Line data Source code
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 operations
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> /* mm_account_reclaimed_pages() */
15 : #include <linux/module.h>
16 : #include <linux/bit_spinlock.h>
17 : #include <linux/interrupt.h>
18 : #include <linux/swab.h>
19 : #include <linux/bitops.h>
20 : #include <linux/slab.h>
21 : #include "slab.h"
22 : #include <linux/proc_fs.h>
23 : #include <linux/seq_file.h>
24 : #include <linux/kasan.h>
25 : #include <linux/kmsan.h>
26 : #include <linux/cpu.h>
27 : #include <linux/cpuset.h>
28 : #include <linux/mempolicy.h>
29 : #include <linux/ctype.h>
30 : #include <linux/stackdepot.h>
31 : #include <linux/debugobjects.h>
32 : #include <linux/kallsyms.h>
33 : #include <linux/kfence.h>
34 : #include <linux/memory.h>
35 : #include <linux/math64.h>
36 : #include <linux/fault-inject.h>
37 : #include <linux/stacktrace.h>
38 : #include <linux/prefetch.h>
39 : #include <linux/memcontrol.h>
40 : #include <linux/random.h>
41 : #include <kunit/test.h>
42 : #include <kunit/test-bug.h>
43 : #include <linux/sort.h>
44 :
45 : #include <linux/debugfs.h>
46 : #include <trace/events/kmem.h>
47 :
48 : #include "internal.h"
49 :
50 : /*
51 : * Lock order:
52 : * 1. slab_mutex (Global Mutex)
53 : * 2. node->list_lock (Spinlock)
54 : * 3. kmem_cache->cpu_slab->lock (Local lock)
55 : * 4. slab_lock(slab) (Only on some arches)
56 : * 5. object_map_lock (Only for debugging)
57 : *
58 : * slab_mutex
59 : *
60 : * The role of the slab_mutex is to protect the list of all the slabs
61 : * and to synchronize major metadata changes to slab cache structures.
62 : * Also synchronizes memory hotplug callbacks.
63 : *
64 : * slab_lock
65 : *
66 : * The slab_lock is a wrapper around the page lock, thus it is a bit
67 : * spinlock.
68 : *
69 : * The slab_lock is only used on arches that do not have the ability
70 : * to do a cmpxchg_double. It only protects:
71 : *
72 : * A. slab->freelist -> List of free objects in a slab
73 : * B. slab->inuse -> Number of objects in use
74 : * C. slab->objects -> Number of objects in slab
75 : * D. slab->frozen -> frozen state
76 : *
77 : * Frozen slabs
78 : *
79 : * If a slab is frozen then it is exempt from list management. It is not
80 : * on any list except per cpu partial list. The processor that froze the
81 : * slab is the one who can perform list operations on the slab. Other
82 : * processors may put objects onto the freelist but the processor that
83 : * froze the slab is the only one that can retrieve the objects from the
84 : * slab's freelist.
85 : *
86 : * list_lock
87 : *
88 : * The list_lock protects the partial and full list on each node and
89 : * the partial slab counter. If taken then no new slabs may be added or
90 : * removed from the lists nor make the number of partial slabs be modified.
91 : * (Note that the total number of slabs is an atomic value that may be
92 : * modified without taking the list lock).
93 : *
94 : * The list_lock is a centralized lock and thus we avoid taking it as
95 : * much as possible. As long as SLUB does not have to handle partial
96 : * slabs, operations can continue without any centralized lock. F.e.
97 : * allocating a long series of objects that fill up slabs does not require
98 : * the list lock.
99 : *
100 : * For debug caches, all allocations are forced to go through a list_lock
101 : * protected region to serialize against concurrent validation.
102 : *
103 : * cpu_slab->lock local lock
104 : *
105 : * This locks protect slowpath manipulation of all kmem_cache_cpu fields
106 : * except the stat counters. This is a percpu structure manipulated only by
107 : * the local cpu, so the lock protects against being preempted or interrupted
108 : * by an irq. Fast path operations rely on lockless operations instead.
109 : *
110 : * On PREEMPT_RT, the local lock neither disables interrupts nor preemption
111 : * which means the lockless fastpath cannot be used as it might interfere with
112 : * an in-progress slow path operations. In this case the local lock is always
113 : * taken but it still utilizes the freelist for the common operations.
114 : *
115 : * lockless fastpaths
116 : *
117 : * The fast path allocation (slab_alloc_node()) and freeing (do_slab_free())
118 : * are fully lockless when satisfied from the percpu slab (and when
119 : * cmpxchg_double is possible to use, otherwise slab_lock is taken).
120 : * They also don't disable preemption or migration or irqs. They rely on
121 : * the transaction id (tid) field to detect being preempted or moved to
122 : * another cpu.
123 : *
124 : * irq, preemption, migration considerations
125 : *
126 : * Interrupts are disabled as part of list_lock or local_lock operations, or
127 : * around the slab_lock operation, in order to make the slab allocator safe
128 : * to use in the context of an irq.
129 : *
130 : * In addition, preemption (or migration on PREEMPT_RT) is disabled in the
131 : * allocation slowpath, bulk allocation, and put_cpu_partial(), so that the
132 : * local cpu doesn't change in the process and e.g. the kmem_cache_cpu pointer
133 : * doesn't have to be revalidated in each section protected by the local lock.
134 : *
135 : * SLUB assigns one slab for allocation to each processor.
136 : * Allocations only occur from these slabs called cpu slabs.
137 : *
138 : * Slabs with free elements are kept on a partial list and during regular
139 : * operations no list for full slabs is used. If an object in a full slab is
140 : * freed then the slab will show up again on the partial lists.
141 : * We track full slabs for debugging purposes though because otherwise we
142 : * cannot scan all objects.
143 : *
144 : * Slabs are freed when they become empty. Teardown and setup is
145 : * minimal so we rely on the page allocators per cpu caches for
146 : * fast frees and allocs.
