 | { | |
1476 | setup_object_debug(s, page, object); | |
1477 | object = kasan_init_slab_obj(s, object); | |
1478 | if (unlikely(s->ctor)) { | |
1479 | kasan_unpoison_object_data(s, object); | |
1480 | s->ctor(object); | |
1481 | kasan_poison_object_data(s, object); | |
1482 | } | |
1483 | return object; | |
1484 | } | |
1485 | ||
1486 | /* | |
1487 | * Slab allocation and freeing | |
1488 | */ | |
1489 | static inline struct page *alloc_slab_page(struct kmem_cache *s, | |
1490 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
1491 | { | |
1492 | struct page *page; | |
1493 | unsigned int order = oo_order(oo); | |
1494 | ||
1495 | if (node == NUMA_NO_NODE) | |
1496 | page = alloc_pages(flags, order); | |
1497 | else | |
1498 | page = __alloc_pages_node(node, flags, order); | |
1499 | ||
1500 | if (page && charge_slab_page(page, flags, order, s)) { | |
1501 | __free_pages(page, order); | |
1502 | page = NULL; | |
1503 | } | |
1504 | ||
1505 | return page; | |
1506 | } | |
1507 | ||
1508 | #ifdef CONFIG_SLAB_FREELIST_RANDOM | |
1509 | /* Pre-initialize the random sequence cache */ | |
1510 | static int init_cache_random_seq(struct kmem_cache *s) | |
1511 | { | |
1512 | unsigned int count = oo_objects(s->oo); | |
1513 | int err; | |
1514 | ||
1515 | /* Bailout if already initialised */ | |
1516 | if (s->random_seq) | |
1517 | return 0; | |
1518 | ||
1519 | err = cache_random_seq_create(s, count, GFP_KERNEL); | |
1520 | if (err) { | |
1521 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1522 | s->name); | |
1523 | return err; | |
1524 | } | |
1525 | ||
1526 | /* Transform to an offset on the set of pages */ | |
1527 | if (s->random_seq) { | |
1528 | unsigned int i; | |
1529 | ||
1530 | for (i = 0; i < count; i++) | |
1531 | s->random_seq[i] *= s->size; | |
1532 | } | |
1533 | return 0; | |
1534 | } | |
1535 | ||
1536 | /* Initialize each random sequence freelist per cache */ | |
1537 | static void __init init_freelist_randomization(void) | |
1538 | { | |
1539 | struct kmem_cache *s; | |
1540 | ||
1541 | mutex_lock(&slab_mutex); | |
1542 | ||
1543 | list_for_each_entry(s, &slab_caches, list) | |
1544 | init_cache_random_seq(s); | |
1545 | ||
1546 | mutex_unlock(&slab_mutex); | |
1547 | } | |
1548 | ||
1549 | /* Get the next entry on the pre-computed freelist randomized */ | |
1550 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1551 | unsigned long *pos, void *start, | |
1552 | unsigned long page_limit, | |
1553 | unsigned long freelist_count) | |
1554 | { | |
1555 | unsigned int idx; | |
1556 | ||
1557 | /* | |
1558 | * If the target page allocation failed, the number of objects on the | |
1559 | * page might be smaller than the usual size defined by the cache. | |
1560 | */ | |
1561 | do { | |
1562 | idx = s->random_seq[*pos]; | |
1563 | *pos += 1; | |
1564 | if (*pos >= freelist_count) | |
1565 | *pos = 0; | |
1566 | } while (unlikely(idx >= page_limit)); | |
1567 | ||
1568 | return (char *)start + idx; | |
1569 | } | |
1570 | ||
1571 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1572 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1573 | { | |
1574 | void *start; | |
1575 | void *cur; | |
1576 | void *next; | |
1577 | unsigned long idx, pos, page_limit, freelist_count; | |
1578 | ||
1579 | if (page->objects < 2 || !s->random_seq) | |
1580 | return false; | |
1581 | ||
1582 | freelist_count = oo_objects(s->oo); | |
1583 | pos = get_random_int() % freelist_count; | |
1584 | ||
1585 | page_limit = page->objects * s->size; | |
1586 | start = fixup_red_left(s, page_address(page)); | |
1587 | ||
1588 | /* First entry is used as the base of the freelist */ | |
1589 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1590 | freelist_count); | |
1591 | cur = setup_object(s, page, cur); | |
1592 | page->freelist = cur; | |
1593 | ||
1594 | for (idx = 1; idx < page->objects; idx++) { | |
1595 | next = next_freelist_entry(s, page, &pos, start, page_limit, | |
1596 | freelist_count); | |
1597 | next = setup_object(s, page, next); | |
1598 | set_freepointer(s, cur, next); | |
1599 | cur = next; | |
1600 | } | |
1601 | set_freepointer(s, cur, NULL); | |
1602 | ||
1603 | return true; | |
1604 | } | |
1605 | #else | |
1606 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1607 | { | |
1608 | return 0; | |
1609 | } | |
1610 | static inline void init_freelist_randomization(void) { } | |
1611 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1612 | { | |
1613 | return false; | |
1614 | } | |
1615 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1616 | ||
1617 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1618 | { | |
1619 | struct page *page; | |
1620 | struct kmem_cache_order_objects oo = s->oo; | |
1621 | gfp_t alloc_gfp; | |
1622 | void *start, *p, *next; | |
1623 | int idx; | |
1624 | bool shuffle; | |
1625 | ||
1626 | flags &= gfp_allowed_mask; | |
1627 | ||
1628 | if (gfpflags_allow_blocking(flags)) | |
1629 | local_irq_enable(); | |
1630 | ||
1631 | flags |= s->allocflags; | |
1632 | ||
1633 | /* | |
1634 | * Let the initial higher-order allocation fail under memory pressure | |
1635 | * so we fall-back to the minimum order allocation. | |
1636 | */ | |
1637 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1638 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) | |
1639 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); | |
1640 | ||
1641 | page = alloc_slab_page(s, alloc_gfp, node, oo); | |
1642 | if (unlikely(!page)) { | |
1643 | oo = s->min; | |
1644 | alloc_gfp = flags; | |
1645 | /* | |
1646 | * Allocation may have failed due to fragmentation. | |
1647 | * Try a lower order alloc if possible | |
1648 | */ | |
1649 | page = alloc_slab_page(s, alloc_gfp, node, oo); | |
1650 | if (unlikely(!page)) | |
1651 | goto out; | |
1652 | stat(s, ORDER_FALLBACK); | |
1653 | } | |
1654 | ||
1655 | page->objects = oo_objects(oo); | |
1656 | ||
1657 | page->slab_cache = s; | |
1658 | __SetPageSlab(page); | |
1659 | if (page_is_pfmemalloc(page)) | |
1660 | SetPageSlabPfmemalloc(page); | |
1661 | ||
1662 | kasan_poison_slab(page); | |
1663 | ||
1664 | start = page_address(page); | |
1665 | ||
1666 | setup_page_debug(s, page, start); | |
1667 | ||
1668 | shuffle = shuffle_freelist(s, page); | |
1669 | ||
1670 | if (!shuffle) { | |
1671 | start = fixup_red_left(s, start); | |
1672 | start = setup_object(s, page, start); | |
1673 | page->freelist = start; | |
1674 | for (idx = 0, p = start; idx < page->objects - 1; idx++) { | |
1675 | next = p + s->size; | |
1676 | next = setup_object(s, page, next); | |
1677 | set_freepointer(s, p, next); | |
1678 | p = next; | |
1679 | } | |
1680 | set_freepointer(s, p, NULL); | |
1681 | } | |
1682 | ||
1683 | page->inuse = page->objects; | |
1684 | page->frozen = 1; | |
1685 | ||
1686 | out: | |
1687 | if (gfpflags_allow_blocking(flags)) | |
1688 | local_irq_disable(); | |
1689 | if (!page) | |
1690 | return NULL; | |
1691 | ||
1692 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1693 | ||
1694 | return page; | |
1695 | } | |
1696 | ||
1697 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1698 | { | |
1699 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
1700 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; | |
1701 | flags &= ~GFP_SLAB_BUG_MASK; | |
1702 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
1703 | invalid_mask, &invalid_mask, flags, &flags); | |
1704 | dump_stack(); | |
1705 | } | |
1706 | ||
1707 | return allocate_slab(s, | |
1708 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1709 | } | |
1710 | ||
1711 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1712 | { | |
1713 | int order = compound_order(page); | |
1714 | int pages = 1 << order; | |
1715 | ||
1716 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1717 | void *p; | |
1718 | ||
1719 | slab_pad_check(s, page); | |
1720 | for_each_object(p, s, page_address(page), | |
1721 | page->objects) | |
1722 | check_object(s, page, p, SLUB_RED_INACTIVE); | |
1723 | } | |
1724 | ||
1725 | __ClearPageSlabPfmemalloc(page); | |
1726 | __ClearPageSlab(page); | |
1727 | ||
1728 | page->mapping = NULL; | |
1729 | if (current->reclaim_state) | |
1730 | current->reclaim_state->reclaimed_slab += pages; | |
1731 | uncharge_slab_page(page, order, s); | |
1732 | __free_pages(page, order); | |
1733 | } | |
1734 | ||
1735 | static void rcu_free_slab(struct rcu_head *h) | |
1736 | { | |
1737 | struct page *page = container_of(h, struct page, rcu_head); | |
1738 | ||
1739 | __free_slab(page->slab_cache, page); | |
1740 | } | |
1741 | ||
1742 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1743 | { | |
1744 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { | |
1745 | call_rcu(&page->rcu_head, rcu_free_slab); | |
1746 | } else | |
1747 | __free_slab(s, page); | |
1748 | } | |
1749 | ||
1750 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1751 | { | |
1752 | dec_slabs_node(s, page_to_nid(page), page->objects); | |
1753 | free_slab(s, page); | |
1754 | } | |
1755 | ||
1756 | /* | |
1757 | * Management of partially allocated slabs. | |
1758 | */ | |
1759 | static inline void | |
1760 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
1761 | { | |
1762 | n->nr_partial++; | |
1763 | if (tail == DEACTIVATE_TO_TAIL) | |
1764 | list_add_tail(&page->slab_list, &n->partial); | |
1765 | else | |
1766 | list_add(&page->slab_list, &n->partial); | |
1767 | } | |
1768 | ||
1769 | static inline void add_partial(struct kmem_cache_node *n, | |
1770 | struct page *page, int tail) | |
1771 | { | |
1772 | lockdep_assert_held(&n->list_lock); | |
1773 | __add_partial(n, page, tail); | |
1774 | } | |
1775 | ||
1776 | static inline void remove_partial(struct kmem_cache_node *n, | |
1777 | struct page *page) | |
1778 | { | |
1779 | lockdep_assert_held(&n->list_lock); | |
1780 | list_del(&page->slab_list); | |
1781 | n->nr_partial--; | |
1782 | } | |
1783 | ||
1784 | /* | |
1785 | * Remove slab from the partial list, freeze it and | |
1786 | * return the pointer to the freelist. | |
1787 | * | |
1788 | * Returns a list of objects or NULL if it fails. | |
1789 | */ | |
1790 | static inline void *acquire_slab(struct kmem_cache *s, | |
1791 | struct kmem_cache_node *n, struct page *page, | |
1792 | int mode, int *objects) | |
1793 | { | |
1794 | void *freelist; | |
1795 | unsigned long counters; | |
1796 | struct page new; | |
1797 | ||
1798 | lockdep_assert_held(&n->list_lock); | |
1799 | ||
1800 | /* | |
1801 | * Zap the freelist and set the frozen bit. | |
1802 | * The old freelist is the list of objects for the | |
1803 | * per cpu allocation list. | |
1804 | */ | |
1805 | freelist = page->freelist; | |
1806 | counters = page->counters; | |
1807 | new.counters = counters; | |
1808 | *objects = new.objects - new.inuse; | |
1809 | if (mode) { | |
1810 | new.inuse = page->objects; | |
1811 | new.freelist = NULL; | |
1812 | } else { | |
1813 | new.freelist = freelist; | |
1814 | } | |
1815 | ||
1816 | VM_BUG_ON(new.frozen); | |
1817 | new.frozen = 1; | |
1818 | ||
1819 | if (!__cmpxchg_double_slab(s, page, | |
1820 | freelist, counters, | |
1821 | new.freelist, new.counters, | |
1822 | "acquire_slab")) | |
1823 | return NULL; | |
1824 | ||
1825 | remove_partial(n, page); | |
1826 | WARN_ON(!freelist); | |
1827 | return freelist; | |
1828 | } | |
1829 | ||
1830 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); | |
1831 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); | |
1832 | ||
1833 | /* | |
1834 | * Try to allocate a partial slab from a specific node. | |
1835 | */ | |
1836 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, | |
1837 | struct kmem_cache_cpu *c, gfp_t flags) | |
1838 | { | |
1839 | struct page *page, *page2; | |
1840 | void *object = NULL; | |
1841 | unsigned int available = 0; | |
1842 | int objects; | |
1843 | ||
1844 | /* | |
1845 | * Racy check. If we mistakenly see no partial slabs then we | |
1846 | * just allocate an empty slab. If we mistakenly try to get a | |
1847 | * partial slab and there is none available then get_partials() | |
1848 | * will return NULL. | |
1849 | */ | |
1850 | if (!n || !n->nr_partial) | |
1851 | return NULL; | |
1852 | ||
1853 | spin_lock(&n->list_lock); | |
1854 | list_for_each_entry_safe(page, page2, &n->partial, slab_list) { | |
1855 | void *t; | |
1856 | ||
1857 | if (!pfmemalloc_match(page, flags)) | |
1858 | continue; | |
1859 | ||
1860 | t = acquire_slab(s, n, page, object == NULL, &objects); | |
1861 | if (!t) | |
1862 | break; | |
1863 | ||
1864 | available += objects; | |
1865 | if (!object) { | |
1866 | c->page = page; | |
1867 | stat(s, ALLOC_FROM_PARTIAL); | |
1868 | object = t; | |
1869 | } else { | |
1870 | put_cpu_partial(s, page, 0); | |
1871 | stat(s, CPU_PARTIAL_NODE); | |
1872 | } | |
1873 | if (!kmem_cache_has_cpu_partial(s) | |
1874 | || available > slub_cpu_partial(s) / 2) | |
1875 | break; | |
1876 | ||
1877 | } | |
1878 | spin_unlock(&n->list_lock); | |
1879 | return object; | |
1880 | } | |
1881 | ||
1882 | /* | |
1883 | * Get a page from somewhere. Search in increasing NUMA distances. | |
1884 | */ | |
1885 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, | |
1886 | struct kmem_cache_cpu *c) | |
1887 | { | |
1888 | #ifdef CONFIG_NUMA | |
1889 | struct zonelist *zonelist; | |
1890 | struct zoneref *z; | |
1891 | struct zone *zone; | |
1892 | enum zone_type high_zoneidx = gfp_zone(flags); | |
1893 | void *object; | |
1894 | unsigned int cpuset_mems_cookie; | |
1895 | ||
1896 | /* | |
1897 | * The defrag ratio allows a configuration of the tradeoffs between | |
1898 | * inter node defragmentation and node local allocations. A lower | |
1899 | * defrag_ratio increases the tendency to do local allocations | |
1900 | * instead of attempting to obtain partial slabs from other nodes. | |
1901 | * | |
1902 | * If the defrag_ratio is set to 0 then kmalloc() always | |
1903 | * returns node local objects. If the ratio is higher then kmalloc() | |
1904 | * may return off node objects because partial slabs are obtained | |
1905 | * from other nodes and filled up. | |
1906 | * | |
1907 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 | |
1908 | * (which makes defrag_ratio = 1000) then every (well almost) | |
1909 | * allocation will first attempt to defrag slab caches on other nodes. | |
1910 | * This means scanning over all nodes to look for partial slabs which | |
1911 | * may be expensive if we do it every time we are trying to find a slab | |
1912 | * with available objects. | |
1913 | */ | |
1914 | if (!s->remote_node_defrag_ratio || | |
1915 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
1916 | return NULL; | |
1917 | ||
1918 | do { | |
1919 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
1920 | zonelist = node_zonelist(mempolicy_slab_node(), flags); | |
1921 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { | |
1922 | struct kmem_cache_node *n; | |
1923 | ||
1924 | n = get_node(s, zone_to_nid(zone)); | |
1925 | ||
1926 | if (n && cpuset_zone_allowed(zone, flags) && | |
1927 | n->nr_partial > s->min_partial) { | |
1928 | object = get_partial_node(s, n, c, flags); | |
1929 | if (object) { | |
1930 | /* | |
1931 | * Don't check read_mems_allowed_retry() | |
1932 | * here - if mems_allowed was updated in | |
1933 | * parallel, that was a harmless race | |
1934 | * between allocation and the cpuset | |
1935 | * update | |
1936 | */ | |
1937 | return object; | |
1938 | } | |
1939 | } | |
1940 | } | |
1941 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); | |
1942 | #endif /* CONFIG_NUMA */ | |
1943 | return NULL; | |
1944 | } | |
1945 | ||
1946 | /* | |
1947 | * Get a partial page, lock it and return it. | |
1948 | */ | |
1949 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, | |
1950 | struct kmem_cache_cpu *c) | |
1951 | { | |
1952 | void *object; | |
1953 | int searchnode = node; | |
1954 | ||
1955 | if (node == NUMA_NO_NODE) | |
1956 | searchnode = numa_mem_id(); | |
1957 | else if (!node_present_pages(node)) | |
1958 | searchnode = node_to_mem_node(node); | |
1959 | ||
1960 | object = get_partial_node(s, get_node(s, searchnode), c, flags); | |
1961 | if (object || node != NUMA_NO_NODE) | |
1962 | return object; | |
1963 | ||
1964 | return get_any_partial(s, flags, c); | |
1965 | } | |
1966 | ||
1967 | #ifdef CONFIG_PREEMPT | |
1968 | /* | |
1969 | * Calculate the next globally unique transaction for disambiguiation | |
1970 | * during cmpxchg. The transactions start with the cpu number and are then | |
1971 | * incremented by CONFIG_NR_CPUS. | |
1972 | */ | |
1973 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1974 | #else | |
1975 | /* | |
1976 | * No preemption supported therefore also no need to check for | |
1977 | * different cpus. | |
1978 | */ | |
1979 | #define TID_STEP 1 | |
1980 | #endif | |
1981 | ||
1982 | static inline unsigned long next_tid(unsigned long tid) | |
1983 | { | |
1984 | return tid + TID_STEP; | |
1985 | } | |
1986 | ||
1987 | #ifdef SLUB_DEBUG_CMPXCHG | |
1988 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1989 | { | |
1990 | return tid % TID_STEP; | |
1991 | } | |
1992 | ||
1993 | static inline unsigned long tid_to_event(unsigned long tid) | |
1994 | { | |
1995 | return tid / TID_STEP; | |
1996 | } | |
1997 | #endif | |
1998 | ||
1999 | static inline unsigned int init_tid(int cpu) | |
2000 | { | |
2001 | return cpu; | |
2002 | } | |
2003 | ||
2004 | static inline void note_cmpxchg_failure(const char *n, | |
2005 | const struct kmem_cache *s, unsigned long tid) | |
2006 | { | |
2007 | #ifdef SLUB_DEBUG_CMPXCHG | |
2008 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2009 | ||
2010 | pr_info("%s %s: cmpxchg redo ", n, s->name); | |
2011 | ||
2012 | #ifdef CONFIG_PREEMPT | |
2013 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
2014 | pr_warn("due to cpu change %d -> %d\n", | |
2015 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
2016 | else | |
2017 | #endif | |
2018 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
2019 | pr_warn("due to cpu running other code. Event %ld->%ld\n", | |
2020 | tid_to_event(tid), tid_to_event(actual_tid)); | |
2021 | else | |
2022 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
2023 | actual_tid, tid, next_tid(tid)); | |
2024 | #endif | |
2025 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); | |
2026 | } | |
2027 | ||
2028 | static void init_kmem_cache_cpus(struct kmem_cache *s) | |