147 : *
148 : * slab->frozen The slab is frozen and exempt from list processing.
149 : * This means that the slab is dedicated to a purpose
150 : * such as satisfying allocations for a specific
151 : * processor. Objects may be freed in the slab while
152 : * it is frozen but slab_free will then skip the usual
153 : * list operations. It is up to the processor holding
154 : * the slab to integrate the slab into the slab lists
155 : * when the slab is no longer needed.
156 : *
157 : * One use of this flag is to mark slabs that are
158 : * used for allocations. Then such a slab becomes a cpu
159 : * slab. The cpu slab may be equipped with an additional
160 : * freelist that allows lockless access to
161 : * free objects in addition to the regular freelist
162 : * that requires the slab lock.
163 : *
164 : * SLAB_DEBUG_FLAGS Slab requires special handling due to debug
165 : * options set. This moves slab handling out of
166 : * the fast path and disables lockless freelists.
167 : */
168 :
169 : /*
170 : * We could simply use migrate_disable()/enable() but as long as it's a
171 : * function call even on !PREEMPT_RT, use inline preempt_disable() there.
172 : */
173 : #ifndef CONFIG_PREEMPT_RT
174 : #define slub_get_cpu_ptr(var) get_cpu_ptr(var)
175 : #define slub_put_cpu_ptr(var) put_cpu_ptr(var)
176 : #define USE_LOCKLESS_FAST_PATH() (true)
177 : #else
178 : #define slub_get_cpu_ptr(var) \
179 : ({ \
180 : migrate_disable(); \
181 : this_cpu_ptr(var); \
182 : })
183 : #define slub_put_cpu_ptr(var) \
184 : do { \
185 : (void)(var); \
186 : migrate_enable(); \
187 : } while (0)
188 : #define USE_LOCKLESS_FAST_PATH() (false)
189 : #endif
190 :
191 : #ifndef CONFIG_SLUB_TINY
192 : #define __fastpath_inline __always_inline
193 : #else
194 : #define __fastpath_inline
195 : #endif
196 :
197 : #ifdef CONFIG_SLUB_DEBUG
198 : #ifdef CONFIG_SLUB_DEBUG_ON
199 : DEFINE_STATIC_KEY_TRUE(slub_debug_enabled);
200 : #else
201 : DEFINE_STATIC_KEY_FALSE(slub_debug_enabled);
202 : #endif
203 : #endif /* CONFIG_SLUB_DEBUG */
204 :
205 : /* Structure holding parameters for get_partial() call chain */
206 : struct partial_context {
207 : struct slab **slab;
208 : gfp_t flags;
209 : unsigned int orig_size;
210 : };
211 :
212 : static inline bool kmem_cache_debug(struct kmem_cache *s)
213 : {
214 2304 : return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS);
215 : }
216 :
217 : static inline bool slub_debug_orig_size(struct kmem_cache *s)
218 : {
219 320 : return (kmem_cache_debug_flags(s, SLAB_STORE_USER) &&
220 0 : (s->flags & SLAB_KMALLOC));
221 : }
222 :
223 0 : void *fixup_red_left(struct kmem_cache *s, void *p)
224 : {
225 870 : if (kmem_cache_debug_flags(s, SLAB_RED_ZONE))
226 0 : p += s->red_left_pad;
227 :
228 0 : return p;
229 : }
230 :
231 : static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
232 : {
233 : #ifdef CONFIG_SLUB_CPU_PARTIAL
234 : return !kmem_cache_debug(s);
235 : #else
236 : return false;
237 : #endif
238 : }
239 :
240 : /*
241 : * Issues still to be resolved:
242 : *
243 : * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
244 : *
245 : * - Variable sizing of the per node arrays
246 : */
247 :
248 : /* Enable to log cmpxchg failures */
249 : #undef SLUB_DEBUG_CMPXCHG
250 :
251 : #ifndef CONFIG_SLUB_TINY
252 : /*
253 : * Minimum number of partial slabs. These will be left on the partial
254 : * lists even if they are empty. kmem_cache_shrink may reclaim them.
255 : */
256 : #define MIN_PARTIAL 5
257 :
258 : /*
259 : * Maximum number of desirable partial slabs.
260 : * The existence of more partial slabs makes kmem_cache_shrink
261 : * sort the partial list by the number of objects in use.
262 : */
263 : #define MAX_PARTIAL 10
264 : #else
265 : #define MIN_PARTIAL 0
266 : #define MAX_PARTIAL 0
267 : #endif
268 :
269 : #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
270 : SLAB_POISON | SLAB_STORE_USER)
271 :
272 : /*
273 : * These debug flags cannot use CMPXCHG because there might be consistency
274 : * issues when checking or reading debug information
275 : */
276 : #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
277 : SLAB_TRACE)
278 :
279 :
280 : /*
281 : * Debugging flags that require metadata to be stored in the slab. These get
282 : * disabled when slub_debug=O is used and a cache's min order increases with
283 : * metadata.
284 : */
285 : #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
286 :
287 : #define OO_SHIFT 16
288 : #define OO_MASK ((1 << OO_SHIFT) - 1)
289 : #define MAX_OBJS_PER_PAGE 32767 /* since slab.objects is u15 */
290 :
291 : /* Internal SLUB flags */
292 : /* Poison object */
293 : #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U)
294 : /* Use cmpxchg_double */
295 :
296 : #ifdef system_has_freelist_aba
297 : #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U)
298 : #else
299 : #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0U)
300 : #endif
301 :
302 : /*
303 : * Tracking user of a slab.
304 : */
305 : #define TRACK_ADDRS_COUNT 16
306 : struct track {
307 : unsigned long addr; /* Called from address */
308 : #ifdef CONFIG_STACKDEPOT
309 : depot_stack_handle_t handle;
310 : #endif
311 : int cpu; /* Was running on cpu */
312 : int pid; /* Pid context */
313 : unsigned long when; /* When did the operation occur */
314 : };
315 :
316 : enum track_item { TRACK_ALLOC, TRACK_FREE };
317 :
318 : #ifdef SLAB_SUPPORTS_SYSFS
319 : static int sysfs_slab_add(struct kmem_cache *);
320 : static int sysfs_slab_alias(struct kmem_cache *, const char *);
321 : #else
322 : static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
323 : static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
324 : { return 0; }
325 : #endif
326 :
327 : #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
328 : static void debugfs_slab_add(struct kmem_cache *);
329 : #else
330 : static inline void debugfs_slab_add(struct kmem_cache *s) { }
331 : #endif
332 :
333 : static inline void stat(const struct kmem_cache *s, enum stat_item si)
334 : {
335 : #ifdef CONFIG_SLUB_STATS
336 : /*
337 : * The rmw is racy on a preemptible kernel but this is acceptable, so
338 : * avoid this_cpu_add()'s irq-disable overhead.
339 : */
340 : raw_cpu_inc(s->cpu_slab->stat[si]);
341 : #endif
342 : }
343 :
344 : /*
345 : * Tracks for which NUMA nodes we have kmem_cache_nodes allocated.
346 : * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily
347 : * differ during memory hotplug/hotremove operations.
348 : * Protected by slab_mutex.
349 : */
350 : static nodemask_t slab_nodes;
351 :
352 : #ifndef CONFIG_SLUB_TINY
353 : /*
354 : * Workqueue used for flush_cpu_slab().
355 : */
356 : static struct workqueue_struct *flushwq;
357 : #endif
358 :
359 : /********************************************************************
360 : * Core slab cache functions
361 : *******************************************************************/
362 :
363 : /*
364 : * Returns freelist pointer (ptr). With hardening, this is obfuscated
365 : * with an XOR of the address where the pointer is held and a per-cache
366 : * random number.
367 : */
368 : static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
369 : unsigned long ptr_addr)
370 : {
371 : #ifdef CONFIG_SLAB_FREELIST_HARDENED
372 : /*
373 : * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged.
374 : * Normally, this doesn't cause any issues, as both set_freepointer()
375 : * and get_freepointer() are called with a pointer with the same tag.
376 : * However, there are some issues with CONFIG_SLUB_DEBUG code. For
377 : * example, when __free_slub() iterates over objects in a cache, it
378 : * passes untagged pointers to check_object(). check_object() in turns
379 : * calls get_freepointer() with an untagged pointer, which causes the
380 : * freepointer to be restored incorrectly.
381 : */
382 : return (void *)((unsigned long)ptr ^ s->random ^
383 : swab((unsigned long)kasan_reset_tag((void *)ptr_addr)));
384 : #else
385 : return ptr;
386 : #endif
387 : }
388 :
389 : /* Returns the freelist pointer recorded at location ptr_addr. */
390 : static inline void *freelist_dereference(const struct kmem_cache *s,
391 : void *ptr_addr)
392 : {
393 21481 : return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr),
394 : (unsigned long)ptr_addr);
395 : }
396 :
397 : static inline void *get_freepointer(struct kmem_cache *s, void *object)
398 : {
399 21481 : object = kasan_reset_tag(object);
400 42962 : return freelist_dereference(s, object + s->offset);
401 : }
402 :
403 : #ifndef CONFIG_SLUB_TINY
404 : static void prefetch_freepointer(const struct kmem_cache *s, void *object)
405 : {
406 16488 : prefetchw(object + s->offset);
407 : }
408 : #endif
409 :
410 : /*
411 : * When running under KMSAN, get_freepointer_safe() may return an uninitialized
412 : * pointer value in the case the current thread loses the race for the next
413 : * memory chunk in the freelist. In that case this_cpu_cmpxchg_double() in
414 : * slab_alloc_node() will fail, so the uninitialized value won't be used, but
415 : * KMSAN will still check all arguments of cmpxchg because of imperfect
416 : * handling of inline assembly.
417 : * To work around this problem, we apply __no_kmsan_checks to ensure that
418 : * get_freepointer_safe() returns initialized memory.
419 : */
420 : __no_kmsan_checks
421 : static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
422 : {
423 : unsigned long freepointer_addr;
424 : void *p;
425 :
426 : if (!debug_pagealloc_enabled_static())
427 32976 : return get_freepointer(s, object);
428 :
429 : object = kasan_reset_tag(object);
430 : freepointer_addr = (unsigned long)object + s->offset;
431 : copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p));
432 : return freelist_ptr(s, p, freepointer_addr);
433 : }
434 :
435 : static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
436 : {
437 23310 : unsigned long freeptr_addr = (unsigned long)object + s->offset;
438 :
439 : #ifdef CONFIG_SLAB_FREELIST_HARDENED
440 : BUG_ON(object == fp); /* naive detection of double free or corruption */
441 : #endif
442 :
443 23310 : freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr);
444 23310 : *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
445 : }
446 :
447 : /* Loop over all objects in a slab */
448 : #define for_each_object(__p, __s, __addr, __objects) \
449 : for (__p = fixup_red_left(__s, __addr); \
450 : __p < (__addr) + (__objects) * (__s)->size; \
451 : __p += (__s)->size)
452 :
453 : static inline unsigned int order_objects(unsigned int order, unsigned int size)
454 : {
455 214 : return ((unsigned int)PAGE_SIZE << order) / size;
456 : }
457 :
458 : static inline struct kmem_cache_order_objects oo_make(unsigned int order,
459 : unsigned int size)
460 : {
461 106 : struct kmem_cache_order_objects x = {
462 212 : (order << OO_SHIFT) + order_objects(order, size)
463 : };
464 :
465 : return x;
466 : }
467 :
468 : static inline unsigned int oo_order(struct kmem_cache_order_objects x)
469 : {
470 1267 : return x.x >> OO_SHIFT;
471 : }
472 :
473 : static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
474 : {
475 53 : return x.x & OO_MASK;
476 : }
477 :
478 : #ifdef CONFIG_SLUB_CPU_PARTIAL
479 : static void slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects)
480 : {
481 : unsigned int nr_slabs;
482 :
483 : s->cpu_partial = nr_objects;
484 :
485 : /*
486 : * We take the number of objects but actually limit the number of
487 : * slabs on the per cpu partial list, in order to limit excessive
488 : * growth of the list. For simplicity we assume that the slabs will
489 : * be half-full.