2029 | { | |
2030 | int cpu; | |
2031 | ||
2032 | for_each_possible_cpu(cpu) | |
2033 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
2034 | } | |
2035 | ||
2036 | /* | |
2037 | * Remove the cpu slab | |
2038 | */ | |
2039 | static void deactivate_slab(struct kmem_cache *s, struct page *page, | |
2040 | void *freelist, struct kmem_cache_cpu *c) | |
2041 | { | |
2042 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; | |
2043 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
2044 | int lock = 0; | |
2045 | enum slab_modes l = M_NONE, m = M_NONE; | |
2046 | void *nextfree; | |
2047 | int tail = DEACTIVATE_TO_HEAD; | |
2048 | struct page new; | |
2049 | struct page old; | |
2050 | ||
2051 | if (page->freelist) { | |
2052 | stat(s, DEACTIVATE_REMOTE_FREES); | |
2053 | tail = DEACTIVATE_TO_TAIL; | |
2054 | } | |
2055 | ||
2056 | /* | |
2057 | * Stage one: Free all available per cpu objects back | |
2058 | * to the page freelist while it is still frozen. Leave the | |
2059 | * last one. | |
2060 | * | |
2061 | * There is no need to take the list->lock because the page | |
2062 | * is still frozen. | |
2063 | */ | |
2064 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2065 | void *prior; | |
2066 | unsigned long counters; | |
2067 | ||
2068 | do { | |
2069 | prior = page->freelist; | |
2070 | counters = page->counters; | |
2071 | set_freepointer(s, freelist, prior); | |
2072 | new.counters = counters; | |
2073 | new.inuse--; | |
2074 | VM_BUG_ON(!new.frozen); | |
2075 | ||
2076 | } while (!__cmpxchg_double_slab(s, page, | |
2077 | prior, counters, | |
2078 | freelist, new.counters, | |
2079 | "drain percpu freelist")); | |
2080 | ||
2081 | freelist = nextfree; | |
2082 | } | |
2083 | ||
2084 | /* | |
2085 | * Stage two: Ensure that the page is unfrozen while the | |
2086 | * list presence reflects the actual number of objects | |
2087 | * during unfreeze. | |
2088 | * | |
2089 | * We setup the list membership and then perform a cmpxchg | |
2090 | * with the count. If there is a mismatch then the page | |
2091 | * is not unfrozen but the page is on the wrong list. | |
2092 | * | |
2093 | * Then we restart the process which may have to remove | |
2094 | * the page from the list that we just put it on again | |
2095 | * because the number of objects in the slab may have | |
2096 | * changed. | |
2097 | */ | |
2098 | redo: | |
2099 | ||
2100 | old.freelist = page->freelist; | |
2101 | old.counters = page->counters; | |
2102 | VM_BUG_ON(!old.frozen); | |
2103 | ||
2104 | /* Determine target state of the slab */ | |
2105 | new.counters = old.counters; | |
2106 | if (freelist) { | |
2107 | new.inuse--; | |
2108 | set_freepointer(s, freelist, old.freelist); | |
2109 | new.freelist = freelist; | |
2110 | } else | |
2111 | new.freelist = old.freelist; | |
2112 | ||
2113 | new.frozen = 0; | |
2114 | ||
2115 | if (!new.inuse && n->nr_partial >= s->min_partial) | |
2116 | m = M_FREE; | |
2117 | else if (new.freelist) { | |
2118 | m = M_PARTIAL; | |
2119 | if (!lock) { | |
2120 | lock = 1; | |
2121 | /* | |
2122 | * Taking the spinlock removes the possibility | |
2123 | * that acquire_slab() will see a slab page that | |
2124 | * is frozen | |
2125 | */ | |
2126 | spin_lock(&n->list_lock); | |
2127 | } | |
2128 | } else { | |
2129 | m = M_FULL; | |
2130 | if (kmem_cache_debug(s) && !lock) { | |
2131 | lock = 1; | |
2132 | /* | |
2133 | * This also ensures that the scanning of full | |
2134 | * slabs from diagnostic functions will not see | |
2135 | * any frozen slabs. | |
2136 | */ | |
2137 | spin_lock(&n->list_lock); | |
2138 | } | |
2139 | } | |
2140 | ||
2141 | if (l != m) { | |
2142 | if (l == M_PARTIAL) | |
2143 | remove_partial(n, page); | |
2144 | else if (l == M_FULL) | |
2145 | remove_full(s, n, page); | |
2146 | ||
2147 | if (m == M_PARTIAL) | |
2148 | add_partial(n, page, tail); | |
2149 | else if (m == M_FULL) | |
2150 | add_full(s, n, page); | |
2151 | } | |
2152 | ||
2153 | l = m; | |
2154 | if (!__cmpxchg_double_slab(s, page, | |
2155 | old.freelist, old.counters, | |
2156 | new.freelist, new.counters, | |
2157 | "unfreezing slab")) | |
2158 | goto redo; | |
2159 | ||
2160 | if (lock) | |
2161 | spin_unlock(&n->list_lock); | |
2162 | ||
2163 | if (m == M_PARTIAL) | |
2164 | stat(s, tail); | |
2165 | else if (m == M_FULL) | |
2166 | stat(s, DEACTIVATE_FULL); | |
2167 | else if (m == M_FREE) { | |
2168 | stat(s, DEACTIVATE_EMPTY); | |
2169 | discard_slab(s, page); | |
2170 | stat(s, FREE_SLAB); | |
2171 | } | |
2172 | ||
2173 | c->page = NULL; | |
2174 | c->freelist = NULL; | |
2175 | } | |
2176 | ||
2177 | /* | |
2178 | * Unfreeze all the cpu partial slabs. | |
2179 | * | |
2180 | * This function must be called with interrupts disabled | |
2181 | * for the cpu using c (or some other guarantee must be there | |
2182 | * to guarantee no concurrent accesses). | |
2183 | */ | |
2184 | static void unfreeze_partials(struct kmem_cache *s, | |
2185 | struct kmem_cache_cpu *c) | |
2186 | { | |
2187 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
2188 | struct kmem_cache_node *n = NULL, *n2 = NULL; | |
2189 | struct page *page, *discard_page = NULL; | |
2190 | ||
2191 | while ((page = c->partial)) { | |
2192 | struct page new; | |
2193 | struct page old; | |
2194 | ||
2195 | c->partial = page->next; | |
2196 | ||
2197 | n2 = get_node(s, page_to_nid(page)); | |
2198 | if (n != n2) { | |
2199 | if (n) | |
2200 | spin_unlock(&n->list_lock); | |
2201 | ||
2202 | n = n2; | |
2203 | spin_lock(&n->list_lock); | |
2204 | } | |
2205 | ||
2206 | do { | |
2207 | ||
2208 | old.freelist = page->freelist; | |
2209 | old.counters = page->counters; | |
2210 | VM_BUG_ON(!old.frozen); | |
2211 | ||
2212 | new.counters = old.counters; | |
2213 | new.freelist = old.freelist; | |
2214 | ||
2215 | new.frozen = 0; | |
2216 | ||
2217 | } while (!__cmpxchg_double_slab(s, page, | |
2218 | old.freelist, old.counters, | |
2219 | new.freelist, new.counters, | |
2220 | "unfreezing slab")); | |
2221 | ||
2222 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { | |
2223 | page->next = discard_page; | |
2224 | discard_page = page; | |
2225 | } else { | |
2226 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2227 | stat(s, FREE_ADD_PARTIAL); | |
2228 | } | |
2229 | } | |
2230 | ||
2231 | if (n) | |
2232 | spin_unlock(&n->list_lock); | |
2233 | ||
2234 | while (discard_page) { | |
2235 | page = discard_page; | |
2236 | discard_page = discard_page->next; | |
2237 | ||
2238 | stat(s, DEACTIVATE_EMPTY); | |
2239 | discard_slab(s, page); | |
2240 | stat(s, FREE_SLAB); | |
2241 | } | |
2242 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ | |
2243 | } | |
2244 | ||
2245 | /* | |
2246 | * Put a page that was just frozen (in __slab_free|get_partial_node) into a | |
2247 | * partial page slot if available. | |
2248 | * | |
2249 | * If we did not find a slot then simply move all the partials to the | |
2250 | * per node partial list. | |
2251 | */ | |
2252 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) | |
2253 | { | |
2254 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
2255 | struct page *oldpage; | |
2256 | int pages; | |
2257 | int pobjects; | |
2258 | ||
2259 | preempt_disable(); | |
2260 | do { | |
2261 | pages = 0; | |
2262 | pobjects = 0; | |
2263 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2264 | ||
2265 | if (oldpage) { | |
2266 | pobjects = oldpage->pobjects; | |
2267 | pages = oldpage->pages; | |
2268 | if (drain && pobjects > s->cpu_partial) { | |
2269 | unsigned long flags; | |
2270 | /* | |
2271 | * partial array is full. Move the existing | |
2272 | * set to the per node partial list. | |
2273 | */ | |
2274 | local_irq_save(flags); | |
2275 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2276 | local_irq_restore(flags); | |
2277 | oldpage = NULL; | |
2278 | pobjects = 0; | |
2279 | pages = 0; | |
2280 | stat(s, CPU_PARTIAL_DRAIN); | |
2281 | } | |
2282 | } | |
2283 | ||
2284 | pages++; | |
2285 | pobjects += page->objects - page->inuse; | |
2286 | ||
2287 | page->pages = pages; | |
2288 | page->pobjects = pobjects; | |
2289 | page->next = oldpage; | |
2290 | ||
2291 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) | |
2292 | != oldpage); | |
2293 | if (unlikely(!s->cpu_partial)) { | |
2294 | unsigned long flags; | |
2295 | ||
2296 | local_irq_save(flags); | |
2297 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2298 | local_irq_restore(flags); | |
2299 | } | |
2300 | preempt_enable(); | |
2301 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ | |
2302 | } | |
2303 | ||
2304 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) | |
2305 | { | |
2306 | stat(s, CPUSLAB_FLUSH); | |
2307 | deactivate_slab(s, c->page, c->freelist, c); | |
2308 | ||
2309 | c->tid = next_tid(c->tid); | |
2310 | } | |
2311 | ||
2312 | /* | |
2313 | * Flush cpu slab. | |
2314 | * | |
2315 | * Called from IPI handler with interrupts disabled. | |
2316 | */ | |
2317 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) | |
2318 | { | |
2319 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2320 | ||
2321 | if (c->page) | |
2322 | flush_slab(s, c); | |
2323 | ||
2324 | unfreeze_partials(s, c); | |
2325 | } | |
2326 | ||
2327 | static void flush_cpu_slab(void *d) | |
2328 | { | |
2329 | struct kmem_cache *s = d; | |
2330 | ||
2331 | __flush_cpu_slab(s, smp_processor_id()); | |
2332 | } | |
2333 | ||
2334 | static bool has_cpu_slab(int cpu, void *info) | |
2335 | { | |
2336 | struct kmem_cache *s = info; | |
2337 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2338 | ||
2339 | return c->page || slub_percpu_partial(c); | |
2340 | } | |
2341 | ||
2342 | static void flush_all(struct kmem_cache *s) | |
2343 | { | |
2344 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); | |
2345 | } | |
2346 | ||
2347 | /* | |
2348 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2349 | * necessary. | |
2350 | */ | |
2351 | static int slub_cpu_dead(unsigned int cpu) | |
2352 | { | |
2353 | struct kmem_cache *s; | |
2354 | unsigned long flags; | |
2355 | ||
2356 | mutex_lock(&slab_mutex); | |
2357 | list_for_each_entry(s, &slab_caches, list) { | |
2358 | local_irq_save(flags); | |
2359 | __flush_cpu_slab(s, cpu); | |
2360 | local_irq_restore(flags); | |
2361 | } | |
2362 | mutex_unlock(&slab_mutex); | |
2363 | return 0; | |
2364 | } | |
2365 | ||
2366 | /* | |
2367 | * Check if the objects in a per cpu structure fit numa | |
2368 | * locality expectations. | |
2369 | */ | |
2370 | static inline int node_match(struct page *page, int node) | |
2371 | { | |
2372 | #ifdef CONFIG_NUMA | |
2373 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) | |
2374 | return 0; | |
2375 | #endif | |
2376 | return 1; | |
2377 | } | |
2378 | ||
2379 | #ifdef CONFIG_SLUB_DEBUG | |
2380 | static int count_free(struct page *page) | |
2381 | { | |
2382 | return page->objects - page->inuse; | |
2383 | } | |
2384 | ||
2385 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) | |
2386 | { | |
2387 | return atomic_long_read(&n->total_objects); | |
2388 | } | |
2389 | #endif /* CONFIG_SLUB_DEBUG */ | |
2390 | ||
2391 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
2392 | static unsigned long count_partial(struct kmem_cache_node *n, | |
2393 | int (*get_count)(struct page *)) | |
2394 | { | |
2395 | unsigned long flags; | |
2396 | unsigned long x = 0; | |
2397 | struct page *page; | |
2398 | ||
2399 | spin_lock_irqsave(&n->list_lock, flags); | |
2400 | list_for_each_entry(page, &n->partial, slab_list) | |
2401 | x += get_count(page); | |
2402 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2403 | return x; | |
2404 | } | |
2405 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ | |
2406 | ||
2407 | static noinline void | |
2408 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2409 | { | |
2410 | #ifdef CONFIG_SLUB_DEBUG | |
2411 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2412 | DEFAULT_RATELIMIT_BURST); | |
2413 | int node; | |
2414 | struct kmem_cache_node *n; | |
2415 | ||
2416 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) | |
2417 | return; | |
2418 | ||
2419 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", | |
2420 | nid, gfpflags, &gfpflags); | |
2421 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", | |
2422 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2423 | oo_order(s->min)); | |
2424 | ||
2425 | if (oo_order(s->min) > get_order(s->object_size)) | |
2426 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", | |
2427 | s->name); | |
2428 | ||
2429 | for_each_kmem_cache_node(s, node, n) { | |
2430 | unsigned long nr_slabs; | |
2431 | unsigned long nr_objs; | |
2432 | unsigned long nr_free; | |
2433 | ||
2434 | nr_free = count_partial(n, count_free); | |
2435 | nr_slabs = node_nr_slabs(n); | |
2436 | nr_objs = node_nr_objs(n); | |
2437 | ||
2438 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
2439 | node, nr_slabs, nr_objs, nr_free); | |
2440 | } | |
2441 | #endif | |
2442 | } | |
2443 | ||
2444 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, | |
2445 | int node, struct kmem_cache_cpu **pc) | |
2446 | { | |
2447 | void *freelist; | |
2448 | struct kmem_cache_cpu *c = *pc; | |
2449 | struct page *page; | |
2450 | ||
2451 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); | |
2452 | ||
2453 | freelist = get_partial(s, flags, node, c); | |
2454 | ||
2455 | if (freelist) | |
2456 | return freelist; | |
2457 | ||
2458 | page = new_slab(s, flags, node); | |
2459 | if (page) { | |
2460 | c = raw_cpu_ptr(s->cpu_slab); | |
2461 | if (c->page) | |
2462 | flush_slab(s, c); | |
2463 | ||
2464 | /* | |
2465 | * No other reference to the page yet so we can | |
2466 | * muck around with it freely without cmpxchg | |
2467 | */ | |
2468 | freelist = page->freelist; | |
2469 | page->freelist = NULL; | |
2470 | ||
2471 | stat(s, ALLOC_SLAB); | |
2472 | c->page = page; | |
2473 | *pc = c; | |
2474 | } | |
2475 | ||
2476 | return freelist; | |
2477 | } | |
2478 | ||
2479 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) | |
2480 | { | |
2481 | if (unlikely(PageSlabPfmemalloc(page))) | |
2482 | return gfp_pfmemalloc_allowed(gfpflags); | |
2483 | ||
2484 | return true; | |
2485 | } | |
2486 | ||
2487 | /* | |
2488 | * Check the page->freelist of a page and either transfer the freelist to the | |
2489 | * per cpu freelist or deactivate the page. | |
2490 | * | |
2491 | * The page is still frozen if the return value is not NULL. | |
2492 | * | |
2493 | * If this function returns NULL then the page has been unfrozen. | |
2494 | * | |
2495 | * This function must be called with interrupt disabled. | |
2496 | */ | |
2497 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2498 | { | |
2499 | struct page new; | |
2500 | unsigned long counters; | |
2501 | void *freelist; | |
2502 | ||
2503 | do { | |
2504 | freelist = page->freelist; | |
2505 | counters = page->counters; | |
2506 | ||
2507 | new.counters = counters; | |
2508 | VM_BUG_ON(!new.frozen); | |
2509 | ||
2510 | new.inuse = page->objects; | |
2511 | new.frozen = freelist != NULL; | |
2512 | ||
2513 | } while (!__cmpxchg_double_slab(s, page, | |
2514 | freelist, counters, | |
2515 | NULL, new.counters, | |
2516 | "get_freelist")); | |
2517 | ||
2518 | return freelist; | |
2519 | } | |
2520 | ||
2521 | /* | |
2522 | * Slow path. The lockless freelist is empty or we need to perform | |
2523 | * debugging duties. | |
2524 | * | |
2525 | * Processing is still very fast if new objects have been freed to the | |
2526 | * regular freelist. In that case we simply take over the regular freelist | |
2527 | * as the lockless freelist and zap the regular freelist. | |
2528 | * | |
2529 | * If that is not working then we fall back to the partial lists. We take the | |
2530 | * first element of the freelist as the object to allocate now and move the | |
2531 | * rest of the freelist to the lockless freelist. | |
2532 | * | |
2533 | * And if we were unable to get a new slab from the partial slab lists then | |
2534 | * we need to allocate a new slab. This is the slowest path since it involves | |
2535 | * a call to the page allocator and the setup of a new slab. | |
2536 | * | |
2537 | * Version of __slab_alloc to use when we know that interrupts are | |
2538 | * already disabled (which is the case for bulk allocation). | |
2539 | */ | |
2540 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2541 | unsigned long addr, struct kmem_cache_cpu *c) | |
2542 | { | |
2543 | void *freelist; | |
2544 | struct page *page; | |
2545 | ||
2546 | page = c->page; | |
2547 | if (!page) | |
2548 | goto new_slab; | |
2549 | redo: | |
2550 | ||
2551 | if (unlikely(!node_match(page, node))) { | |
2552 | int searchnode = node; | |
2553 | ||
2554 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2555 | searchnode = node_to_mem_node(node); | |
2556 | ||
2557 | if (unlikely(!node_match(page, searchnode))) { | |
2558 | stat(s, ALLOC_NODE_MISMATCH); | |
2559 | deactivate_slab(s, page, c->freelist, c); | |
2560 | goto new_slab; | |
2561 | } | |
2562 | } | |
2563 | ||
2564 | /* | |
2565 | * By rights, we should be searching for a slab page that was | |
2566 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2567 | * information when the page leaves the per-cpu allocator | |
2568 | */ | |
2569 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2570 | deactivate_slab(s, page, c->freelist, c); | |
2571 | goto new_slab; | |
2572 | } | |
2573 | ||
2574 | /* must check again c->freelist in case of cpu migration or IRQ */ | |
2575 | freelist = c->freelist; | |
2576 | if (freelist) | |
2577 | goto load_freelist; | |
2578 | ||
2579 | freelist = get_freelist(s, page); | |
2580 | ||
2581 | if (!freelist) { | |
2582 | c->page = NULL; | |
2583 | stat(s, DEACTIVATE_BYPASS); | |
2584 | goto new_slab; | |
2585 | } | |
2586 | ||
2587 | stat(s, ALLOC_REFILL); | |
2588 | ||
2589 | load_freelist: | |
2590 | /* | |
2591 | * freelist is pointing to the list of objects to be used. | |
2592 | * page is pointing to the page from which the objects are obtained. | |
2593 | * That page must be frozen for per cpu allocations to work. | |
2594 | */ | |
2595 | VM_BUG_ON(!c->page->frozen); | |
2596 | c->freelist = get_freepointer(s, freelist); | |
2597 | c->tid = next_tid(c->tid); | |
2598 | return freelist; | |
2599 | ||
2600 | new_slab: | |
2601 | ||
2602 | if (slub_percpu_partial(c)) { | |