490 : */
491 : nr_slabs = DIV_ROUND_UP(nr_objects * 2, oo_objects(s->oo));
492 : s->cpu_partial_slabs = nr_slabs;
493 : }
494 : #else
495 : static inline void
496 : slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects)
497 : {
498 : }
499 : #endif /* CONFIG_SLUB_CPU_PARTIAL */
500 :
501 : /*
502 : * Per slab locking using the pagelock
503 : */
504 : static __always_inline void slab_lock(struct slab *slab)
505 : {
506 690 : struct page *page = slab_page(slab);
507 :
508 : VM_BUG_ON_PAGE(PageTail(page), page);
509 690 : bit_spin_lock(PG_locked, &page->flags);
510 : }
511 :
512 : static __always_inline void slab_unlock(struct slab *slab)
513 : {
514 690 : struct page *page = slab_page(slab);
515 :
516 : VM_BUG_ON_PAGE(PageTail(page), page);
517 690 : __bit_spin_unlock(PG_locked, &page->flags);
518 : }
519 :
520 : static inline bool
521 : __update_freelist_fast(struct slab *slab,
522 : void *freelist_old, unsigned long counters_old,
523 : void *freelist_new, unsigned long counters_new)
524 : {
525 : #ifdef system_has_freelist_aba
526 : freelist_aba_t old = { .freelist = freelist_old, .counter = counters_old };
527 : freelist_aba_t new = { .freelist = freelist_new, .counter = counters_new };
528 :
529 : return try_cmpxchg_freelist(&slab->freelist_counter.full, &old.full, new.full);
530 : #else
531 : return false;
532 : #endif
533 : }
534 :
535 : static inline bool
536 : __update_freelist_slow(struct slab *slab,
537 : void *freelist_old, unsigned long counters_old,
538 : void *freelist_new, unsigned long counters_new)
539 : {
540 690 : bool ret = false;
541 :
542 690 : slab_lock(slab);
543 1380 : if (slab->freelist == freelist_old &&
544 690 : slab->counters == counters_old) {
545 690 : slab->freelist = freelist_new;
546 690 : slab->counters = counters_new;
547 690 : ret = true;
548 : }
549 690 : slab_unlock(slab);
550 :
551 : return ret;
552 : }
553 :
554 : /*
555 : * Interrupts must be disabled (for the fallback code to work right), typically
556 : * by an _irqsave() lock variant. On PREEMPT_RT the preempt_disable(), which is
557 : * part of bit_spin_lock(), is sufficient because the policy is not to allow any
558 : * allocation/ free operation in hardirq context. Therefore nothing can
559 : * interrupt the operation.
560 : */
561 : static inline bool __slab_update_freelist(struct kmem_cache *s, struct slab *slab,
562 : void *freelist_old, unsigned long counters_old,
563 : void *freelist_new, unsigned long counters_new,
564 : const char *n)
565 : {
566 : bool ret;
567 :
568 : if (USE_LOCKLESS_FAST_PATH())
569 : lockdep_assert_irqs_disabled();
570 :
571 : if (s->flags & __CMPXCHG_DOUBLE) {
572 : ret = __update_freelist_fast(slab, freelist_old, counters_old,
573 : freelist_new, counters_new);
574 : } else {
575 920 : ret = __update_freelist_slow(slab, freelist_old, counters_old,
576 : freelist_new, counters_new);
577 : }
578 460 : if (likely(ret))
579 : return true;
580 :
581 : cpu_relax();
582 0 : stat(s, CMPXCHG_DOUBLE_FAIL);
583 :
584 : #ifdef SLUB_DEBUG_CMPXCHG
585 : pr_info("%s %s: cmpxchg double redo ", n, s->name);
586 : #endif
587 :
588 : return false;
589 : }
590 :
591 230 : static inline bool slab_update_freelist(struct kmem_cache *s, struct slab *slab,
592 : void *freelist_old, unsigned long counters_old,
593 : void *freelist_new, unsigned long counters_new,
594 : const char *n)
595 : {
596 : bool ret;
597 :
598 : if (s->flags & __CMPXCHG_DOUBLE) {
599 : ret = __update_freelist_fast(slab, freelist_old, counters_old,
600 : freelist_new, counters_new);
601 : } else {
602 : unsigned long flags;
603 :
604 230 : local_irq_save(flags);
605 460 : ret = __update_freelist_slow(slab, freelist_old, counters_old,
606 : freelist_new, counters_new);
607 460 : local_irq_restore(flags);
608 : }
609 230 : if (likely(ret))
610 : return true;
611 :
612 : cpu_relax();
613 0 : stat(s, CMPXCHG_DOUBLE_FAIL);
614 :
615 : #ifdef SLUB_DEBUG_CMPXCHG
616 : pr_info("%s %s: cmpxchg double redo ", n, s->name);
617 : #endif
618 :
619 : return false;
620 : }
621 :
622 : #ifdef CONFIG_SLUB_DEBUG
623 : static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)];
624 : static DEFINE_SPINLOCK(object_map_lock);
625 :
626 0 : static void __fill_map(unsigned long *obj_map, struct kmem_cache *s,
627 : struct slab *slab)
628 : {
629 0 : void *addr = slab_address(slab);
630 : void *p;
631 :
632 0 : bitmap_zero(obj_map, slab->objects);
633 :
634 0 : for (p = slab->freelist; p; p = get_freepointer(s, p))
635 0 : set_bit(__obj_to_index(s, addr, p), obj_map);
636 0 : }
637 :
638 : #if IS_ENABLED(CONFIG_KUNIT)
639 0 : static bool slab_add_kunit_errors(void)
640 : {
641 : struct kunit_resource *resource;
642 :
643 0 : if (!kunit_get_current_test())
644 : return false;
645 :
646 0 : resource = kunit_find_named_resource(current->kunit_test, "slab_errors");
647 0 : if (!resource)
648 : return false;
649 :
650 0 : (*(int *)resource->data)++;
651 0 : kunit_put_resource(resource);
652 0 : return true;
653 : }
654 : #else
655 : static inline bool slab_add_kunit_errors(void) { return false; }
656 : #endif
657 :
658 : static inline unsigned int size_from_object(struct kmem_cache *s)
659 : {
660 0 : if (s->flags & SLAB_RED_ZONE)
661 0 : return s->size - s->red_left_pad;
662 :
663 : return s->size;
664 : }
665 :
666 : static inline void *restore_red_left(struct kmem_cache *s, void *p)
667 : {
668 0 : if (s->flags & SLAB_RED_ZONE)
669 0 : p -= s->red_left_pad;
670 :
671 : return p;
672 : }
673 :
674 : /*
675 : * Debug settings:
676 : */
677 : #if defined(CONFIG_SLUB_DEBUG_ON)
678 : static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
679 : #else
680 : static slab_flags_t slub_debug;
681 : #endif
682 :
683 : static char *slub_debug_string;
684 : static int disable_higher_order_debug;
685 :
686 : /*
687 : * slub is about to manipulate internal object metadata. This memory lies
688 : * outside the range of the allocated object, so accessing it would normally
689 : * be reported by kasan as a bounds error. metadata_access_enable() is used
690 : * to tell kasan that these accesses are OK.
691 : */
692 : static inline void metadata_access_enable(void)
693 : {
694 : kasan_disable_current();
695 : }
696 :
697 : static inline void metadata_access_disable(void)
698 : {
699 : kasan_enable_current();
700 : }
701 :
702 : /*
703 : * Object debugging
704 : */
705 :
706 : /* Verify that a pointer has an address that is valid within a slab page */
707 0 : static inline int check_valid_pointer(struct kmem_cache *s,
708 : struct slab *slab, void *object)
709 : {
710 : void *base;
711 :
712 0 : if (!object)
713 : return 1;
714 :
715 0 : base = slab_address(slab);
716 0 : object = kasan_reset_tag(object);
717 0 : object = restore_red_left(s, object);
718 0 : if (object < base || object >= base + slab->objects * s->size ||
719 0 : (object - base) % s->size) {
720 : return 0;
721 : }
722 :
723 0 : return 1;
724 : }
725 :
726 : static void print_section(char *level, char *text, u8 *addr,
727 : unsigned int length)
728 : {
729 : metadata_access_enable();
730 0 : print_hex_dump(level, text, DUMP_PREFIX_ADDRESS,
731 0 : 16, 1, kasan_reset_tag((void *)addr), length, 1);
732 : metadata_access_disable();
733 : }
734 :
735 : /*
736 : * See comment in calculate_sizes().
737 : */
738 : static inline bool freeptr_outside_object(struct kmem_cache *s)
739 : {
740 : return s->offset >= s->inuse;
741 : }
742 :
743 : /*
744 : * Return offset of the end of info block which is inuse + free pointer if
745 : * not overlapping with object.
746 : */
747 : static inline unsigned int get_info_end(struct kmem_cache *s)
748 : {
749 0 : if (freeptr_outside_object(s))
750 0 : return s->inuse + sizeof(void *);
751 : else
752 : return s->inuse;
753 : }
754 :
755 : static struct track *get_track(struct kmem_cache *s, void *object,
756 : enum track_item alloc)
757 : {
758 : struct track *p;
759 :
760 0 : p = object + get_info_end(s);
761 :
762 0 : return kasan_reset_tag(p + alloc);
763 : }
764 :
765 : #ifdef CONFIG_STACKDEPOT
766 0 : static noinline depot_stack_handle_t set_track_prepare(void)
767 : {
768 : depot_stack_handle_t handle;
769 : unsigned long entries[TRACK_ADDRS_COUNT];
770 : unsigned int nr_entries;
771 :
772 0 : nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 3);
773 0 : handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT);
774 :
775 0 : return handle;
776 : }
777 : #else
778 : static inline depot_stack_handle_t set_track_prepare(void)
779 : {
780 : return 0;
781 : }
782 : #endif
783 :
784 : static void set_track_update(struct kmem_cache *s, void *object,
785 : enum track_item alloc, unsigned long addr,
786 : depot_stack_handle_t handle)
787 : {
788 0 : struct track *p = get_track(s, object, alloc);
789 :
790 : #ifdef CONFIG_STACKDEPOT
791 0 : p->handle = handle;
792 : #endif
793 0 : p->addr = addr;
794 0 : p->cpu = smp_processor_id();
795 0 : p->pid = current->pid;
796 0 : p->when = jiffies;
797 : }
798 :
799 : static __always_inline void set_track(struct kmem_cache *s, void *object,
800 : enum track_item alloc, unsigned long addr)
801 : {
802 0 : depot_stack_handle_t handle = set_track_prepare();
803 :
804 : set_track_update(s, object, alloc, addr, handle);
805 : }
806 :
807 1 : static void init_tracking(struct kmem_cache *s, void *object)
808 : {
809 : struct track *p;
810 :
811 1 : if (!(s->flags & SLAB_STORE_USER))
812 : return;
813 :
814 0 : p = get_track(s, object, TRACK_ALLOC);
815 0 : memset(p, 0, 2*sizeof(struct track));
816 : }
817 :
818 0 : static void print_track(const char *s, struct track *t, unsigned long pr_time)
819 : {
820 : depot_stack_handle_t handle __maybe_unused;
821 :
822 0 : if (!t->addr)
823 : return;
824 :
825 0 : pr_err("%s in %pS age=%lu cpu=%u pid=%d\n",
826 : s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
827 : #ifdef CONFIG_STACKDEPOT
828 0 : handle = READ_ONCE(t->handle);
829 0 : if (handle)
830 0 : stack_depot_print(handle);
831 : else
832 0 : pr_err("object allocation/free stack trace missing\n");
833 : #endif
834 : }
835 :
836 0 : void print_tracking(struct kmem_cache *s, void *object)
837 : {
838 0 : unsigned long pr_time = jiffies;
839 0 : if (!(s->flags & SLAB_STORE_USER))
840 : return;
841 :
842 0 : print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
843 0 : print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
844 : }
845 :
846 : static void print_slab_info(const struct slab *slab)
847 : {
848 0 : struct folio *folio = (struct folio *)slab_folio(slab);
849 :
850 0 : pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%pGp\n",
851 : slab, slab->objects, slab->inuse, slab->freelist,
852 : folio_flags(folio, 0));
853 : }
854 :
855 : /*
856 : * kmalloc caches has fixed sizes (mostly power of 2), and kmalloc() API
857 : * family will round up the real request size to these fixed ones, so
858 : * there could be an extra area than what is requested. Save the original
859 : * request size in the meta data area, for better debug and sanity check.