2603 | page = c->page = slub_percpu_partial(c); | |
2604 | slub_set_percpu_partial(c, page); | |
2605 | stat(s, CPU_PARTIAL_ALLOC); | |
2606 | goto redo; | |
2607 | } | |
2608 | ||
2609 | freelist = new_slab_objects(s, gfpflags, node, &c); | |
2610 | ||
2611 | if (unlikely(!freelist)) { | |
2612 | slab_out_of_memory(s, gfpflags, node); | |
2613 | return NULL; | |
2614 | } | |
2615 | ||
2616 | page = c->page; | |
2617 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) | |
2618 | goto load_freelist; | |
2619 | ||
2620 | /* Only entered in the debug case */ | |
2621 | if (kmem_cache_debug(s) && | |
2622 | !alloc_debug_processing(s, page, freelist, addr)) | |
2623 | goto new_slab; /* Slab failed checks. Next slab needed */ | |
2624 | ||
2625 | deactivate_slab(s, page, get_freepointer(s, freelist), c); | |
2626 | return freelist; | |
2627 | } | |
2628 | ||
2629 | /* | |
2630 | * Another one that disabled interrupt and compensates for possible | |
2631 | * cpu changes by refetching the per cpu area pointer. | |
2632 | */ | |
2633 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2634 | unsigned long addr, struct kmem_cache_cpu *c) | |
2635 | { | |
2636 | void *p; | |
2637 | unsigned long flags; | |
2638 | ||
2639 | local_irq_save(flags); | |
2640 | #ifdef CONFIG_PREEMPT | |
2641 | /* | |
2642 | * We may have been preempted and rescheduled on a different | |
2643 | * cpu before disabling interrupts. Need to reload cpu area | |
2644 | * pointer. | |
2645 | */ | |
2646 | c = this_cpu_ptr(s->cpu_slab); | |
2647 | #endif | |
2648 | ||
2649 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2650 | local_irq_restore(flags); | |
2651 | return p; | |
2652 | } | |
2653 | ||
2654 | /* | |
2655 | * If the object has been wiped upon free, make sure it's fully initialized by | |
2656 | * zeroing out freelist pointer. | |
2657 | */ | |
2658 | static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s, | |
2659 | void *obj) | |
2660 | { | |
2661 | if (unlikely(slab_want_init_on_free(s)) && obj) | |
2662 | memset((void *)((char *)obj + s->offset), 0, sizeof(void *)); | |
2663 | } | |
2664 | ||
2665 | /* | |
2666 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2667 | * have the fastpath folded into their functions. So no function call | |
2668 | * overhead for requests that can be satisfied on the fastpath. | |
2669 | * | |
2670 | * The fastpath works by first checking if the lockless freelist can be used. | |
2671 | * If not then __slab_alloc is called for slow processing. | |
2672 | * | |
2673 | * Otherwise we can simply pick the next object from the lockless free list. | |
2674 | */ | |
2675 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, | |
2676 | gfp_t gfpflags, int node, unsigned long addr) | |
2677 | { | |
2678 | void *object; | |
2679 | struct kmem_cache_cpu *c; | |
2680 | struct page *page; | |
2681 | unsigned long tid; | |
2682 | ||
2683 | s = slab_pre_alloc_hook(s, gfpflags); | |
2684 | if (!s) | |
2685 | return NULL; | |
2686 | redo: | |
2687 | /* | |
2688 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2689 | * enabled. We may switch back and forth between cpus while | |
2690 | * reading from one cpu area. That does not matter as long | |
2691 | * as we end up on the original cpu again when doing the cmpxchg. | |
2692 | * | |
2693 | * We should guarantee that tid and kmem_cache are retrieved on | |
2694 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2695 | * to check if it is matched or not. | |
2696 | */ | |
2697 | do { | |
2698 | tid = this_cpu_read(s->cpu_slab->tid); | |
2699 | c = raw_cpu_ptr(s->cpu_slab); | |
2700 | } while (IS_ENABLED(CONFIG_PREEMPT) && | |
2701 | unlikely(tid != READ_ONCE(c->tid))); | |
2702 | ||
2703 | /* | |
2704 | * Irqless object alloc/free algorithm used here depends on sequence | |
2705 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2706 | * on c to guarantee that object and page associated with previous tid | |
2707 | * won't be used with current tid. If we fetch tid first, object and | |
2708 | * page could be one associated with next tid and our alloc/free | |
2709 | * request will be failed. In this case, we will retry. So, no problem. | |
2710 | */ | |
2711 | barrier(); | |
2712 | ||
2713 | /* | |
2714 | * The transaction ids are globally unique per cpu and per operation on | |
2715 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2716 | * occurs on the right processor and that there was no operation on the | |
2717 | * linked list in between. | |
2718 | */ | |
2719 | ||
2720 | object = c->freelist; | |
2721 | page = c->page; | |
2722 | if (unlikely(!object || !node_match(page, node))) { | |
2723 | object = __slab_alloc(s, gfpflags, node, addr, c); | |
2724 | stat(s, ALLOC_SLOWPATH); | |
2725 | } else { | |
2726 | void *next_object = get_freepointer_safe(s, object); | |
2727 | ||
2728 | /* | |
2729 | * The cmpxchg will only match if there was no additional | |
2730 | * operation and if we are on the right processor. | |
2731 | * | |
2732 | * The cmpxchg does the following atomically (without lock | |
2733 | * semantics!) | |
2734 | * 1. Relocate first pointer to the current per cpu area. | |
2735 | * 2. Verify that tid and freelist have not been changed | |
2736 | * 3. If they were not changed replace tid and freelist | |
2737 | * | |
2738 | * Since this is without lock semantics the protection is only | |
2739 | * against code executing on this cpu *not* from access by | |
2740 | * other cpus. | |
2741 | */ | |
2742 | if (unlikely(!this_cpu_cmpxchg_double( | |
2743 | s->cpu_slab->freelist, s->cpu_slab->tid, | |
2744 | object, tid, | |
2745 | next_object, next_tid(tid)))) { | |
2746 | ||
2747 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2748 | goto redo; | |
2749 | } | |
2750 | prefetch_freepointer(s, next_object); | |
2751 | stat(s, ALLOC_FASTPATH); | |
2752 | } | |
2753 | ||
2754 | maybe_wipe_obj_freeptr(s, object); | |
2755 | ||
2756 | if (unlikely(slab_want_init_on_alloc(gfpflags, s)) && object) | |
2757 | memset(object, 0, s->object_size); | |
2758 | ||
2759 | slab_post_alloc_hook(s, gfpflags, 1, &object); | |
2760 | ||
2761 | return object; | |
2762 | } | |
2763 | ||
2764 | static __always_inline void *slab_alloc(struct kmem_cache *s, | |
2765 | gfp_t gfpflags, unsigned long addr) | |
2766 | { | |
2767 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2768 | } | |
2769 | ||
2770 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
2771 | { | |
2772 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); | |
2773 | ||
2774 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, | |
2775 | s->size, gfpflags); | |
2776 | ||
2777 | return ret; | |
2778 | } | |
2779 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2780 | ||
2781 | #ifdef CONFIG_TRACING | |
2782 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) | |
2783 | { | |
2784 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); | |
2785 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
2786 | ret = kasan_kmalloc(s, ret, size, gfpflags); | |
2787 | return ret; | |
2788 | } | |
2789 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
2790 | #endif | |
2791 | ||
2792 | #ifdef CONFIG_NUMA | |
2793 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2794 | { | |
2795 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); | |
2796 | ||
2797 | trace_kmem_cache_alloc_node(_RET_IP_, ret, | |
2798 | s->object_size, s->size, gfpflags, node); | |
2799 | ||
2800 | return ret; | |
2801 | } | |
2802 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
2803 | ||
2804 | #ifdef CONFIG_TRACING | |
2805 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, | |
2806 | gfp_t gfpflags, | |
2807 | int node, size_t size) | |
2808 | { | |
2809 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); | |
2810 | ||
2811 | trace_kmalloc_node(_RET_IP_, ret, | |
2812 | size, s->size, gfpflags, node); | |
2813 | ||
2814 | ret = kasan_kmalloc(s, ret, size, gfpflags); | |
2815 | return ret; | |
2816 | } | |
2817 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); | |
2818 | #endif | |
2819 | #endif /* CONFIG_NUMA */ | |
2820 | ||
2821 | /* | |
2822 | * Slow path handling. This may still be called frequently since objects | |
2823 | * have a longer lifetime than the cpu slabs in most processing loads. | |
2824 | * | |
2825 | * So we still attempt to reduce cache line usage. Just take the slab | |
2826 | * lock and free the item. If there is no additional partial page | |
2827 | * handling required then we can return immediately. | |
2828 | */ | |
2829 | static void __slab_free(struct kmem_cache *s, struct page *page, | |
2830 | void *head, void *tail, int cnt, | |
2831 | unsigned long addr) | |
2832 | ||
2833 | { | |
2834 | void *prior; | |
2835 | int was_frozen; | |
2836 | struct page new; | |
2837 | unsigned long counters; | |
2838 | struct kmem_cache_node *n = NULL; | |
2839 | unsigned long uninitialized_var(flags); | |
2840 | ||
2841 | stat(s, FREE_SLOWPATH); | |
2842 | ||
2843 | if (kmem_cache_debug(s) && | |
2844 | !free_debug_processing(s, page, head, tail, cnt, addr)) | |
2845 | return; | |
2846 | ||
2847 | do { | |
2848 | if (unlikely(n)) { | |
2849 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2850 | n = NULL; | |
2851 | } | |
2852 | prior = page->freelist; | |
2853 | counters = page->counters; | |
2854 | set_freepointer(s, tail, prior); | |
2855 | new.counters = counters; | |
2856 | was_frozen = new.frozen; | |
2857 | new.inuse -= cnt; | |
2858 | if ((!new.inuse || !prior) && !was_frozen) { | |
2859 | ||
2860 | if (kmem_cache_has_cpu_partial(s) && !prior) { | |
2861 | ||
2862 | /* | |
2863 | * Slab was on no list before and will be | |
2864 | * partially empty | |
2865 | * We can defer the list move and instead | |
2866 | * freeze it. | |
2867 | */ | |
2868 | new.frozen = 1; | |
2869 | ||
2870 | } else { /* Needs to be taken off a list */ | |
2871 | ||
2872 | n = get_node(s, page_to_nid(page)); | |
2873 | /* | |
2874 | * Speculatively acquire the list_lock. | |
2875 | * If the cmpxchg does not succeed then we may | |
2876 | * drop the list_lock without any processing. | |
2877 | * | |
2878 | * Otherwise the list_lock will synchronize with | |
2879 | * other processors updating the list of slabs. | |
2880 | */ | |
2881 | spin_lock_irqsave(&n->list_lock, flags); | |
2882 | ||
2883 | } | |
2884 | } | |
2885 | ||
2886 | } while (!cmpxchg_double_slab(s, page, | |
2887 | prior, counters, | |
2888 | head, new.counters, | |
2889 | "__slab_free")); | |
2890 | ||
2891 | if (likely(!n)) { | |
2892 | ||
2893 | /* | |
2894 | * If we just froze the page then put it onto the | |
2895 | * per cpu partial list. | |
2896 | */ | |
2897 | if (new.frozen && !was_frozen) { | |
2898 | put_cpu_partial(s, page, 1); | |
2899 | stat(s, CPU_PARTIAL_FREE); | |
2900 | } | |
2901 | /* | |
2902 | * The list lock was not taken therefore no list | |
2903 | * activity can be necessary. | |
2904 | */ | |
2905 | if (was_frozen) | |
2906 | stat(s, FREE_FROZEN); | |
2907 | return; | |
2908 | } | |
2909 | ||
2910 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) | |
2911 | goto slab_empty; | |
2912 | ||
2913 | /* | |
2914 | * Objects left in the slab. If it was not on the partial list before | |
2915 | * then add it. | |
2916 | */ | |
2917 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { | |
2918 | remove_full(s, n, page); | |
2919 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2920 | stat(s, FREE_ADD_PARTIAL); | |
2921 | } | |
2922 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2923 | return; | |
2924 | ||
2925 | slab_empty: | |
2926 | if (prior) { | |
2927 | /* | |
2928 | * Slab on the partial list. | |
2929 | */ | |
2930 | remove_partial(n, page); | |
2931 | stat(s, FREE_REMOVE_PARTIAL); | |
2932 | } else { | |
2933 | /* Slab must be on the full list */ | |
2934 | remove_full(s, n, page); | |
2935 | } | |
2936 | ||
2937 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2938 | stat(s, FREE_SLAB); | |
2939 | discard_slab(s, page); | |
2940 | } | |
2941 | ||
2942 | /* | |
2943 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2944 | * can perform fastpath freeing without additional function calls. | |
2945 | * | |
2946 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2947 | * of this processor. This typically the case if we have just allocated | |
2948 | * the item before. | |
2949 | * | |
2950 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2951 | * with all sorts of special processing. | |
2952 | * | |
2953 | * Bulk free of a freelist with several objects (all pointing to the | |
2954 | * same page) possible by specifying head and tail ptr, plus objects | |
2955 | * count (cnt). Bulk free indicated by tail pointer being set. | |
2956 | */ | |
2957 | static __always_inline void do_slab_free(struct kmem_cache *s, | |
2958 | struct page *page, void *head, void *tail, | |
2959 | int cnt, unsigned long addr) | |
2960 | { | |
2961 | void *tail_obj = tail ? : head; | |
2962 | struct kmem_cache_cpu *c; | |
2963 | unsigned long tid; | |
2964 | redo: | |
2965 | /* | |
2966 | * Determine the currently cpus per cpu slab. | |
2967 | * The cpu may change afterward. However that does not matter since | |
2968 | * data is retrieved via this pointer. If we are on the same cpu | |
2969 | * during the cmpxchg then the free will succeed. | |
2970 | */ | |
2971 | do { | |
2972 | tid = this_cpu_read(s->cpu_slab->tid); | |
2973 | c = raw_cpu_ptr(s->cpu_slab); | |
2974 | } while (IS_ENABLED(CONFIG_PREEMPT) && | |
2975 | unlikely(tid != READ_ONCE(c->tid))); | |
2976 | ||
2977 | /* Same with comment on barrier() in slab_alloc_node() */ | |
2978 | barrier(); | |
2979 | ||
2980 | if (likely(page == c->page)) { | |
2981 | set_freepointer(s, tail_obj, c->freelist); | |
2982 | ||
2983 | if (unlikely(!this_cpu_cmpxchg_double( | |
2984 | s->cpu_slab->freelist, s->cpu_slab->tid, | |
2985 | c->freelist, tid, | |
2986 | head, next_tid(tid)))) { | |
2987 | ||
2988 | note_cmpxchg_failure("slab_free", s, tid); | |
2989 | goto redo; | |
2990 | } | |
2991 | stat(s, FREE_FASTPATH); | |
2992 | } else | |
2993 | __slab_free(s, page, head, tail_obj, cnt, addr); | |
2994 | ||
2995 | } | |
2996 | ||
2997 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, | |
2998 | void *head, void *tail, int cnt, | |
2999 | unsigned long addr) | |
3000 | { | |
3001 | /* | |
3002 | * With KASAN enabled slab_free_freelist_hook modifies the freelist | |
3003 | * to remove objects, whose reuse must be delayed. | |
3004 | */ | |
3005 | if (slab_free_freelist_hook(s, &head, &tail)) | |
3006 | do_slab_free(s, page, head, tail, cnt, addr); | |
3007 | } | |
3008 | ||
3009 | #ifdef CONFIG_KASAN_GENERIC | |
3010 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) | |
3011 | { | |
3012 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
3013 | } | |
3014 | #endif | |
3015 | ||
3016 | void kmem_cache_free(struct kmem_cache *s, void *x) | |
3017 | { | |
3018 | s = cache_from_obj(s, x); | |
3019 | if (!s) | |
3020 | return; | |
3021 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); | |
3022 | trace_kmem_cache_free(_RET_IP_, x); | |
3023 | } | |
3024 | EXPORT_SYMBOL(kmem_cache_free); | |
3025 | ||
3026 | struct detached_freelist { | |
3027 | struct page *page; | |
3028 | void *tail; | |
3029 | void *freelist; | |
3030 | int cnt; | |
3031 | struct kmem_cache *s; | |
3032 | }; | |
3033 | ||
3034 | /* | |
3035 | * This function progressively scans the array with free objects (with | |
3036 | * a limited look ahead) and extract objects belonging to the same | |
3037 | * page. It builds a detached freelist directly within the given | |
3038 | * page/objects. This can happen without any need for | |
3039 | * synchronization, because the objects are owned by running process. | |
3040 | * The freelist is build up as a single linked list in the objects. | |
3041 | * The idea is, that this detached freelist can then be bulk | |
3042 | * transferred to the real freelist(s), but only requiring a single | |
3043 | * synchronization primitive. Look ahead in the array is limited due | |
3044 | * to performance reasons. | |
3045 | */ | |
3046 | static inline | |
3047 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3048 | void **p, struct detached_freelist *df) | |
3049 | { | |
3050 | size_t first_skipped_index = 0; | |
3051 | int lookahead = 3; | |
3052 | void *object; | |
3053 | struct page *page; | |
3054 | ||
3055 | /* Always re-init detached_freelist */ | |
3056 | df->page = NULL; | |
3057 | ||
3058 | do { | |
3059 | object = p[--size]; | |
3060 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ | |
3061 | } while (!object && size); | |
3062 | ||
3063 | if (!object) | |
3064 | return 0; | |
3065 | ||
3066 | page = virt_to_head_page(object); | |
3067 | if (!s) { | |
3068 | /* Handle kalloc'ed objects */ | |
3069 | if (unlikely(!PageSlab(page))) { | |
3070 | BUG_ON(!PageCompound(page)); | |
3071 | kfree_hook(object); | |
3072 | __free_pages(page, compound_order(page)); | |
3073 | p[size] = NULL; /* mark object processed */ | |
3074 | return size; | |
3075 | } | |
3076 | /* Derive kmem_cache from object */ | |
3077 | df->s = page->slab_cache; | |
3078 | } else { | |
3079 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3080 | } | |
3081 | ||
3082 | /* Start new detached freelist */ | |
3083 | df->page = page; | |
3084 | set_freepointer(df->s, object, NULL); | |
3085 | df->tail = object; | |
3086 | df->freelist = object; | |
3087 | p[size] = NULL; /* mark object processed */ | |
3088 | df->cnt = 1; | |
3089 | ||
3090 | while (size) { | |
3091 | object = p[--size]; | |
3092 | if (!object) | |
3093 | continue; /* Skip processed objects */ | |
3094 | ||
3095 | /* df->page is always set at this point */ | |
3096 | if (df->page == virt_to_head_page(object)) { | |
3097 | /* Opportunity build freelist */ | |
3098 | set_freepointer(df->s, object, df->freelist); | |
3099 | df->freelist = object; | |
3100 | df->cnt++; | |
3101 | p[size] = NULL; /* mark object processed */ | |
3102 | ||
3103 | continue; | |
3104 | } | |
3105 | ||
3106 | /* Limit look ahead search */ | |
3107 | if (!--lookahead) | |
3108 | break; | |
3109 | ||
3110 | if (!first_skipped_index) | |
3111 | first_skipped_index = size + 1; | |
3112 | } | |
3113 | ||
3114 | return first_skipped_index; | |
3115 | } | |
3116 | ||
3117 | /* Note that interrupts must be enabled when calling this function. */ | |
3118 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) | |
3119 | { | |
3120 | if (WARN_ON(!size)) | |
3121 | return; | |
3122 | ||
3123 | do { | |
3124 | struct detached_freelist df; | |
3125 | ||
3126 | size = build_detached_freelist(s, size, p, &df); | |
3127 | if (!df.page) | |
3128 | continue; | |
3129 | ||
3130 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); | |
3131 | } while (likely(size)); | |
3132 | } | |
3133 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3134 | ||
3135 | /* Note that interrupts must be enabled when calling this function. */ | |
3136 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, | |
3137 | void **p) | |
3138 | { | |
3139 | struct kmem_cache_cpu *c; | |
3140 | int i; | |
3141 | ||
3142 | /* memcg and kmem_cache debug support */ | |
3143 | s = slab_pre_alloc_hook(s, flags); | |
3144 | if (unlikely(!s)) | |
3145 | return false; | |
3146 | /* | |
3147 | * Drain objects in the per cpu slab, while disabling local | |
3148 | * IRQs, which protects against PREEMPT and interrupts | |
3149 | * handlers invoking normal fastpath. | |
3150 | */ | |
3151 | local_irq_disable(); | |
3152 | c = this_cpu_ptr(s->cpu_slab); | |
3153 | ||
3154 | for (i = 0; i < size; i++) { | |
3155 | void *object = c->freelist; | |
3156 | ||
3157 | if (unlikely(!object)) { | |
3158 | /* | |
3159 | * Invoking slow path likely have side-effect | |
3160 | * of re-populating per CPU c->freelist | |
3161 | */ | |
3162 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, | |
3163 | _RET_IP_, c); | |
3164 | if (unlikely(!p[i])) | |
3165 | goto error; | |
3166 | ||
3167 | c = this_cpu_ptr(s->cpu_slab); | |
3168 | maybe_wipe_obj_freeptr(s, p[i]); | |
3169 | ||
3170 | continue; /* goto for-loop */ | |
3171 | } | |
3172 | c->freelist = get_freepointer(s, object); | |
3173 | p[i] = object; | |
3174 | maybe_wipe_obj_freeptr(s, p[i]); | |
3175 | } | |
3176 | c->tid = next_tid(c->tid); | |
3177 | local_irq_enable(); | |
3178 | ||
3179 | /* Clear memory outside IRQ disabled fastpath loop */ | |
3180 | if (unlikely(slab_want_init_on_alloc(flags, s))) { | |
3181 | int j; | |
3182 | ||
3183 | for (j = 0; j < i; j++) | |
3184 | memset(p[j], 0, s->object_size); | |
3185 | } | |
3186 | ||
3187 | /* memcg and kmem_cache debug support */ | |
3188 | slab_post_alloc_hook(s, flags, size, p); | |
3189 | return i; | |
3190 | error: | |
3191 | local_irq_enable(); | |
3192 | slab_post_alloc_hook(s, flags, i, p); | |
3193 | __kmem_cache_free_bulk(s, i, p); | |
3194 | return 0; | |
3195 | } | |
3196 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3197 | ||
3198 | ||
3199 | /* | |
3200 | * Object placement in a slab is made very easy because we always start at | |
3201 | * offset 0. If we tune the size of the object to the alignment then we can | |
3202 | * get the required alignment by putting one properly sized object after | |
3203 | * another. | |
3204 | * | |
3205 | * Notice that the allocation order determines the sizes of the per cpu | |
3206 | * caches. Each processor has always one slab available for allocations. | |
3207 | * Increasing the allocation order reduces the number of times that slabs | |
3208 | * must be moved on and off the partial lists and is therefore a factor in | |
3209 | * locking overhead. | |
3210 | */ | |
3211 | ||
3212 | /* | |
3213 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3214 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3215 | * and increases the number of allocations possible without having to | |
3216 | * take the list_lock. | |
3217 | */ | |
3218 | static unsigned int slub_min_order; | |
3219 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3220 | static unsigned int slub_min_objects; | |
3221 | ||
3222 | /* | |
3223 | * Calculate the order of allocation given an slab object size. | |
3224 | * | |
3225 | * The order of allocation has significant impact on performance and other | |
3226 | * system components. Generally order 0 allocations should be preferred since | |
3227 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3228 | * be problematic to put into order 0 slabs because there may be too much | |
3229 | * unused space left. We go to a higher order if more than 1/16th of the slab | |
3230 | * would be wasted. | |
3231 | * | |
3232 | * In order to reach satisfactory performance we must ensure that a minimum | |
3233 | * number of objects is in one slab. Otherwise we may generate too much | |
3234 | * activity on the partial lists which requires taking the list_lock. This is | |
3235 | * less a concern for large slabs though which are rarely used. | |
3236 | * | |
3237 | * slub_max_order specifies the order where we begin to stop considering the | |
3238 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3239 | * we try to keep the page order as low as possible. So we accept more waste | |
3240 | * of space in favor of a small page order. | |
3241 | * | |
3242 | * Higher order allocations also allow the placement of more objects in a | |
3243 | * slab and thereby reduce object handling overhead. If the user has | |
3244 | * requested a higher mininum order then we start with that one instead of | |
3245 | * the smallest order which will fit the object. | |
3246 | */ | |
3247 | static inline unsigned int slab_order(unsigned int size, | |
3248 | unsigned int min_objects, unsigned int max_order, | |
3249 | unsigned int fract_leftover) | |
3250 | { | |
3251 | unsigned int min_order = slub_min_order; | |
3252 | unsigned int order; | |
3253 | ||
3254 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) | |
3255 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
3256 | ||
3257 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); | |
3258 | order <= max_order; order++) { | |
3259 | ||
3260 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; | |
3261 | unsigned int rem; | |
3262 | ||
3263 | rem = slab_size % size; | |
3264 | ||
3265 | if (rem <= slab_size / fract_leftover) | |
3266 | break; | |
3267 | } | |
3268 | ||
3269 | return order; | |
3270 | } | |
3271 | ||
3272 | static inline int calculate_order(unsigned int size) | |
3273 | { | |
3274 | unsigned int order; | |
3275 | unsigned int min_objects; | |
3276 | unsigned int max_objects; | |
3277 | ||
3278 | /* | |
3279 | * Attempt to find best configuration for a slab. This | |
3280 | * works by first attempting to generate a layout with | |
3281 | * the best configuration and backing off gradually. | |
3282 | * | |
3283 | * First we increase the acceptable waste in a slab. Then | |
3284 | * we reduce the minimum objects required in a slab. | |
3285 | */ | |
3286 | min_objects = slub_min_objects; | |
3287 | if (!min_objects) | |
3288 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
3289 | max_objects = order_objects(slub_max_order, size); | |
3290 | min_objects = min(min_objects, max_objects); | |
3291 | ||
3292 | while (min_objects > 1) { | |
3293 | unsigned int fraction; | |
3294 | ||
3295 | fraction = 16; | |
3296 | while (fraction >= 4) { | |
3297 | order = slab_order(size, min_objects, | |
3298 | slub_max_order, fraction); | |
3299 | if (order <= slub_max_order) | |
3300 | return order; | |
3301 | fraction /= 2; | |
3302 | } | |
3303 | min_objects--; | |
3304 | } | |
3305 | ||
3306 | /* | |
3307 | * We were unable to place multiple objects in a slab. Now | |
3308 | * lets see if we can place a single object there. | |
3309 | */ | |
3310 | order = slab_order(size, 1, slub_max_order, 1); | |
3311 | if (order <= slub_max_order) | |
3312 | return order; | |
3313 | ||
3314 | /* | |
3315 | * Doh this slab cannot be placed using slub_max_order. | |
3316 | */ | |
3317 | order = slab_order(size, 1, MAX_ORDER, 1); | |
3318 | if (order < MAX_ORDER) | |
3319 | return order; | |
3320 | return -ENOSYS; | |
3321 | } | |
3322 | ||
3323 | static void | |
3324 | init_kmem_cache_node(struct kmem_cache_node *n) | |
3325 | { | |
3326 | n->nr_partial = 0; | |
3327 | spin_lock_init(&n->list_lock); | |
3328 | INIT_LIST_HEAD(&n->partial); | |
3329 | #ifdef CONFIG_SLUB_DEBUG | |
3330 | atomic_long_set(&n->nr_slabs, 0); | |
3331 | atomic_long_set(&n->total_objects, 0); | |
3332 | INIT_LIST_HEAD(&n->full); | |
3333 | #endif | |
3334 | } | |
3335 | ||
3336 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) | |
3337 | { | |
3338 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < | |
3339 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); | |
3340 | ||
3341 | /* | |
3342 | * Must align to double word boundary for the double cmpxchg | |
3343 | * instructions to work; see __pcpu_double_call_return_bool(). | |
3344 | */ | |
3345 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), | |
3346 | 2 * sizeof(void *)); | |
3347 | ||
3348 | if (!s->cpu_slab) | |
3349 | return 0; | |
3350 | ||
3351 | init_kmem_cache_cpus(s); | |
3352 | ||
3353 | return 1; | |
3354 | } | |
3355 | ||
3356 | static struct kmem_cache *kmem_cache_node; | |
3357 | ||
3358 | /* | |
3359 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3360 | * slab on the node for this slabcache. There are no concurrent accesses | |
3361 | * possible. | |
3362 | * | |
3363 | * Note that this function only works on the kmem_cache_node | |
3364 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
3365 | * memory on a fresh node that has no slab structures yet. | |
3366 | */ | |
3367 | static void early_kmem_cache_node_alloc(int node) | |
3368 | { | |
3369 | struct page *page; | |
3370 | struct kmem_cache_node *n; | |
3371 | ||
3372 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); | |
3373 | ||
3374 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); | |
3375 | ||
3376 | BUG_ON(!page); | |
3377 | if (page_to_nid(page) != node) { | |
3378 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); | |
3379 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
3380 | } | |
3381 | ||
3382 | n = page->freelist; | |
3383 | BUG_ON(!n); | |
3384 | #ifdef CONFIG_SLUB_DEBUG | |
3385 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); | |
3386 | init_tracking(kmem_cache_node, n); | |
3387 | #endif | |
3388 | n = kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), | |
3389 | GFP_KERNEL); | |
3390 | page->freelist = get_freepointer(kmem_cache_node, n); | |
3391 | page->inuse = 1; | |
3392 | page->frozen = 0; | |
3393 | kmem_cache_node->node[node] = n; | |
3394 | init_kmem_cache_node(n); | |
3395 | inc_slabs_node(kmem_cache_node, node, page->objects); | |
3396 | ||
3397 | /* | |
3398 | * No locks need to be taken here as it has just been | |
3399 | * initialized and there is no concurrent access. | |
3400 | */ | |
3401 | __add_partial(n, page, DEACTIVATE_TO_HEAD); | |
3402 | } | |
3403 | ||
3404 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3405 | { | |
3406 | int node; | |
3407 | struct kmem_cache_node *n; | |
3408 | ||
3409 | for_each_kmem_cache_node(s, node, n) { | |
3410 | s->node[node] = NULL; | |
3411 | kmem_cache_free(kmem_cache_node, n); | |
3412 | } | |
3413 | } | |
3414 | ||
3415 | void __kmem_cache_release(struct kmem_cache *s) | |
3416 | { | |
3417 | cache_random_seq_destroy(s); | |
3418 | free_percpu(s->cpu_slab); | |
3419 | free_kmem_cache_nodes(s); | |
3420 | } | |
3421 | ||
3422 | static int init_kmem_cache_nodes(struct kmem_cache *s) | |
3423 | { | |
3424 | int node; | |
3425 | ||
3426 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3427 | struct kmem_cache_node *n; | |
3428 | ||
3429 | if (slab_state == DOWN) { | |
3430 | early_kmem_cache_node_alloc(node); | |
3431 | continue; | |
3432 | } | |
3433 | n = kmem_cache_alloc_node(kmem_cache_node, | |
3434 | GFP_KERNEL, node); | |
3435 | ||
3436 | if (!n) { | |
3437 | free_kmem_cache_nodes(s); | |
3438 | return 0; | |
3439 | } | |
3440 | ||
3441 | init_kmem_cache_node(n); | |
3442 | s->node[node] = n; | |
3443 | } | |
3444 | return 1; | |
3445 | } | |
3446 | ||
3447 | static void set_min_partial(struct kmem_cache *s, unsigned long min) | |
3448 | { | |
3449 | if (min < MIN_PARTIAL) | |
3450 | min = MIN_PARTIAL; | |
3451 | else if (min > MAX_PARTIAL) | |
3452 | min = MAX_PARTIAL; | |
3453 | s->min_partial = min; | |
3454 | } | |
3455 | ||
3456 | static void set_cpu_partial(struct kmem_cache *s) | |
3457 | { | |
3458 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3459 | /* | |
3460 | * cpu_partial determined the maximum number of objects kept in the | |
3461 | * per cpu partial lists of a processor. | |
3462 | * | |
3463 | * Per cpu partial lists mainly contain slabs that just have one | |
3464 | * object freed. If they are used for allocation then they can be | |
3465 | * filled up again with minimal effort. The slab will never hit the | |
3466 | * per node partial lists and therefore no locking will be required. | |
3467 | * | |
3468 | * This setting also determines | |
3469 | * | |
3470 | * A) The number of objects from per cpu partial slabs dumped to the | |
3471 | * per node list when we reach the limit. | |
3472 | * B) The number of objects in cpu partial slabs to extract from the | |
3473 | * per node list when we run out of per cpu objects. We only fetch | |
3474 | * 50% to keep some capacity around for frees. | |
3475 | */ | |
3476 | if (!kmem_cache_has_cpu_partial(s)) | |
3477 | s->cpu_partial = 0; | |
3478 | else if (s->size >= PAGE_SIZE) | |
3479 | s->cpu_partial = 2; | |
3480 | else if (s->size >= 1024) | |
3481 | s->cpu_partial = 6; | |
3482 | else if (s->size >= 256) | |
3483 | s->cpu_partial = 13; | |
3484 | else | |
3485 | s->cpu_partial = 30; | |
3486 | #endif | |
3487 | } | |
3488 | ||
3489 | /* | |
3490 | * calculate_sizes() determines the order and the distribution of data within | |
3491 | * a slab object. | |
3492 | */ | |
3493 | static int calculate_sizes(struct kmem_cache *s, int forced_order) | |
3494 | { | |
3495 | slab_flags_t flags = s->flags; | |
3496 | unsigned int size = s->object_size; | |
3497 | unsigned int order; | |
3498 | ||
3499 | /* | |
3500 | * Round up object size to the next word boundary. We can only | |
3501 | * place the free pointer at word boundaries and this determines | |
3502 | * the possible location of the free pointer. | |
3503 | */ | |
3504 | size = ALIGN(size, sizeof(void *)); | |
3505 | ||
3506 | #ifdef CONFIG_SLUB_DEBUG | |
3507 | /* | |
3508 | * Determine if we can poison the object itself. If the user of | |
3509 | * the slab may touch the object after free or before allocation | |
3510 | * then we should never poison the object itself. | |
3511 | */ | |
3512 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && | |
3513 | !s->ctor) | |
3514 | s->flags |= __OBJECT_POISON; | |
3515 | else | |
3516 | s->flags &= ~__OBJECT_POISON; | |
3517 | ||
3518 | ||
3519 | /* | |
3520 | * If we are Redzoning then check if there is some space between the | |
3521 | * end of the object and the free pointer. If not then add an | |
3522 | * additional word to have some bytes to store Redzone information. | |
3523 | */ | |
3524 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) | |
3525 | size += sizeof(void *); | |
3526 | #endif | |
3527 | ||
3528 | /* | |
3529 | * With that we have determined the number of bytes in actual use | |
3530 | * by the object. This is the potential offset to the free pointer. | |
3531 | */ | |
3532 | s->inuse = size; | |
3533 | ||
3534 | if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || | |
3535 | s->ctor)) { | |
3536 | /* | |
3537 | * Relocate free pointer after the object if it is not | |
3538 | * permitted to overwrite the first word of the object on | |
3539 | * kmem_cache_free. | |
3540 | * | |
3541 | * This is the case if we do RCU, have a constructor or | |
3542 | * destructor or are poisoning the objects. | |
3543 | */ | |
3544 | s->offset = size; | |
3545 | size += sizeof(void *); | |
3546 | } | |
3547 | ||
3548 | #ifdef CONFIG_SLUB_DEBUG | |
3549 | if (flags & SLAB_STORE_USER) | |
3550 | /* | |
3551 | * Need to store information about allocs and frees after | |
3552 | * the object. | |
3553 | */ | |
3554 | size += 2 * sizeof(struct track); | |
3555 | #endif | |
3556 | ||
3557 | kasan_cache_create(s, &size, &s->flags); | |
3558 | #ifdef CONFIG_SLUB_DEBUG | |
3559 | if (flags & SLAB_RED_ZONE) { | |
3560 | /* | |
3561 | * Add some empty padding so that we can catch | |
3562 | * overwrites from earlier objects rather than let | |
3563 | * tracking information or the free pointer be | |
3564 | * corrupted if a user writes before the start | |
3565 | * of the object. | |
3566 | */ | |
3567 | size += sizeof(void *); | |
3568 | ||
3569 | s->red_left_pad = sizeof(void *); | |
3570 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3571 | size += s->red_left_pad; | |
3572 | } | |
3573 | #endif | |
3574 | ||
3575 | /* | |
3576 | * SLUB stores one object immediately after another beginning from | |
3577 | * offset 0. In order to align the objects we have to simply size | |
3578 | * each object to conform to the alignment. | |
3579 | */ | |
3580 | size = ALIGN(size, s->align); | |
3581 | s->size = size; | |
3582 | if (forced_order >= 0) | |
3583 | order = forced_order; | |
3584 | else | |
3585 | order = calculate_order(size); | |
3586 | ||
3587 | if ((int)order < 0) | |
3588 | return 0; | |
3589 | ||
3590 | s->allocflags = 0; | |
3591 | if (order) | |
3592 | s->allocflags |= __GFP_COMP; | |
3593 | ||
3594 | if (s->flags & SLAB_CACHE_DMA) | |
3595 | s->allocflags |= GFP_DMA; | |
3596 | ||
3597 | if (s->flags & SLAB_CACHE_DMA32) | |
3598 | s->allocflags |= GFP_DMA32; | |
3599 | ||
3600 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3601 | s->allocflags |= __GFP_RECLAIMABLE; | |
3602 | ||
3603 | /* | |
3604 | * Determine the number of objects per slab | |
3605 | */ | |
3606 | s->oo = oo_make(order, size); | |
3607 | s->min = oo_make(get_order(size), size); | |
3608 | if (oo_objects(s->oo) > oo_objects(s->max)) | |
3609 | s->max = s->oo; | |
3610 | ||
3611 | return !!oo_objects(s->oo); | |
3612 | } | |
3613 | ||
3614 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) | |
3615 | { | |
3616 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); | |
3617 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
3618 | s->random = get_random_long(); | |
3619 | #endif | |
3620 | ||
3621 | if (!calculate_sizes(s, -1)) | |
3622 | goto error; | |
3623 | if (disable_higher_order_debug) { | |
3624 | /* | |
3625 | * Disable debugging flags that store metadata if the min slab | |
3626 | * order increased. | |
3627 | */ | |