860 : */
861 107 : static inline void set_orig_size(struct kmem_cache *s,
862 : void *object, unsigned int orig_size)
863 : {
864 107 : void *p = kasan_reset_tag(object);
865 :
866 107 : if (!slub_debug_orig_size(s))
867 : return;
868 :
869 : #ifdef CONFIG_KASAN_GENERIC
870 : /*
871 : * KASAN could save its free meta data in object's data area at
872 : * offset 0, if the size is larger than 'orig_size', it will
873 : * overlap the data redzone in [orig_size+1, object_size], and
874 : * the check should be skipped.
875 : */
876 : if (kasan_metadata_size(s, true) > orig_size)
877 : orig_size = s->object_size;
878 : #endif
879 :
880 0 : p += get_info_end(s);
881 0 : p += sizeof(struct track) * 2;
882 :
883 0 : *(unsigned int *)p = orig_size;
884 : }
885 :
886 0 : static inline unsigned int get_orig_size(struct kmem_cache *s, void *object)
887 : {
888 0 : void *p = kasan_reset_tag(object);
889 :
890 0 : if (!slub_debug_orig_size(s))
891 0 : return s->object_size;
892 :
893 0 : p += get_info_end(s);
894 0 : p += sizeof(struct track) * 2;
895 :
896 0 : return *(unsigned int *)p;
897 : }
898 :
899 107 : void skip_orig_size_check(struct kmem_cache *s, const void *object)
900 : {
901 107 : set_orig_size(s, (void *)object, s->object_size);
902 107 : }
903 :
904 0 : static void slab_bug(struct kmem_cache *s, char *fmt, ...)
905 : {
906 : struct va_format vaf;
907 : va_list args;
908 :
909 0 : va_start(args, fmt);
910 0 : vaf.fmt = fmt;
911 0 : vaf.va = &args;
912 0 : pr_err("=============================================================================\n");
913 0 : pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
914 0 : pr_err("-----------------------------------------------------------------------------\n\n");
915 0 : va_end(args);
916 0 : }
917 :
918 : __printf(2, 3)
919 0 : static void slab_fix(struct kmem_cache *s, char *fmt, ...)
920 : {
921 : struct va_format vaf;
922 : va_list args;
923 :
924 0 : if (slab_add_kunit_errors())
925 0 : return;
926 :
927 0 : va_start(args, fmt);
928 0 : vaf.fmt = fmt;
929 0 : vaf.va = &args;
930 0 : pr_err("FIX %s: %pV\n", s->name, &vaf);
931 0 : va_end(args);
932 : }
933 :
934 0 : static void print_trailer(struct kmem_cache *s, struct slab *slab, u8 *p)
935 : {
936 : unsigned int off; /* Offset of last byte */
937 0 : u8 *addr = slab_address(slab);
938 :
939 0 : print_tracking(s, p);
940 :
941 0 : print_slab_info(slab);
942 :
943 0 : pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n",
944 : p, p - addr, get_freepointer(s, p));
945 :
946 0 : if (s->flags & SLAB_RED_ZONE)
947 0 : print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
948 : s->red_left_pad);
949 0 : else if (p > addr + 16)
950 0 : print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
951 :
952 0 : print_section(KERN_ERR, "Object ", p,
953 0 : min_t(unsigned int, s->object_size, PAGE_SIZE));
954 0 : if (s->flags & SLAB_RED_ZONE)
955 0 : print_section(KERN_ERR, "Redzone ", p + s->object_size,
956 0 : s->inuse - s->object_size);
957 :
958 0 : off = get_info_end(s);
959 :
960 0 : if (s->flags & SLAB_STORE_USER)
961 0 : off += 2 * sizeof(struct track);
962 :
963 0 : if (slub_debug_orig_size(s))
964 0 : off += sizeof(unsigned int);
965 :
966 0 : off += kasan_metadata_size(s, false);
967 :
968 0 : if (off != size_from_object(s))
969 : /* Beginning of the filler is the free pointer */
970 0 : print_section(KERN_ERR, "Padding ", p + off,
971 0 : size_from_object(s) - off);
972 :
973 0 : dump_stack();
974 0 : }
975 :
976 0 : static void object_err(struct kmem_cache *s, struct slab *slab,
977 : u8 *object, char *reason)
978 : {
979 0 : if (slab_add_kunit_errors())
980 : return;
981 :
982 0 : slab_bug(s, "%s", reason);
983 0 : print_trailer(s, slab, object);
984 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
985 : }
986 :
987 82 : static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab,
988 : void **freelist, void *nextfree)
989 : {
990 82 : if ((s->flags & SLAB_CONSISTENCY_CHECKS) &&
991 0 : !check_valid_pointer(s, slab, nextfree) && freelist) {
992 0 : object_err(s, slab, *freelist, "Freechain corrupt");
993 0 : *freelist = NULL;
994 0 : slab_fix(s, "Isolate corrupted freechain");
995 0 : return true;
996 : }
997 :
998 : return false;
999 : }
1000 :
1001 0 : static __printf(3, 4) void slab_err(struct kmem_cache *s, struct slab *slab,
1002 : const char *fmt, ...)
1003 : {
1004 : va_list args;
1005 : char buf[100];
1006 :
1007 0 : if (slab_add_kunit_errors())
1008 0 : return;
1009 :
1010 0 : va_start(args, fmt);
1011 0 : vsnprintf(buf, sizeof(buf), fmt, args);
1012 0 : va_end(args);
1013 0 : slab_bug(s, "%s", buf);
1014 0 : print_slab_info(slab);
1015 0 : dump_stack();
1016 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
1017 : }
1018 :
1019 1 : static void init_object(struct kmem_cache *s, void *object, u8 val)
1020 : {
1021 1 : u8 *p = kasan_reset_tag(object);
1022 1 : unsigned int poison_size = s->object_size;
1023 :
1024 1 : if (s->flags & SLAB_RED_ZONE) {
1025 0 : memset(p - s->red_left_pad, val, s->red_left_pad);
1026 :
1027 0 : if (slub_debug_orig_size(s) && val == SLUB_RED_ACTIVE) {
1028 : /*
1029 : * Redzone the extra allocated space by kmalloc than
1030 : * requested, and the poison size will be limited to
1031 : * the original request size accordingly.
1032 : */
1033 0 : poison_size = get_orig_size(s, object);
1034 : }
1035 : }
1036 :
1037 1 : if (s->flags & __OBJECT_POISON) {
1038 0 : memset(p, POISON_FREE, poison_size - 1);
1039 0 : p[poison_size - 1] = POISON_END;
1040 : }
1041 :
1042 1 : if (s->flags & SLAB_RED_ZONE)
1043 0 : memset(p + poison_size, val, s->inuse - poison_size);
1044 1 : }
1045 :
1046 0 : static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
1047 : void *from, void *to)
1048 : {
1049 0 : slab_fix(s, "Restoring %s 0x%p-0x%p=0x%x", message, from, to - 1, data);
1050 0 : memset(from, data, to - from);
1051 0 : }
1052 :
1053 0 : static int check_bytes_and_report(struct kmem_cache *s, struct slab *slab,
1054 : u8 *object, char *what,
1055 : u8 *start, unsigned int value, unsigned int bytes)
1056 : {
1057 : u8 *fault;
1058 : u8 *end;
1059 0 : u8 *addr = slab_address(slab);
1060 :
1061 : metadata_access_enable();
1062 0 : fault = memchr_inv(kasan_reset_tag(start), value, bytes);
1063 : metadata_access_disable();
1064 0 : if (!fault)
1065 : return 1;
1066 :
1067 0 : end = start + bytes;
1068 0 : while (end > fault && end[-1] == value)
1069 0 : end--;
1070 :
1071 0 : if (slab_add_kunit_errors())
1072 : goto skip_bug_print;
1073 :
1074 0 : slab_bug(s, "%s overwritten", what);
1075 0 : pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n",
1076 : fault, end - 1, fault - addr,
1077 : fault[0], value);
1078 0 : print_trailer(s, slab, object);
1079 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
1080 :
1081 : skip_bug_print:
1082 0 : restore_bytes(s, what, value, fault, end);
1083 0 : return 0;
1084 : }
1085 :
1086 : /*
1087 : * Object layout:
1088 : *
1089 : * object address
1090 : * Bytes of the object to be managed.
1091 : * If the freepointer may overlay the object then the free
1092 : * pointer is at the middle of the object.
1093 : *
1094 : * Poisoning uses 0x6b (POISON_FREE) and the last byte is
1095 : * 0xa5 (POISON_END)
1096 : *
1097 : * object + s->object_size
1098 : * Padding to reach word boundary. This is also used for Redzoning.
1099 : * Padding is extended by another word if Redzoning is enabled and
1100 : * object_size == inuse.
1101 : *
1102 : * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
1103 : * 0xcc (RED_ACTIVE) for objects in use.
1104 : *
1105 : * object + s->inuse
1106 : * Meta data starts here.
1107 : *
1108 : * A. Free pointer (if we cannot overwrite object on free)
1109 : * B. Tracking data for SLAB_STORE_USER
1110 : * C. Original request size for kmalloc object (SLAB_STORE_USER enabled)
1111 : * D. Padding to reach required alignment boundary or at minimum
1112 : * one word if debugging is on to be able to detect writes
1113 : * before the word boundary.
1114 : *
1115 : * Padding is done using 0x5a (POISON_INUSE)
1116 : *
1117 : * object + s->size
1118 : * Nothing is used beyond s->size.
1119 : *
1120 : * If slabcaches are merged then the object_size and inuse boundaries are mostly
1121 : * ignored. And therefore no slab options that rely on these boundaries
1122 : * may be used with merged slabcaches.