3628 | if (get_order(s->size) > get_order(s->object_size)) { | |
3629 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
3630 | s->offset = 0; | |
3631 | if (!calculate_sizes(s, -1)) | |
3632 | goto error; | |
3633 | } | |
3634 | } | |
3635 | ||
3636 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ | |
3637 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
3638 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) | |
3639 | /* Enable fast mode */ | |
3640 | s->flags |= __CMPXCHG_DOUBLE; | |
3641 | #endif | |
3642 | ||
3643 | /* | |
3644 | * The larger the object size is, the more pages we want on the partial | |
3645 | * list to avoid pounding the page allocator excessively. | |
3646 | */ | |
3647 | set_min_partial(s, ilog2(s->size) / 2); | |
3648 | ||
3649 | set_cpu_partial(s); | |
3650 | ||
3651 | #ifdef CONFIG_NUMA | |
3652 | s->remote_node_defrag_ratio = 1000; | |
3653 | #endif | |
3654 | ||
3655 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3656 | if (slab_state >= UP) { | |
3657 | if (init_cache_random_seq(s)) | |
3658 | goto error; | |
3659 | } | |
3660 | ||
3661 | if (!init_kmem_cache_nodes(s)) | |
3662 | goto error; | |
3663 | ||
3664 | if (alloc_kmem_cache_cpus(s)) | |
3665 | return 0; | |
3666 | ||
3667 | free_kmem_cache_nodes(s); | |
3668 | error: | |
3669 | return -EINVAL; | |
3670 | } | |
3671 | ||
3672 | static void list_slab_objects(struct kmem_cache *s, struct page *page, | |
3673 | const char *text) | |
3674 | { | |
3675 | #ifdef CONFIG_SLUB_DEBUG | |
3676 | void *addr = page_address(page); | |
3677 | void *p; | |
3678 | unsigned long *map = bitmap_zalloc(page->objects, GFP_ATOMIC); | |
3679 | if (!map) | |
3680 | return; | |
3681 | slab_err(s, page, text, s->name); | |
3682 | slab_lock(page); | |
3683 | ||
3684 | get_map(s, page, map); | |
3685 | for_each_object(p, s, addr, page->objects) { | |
3686 | ||
3687 | if (!test_bit(slab_index(p, s, addr), map)) { | |
3688 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); | |
3689 | print_tracking(s, p); | |
3690 | } | |
3691 | } | |
3692 | slab_unlock(page); | |
3693 | bitmap_free(map); | |
3694 | #endif | |
3695 | } | |
3696 | ||
3697 | /* | |
3698 | * Attempt to free all partial slabs on a node. | |
3699 | * This is called from __kmem_cache_shutdown(). We must take list_lock | |
3700 | * because sysfs file might still access partial list after the shutdowning. | |
3701 | */ | |
3702 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) | |
3703 | { | |
3704 | LIST_HEAD(discard); | |
3705 | struct page *page, *h; | |
3706 | ||
3707 | BUG_ON(irqs_disabled()); | |
3708 | spin_lock_irq(&n->list_lock); | |
3709 | list_for_each_entry_safe(page, h, &n->partial, slab_list) { | |
3710 | if (!page->inuse) { | |
3711 | remove_partial(n, page); | |
3712 | list_add(&page->slab_list, &discard); | |
3713 | } else { | |
3714 | list_slab_objects(s, page, | |
3715 | "Objects remaining in %s on __kmem_cache_shutdown()"); | |
3716 | } | |
3717 | } | |
3718 | spin_unlock_irq(&n->list_lock); | |
3719 | ||
3720 | list_for_each_entry_safe(page, h, &discard, slab_list) | |
3721 | discard_slab(s, page); | |
3722 | } | |
3723 | ||
3724 | bool __kmem_cache_empty(struct kmem_cache *s) | |
3725 | { | |
3726 | int node; | |
3727 | struct kmem_cache_node *n; | |
3728 | ||
3729 | for_each_kmem_cache_node(s, node, n) | |
3730 | if (n->nr_partial || slabs_node(s, node)) | |
3731 | return false; | |
3732 | return true; | |
3733 | } | |
3734 | ||
3735 | /* | |
3736 | * Release all resources used by a slab cache. | |
3737 | */ | |
3738 | int __kmem_cache_shutdown(struct kmem_cache *s) | |
3739 | { | |
3740 | int node; | |
3741 | struct kmem_cache_node *n; | |
3742 | ||
3743 | flush_all(s); | |
3744 | /* Attempt to free all objects */ | |
3745 | for_each_kmem_cache_node(s, node, n) { | |
3746 | free_partial(s, n); | |
3747 | if (n->nr_partial || slabs_node(s, node)) | |
3748 | return 1; | |
3749 | } | |
3750 | sysfs_slab_remove(s); | |
3751 | return 0; | |
3752 | } | |
3753 | ||
3754 | /******************************************************************** | |
3755 | * Kmalloc subsystem | |
3756 | *******************************************************************/ | |
3757 | ||
3758 | static int __init setup_slub_min_order(char *str) | |
3759 | { | |
3760 | get_option(&str, (int *)&slub_min_order); | |
3761 | ||
3762 | return 1; | |
3763 | } | |
3764 | ||
3765 | __setup("slub_min_order=", setup_slub_min_order); | |
3766 | ||
3767 | static int __init setup_slub_max_order(char *str) | |
3768 | { | |
3769 | get_option(&str, (int *)&slub_max_order); | |
3770 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
3771 | ||
3772 | return 1; | |
3773 | } | |
3774 | ||
3775 | __setup("slub_max_order=", setup_slub_max_order); | |
3776 | ||
3777 | static int __init setup_slub_min_objects(char *str) | |
3778 | { | |
3779 | get_option(&str, (int *)&slub_min_objects); | |
3780 | ||
3781 | return 1; | |
3782 | } | |
3783 | ||
3784 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3785 | ||
3786 | void *__kmalloc(size_t size, gfp_t flags) | |
3787 | { | |
3788 | struct kmem_cache *s; | |
3789 | void *ret; | |
3790 | ||
3791 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) | |
3792 | return kmalloc_large(size, flags); | |
3793 | ||
3794 | s = kmalloc_slab(size, flags); | |
3795 | ||
3796 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
3797 | return s; | |
3798 | ||
3799 | ret = slab_alloc(s, flags, _RET_IP_); | |
3800 | ||
3801 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); | |
3802 | ||
3803 | ret = kasan_kmalloc(s, ret, size, flags); | |
3804 | ||
3805 | return ret; | |
3806 | } | |
3807 | EXPORT_SYMBOL(__kmalloc); | |
3808 | ||
3809 | #ifdef CONFIG_NUMA | |
3810 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) | |
3811 | { | |
3812 | struct page *page; | |
3813 | void *ptr = NULL; | |
3814 | unsigned int order = get_order(size); | |
3815 | ||
3816 | flags |= __GFP_COMP; | |
3817 | page = alloc_pages_node(node, flags, order); | |
3818 | if (page) { | |
3819 | ptr = page_address(page); | |
3820 | mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE, | |
3821 | 1 << order); | |
3822 | } | |
3823 | ||
3824 | return kmalloc_large_node_hook(ptr, size, flags); | |
3825 | } | |
3826 | ||
3827 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
3828 | { | |
3829 | struct kmem_cache *s; | |
3830 | void *ret; | |
3831 | ||
3832 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { | |
3833 | ret = kmalloc_large_node(size, flags, node); | |
3834 | ||
3835 | trace_kmalloc_node(_RET_IP_, ret, | |
3836 | size, PAGE_SIZE << get_order(size), | |
3837 | flags, node); | |
3838 | ||
3839 | return ret; | |
3840 | } | |
3841 | ||
3842 | s = kmalloc_slab(size, flags); | |
3843 | ||
3844 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
3845 | return s; | |
3846 | ||
3847 | ret = slab_alloc_node(s, flags, node, _RET_IP_); | |
3848 | ||
3849 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); | |
3850 | ||
3851 | ret = kasan_kmalloc(s, ret, size, flags); | |
3852 | ||
3853 | return ret; | |
3854 | } | |
3855 | EXPORT_SYMBOL(__kmalloc_node); | |
3856 | #endif /* CONFIG_NUMA */ | |
3857 | ||
3858 | #ifdef CONFIG_HARDENED_USERCOPY | |
3859 | /* | |
3860 | * Rejects incorrectly sized objects and objects that are to be copied | |
3861 | * to/from userspace but do not fall entirely within the containing slab | |
3862 | * cache's usercopy region. | |
3863 | * | |
3864 | * Returns NULL if check passes, otherwise const char * to name of cache | |
3865 | * to indicate an error. | |
3866 | */ | |
3867 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, | |
3868 | bool to_user) | |
3869 | { | |
3870 | struct kmem_cache *s; | |
3871 | unsigned int offset; | |
3872 | size_t object_size; | |
3873 | ||
3874 | ptr = kasan_reset_tag(ptr); | |
3875 | ||
3876 | /* Find object and usable object size. */ | |
3877 | s = page->slab_cache; | |
3878 | ||
3879 | /* Reject impossible pointers. */ | |
3880 | if (ptr < page_address(page)) | |
3881 | usercopy_abort("SLUB object not in SLUB page?!", NULL, | |
3882 | to_user, 0, n); | |
3883 | ||
3884 | /* Find offset within object. */ | |
3885 | offset = (ptr - page_address(page)) % s->size; | |
3886 | ||
3887 | /* Adjust for redzone and reject if within the redzone. */ | |
3888 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) { | |
3889 | if (offset < s->red_left_pad) | |
3890 | usercopy_abort("SLUB object in left red zone", | |
3891 | s->name, to_user, offset, n); | |
3892 | offset -= s->red_left_pad; | |
3893 | } | |
3894 | ||
3895 | /* Allow address range falling entirely within usercopy region. */ | |
3896 | if (offset >= s->useroffset && | |
3897 | offset - s->useroffset <= s->usersize && | |
3898 | n <= s->useroffset - offset + s->usersize) | |
3899 | return; | |
3900 | ||
3901 | /* | |
3902 | * If the copy is still within the allocated object, produce | |
3903 | * a warning instead of rejecting the copy. This is intended | |
3904 | * to be a temporary method to find any missing usercopy | |
3905 | * whitelists. | |
3906 | */ | |
3907 | object_size = slab_ksize(s); | |
3908 | if (usercopy_fallback && | |
3909 | offset <= object_size && n <= object_size - offset) { | |
3910 | usercopy_warn("SLUB object", s->name, to_user, offset, n); | |
3911 | return; | |
3912 | } | |
3913 | ||
3914 | usercopy_abort("SLUB object", s->name, to_user, offset, n); | |
3915 | } | |
3916 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
3917 | ||
3918 | size_t __ksize(const void *object) | |
3919 | { | |
3920 | struct page *page; | |
3921 | ||
3922 | if (unlikely(object == ZERO_SIZE_PTR)) | |
3923 | return 0; | |
3924 | ||
3925 | page = virt_to_head_page(object); | |
3926 | ||
3927 | if (unlikely(!PageSlab(page))) { | |
3928 | WARN_ON(!PageCompound(page)); | |
3929 | return page_size(page); | |
3930 | } | |
3931 | ||
3932 | return slab_ksize(page->slab_cache); | |
3933 | } | |
3934 | EXPORT_SYMBOL(__ksize); | |
3935 | ||
3936 | void kfree(const void *x) | |
3937 | { | |
3938 | struct page *page; | |
3939 | void *object = (void *)x; | |
3940 | ||
3941 | trace_kfree(_RET_IP_, x); | |
3942 | ||
3943 | if (unlikely(ZERO_OR_NULL_PTR(x))) | |
3944 | return; | |
3945 | ||
3946 | page = virt_to_head_page(x); | |
3947 | if (unlikely(!PageSlab(page))) { | |
3948 | unsigned int order = compound_order(page); | |
3949 | ||
3950 | BUG_ON(!PageCompound(page)); | |
3951 | kfree_hook(object); | |
3952 | mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE, | |
3953 | -(1 << order)); | |
3954 | __free_pages(page, order); | |
3955 | return; | |
3956 | } | |
3957 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); | |
3958 | } | |
3959 | EXPORT_SYMBOL(kfree); | |
3960 | ||
3961 | #define SHRINK_PROMOTE_MAX 32 | |
3962 | ||
3963 | /* | |
3964 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled | |
3965 | * up most to the head of the partial lists. New allocations will then | |
3966 | * fill those up and thus they can be removed from the partial lists. | |
3967 | * | |
3968 | * The slabs with the least items are placed last. This results in them | |
3969 | * being allocated from last increasing the chance that the last objects | |
3970 | * are freed in them. | |
3971 | */ | |
3972 | int __kmem_cache_shrink(struct kmem_cache *s) | |
3973 | { | |
3974 | int node; | |
3975 | int i; | |
3976 | struct kmem_cache_node *n; | |
3977 | struct page *page; | |
3978 | struct page *t; | |
3979 | struct list_head discard; | |
3980 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
3981 | unsigned long flags; | |
3982 | int ret = 0; | |
3983 | ||
3984 | flush_all(s); | |
3985 | for_each_kmem_cache_node(s, node, n) { | |
3986 | INIT_LIST_HEAD(&discard); | |
3987 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3988 | INIT_LIST_HEAD(promote + i); | |
3989 | ||
3990 | spin_lock_irqsave(&n->list_lock, flags); | |
3991 | ||
3992 | /* | |
3993 | * Build lists of slabs to discard or promote. | |
3994 | * | |
3995 | * Note that concurrent frees may occur while we hold the | |
3996 | * list_lock. page->inuse here is the upper limit. | |
3997 | */ | |
3998 | list_for_each_entry_safe(page, t, &n->partial, slab_list) { | |
3999 | int free = page->objects - page->inuse; | |
4000 | ||
4001 | /* Do not reread page->inuse */ | |
4002 | barrier(); | |
4003 | ||
4004 | /* We do not keep full slabs on the list */ | |
4005 | BUG_ON(free <= 0); | |
4006 | ||
4007 | if (free == page->objects) { | |
4008 | list_move(&page->slab_list, &discard); | |
4009 | n->nr_partial--; | |
4010 | } else if (free <= SHRINK_PROMOTE_MAX) | |
4011 | list_move(&page->slab_list, promote + free - 1); | |
4012 | } | |
4013 | ||
4014 | /* | |
4015 | * Promote the slabs filled up most to the head of the | |
4016 | * partial list. | |
4017 | */ | |
4018 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) | |
4019 | list_splice(promote + i, &n->partial); | |
4020 | ||
4021 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4022 | ||
4023 | /* Release empty slabs */ | |
4024 | list_for_each_entry_safe(page, t, &discard, slab_list) | |
4025 | discard_slab(s, page); | |
4026 | ||
4027 | if (slabs_node(s, node)) | |
4028 | ret = 1; | |
4029 | } | |
4030 | ||
4031 | return ret; | |
4032 | } | |
4033 | ||
4034 | #ifdef CONFIG_MEMCG | |
4035 | void __kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s) | |
4036 | { | |
4037 | /* | |
4038 | * Called with all the locks held after a sched RCU grace period. | |
4039 | * Even if @s becomes empty after shrinking, we can't know that @s | |
4040 | * doesn't have allocations already in-flight and thus can't | |
4041 | * destroy @s until the associated memcg is released. | |
4042 | * | |
4043 | * However, let's remove the sysfs files for empty caches here. | |
4044 | * Each cache has a lot of interface files which aren't | |
4045 | * particularly useful for empty draining caches; otherwise, we can | |
4046 | * easily end up with millions of unnecessary sysfs files on | |
4047 | * systems which have a lot of memory and transient cgroups. | |
4048 | */ | |
4049 | if (!__kmem_cache_shrink(s)) | |
4050 | sysfs_slab_remove(s); | |
4051 | } | |
4052 | ||
4053 | void __kmemcg_cache_deactivate(struct kmem_cache *s) | |
4054 | { | |
4055 | /* | |
4056 | * Disable empty slabs caching. Used to avoid pinning offline | |
4057 | * memory cgroups by kmem pages that can be freed. | |
4058 | */ | |
4059 | slub_set_cpu_partial(s, 0); | |
4060 | s->min_partial = 0; | |
4061 | } | |
4062 | #endif /* CONFIG_MEMCG */ | |
4063 | ||
4064 | static int slab_mem_going_offline_callback(void *arg) | |
4065 | { | |
4066 | struct kmem_cache *s; | |
4067 | ||
4068 | mutex_lock(&slab_mutex); | |
4069 | list_for_each_entry(s, &slab_caches, list) | |
4070 | __kmem_cache_shrink(s); | |
4071 | mutex_unlock(&slab_mutex); | |
4072 | ||
4073 | return 0; | |
4074 | } | |
4075 | ||
4076 | static void slab_mem_offline_callback(void *arg) | |
4077 | { | |
4078 | struct kmem_cache_node *n; | |
4079 | struct kmem_cache *s; | |
4080 | struct memory_notify *marg = arg; | |
4081 | int offline_node; | |
4082 | ||
4083 | offline_node = marg->status_change_nid_normal; | |
4084 | ||
4085 | /* | |
4086 | * If the node still has available memory. we need kmem_cache_node | |
4087 | * for it yet. | |
4088 | */ | |
4089 | if (offline_node < 0) | |
4090 | return; | |
4091 | ||
4092 | mutex_lock(&slab_mutex); | |
4093 | list_for_each_entry(s, &slab_caches, list) { | |
4094 | n = get_node(s, offline_node); | |
4095 | if (n) { | |
4096 | /* | |
4097 | * if n->nr_slabs > 0, slabs still exist on the node | |
4098 | * that is going down. We were unable to free them, | |
4099 | * and offline_pages() function shouldn't call this | |
4100 | * callback. So, we must fail. | |
4101 | */ | |
4102 | BUG_ON(slabs_node(s, offline_node)); | |
4103 | ||
4104 | s->node[offline_node] = NULL; | |
4105 | kmem_cache_free(kmem_cache_node, n); | |
4106 | } | |
4107 | } | |
4108 | mutex_unlock(&slab_mutex); | |
4109 | } | |
4110 | ||
4111 | static int slab_mem_going_online_callback(void *arg) | |
4112 | { | |
4113 | struct kmem_cache_node *n; | |
4114 | struct kmem_cache *s; | |
4115 | struct memory_notify *marg = arg; | |
4116 | int nid = marg->status_change_nid_normal; | |
4117 | int ret = 0; | |
4118 | ||
4119 | /* | |
4120 | * If the node's memory is already available, then kmem_cache_node is | |
4121 | * already created. Nothing to do. | |
4122 | */ | |
4123 | if (nid < 0) | |
4124 | return 0; | |
4125 | ||
4126 | /* | |
4127 | * We are bringing a node online. No memory is available yet. We must | |
4128 | * allocate a kmem_cache_node structure in order to bring the node | |
4129 | * online. | |
4130 | */ | |
4131 | mutex_lock(&slab_mutex); | |
4132 | list_for_each_entry(s, &slab_caches, list) { | |
4133 | /* | |
4134 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4135 | * since memory is not yet available from the node that | |
4136 | * is brought up. | |
4137 | */ | |
4138 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); | |
4139 | if (!n) { | |
4140 | ret = -ENOMEM; | |
4141 | goto out; | |
4142 | } | |
4143 | init_kmem_cache_node(n); | |
4144 | s->node[nid] = n; | |
4145 | } | |
4146 | out: | |
4147 | mutex_unlock(&slab_mutex); | |
4148 | return ret; | |
4149 | } | |
4150 | ||
4151 | static int slab_memory_callback(struct notifier_block *self, | |
4152 | unsigned long action, void *arg) | |
4153 | { | |
4154 | int ret = 0; | |
4155 | ||
4156 | switch (action) { | |
4157 | case MEM_GOING_ONLINE: | |
4158 | ret = slab_mem_going_online_callback(arg); | |
4159 | break; | |
4160 | case MEM_GOING_OFFLINE: | |
4161 | ret = slab_mem_going_offline_callback(arg); | |
4162 | break; | |
4163 | case MEM_OFFLINE: | |
4164 | case MEM_CANCEL_ONLINE: | |
4165 | slab_mem_offline_callback(arg); | |
4166 | break; | |
4167 | case MEM_ONLINE: | |
4168 | case MEM_CANCEL_OFFLINE: | |
4169 | break; | |
4170 | } | |
4171 | if (ret) | |
4172 | ret = notifier_from_errno(ret); | |
4173 | else | |
4174 | ret = NOTIFY_OK; | |
4175 | return ret; | |
4176 | } | |
4177 | ||
4178 | static struct notifier_block slab_memory_callback_nb = { | |
4179 | .notifier_call = slab_memory_callback, | |
4180 | .priority = SLAB_CALLBACK_PRI, | |
4181 | }; | |
4182 | ||
4183 | /******************************************************************** | |
4184 | * Basic setup of slabs | |
4185 | *******************************************************************/ | |
4186 | ||
4187 | /* | |
4188 | * Used for early kmem_cache structures that were allocated using | |
4189 | * the page allocator. Allocate them properly then fix up the pointers | |
4190 | * that may be pointing to the wrong kmem_cache structure. | |
4191 | */ | |
4192 | ||
4193 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) | |
4194 | { | |
4195 | int node; | |
4196 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
4197 | struct kmem_cache_node *n; | |
4198 | ||
4199 | memcpy(s, static_cache, kmem_cache->object_size); | |
4200 | ||
4201 | /* | |
4202 | * This runs very early, and only the boot processor is supposed to be | |
4203 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4204 | * IPIs around. | |
4205 | */ | |
4206 | __flush_cpu_slab(s, smp_processor_id()); | |
4207 | for_each_kmem_cache_node(s, node, n) { | |
4208 | struct page *p; | |
4209 | ||
4210 | list_for_each_entry(p, &n->partial, slab_list) | |
4211 | p->slab_cache = s; | |
4212 | ||
4213 | #ifdef CONFIG_SLUB_DEBUG | |
4214 | list_for_each_entry(p, &n->full, slab_list) | |
4215 | p->slab_cache = s; | |
4216 | #endif | |
4217 | } | |
4218 | slab_init_memcg_params(s); | |
4219 | list_add(&s->list, &slab_caches); | |
4220 | memcg_link_cache(s, NULL); | |
4221 | return s; | |
4222 | } | |
4223 | ||
4224 | void __init kmem_cache_init(void) | |
4225 | { | |
4226 | static __initdata struct kmem_cache boot_kmem_cache, | |
4227 | boot_kmem_cache_node; | |
4228 | ||
4229 | if (debug_guardpage_minorder()) | |
4230 | slub_max_order = 0; | |
4231 | ||
4232 | kmem_cache_node = &boot_kmem_cache_node; | |
4233 | kmem_cache = &boot_kmem_cache; | |
4234 | ||
4235 | create_boot_cache(kmem_cache_node, "kmem_cache_node", | |
4236 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); | |
4237 | ||
4238 | register_hotmemory_notifier(&slab_memory_callback_nb); | |
4239 | ||
4240 | /* Able to allocate the per node structures */ | |
4241 | slab_state = PARTIAL; | |
4242 | ||
4243 | create_boot_cache(kmem_cache, "kmem_cache", | |
4244 | offsetof(struct kmem_cache, node) + | |
4245 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
4246 | SLAB_HWCACHE_ALIGN, 0, 0); | |
4247 | ||
4248 | kmem_cache = bootstrap(&boot_kmem_cache); | |
4249 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); | |
4250 | ||
4251 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
4252 | setup_kmalloc_cache_index_table(); | |
4253 | create_kmalloc_caches(0); | |
4254 | ||
4255 | /* Setup random freelists for each cache */ | |
4256 | init_freelist_randomization(); | |
4257 | ||
4258 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, | |
4259 | slub_cpu_dead); | |
4260 | ||
4261 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", | |
4262 | cache_line_size(), | |
4263 | slub_min_order, slub_max_order, slub_min_objects, | |
4264 | nr_cpu_ids, nr_node_ids); | |
4265 | } | |
4266 | ||
4267 | void __init kmem_cache_init_late(void) | |
4268 | { | |
4269 | } | |
4270 | ||
4271 | struct kmem_cache * | |
4272 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, | |
4273 | slab_flags_t flags, void (*ctor)(void *)) | |
4274 | { | |
4275 | struct kmem_cache *s, *c; | |
4276 | ||
4277 | s = find_mergeable(size, align, flags, name, ctor); | |
4278 | if (s) { | |
4279 | s->refcount++; | |
4280 | ||
4281 | /* | |
4282 | * Adjust the object sizes so that we clear | |
4283 | * the complete object on kzalloc. | |
4284 | */ | |
4285 | s->object_size = max(s->object_size, size); | |
4286 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); | |
4287 | ||
4288 | for_each_memcg_cache(c, s) { | |
4289 | c->object_size = s->object_size; | |
4290 | c->inuse = max(c->inuse, ALIGN(size, sizeof(void *))); | |
4291 | } | |
4292 | ||
4293 | if (sysfs_slab_alias(s, name)) { | |
4294 | s->refcount--; | |
4295 | s = NULL; | |
4296 | } | |
4297 | } | |
4298 | ||
4299 | return s; | |
4300 | } | |
4301 | ||
4302 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) | |
4303 | { | |
4304 | int err; | |
4305 | ||
4306 | err = kmem_cache_open(s, flags); | |
4307 | if (err) | |
4308 | return err; | |
4309 | ||
4310 | /* Mutex is not taken during early boot */ | |
4311 | if (slab_state <= UP) | |
4312 | return 0; | |
4313 | ||
4314 | memcg_propagate_slab_attrs(s); | |
4315 | err = sysfs_slab_add(s); | |
4316 | if (err) | |
4317 | __kmem_cache_release(s); | |
4318 | ||
4319 | return err; | |
4320 | } | |
4321 | ||
4322 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) | |
4323 | { | |
4324 | struct kmem_cache *s; | |
4325 | void *ret; | |
4326 | ||
4327 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) | |
4328 | return kmalloc_large(size, gfpflags); | |
4329 | ||
4330 | s = kmalloc_slab(size, gfpflags); | |
4331 | ||
4332 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
4333 | return s; | |
4334 | ||
4335 | ret = slab_alloc(s, gfpflags, caller); | |
4336 | ||
4337 | /* Honor the call site pointer we received. */ | |
4338 | trace_kmalloc(caller, ret, size, s->size, gfpflags); | |
4339 | ||
4340 | return ret; | |
4341 | } | |
4342 | ||
4343 | #ifdef CONFIG_NUMA | |
4344 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
4345 | int node, unsigned long caller) | |
4346 | { | |
4347 | struct kmem_cache *s; | |
4348 | void *ret; | |
4349 | ||
4350 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { | |
4351 | ret = kmalloc_large_node(size, gfpflags, node); | |
4352 | ||
4353 | trace_kmalloc_node(caller, ret, | |
4354 | size, PAGE_SIZE << get_order(size), | |
4355 | gfpflags, node); | |
4356 | ||
4357 | return ret; | |
4358 | } | |
4359 | ||
4360 | s = kmalloc_slab(size, gfpflags); | |
4361 | ||
4362 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
4363 | return s; | |
4364 | ||
4365 | ret = slab_alloc_node(s, gfpflags, node, caller); | |
4366 | ||
4367 | /* Honor the call site pointer we received. */ | |
4368 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); | |
4369 | ||
4370 | return ret; | |
4371 | } | |
4372 | #endif | |
4373 | ||
4374 | #ifdef CONFIG_SYSFS | |
4375 | static int count_inuse(struct page *page) | |
4376 | { | |
4377 | return page->inuse; | |
4378 | } | |
4379 | ||
4380 | static int count_total(struct page *page) | |
4381 | { | |
4382 | return page->objects; | |
4383 | } | |
4384 | #endif | |
4385 | ||
4386 | #ifdef CONFIG_SLUB_DEBUG | |
4387 | static int validate_slab(struct kmem_cache *s, struct page *page, | |
4388 | unsigned long *map) | |
4389 | { | |
4390 | void *p; | |
4391 | void *addr = page_address(page); | |
4392 | ||
4393 | if (!check_slab(s, page) || | |
4394 | !on_freelist(s, page, NULL)) | |
4395 | return 0; | |
4396 | ||
4397 | /* Now we know that a valid freelist exists */ | |
4398 | bitmap_zero(map, page->objects); | |
4399 | ||
4400 | get_map(s, page, map); | |
4401 | for_each_object(p, s, addr, page->objects) { | |
4402 | if (test_bit(slab_index(p, s, addr), map)) | |
4403 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4404 | return 0; | |
4405 | } | |
4406 | ||
4407 | for_each_object(p, s, addr, page->objects) | |
4408 | if (!test_bit(slab_index(p, s, addr), map)) | |
4409 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) | |
4410 | return 0; | |
4411 | return 1; | |
4412 | } | |
4413 | ||
4414 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, | |
4415 | unsigned long *map) | |
4416 | { | |
4417 | slab_lock(page); | |
4418 | validate_slab(s, page, map); | |
4419 | slab_unlock(page); | |
4420 | } | |
4421 | ||
4422 | static int validate_slab_node(struct kmem_cache *s, | |
4423 | struct kmem_cache_node *n, unsigned long *map) | |
4424 | { | |
4425 | unsigned long count = 0; | |
4426 | struct page *page; | |
4427 | unsigned long flags; | |
4428 | ||
4429 | spin_lock_irqsave(&n->list_lock, flags); | |
4430 | ||
4431 | list_for_each_entry(page, &n->partial, slab_list) { | |
4432 | validate_slab_slab(s, page, map); | |
4433 | count++; | |
4434 | } | |
4435 | if (count != n->nr_partial) | |
4436 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", | |
4437 | s->name, count, n->nr_partial); | |
4438 | ||
4439 | if (!(s->flags & SLAB_STORE_USER)) | |
4440 | goto out; | |
4441 | ||
4442 | list_for_each_entry(page, &n->full, slab_list) { | |
4443 | validate_slab_slab(s, page, map); | |
4444 | count++; | |
4445 | } | |
4446 | if (count != atomic_long_read(&n->nr_slabs)) | |
4447 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", | |
4448 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
4449 | ||
4450 | out: | |
4451 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4452 | return count; | |
4453 | } | |
4454 | ||
4455 | static long validate_slab_cache(struct kmem_cache *s) | |
4456 | { | |
4457 | int node; | |
4458 | unsigned long count = 0; | |
4459 | struct kmem_cache_node *n; | |
4460 | unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL); | |
4461 | ||
4462 | if (!map) | |
4463 | return -ENOMEM; | |
4464 | ||
4465 | flush_all(s); | |
4466 | for_each_kmem_cache_node(s, node, n) | |
4467 | count += validate_slab_node(s, n, map); | |
4468 | bitmap_free(map); | |
4469 | return count; | |
4470 | } | |
4471 | /* | |
4472 | * Generate lists of code addresses where slabcache objects are allocated | |
4473 | * and freed. | |
4474 | */ | |
4475 | ||
4476 | struct location { | |
4477 | unsigned long count; | |
4478 | unsigned long addr; | |
4479 | long long sum_time; | |
4480 | long min_time; | |
4481 | long max_time; | |
4482 | long min_pid; | |
4483 | long max_pid; | |
4484 | DECLARE_BITMAP(cpus, NR_CPUS); | |
4485 | nodemask_t nodes; | |
4486 | }; | |
4487 | ||
4488 | struct loc_track { | |
4489 | unsigned long max; | |
4490 | unsigned long count; | |
4491 | struct location *loc; | |
4492 | }; | |
4493 | ||
4494 | static void free_loc_track(struct loc_track *t) | |
4495 | { | |
4496 | if (t->max) | |
4497 | free_pages((unsigned long)t->loc, | |
4498 | get_order(sizeof(struct location) * t->max)); | |
4499 | } | |
4500 | ||
4501 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) | |
4502 | { | |
4503 | struct location *l; | |
4504 | int order; | |
4505 | ||
4506 | order = get_order(sizeof(struct location) * max); | |
4507 | ||
4508 | l = (void *)__get_free_pages(flags, order); | |
4509 | if (!l) | |
4510 | return 0; | |
4511 | ||
4512 | if (t->count) { | |
4513 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4514 | free_loc_track(t); | |
4515 | } | |
4516 | t->max = max; | |
4517 | t->loc = l; | |
4518 | return 1; | |
4519 | } | |
4520 | ||
4521 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
4522 | const struct track *track) | |
4523 | { | |
4524 | long start, end, pos; | |
4525 | struct location *l; | |
4526 | unsigned long caddr; | |
4527 | unsigned long age = jiffies - track->when; | |
4528 | ||
4529 | start = -1; | |
4530 | end = t->count; | |
4531 | ||
4532 | for ( ; ; ) { | |
4533 | pos = start + (end - start + 1) / 2; | |
4534 | ||
4535 | /* | |
4536 | * There is nothing at "end". If we end up there | |
4537 | * we need to add something to before end. | |
4538 | */ | |
4539 | if (pos == end) | |
4540 | break; | |
4541 | ||
4542 | caddr = t->loc[pos].addr; | |
4543 | if (track->addr == caddr) { | |
4544 | ||
4545 | l = &t->loc[pos]; | |
4546 | l->count++; | |
4547 | if (track->when) { | |
4548 | l->sum_time += age; | |
4549 | if (age < l->min_time) | |
4550 | l->min_time = age; | |
4551 | if (age > l->max_time) | |
4552 | l->max_time = age; | |
4553 | ||
4554 | if (track->pid < l->min_pid) | |
4555 | l->min_pid = track->pid; | |
4556 | if (track->pid > l->max_pid) | |
4557 | l->max_pid = track->pid; | |
4558 | ||
4559 | cpumask_set_cpu(track->cpu, | |
4560 | to_cpumask(l->cpus)); | |
4561 | } | |
4562 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
4563 | return 1; | |
4564 | } | |
4565 | ||
4566 | if (track->addr < caddr) | |
4567 | end = pos; | |
4568 | else | |
4569 | start = pos; | |
4570 | } | |
4571 | ||
4572 | /* | |
4573 | * Not found. Insert new tracking element. | |
4574 | */ | |
4575 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) | |
4576 | return 0; | |
4577 | ||
4578 | l = t->loc + pos; | |
4579 | if (pos < t->count) | |
4580 | memmove(l + 1, l, | |
4581 | (t->count - pos) * sizeof(struct location)); | |
4582 | t->count++; | |
4583 | l->count = 1; | |
4584 | l->addr = track->addr; | |
4585 | l->sum_time = age; | |
4586 | l->min_time = age; | |
4587 | l->max_time = age; | |
4588 | l->min_pid = track->pid; | |
4589 | l->max_pid = track->pid; | |
4590 | cpumask_clear(to_cpumask(l->cpus)); | |
4591 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
4592 | nodes_clear(l->nodes); | |
4593 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
4594 | return 1; | |
4595 | } | |
4596 | ||
4597 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
4598 | struct page *page, enum track_item alloc, | |
4599 | unsigned long *map) | |
4600 | { | |
4601 | void *addr = page_address(page); | |
4602 | void *p; | |
4603 | ||
4604 | bitmap_zero(map, page->objects); | |
4605 | get_map(s, page, map); | |
4606 | ||
4607 | for_each_object(p, s, addr, page->objects) | |
4608 | if (!test_bit(slab_index(p, s, addr), map)) | |
4609 | add_location(t, s, get_track(s, p, alloc)); | |
4610 | } | |
4611 | ||
4612 | static int list_locations(struct kmem_cache *s, char *buf, | |
4613 | enum track_item alloc) | |
4614 | { | |
4615 | int len = 0; | |
4616 | unsigned long i; | |
4617 | struct loc_track t = { 0, 0, NULL }; | |
4618 | int node; | |
4619 | struct kmem_cache_node *n; | |
4620 | unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL); | |
4621 | ||
4622 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), | |
4623 | GFP_KERNEL)) { | |
4624 | bitmap_free(map); | |
4625 | return sprintf(buf, "Out of memory\n"); | |
4626 | } | |
4627 | /* Push back cpu slabs */ | |
4628 | flush_all(s); | |
4629 | ||
4630 | for_each_kmem_cache_node(s, node, n) { | |
4631 | unsigned long flags; | |
4632 | struct page *page; | |
4633 | ||
4634 | if (!atomic_long_read(&n->nr_slabs)) | |
4635 | continue; | |
4636 | ||
4637 | spin_lock_irqsave(&n->list_lock, flags); | |
4638 | list_for_each_entry(page, &n->partial, slab_list) | |
4639 | process_slab(&t, s, page, alloc, map); | |
4640 | list_for_each_entry(page, &n->full, slab_list) | |
4641 | process_slab(&t, s, page, alloc, map); | |
4642 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4643 | } | |
4644 | ||
4645 | for (i = 0; i < t.count; i++) { | |
4646 | struct location *l = &t.loc[i]; | |
4647 | ||
4648 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) | |
4649 | break; | |
4650 | len += sprintf(buf + len, "%7ld ", l->count); | |
4651 | ||
4652 | if (l->addr) | |
4653 | len += sprintf(buf + len, "%pS", (void *)l->addr); | |
4654 | else | |
4655 | len += sprintf(buf + len, "<not-available>"); | |
4656 | ||
4657 | if (l->sum_time != l->min_time) { | |
4658 | len += sprintf(buf + len, " age=%ld/%ld/%ld", | |
4659 | l->min_time, | |
4660 | (long)div_u64(l->sum_time, l->count), | |
4661 | l->max_time); | |
4662 | } else | |
4663 | len += sprintf(buf + len, " age=%ld", | |
4664 | l->min_time); | |
4665 | ||
4666 | if (l->min_pid != l->max_pid) | |
4667 | len += sprintf(buf + len, " pid=%ld-%ld", | |
4668 | l->min_pid, l->max_pid); | |
4669 | else | |
4670 | len += sprintf(buf + len, " pid=%ld", | |
4671 | l->min_pid); | |
4672 | ||
4673 | if (num_online_cpus() > 1 && | |
4674 | !cpumask_empty(to_cpumask(l->cpus)) && | |
4675 | len < PAGE_SIZE - 60) | |
4676 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4677 | " cpus=%*pbl", | |
4678 | cpumask_pr_args(to_cpumask(l->cpus))); | |
4679 | ||
4680 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && | |
4681 | len < PAGE_SIZE - 60) | |
4682 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4683 | " nodes=%*pbl", | |
4684 | nodemask_pr_args(&l->nodes)); | |
4685 | ||
4686 | len += sprintf(buf + len, "\n"); | |
4687 | } | |
4688 | ||
4689 | free_loc_track(&t); | |
4690 | bitmap_free(map); | |
4691 | if (!t.count) | |
4692 | len += sprintf(buf, "No data\n"); | |
4693 | return len; | |
4694 | } | |
4695 | #endif /* CONFIG_SLUB_DEBUG */ | |
4696 | ||
4697 | #ifdef SLUB_RESILIENCY_TEST | |
4698 | static void __init resiliency_test(void) | |
4699 | { | |
4700 | u8 *p; | |
4701 | int type = KMALLOC_NORMAL; | |
4702 | ||
4703 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); | |
4704 | ||
4705 | pr_err("SLUB resiliency testing\n"); | |
4706 | pr_err("-----------------------\n"); | |
4707 | pr_err("A. Corruption after allocation\n"); | |
4708 | ||
4709 | p = kzalloc(16, GFP_KERNEL); | |
4710 | p[16] = 0x12; | |
4711 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", | |
4712 | p + 16); | |
4713 | ||
4714 | validate_slab_cache(kmalloc_caches[type][4]); | |
4715 | ||
4716 | /* Hmmm... The next two are dangerous */ | |
4717 | p = kzalloc(32, GFP_KERNEL); | |
4718 | p[32 + sizeof(void *)] = 0x34; | |
4719 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", | |
4720 | p); | |
4721 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
4722 | ||
4723 | validate_slab_cache(kmalloc_caches[type][5]); | |
4724 | p = kzalloc(64, GFP_KERNEL); | |
4725 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4726 | *p = 0x56; | |
4727 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4728 | p); | |
4729 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
4730 | validate_slab_cache(kmalloc_caches[type][6]); | |
4731 | ||
4732 | pr_err("\nB. Corruption after free\n"); | |
4733 | p = kzalloc(128, GFP_KERNEL); | |
4734 | kfree(p); | |
4735 | *p = 0x78; | |
4736 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4737 | validate_slab_cache(kmalloc_caches[type][7]); | |
4738 | ||
4739 | p = kzalloc(256, GFP_KERNEL); | |
4740 | kfree(p); | |
4741 | p[50] = 0x9a; | |
4742 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); | |
4743 | validate_slab_cache(kmalloc_caches[type][8]); | |
4744 | ||
4745 | p = kzalloc(512, GFP_KERNEL); | |
4746 | kfree(p); | |
4747 | p[512] = 0xab; | |
4748 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4749 | validate_slab_cache(kmalloc_caches[type][9]); | |
4750 | } | |
4751 | #else | |
4752 | #ifdef CONFIG_SYSFS | |
4753 | static void resiliency_test(void) {}; | |
4754 | #endif | |
4755 | #endif /* SLUB_RESILIENCY_TEST */ | |
4756 | ||
4757 | #ifdef CONFIG_SYSFS | |
4758 | enum slab_stat_type { | |
4759 | SL_ALL, /* All slabs */ | |
4760 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4761 | SL_CPU, /* Only slabs used for cpu caches */ | |
4762 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4763 | SL_TOTAL /* Determine object capacity not slabs */ | |
4764 | }; | |
4765 | ||
4766 | #define SO_ALL (1 << SL_ALL) | |
4767 | #define SO_PARTIAL (1 << SL_PARTIAL) | |
4768 | #define SO_CPU (1 << SL_CPU) | |
4769 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
4770 | #define SO_TOTAL (1 << SL_TOTAL) | |
4771 | ||
4772 | #ifdef CONFIG_MEMCG | |
4773 | static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON); | |
4774 | ||
4775 | static int __init setup_slub_memcg_sysfs(char *str) | |
4776 | { | |
4777 | int v; | |
4778 | ||
4779 | if (get_option(&str, &v) > 0) | |
4780 | memcg_sysfs_enabled = v; | |
4781 | ||
4782 | return 1; | |
4783 | } | |
4784 | ||
4785 | __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs); | |
4786 | #endif | |
4787 | ||
4788 | static ssize_t show_slab_objects(struct kmem_cache *s, | |
4789 | char *buf, unsigned long flags) | |
4790 | { | |
4791 | unsigned long total = 0; | |
4792 | int node; | |
4793 | int x; | |
4794 | unsigned long *nodes; | |
4795 | ||
4796 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); | |
4797 | if (!nodes) | |
4798 | return -ENOMEM; | |
4799 | ||
4800 | if (flags & SO_CPU) { | |
4801 | int cpu; | |
4802 | ||
4803 | for_each_possible_cpu(cpu) { | |
4804 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, | |
4805 | cpu); | |
4806 | int node; | |
4807 | struct page *page; | |
4808 | ||
4809 | page = READ_ONCE(c->page); | |
4810 | if (!page) | |
4811 | continue; | |
4812 | ||
4813 | node = page_to_nid(page); | |
4814 | if (flags & SO_TOTAL) | |
4815 | x = page->objects; | |
4816 | else if (flags & SO_OBJECTS) | |
4817 | x = page->inuse; | |
4818 | else | |
4819 | x = 1; | |
4820 | ||
4821 | total += x; | |
4822 | nodes[node] += x; | |
4823 | ||
4824 | page = slub_percpu_partial_read_once(c); | |
4825 | if (page) { | |
4826 | node = page_to_nid(page); | |
4827 | if (flags & SO_TOTAL) | |
4828 | WARN_ON_ONCE(1); | |
4829 | else if (flags & SO_OBJECTS) | |
4830 | WARN_ON_ONCE(1); | |
4831 | else | |
4832 | x = page->pages; | |
4833 | total += x; | |
4834 | nodes[node] += x; | |
4835 | } | |
4836 | } | |
4837 | } | |
4838 | ||
4839 | /* | |
4840 | * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex" | |
4841 | * already held which will conflict with an existing lock order: | |
4842 | * | |
4843 | * mem_hotplug_lock->slab_mutex->kernfs_mutex | |
4844 | * | |
4845 | * We don't really need mem_hotplug_lock (to hold off | |
4846 | * slab_mem_going_offline_callback) here because slab's memory hot | |
4847 | * unplug code doesn't destroy the kmem_cache->node[] data. | |
4848 | */ | |
4849 | ||
4850 | #ifdef CONFIG_SLUB_DEBUG | |
4851 | if (flags & SO_ALL) { | |
4852 | struct kmem_cache_node *n; | |
4853 | ||
4854 | for_each_kmem_cache_node(s, node, n) { | |
4855 | ||
4856 | if (flags & SO_TOTAL) | |
4857 | x = atomic_long_read(&n->total_objects); | |
4858 | else if (flags & SO_OBJECTS) | |
4859 | x = atomic_long_read(&n->total_objects) - | |
4860 | count_partial(n, count_free); | |
4861 | else | |
4862 | x = atomic_long_read(&n->nr_slabs); | |
4863 | total += x; | |
4864 | nodes[node] += x; | |
4865 | } | |
4866 | ||
4867 | } else | |
4868 | #endif | |
4869 | if (flags & SO_PARTIAL) { | |
4870 | struct kmem_cache_node *n; | |
4871 | ||
4872 | for_each_kmem_cache_node(s, node, n) { | |
4873 | if (flags & SO_TOTAL) | |
4874 | x = count_partial(n, count_total); | |
4875 | else if (flags & SO_OBJECTS) | |
4876 | x = count_partial(n, count_inuse); | |
4877 | else | |
4878 | x = n->nr_partial; | |
4879 | total += x; | |
4880 | nodes[node] += x; | |
4881 | } | |
4882 | } | |
4883 | x = sprintf(buf, "%lu", total); | |
4884 | #ifdef CONFIG_NUMA | |
4885 | for (node = 0; node < nr_node_ids; node++) | |
4886 | if (nodes[node]) | |
4887 | x += sprintf(buf + x, " N%d=%lu", | |
4888 | node, nodes[node]); | |
4889 | #endif | |
4890 | kfree(nodes); | |
4891 | return x + sprintf(buf + x, "\n"); | |
4892 | } | |
4893 | ||
4894 | #ifdef CONFIG_SLUB_DEBUG | |
4895 | static int any_slab_objects(struct kmem_cache *s) | |
4896 | { | |
4897 | int node; | |
4898 | struct kmem_cache_node *n; | |
4899 | ||
4900 | for_each_kmem_cache_node(s, node, n) | |
4901 | if (atomic_long_read(&n->total_objects)) | |
4902 | return 1; | |
4903 | ||
4904 | return 0; | |
4905 | } | |
4906 | #endif | |
4907 | ||
4908 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
4909 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) | |
4910 | ||
4911 | struct slab_attribute { | |
4912 | struct attribute attr; | |
4913 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4914 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4915 | }; | |
4916 | ||
4917 | #define SLAB_ATTR_RO(_name) \ | |
4918 | static struct slab_attribute _name##_attr = \ | |
4919 | __ATTR(_name, 0400, _name##_show, NULL) | |
4920 | ||
4921 | #define SLAB_ATTR(_name) \ | |
4922 | static struct slab_attribute _name##_attr = \ | |
4923 | __ATTR(_name, 0600, _name##_show, _name##_store) | |
4924 | ||
4925 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) | |
4926 | { | |
4927 | return sprintf(buf, "%u\n", s->size); | |
4928 | } | |
4929 | SLAB_ATTR_RO(slab_size); | |
4930 | ||
4931 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4932 | { | |
4933 | return sprintf(buf, "%u\n", s->align); | |
4934 | } | |
4935 | SLAB_ATTR_RO(align); | |
4936 | ||
4937 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4938 | { | |
4939 | return sprintf(buf, "%u\n", s->object_size); | |
4940 | } | |
4941 | SLAB_ATTR_RO(object_size); | |
4942 | ||
4943 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4944 | { | |
4945 | return sprintf(buf, "%u\n", oo_objects(s->oo)); | |
4946 | } | |
4947 | SLAB_ATTR_RO(objs_per_slab); | |
4948 | ||
4949 | static ssize_t order_store(struct kmem_cache *s, | |
4950 | const char *buf, size_t length) | |
4951 | { | |
4952 | unsigned int order; | |
4953 | int err; | |
4954 | ||
4955 | err = kstrtouint(buf, 10, &order); | |
4956 | if (err) | |
4957 | return err; | |
4958 | ||
4959 | if (order > slub_max_order || order < slub_min_order) | |
4960 | return -EINVAL; | |
4961 | ||
4962 | calculate_sizes(s, order); | |
4963 | return length; | |
4964 | } | |
4965 | ||
4966 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
4967 | { | |
4968 | return sprintf(buf, "%u\n", oo_order(s->oo)); | |
4969 | } | |
4970 | SLAB_ATTR(order); | |
4971 | ||
4972 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) | |
4973 | { | |
4974 | return sprintf(buf, "%lu\n", s->min_partial); | |
4975 | } | |
4976 | ||
4977 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4978 | size_t length) | |
4979 | { | |
4980 | unsigned long min; | |
4981 | int err; | |
4982 | ||
4983 | err = kstrtoul(buf, 10, &min); | |
4984 | if (err) | |
4985 | return err; | |
4986 | ||
4987 | set_min_partial(s, min); | |
4988 | return length; | |
4989 | } | |
4990 | SLAB_ATTR(min_partial); | |
4991 | ||
4992 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) | |
4993 | { | |
4994 | return sprintf(buf, "%u\n", slub_cpu_partial(s)); | |
4995 | } | |
4996 | ||
4997 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4998 | size_t length) | |
4999 | { | |
5000 | unsigned int objects; | |
5001 | int err; | |
5002 | ||
5003 | err = kstrtouint(buf, 10, &objects); | |
5004 | if (err) | |
5005 | return err; | |
5006 | if (objects && !kmem_cache_has_cpu_partial(s)) | |
5007 | return -EINVAL; | |
5008 | ||
5009 | slub_set_cpu_partial(s, objects); | |
5010 | flush_all(s); | |
5011 | return length; | |
5012 | } | |
5013 | SLAB_ATTR(cpu_partial); | |
5014 | ||
5015 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) | |
5016 | { | |
5017 | if (!s->ctor) | |
5018 | return 0; | |
5019 | return sprintf(buf, "%pS\n", s->ctor); | |
5020 | } | |
5021 | SLAB_ATTR_RO(ctor); | |
5022 | ||
5023 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) | |
5024 | { | |
5025 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); | |
5026 | } | |
5027 | SLAB_ATTR_RO(aliases); | |
5028 | ||
5029 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
5030 | { | |
5031 | return show_slab_objects(s, buf, SO_PARTIAL); | |
5032 | } | |
5033 | SLAB_ATTR_RO(partial); | |
5034 | ||
5035 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5036 | { | |
5037 | return show_slab_objects(s, buf, SO_CPU); | |
5038 | } | |
5039 | SLAB_ATTR_RO(cpu_slabs); | |
5040 | ||
5041 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5042 | { | |
5043 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); | |
5044 | } | |
5045 | SLAB_ATTR_RO(objects); | |
5046 | ||
5047 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) | |
5048 | { | |
5049 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5050 | } | |
5051 | SLAB_ATTR_RO(objects_partial); | |
5052 | ||
5053 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) | |
5054 | { | |
5055 | int objects = 0; | |
5056 | int pages = 0; | |
5057 | int cpu; | |
5058 | int len; | |
5059 | ||
5060 | for_each_online_cpu(cpu) { | |
5061 | struct page *page; | |
5062 | ||
5063 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
5064 | ||
5065 | if (page) { | |
5066 | pages += page->pages; | |
5067 | objects += page->pobjects; | |
5068 | } | |
5069 | } | |
5070 | ||
5071 | len = sprintf(buf, "%d(%d)", objects, pages); | |
5072 | ||
5073 | #ifdef CONFIG_SMP | |
5074 | for_each_online_cpu(cpu) { | |
5075 | struct page *page; | |
5076 | ||
5077 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
5078 | ||
5079 | if (page && len < PAGE_SIZE - 20) | |
5080 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
5081 | page->pobjects, page->pages); | |
5082 | } | |
5083 | #endif | |
5084 | return len + sprintf(buf + len, "\n"); | |
5085 | } | |
5086 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5087 | ||
5088 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
5089 | { | |
5090 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
5091 | } | |
5092 | ||
5093 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
5094 | const char *buf, size_t length) | |
5095 | { | |
5096 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
5097 | if (buf[0] == '1') | |
5098 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
5099 | return length; | |
5100 | } | |
5101 | SLAB_ATTR(reclaim_account); | |
5102 | ||
5103 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5104 | { | |
5105 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
5106 | } | |
5107 | SLAB_ATTR_RO(hwcache_align); | |
5108 | ||
5109 | #ifdef CONFIG_ZONE_DMA | |
5110 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5111 | { | |
5112 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
5113 | } | |
5114 | SLAB_ATTR_RO(cache_dma); | |
5115 | #endif | |
5116 | ||
5117 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) | |
5118 | { | |
5119 | return sprintf(buf, "%u\n", s->usersize); | |
5120 | } | |
5121 | SLAB_ATTR_RO(usersize); | |
5122 | ||
5123 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
5124 | { | |
5125 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); | |
5126 | } | |
5127 | SLAB_ATTR_RO(destroy_by_rcu); | |
5128 | ||
5129 | #ifdef CONFIG_SLUB_DEBUG | |
5130 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
5131 | { | |
5132 | return show_slab_objects(s, buf, SO_ALL); | |
5133 | } | |
5134 | SLAB_ATTR_RO(slabs); | |
5135 | ||
5136 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) | |
5137 | { | |
5138 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5139 | } | |
5140 | SLAB_ATTR_RO(total_objects); | |
5141 | ||
5142 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) | |
5143 | { | |
5144 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); | |
5145 | } | |
5146 | ||
5147 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
5148 | const char *buf, size_t length) | |
5149 | { | |
5150 | s->flags &= ~SLAB_CONSISTENCY_CHECKS; | |
5151 | if (buf[0] == '1') { | |
5152 | s->flags &= ~__CMPXCHG_DOUBLE; | |
5153 | s->flags |= SLAB_CONSISTENCY_CHECKS; | |
5154 | } | |
5155 | return length; | |
5156 | } | |
5157 | SLAB_ATTR(sanity_checks); | |
5158 | ||
5159 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5160 | { | |
5161 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
5162 | } | |
5163 | ||
5164 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
5165 | size_t length) | |
5166 | { | |
5167 | /* | |
5168 | * Tracing a merged cache is going to give confusing results | |
5169 | * as well as cause other issues like converting a mergeable | |
5170 | * cache into an umergeable one. | |
5171 | */ | |
5172 | if (s->refcount > 1) | |
5173 | return -EINVAL; | |
5174 | ||
5175 | s->flags &= ~SLAB_TRACE; | |
5176 | if (buf[0] == '1') { | |
5177 | s->flags &= ~__CMPXCHG_DOUBLE; | |
5178 | s->flags |= SLAB_TRACE; | |
5179 | } | |
5180 | return length; | |
5181 | } | |
5182 | SLAB_ATTR(trace); | |
5183 | ||
5184 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
5185 | { | |
5186 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
5187 | } | |
5188 | ||
5189 | static ssize_t red_zone_store(struct kmem_cache *s, | |
5190 | const char *buf, size_t length) | |
5191 | { | |
5192 | if (any_slab_objects(s)) | |
5193 | return -EBUSY; | |
5194 | ||
5195 | s->flags &= ~SLAB_RED_ZONE; | |
5196 | if (buf[0] == '1') { | |
5197 | s->flags |= SLAB_RED_ZONE; | |
5198 | } | |
5199 | calculate_sizes(s, -1); | |
5200 | return length; | |
5201 | } | |
5202 | SLAB_ATTR(red_zone); | |
5203 | ||
5204 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5205 | { | |
5206 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
5207 | } | |
5208 | ||
5209 | static ssize_t poison_store(struct kmem_cache *s, | |
5210 | const char *buf, size_t length) | |
5211 | { | |
5212 | if (any_slab_objects(s)) | |
5213 | return -EBUSY; | |
5214 | ||
5215 | s->flags &= ~SLAB_POISON; | |
5216 | if (buf[0] == '1') { | |
5217 | s->flags |= SLAB_POISON; | |
5218 | } | |
5219 | calculate_sizes(s, -1); | |
5220 | return length; | |
5221 | } | |
5222 | SLAB_ATTR(poison); | |
5223 | ||
5224 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5225 | { | |
5226 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
5227 | } | |
5228 | ||
5229 | static ssize_t store_user_store(struct kmem_cache *s, | |
5230 | const char *buf, size_t length) | |
5231 | { | |
5232 | if (any_slab_objects(s)) | |
5233 | return -EBUSY; | |
5234 | ||
5235 | s->flags &= ~SLAB_STORE_USER; | |
5236 | if (buf[0] == '1') { | |
5237 | s->flags &= ~__CMPXCHG_DOUBLE; | |
5238 | s->flags |= SLAB_STORE_USER; | |
5239 | } | |
5240 | calculate_sizes(s, -1); | |
5241 | return length; | |
5242 | } | |
5243 | SLAB_ATTR(store_user); | |
5244 | ||
5245 | static ssize_t validate_show(struct kmem_cache *s, char *buf) | |
5246 | { | |
5247 | return 0; | |
5248 | } | |
5249 | ||
5250 | static ssize_t validate_store(struct kmem_cache *s, | |
5251 | const char *buf, size_t length) | |
5252 | { | |
5253 | int ret = -EINVAL; | |
5254 | ||
5255 | if (buf[0] == '1') { | |
5256 | ret = validate_slab_cache(s); | |
5257 | if (ret >= 0) | |
5258 | ret = length; | |
5259 | } | |
5260 | return ret; | |
5261 | } | |
5262 | SLAB_ATTR(validate); | |
5263 | ||
5264 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5265 | { | |
5266 | if (!(s->flags & SLAB_STORE_USER)) | |
5267 | return -ENOSYS; | |
5268 | return list_locations(s, buf, TRACK_ALLOC); | |
5269 | } | |
5270 | SLAB_ATTR_RO(alloc_calls); | |
5271 | ||
5272 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5273 | { | |
5274 | if (!(s->flags & SLAB_STORE_USER)) | |
5275 | return -ENOSYS; | |
5276 | return list_locations(s, buf, TRACK_FREE); | |
5277 | } | |
5278 | SLAB_ATTR_RO(free_calls); | |
5279 | #endif /* CONFIG_SLUB_DEBUG */ | |
5280 | ||
5281 | #ifdef CONFIG_FAILSLAB | |
5282 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5283 | { | |
5284 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
5285 | } | |
5286 | ||
5287 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
5288 | size_t length) | |
5289 | { | |
5290 | if (s->refcount > 1) | |
5291 | return -EINVAL; | |
5292 | ||
5293 | s->flags &= ~SLAB_FAILSLAB; | |
5294 | if (buf[0] == '1') | |
5295 | s->flags |= SLAB_FAILSLAB; | |
5296 | return length; | |
5297 | } | |
5298 | SLAB_ATTR(failslab); | |
5299 | #endif | |
5300 | ||
5301 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) | |
5302 | { | |
5303 | return 0; | |
5304 | } | |
5305 | ||
5306 | static ssize_t shrink_store(struct kmem_cache *s, | |
5307 | const char *buf, size_t length) | |
5308 | { | |
5309 | if (buf[0] == '1') | |
5310 | kmem_cache_shrink_all(s); | |
5311 | else | |
5312 | return -EINVAL; | |
5313 | return length; | |
5314 | } | |
5315 | SLAB_ATTR(shrink); | |
5316 | ||
5317 | #ifdef CONFIG_NUMA | |
5318 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) | |
5319 | { | |
5320 | return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10); | |
5321 | } | |
5322 | ||
5323 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, | |
5324 | const char *buf, size_t length) | |
5325 | { | |
5326 | unsigned int ratio; | |
5327 | int err; | |
5328 | ||
5329 | err = kstrtouint(buf, 10, &ratio); | |
5330 | if (err) | |
5331 | return err; | |
5332 | if (ratio > 100) | |
5333 | return -ERANGE; | |
5334 | ||
5335 | s->remote_node_defrag_ratio = ratio * 10; | |
5336 | ||
5337 | return length; | |
5338 | } | |
5339 | SLAB_ATTR(remote_node_defrag_ratio); | |
5340 | #endif | |
5341 | ||
5342 | #ifdef CONFIG_SLUB_STATS | |
5343 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) | |
5344 | { | |
5345 | unsigned long sum = 0; | |
5346 | int cpu; | |
5347 | int len; | |
5348 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); | |
5349 | ||
5350 | if (!data) | |
5351 | return -ENOMEM; | |
5352 | ||
5353 | for_each_online_cpu(cpu) { | |
5354 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; | |
5355 | ||
5356 | data[cpu] = x; | |
5357 | sum += x; | |
5358 | } | |
5359 | ||
5360 | len = sprintf(buf, "%lu", sum); | |
5361 | ||
5362 | #ifdef CONFIG_SMP | |
5363 | for_each_online_cpu(cpu) { | |
5364 | if (data[cpu] && len < PAGE_SIZE - 20) | |
5365 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); | |
5366 | } | |
5367 | #endif | |
5368 | kfree(data); | |
5369 | return len + sprintf(buf + len, "\n"); | |
5370 | } | |
5371 | ||
5372 | static void clear_stat(struct kmem_cache *s, enum stat_item si) | |
5373 | { | |
5374 | int cpu; | |
5375 | ||
5376 | for_each_online_cpu(cpu) | |
5377 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; | |
5378 | } | |
5379 | ||
5380 | #define STAT_ATTR(si, text) \ | |
5381 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5382 | { \ | |
5383 | return show_stat(s, buf, si); \ | |
5384 | } \ | |
5385 | static ssize_t text##_store(struct kmem_cache *s, \ | |
5386 | const char *buf, size_t length) \ | |
5387 | { \ | |
5388 | if (buf[0] != '0') \ | |
5389 | return -EINVAL; \ | |
5390 | clear_stat(s, si); \ | |
5391 | return length; \ | |
5392 | } \ | |
5393 | SLAB_ATTR(text); \ | |
5394 | ||
5395 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5396 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5397 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5398 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5399 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5400 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5401 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5402 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5403 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5404 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
5405 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); | |
5406 | STAT_ATTR(FREE_SLAB, free_slab); | |
5407 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5408 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5409 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5410 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5411 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5412 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
5413 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); | |
5414 | STAT_ATTR(ORDER_FALLBACK, order_fallback); | |
5415 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); | |
5416 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
5417 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); | |
5418 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
5419 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); | |
5420 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
5421 | #endif /* CONFIG_SLUB_STATS */ | |
5422 | ||
5423 | static struct attribute *slab_attrs[] = { | |
5424 | &slab_size_attr.attr, | |
5425 | &object_size_attr.attr, | |
5426 | &objs_per_slab_attr.attr, | |
5427 | &order_attr.attr, | |
5428 | &min_partial_attr.attr, | |
5429 | &cpu_partial_attr.attr, | |
5430 | &objects_attr.attr, | |
5431 | &objects_partial_attr.attr, | |
5432 | &partial_attr.attr, | |
5433 | &cpu_slabs_attr.attr, | |
5434 | &ctor_attr.attr, | |
5435 | &aliases_attr.attr, | |
5436 | &align_attr.attr, | |
5437 | &hwcache_align_attr.attr, | |
5438 | &reclaim_account_attr.attr, | |
5439 | &destroy_by_rcu_attr.attr, | |
5440 | &shrink_attr.attr, | |
5441 | &slabs_cpu_partial_attr.attr, | |
5442 | #ifdef CONFIG_SLUB_DEBUG | |
5443 | &total_objects_attr.attr, | |
5444 | &slabs_attr.attr, | |
5445 | &sanity_checks_attr.attr, | |
5446 | &trace_attr.attr, | |
5447 | &red_zone_attr.attr, | |
5448 | &poison_attr.attr, | |
5449 | &store_user_attr.attr, | |
5450 | &validate_attr.attr, | |
5451 | &alloc_calls_attr.attr, | |
5452 | &free_calls_attr.attr, | |
5453 | #endif | |
5454 | #ifdef CONFIG_ZONE_DMA | |
5455 | &cache_dma_attr.attr, | |
5456 | #endif | |
5457 | #ifdef CONFIG_NUMA | |
5458 | &remote_node_defrag_ratio_attr.attr, | |
5459 | #endif | |
5460 | #ifdef CONFIG_SLUB_STATS | |
5461 | &alloc_fastpath_attr.attr, | |
5462 | &alloc_slowpath_attr.attr, | |
5463 | &free_fastpath_attr.attr, | |
5464 | &free_slowpath_attr.attr, | |
5465 | &free_frozen_attr.attr, | |
5466 | &free_add_partial_attr.attr, | |
5467 | &free_remove_partial_attr.attr, | |
5468 | &alloc_from_partial_attr.attr, | |
5469 | &alloc_slab_attr.attr, | |
5470 | &alloc_refill_attr.attr, | |
5471 | &alloc_node_mismatch_attr.attr, | |
5472 | &free_slab_attr.attr, | |
5473 | &cpuslab_flush_attr.attr, | |
5474 | &deactivate_full_attr.attr, | |
5475 | &deactivate_empty_attr.attr, | |
5476 | &deactivate_to_head_attr.attr, | |
5477 | &deactivate_to_tail_attr.attr, | |
5478 | &deactivate_remote_frees_attr.attr, | |
5479 | &deactivate_bypass_attr.attr, | |
5480 | &order_fallback_attr.attr, | |
5481 | &cmpxchg_double_fail_attr.attr, | |
5482 | &cmpxchg_double_cpu_fail_attr.attr, | |
5483 | &cpu_partial_alloc_attr.attr, | |
5484 | &cpu_partial_free_attr.attr, | |
5485 | &cpu_partial_node_attr.attr, | |
5486 | &cpu_partial_drain_attr.attr, | |
5487 | #endif | |
5488 | #ifdef CONFIG_FAILSLAB | |
5489 | &failslab_attr.attr, | |
5490 | #endif | |
5491 | &usersize_attr.attr, | |
5492 | ||
5493 | NULL | |
5494 | }; | |
5495 | ||
5496 | static const struct attribute_group slab_attr_group = { | |
5497 | .attrs = slab_attrs, | |
5498 | }; | |
5499 | ||
5500 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5501 | struct attribute *attr, | |
5502 | char *buf) | |
5503 | { | |
5504 | struct slab_attribute *attribute; | |
5505 | struct kmem_cache *s; | |
5506 | int err; | |
5507 | ||
5508 | attribute = to_slab_attr(attr); | |
5509 | s = to_slab(kobj); | |
5510 | ||
5511 | if (!attribute->show) | |
5512 | return -EIO; | |
5513 | ||
5514 | err = attribute->show(s, buf); | |
5515 | ||
5516 | return err; | |
5517 | } | |
5518 | ||
5519 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5520 | struct attribute *attr, | |
5521 | const char *buf, size_t len) | |
5522 | { | |
5523 | struct slab_attribute *attribute; | |
5524 | struct kmem_cache *s; | |
5525 | int err; | |
5526 | ||
5527 | attribute = to_slab_attr(attr); | |
5528 | s = to_slab(kobj); | |
5529 | ||
5530 | if (!attribute->store) | |
5531 | return -EIO; | |
5532 | ||
5533 | err = attribute->store(s, buf, len); | |
5534 | #ifdef CONFIG_MEMCG | |
5535 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { | |
5536 | struct kmem_cache *c; | |
5537 | ||
5538 | mutex_lock(&slab_mutex); | |
5539 | if (s->max_attr_size < len) | |
5540 | s->max_attr_size = len; | |
5541 | ||
5542 | /* | |
5543 | * This is a best effort propagation, so this function's return | |
5544 | * value will be determined by the parent cache only. This is | |
5545 | * basically because not all attributes will have a well | |
5546 | * defined semantics for rollbacks - most of the actions will | |
5547 | * have permanent effects. | |
5548 | * | |
5549 | * Returning the error value of any of the children that fail | |
5550 | * is not 100 % defined, in the sense that users seeing the | |
5551 | * error code won't be able to know anything about the state of | |
5552 | * the cache. | |
5553 | * | |
5554 | * Only returning the error code for the parent cache at least | |
5555 | * has well defined semantics. The cache being written to | |
5556 | * directly either failed or succeeded, in which case we loop | |
5557 | * through the descendants with best-effort propagation. | |
5558 | */ | |
5559 | for_each_memcg_cache(c, s) | |
5560 | attribute->store(c, buf, len); | |
5561 | mutex_unlock(&slab_mutex); | |
5562 | } | |
5563 | #endif | |
5564 | return err; | |
5565 | } | |
5566 | ||
5567 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) | |
5568 | { | |
5569 | #ifdef CONFIG_MEMCG | |
5570 | int i; | |
5571 | char *buffer = NULL; | |
5572 | struct kmem_cache *root_cache; | |
5573 | ||
5574 | if (is_root_cache(s)) | |
5575 | return; | |
5576 | ||
5577 | root_cache = s->memcg_params.root_cache; | |
5578 | ||
5579 | /* | |
5580 | * This mean this cache had no attribute written. Therefore, no point | |
5581 | * in copying default values around | |
5582 | */ | |
5583 | if (!root_cache->max_attr_size) | |
5584 | return; | |
5585 | ||
5586 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5587 | char mbuf[64]; | |
5588 | char *buf; | |
5589 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5590 | ssize_t len; | |
5591 | ||
5592 | if (!attr || !attr->store || !attr->show) | |
5593 | continue; | |
5594 | ||
5595 | /* | |
5596 | * It is really bad that we have to allocate here, so we will | |
5597 | * do it only as a fallback. If we actually allocate, though, | |
5598 | * we can just use the allocated buffer until the end. | |
5599 | * | |
5600 | * Most of the slub attributes will tend to be very small in | |
5601 | * size, but sysfs allows buffers up to a page, so they can | |
5602 | * theoretically happen. | |
5603 | */ | |
5604 | if (buffer) | |
5605 | buf = buffer; | |
5606 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) | |
5607 | buf = mbuf; | |
5608 | else { | |
5609 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5610 | if (WARN_ON(!buffer)) | |
5611 | continue; | |
5612 | buf = buffer; | |
5613 | } | |
5614 | ||
5615 | len = attr->show(root_cache, buf); | |
5616 | if (len > 0) | |
5617 | attr->store(s, buf, len); | |
5618 | } | |
5619 | ||
5620 | if (buffer) | |
5621 | free_page((unsigned long)buffer); | |
5622 | #endif /* CONFIG_MEMCG */ | |
5623 | } | |
5624 | ||
5625 | static void kmem_cache_release(struct kobject *k) | |
5626 | { | |
5627 | slab_kmem_cache_release(to_slab(k)); | |
5628 | } | |
5629 | ||
5630 | static const struct sysfs_ops slab_sysfs_ops = { | |
5631 | .show = slab_attr_show, | |
5632 | .store = slab_attr_store, | |
5633 | }; | |
5634 | ||
5635 | static struct kobj_type slab_ktype = { | |
5636 | .sysfs_ops = &slab_sysfs_ops, | |
5637 | .release = kmem_cache_release, | |
5638 | }; | |
5639 | ||
5640 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5641 | { | |
5642 | struct kobj_type *ktype = get_ktype(kobj); | |
5643 | ||
5644 | if (ktype == &slab_ktype) | |
5645 | return 1; | |
5646 | return 0; | |
5647 | } | |
5648 | ||
5649 | static const struct kset_uevent_ops slab_uevent_ops = { | |
5650 | .filter = uevent_filter, | |
5651 | }; | |
5652 | ||
5653 | static struct kset *slab_kset; | |
5654 | ||
5655 | static inline struct kset *cache_kset(struct kmem_cache *s) | |
5656 | { | |
5657 | #ifdef CONFIG_MEMCG | |
5658 | if (!is_root_cache(s)) | |
5659 | return s->memcg_params.root_cache->memcg_kset; | |
5660 | #endif | |
5661 | return slab_kset; | |
5662 | } | |
5663 | ||
5664 | #define ID_STR_LENGTH 64 | |
5665 | ||
5666 | /* Create a unique string id for a slab cache: | |
5667 | * | |
5668 | * Format :[flags-]size | |
5669 | */ | |
5670 | static char *create_unique_id(struct kmem_cache *s) | |
5671 | { | |
5672 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5673 | char *p = name; | |
5674 | ||
5675 | BUG_ON(!name); | |
5676 | ||
5677 | *p++ = ':'; | |
5678 | /* | |
5679 | * First flags affecting slabcache operations. We will only | |
5680 | * get here for aliasable slabs so we do not need to support | |
5681 | * too many flags. The flags here must cover all flags that | |
5682 | * are matched during merging to guarantee that the id is | |
5683 | * unique. | |
5684 | */ | |
5685 | if (s->flags & SLAB_CACHE_DMA) | |
5686 | *p++ = 'd'; | |
5687 | if (s->flags & SLAB_CACHE_DMA32) | |
5688 | *p++ = 'D'; | |
5689 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5690 | *p++ = 'a'; | |
5691 | if (s->flags & SLAB_CONSISTENCY_CHECKS) | |
5692 | *p++ = 'F'; | |
5693 | if (s->flags & SLAB_ACCOUNT) | |
5694 | *p++ = 'A'; | |
5695 | if (p != name + 1) | |
5696 | *p++ = '-'; | |
5697 | p += sprintf(p, "%07u", s->size); | |
5698 | ||
5699 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
5700 | return name; | |
5701 | } | |
5702 | ||
5703 | static void sysfs_slab_remove_workfn(struct work_struct *work) | |
5704 | { | |
5705 | struct kmem_cache *s = | |
5706 | container_of(work, struct kmem_cache, kobj_remove_work); | |
5707 | ||
5708 | if (!s->kobj.state_in_sysfs) | |
5709 | /* | |
5710 | * For a memcg cache, this may be called during | |
5711 | * deactivation and again on shutdown. Remove only once. | |
5712 | * A cache is never shut down before deactivation is | |
5713 | * complete, so no need to worry about synchronization. | |
5714 | */ | |
5715 | goto out; | |
5716 | ||
5717 | #ifdef CONFIG_MEMCG | |
5718 | kset_unregister(s->memcg_kset); | |
5719 | #endif | |
5720 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
5721 | out: | |
5722 | kobject_put(&s->kobj); | |
5723 | } | |
5724 | ||
5725 | static int sysfs_slab_add(struct kmem_cache *s) | |
5726 | { | |
5727 | int err; | |
5728 | const char *name; | |
5729 | struct kset *kset = cache_kset(s); | |
5730 | int unmergeable = slab_unmergeable(s); | |
5731 | ||
5732 | INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn); | |
5733 | ||
5734 | if (!kset) { | |
5735 | kobject_init(&s->kobj, &slab_ktype); | |
5736 | return 0; | |
5737 | } | |
5738 | ||
5739 | if (!unmergeable && disable_higher_order_debug && | |
5740 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5741 | unmergeable = 1; | |
5742 | ||
5743 | if (unmergeable) { | |
5744 | /* | |
5745 | * Slabcache can never be merged so we can use the name proper. | |
5746 | * This is typically the case for debug situations. In that | |
5747 | * case we can catch duplicate names easily. | |
5748 | */ | |
5749 | sysfs_remove_link(&slab_kset->kobj, s->name); | |
5750 | name = s->name; | |
5751 | } else { | |
5752 | /* | |
5753 | * Create a unique name for the slab as a target | |
5754 | * for the symlinks. | |
5755 | */ | |
5756 | name = create_unique_id(s); | |
5757 | } | |
5758 | ||
5759 | s->kobj.kset = kset; | |
5760 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); | |
5761 | if (err) | |
5762 | goto out; | |
5763 | ||
5764 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5765 | if (err) | |
5766 | goto out_del_kobj; | |
5767 | ||
5768 | #ifdef CONFIG_MEMCG | |
5769 | if (is_root_cache(s) && memcg_sysfs_enabled) { | |
5770 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5771 | if (!s->memcg_kset) { | |
5772 | err = -ENOMEM; | |
5773 | goto out_del_kobj; | |
5774 | } | |
5775 | } | |
5776 | #endif | |
5777 | ||
5778 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
5779 | if (!unmergeable) { | |
5780 | /* Setup first alias */ | |
5781 | sysfs_slab_alias(s, s->name); | |
5782 | } | |
5783 | out: | |
5784 | if (!unmergeable) | |
5785 | kfree(name); | |
5786 | return err; | |
5787 | out_del_kobj: | |
5788 | kobject_del(&s->kobj); | |
5789 | goto out; | |
5790 | } | |
5791 | ||
5792 | static void sysfs_slab_remove(struct kmem_cache *s) | |
5793 | { | |
5794 | if (slab_state < FULL) | |
5795 | /* | |
5796 | * Sysfs has not been setup yet so no need to remove the | |
5797 | * cache from sysfs. | |
5798 | */ | |
5799 | return; | |
5800 | ||
5801 | kobject_get(&s->kobj); | |
5802 | schedule_work(&s->kobj_remove_work); | |
5803 | } | |
5804 | ||
5805 | void sysfs_slab_unlink(struct kmem_cache *s) | |
5806 | { | |
5807 | if (slab_state >= FULL) | |
5808 | kobject_del(&s->kobj); | |
5809 | } | |
5810 | ||
5811 | void sysfs_slab_release(struct kmem_cache *s) | |
5812 | { | |
5813 | if (slab_state >= FULL) | |
5814 | kobject_put(&s->kobj); | |
5815 | } | |
5816 | ||
5817 | /* | |
5818 | * Need to buffer aliases during bootup until sysfs becomes | |
5819 | * available lest we lose that information. | |
5820 | */ | |
5821 | struct saved_alias { | |
5822 | struct kmem_cache *s; | |
5823 | const char *name; | |
5824 | struct saved_alias *next; | |
5825 | }; | |
5826 | ||
5827 | static struct saved_alias *alias_list; | |
5828 | ||
5829 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5830 | { | |
5831 | struct saved_alias *al; | |
5832 | ||
5833 | if (slab_state == FULL) { | |
5834 | /* | |
5835 | * If we have a leftover link then remove it. | |
5836 | */ | |
5837 | sysfs_remove_link(&slab_kset->kobj, name); | |
5838 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
5839 | } | |
5840 | ||
5841 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5842 | if (!al) | |
5843 | return -ENOMEM; | |
5844 | ||
5845 | al->s = s; | |
5846 | al->name = name; | |
5847 | al->next = alias_list; | |
5848 | alias_list = al; | |
5849 | return 0; | |
5850 | } | |
5851 | ||
5852 | static int __init slab_sysfs_init(void) | |
5853 | { | |
5854 | struct kmem_cache *s; | |
5855 | int err; | |
5856 | ||
5857 | mutex_lock(&slab_mutex); | |
5858 | ||
5859 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); | |
5860 | if (!slab_kset) { | |
5861 | mutex_unlock(&slab_mutex); | |
5862 | pr_err("Cannot register slab subsystem.\n"); | |
5863 | return -ENOSYS; | |
5864 | } | |
5865 | ||
5866 | slab_state = FULL; | |
5867 | ||
5868 | list_for_each_entry(s, &slab_caches, list) { | |
5869 | err = sysfs_slab_add(s); | |
5870 | if (err) | |
5871 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", | |
5872 | s->name); | |
5873 | } | |
5874 | ||
5875 | while (alias_list) { | |
5876 | struct saved_alias *al = alias_list; | |
5877 | ||
5878 | alias_list = alias_list->next; | |
5879 | err = sysfs_slab_alias(al->s, al->name); | |
5880 | if (err) | |
5881 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", | |
5882 | al->name); | |
5883 | kfree(al); | |
5884 | } | |
5885 | ||
5886 | mutex_unlock(&slab_mutex); | |
5887 | resiliency_test(); | |
5888 | return 0; | |
5889 | } | |
5890 | ||
5891 | __initcall(slab_sysfs_init); | |
5892 | #endif /* CONFIG_SYSFS */ | |
5893 | ||
5894 | /* | |
5895 | * The /proc/slabinfo ABI | |
5896 | */ | |
5897 | #ifdef CONFIG_SLUB_DEBUG | |
5898 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) | |
5899 | { | |
5900 | unsigned long nr_slabs = 0; | |
5901 | unsigned long nr_objs = 0; | |
5902 | unsigned long nr_free = 0; | |
5903 | int node; | |
5904 | struct kmem_cache_node *n; | |
5905 | ||
5906 | for_each_kmem_cache_node(s, node, n) { | |
5907 | nr_slabs += node_nr_slabs(n); | |
5908 | nr_objs += node_nr_objs(n); | |
5909 | nr_free += count_partial(n, count_free); | |
5910 | } | |
5911 | ||
5912 | sinfo->active_objs = nr_objs - nr_free; | |
5913 | sinfo->num_objs = nr_objs; | |
5914 | sinfo->active_slabs = nr_slabs; | |
5915 | sinfo->num_slabs = nr_slabs; | |
5916 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5917 | sinfo->cache_order = oo_order(s->oo); | |
5918 | } | |
5919 | ||
5920 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) | |
5921 | { | |
5922 | } | |
5923 | ||
5924 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, | |
5925 | size_t count, loff_t *ppos) | |
5926 | { | |
5927 | return -EIO; | |
5928 | } | |
5929 | #endif /* CONFIG_SLUB_DEBUG */ |