1123 : */
1124 :
1125 0 : static int check_pad_bytes(struct kmem_cache *s, struct slab *slab, u8 *p)
1126 : {
1127 0 : unsigned long off = get_info_end(s); /* The end of info */
1128 :
1129 0 : if (s->flags & SLAB_STORE_USER) {
1130 : /* We also have user information there */
1131 0 : off += 2 * sizeof(struct track);
1132 :
1133 0 : if (s->flags & SLAB_KMALLOC)
1134 0 : off += sizeof(unsigned int);
1135 : }
1136 :
1137 0 : off += kasan_metadata_size(s, false);
1138 :
1139 0 : if (size_from_object(s) == off)
1140 : return 1;
1141 :
1142 0 : return check_bytes_and_report(s, slab, p, "Object padding",
1143 0 : p + off, POISON_INUSE, size_from_object(s) - off);
1144 : }
1145 :
1146 : /* Check the pad bytes at the end of a slab page */
1147 0 : static void slab_pad_check(struct kmem_cache *s, struct slab *slab)
1148 : {
1149 : u8 *start;
1150 : u8 *fault;
1151 : u8 *end;
1152 : u8 *pad;
1153 : int length;
1154 : int remainder;
1155 :
1156 0 : if (!(s->flags & SLAB_POISON))
1157 : return;
1158 :
1159 0 : start = slab_address(slab);
1160 0 : length = slab_size(slab);
1161 0 : end = start + length;
1162 0 : remainder = length % s->size;
1163 0 : if (!remainder)
1164 : return;
1165 :
1166 0 : pad = end - remainder;
1167 : metadata_access_enable();
1168 0 : fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder);
1169 : metadata_access_disable();
1170 0 : if (!fault)
1171 : return;
1172 0 : while (end > fault && end[-1] == POISON_INUSE)
1173 0 : end--;
1174 :
1175 0 : slab_err(s, slab, "Padding overwritten. 0x%p-0x%p @offset=%tu",
1176 : fault, end - 1, fault - start);
1177 0 : print_section(KERN_ERR, "Padding ", pad, remainder);
1178 :
1179 0 : restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
1180 : }
1181 :
1182 0 : static int check_object(struct kmem_cache *s, struct slab *slab,
1183 : void *object, u8 val)
1184 : {
1185 0 : u8 *p = object;
1186 0 : u8 *endobject = object + s->object_size;
1187 : unsigned int orig_size;
1188 :
1189 0 : if (s->flags & SLAB_RED_ZONE) {
1190 0 : if (!check_bytes_and_report(s, slab, object, "Left Redzone",
1191 0 : object - s->red_left_pad, val, s->red_left_pad))
1192 : return 0;
1193 :
1194 0 : if (!check_bytes_and_report(s, slab, object, "Right Redzone",
1195 0 : endobject, val, s->inuse - s->object_size))
1196 : return 0;
1197 :
1198 0 : if (slub_debug_orig_size(s) && val == SLUB_RED_ACTIVE) {
1199 0 : orig_size = get_orig_size(s, object);
1200 :
1201 0 : if (s->object_size > orig_size &&
1202 0 : !check_bytes_and_report(s, slab, object,
1203 : "kmalloc Redzone", p + orig_size,
1204 : val, s->object_size - orig_size)) {
1205 : return 0;
1206 : }
1207 : }
1208 : } else {
1209 0 : if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
1210 0 : check_bytes_and_report(s, slab, p, "Alignment padding",
1211 : endobject, POISON_INUSE,
1212 : s->inuse - s->object_size);
1213 : }
1214 : }
1215 :
1216 0 : if (s->flags & SLAB_POISON) {
1217 0 : if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
1218 0 : (!check_bytes_and_report(s, slab, p, "Poison", p,
1219 0 : POISON_FREE, s->object_size - 1) ||
1220 0 : !check_bytes_and_report(s, slab, p, "End Poison",
1221 0 : p + s->object_size - 1, POISON_END, 1)))
1222 : return 0;
1223 : /*
1224 : * check_pad_bytes cleans up on its own.
1225 : */
1226 0 : check_pad_bytes(s, slab, p);
1227 : }
1228 :
1229 0 : if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE)
1230 : /*
1231 : * Object and freepointer overlap. Cannot check
1232 : * freepointer while object is allocated.
1233 : */
1234 : return 1;
1235 :
1236 : /* Check free pointer validity */
1237 0 : if (!check_valid_pointer(s, slab, get_freepointer(s, p))) {
1238 0 : object_err(s, slab, p, "Freepointer corrupt");
1239 : /*
1240 : * No choice but to zap it and thus lose the remainder
1241 : * of the free objects in this slab. May cause
1242 : * another error because the object count is now wrong.
1243 : */
1244 0 : set_freepointer(s, p, NULL);
1245 0 : return 0;
1246 : }
1247 : return 1;
1248 : }
1249 :
1250 0 : static int check_slab(struct kmem_cache *s, struct slab *slab)
1251 : {
1252 : int maxobj;
1253 :
1254 0 : if (!folio_test_slab(slab_folio(slab))) {
1255 0 : slab_err(s, slab, "Not a valid slab page");
1256 0 : return 0;
1257 : }
1258 :
1259 0 : maxobj = order_objects(slab_order(slab), s->size);
1260 0 : if (slab->objects > maxobj) {
1261 0 : slab_err(s, slab, "objects %u > max %u",
1262 : slab->objects, maxobj);
1263 0 : return 0;
1264 : }
1265 0 : if (slab->inuse > slab->objects) {
1266 0 : slab_err(s, slab, "inuse %u > max %u",
1267 : slab->inuse, slab->objects);
1268 0 : return 0;
1269 : }
1270 : /* Slab_pad_check fixes things up after itself */
1271 0 : slab_pad_check(s, slab);
1272 0 : return 1;
1273 : }
1274 :
1275 : /*
1276 : * Determine if a certain object in a slab is on the freelist. Must hold the
1277 : * slab lock to guarantee that the chains are in a consistent state.
1278 : */
1279 0 : static int on_freelist(struct kmem_cache *s, struct slab *slab, void *search)
1280 : {
1281 0 : int nr = 0;
1282 : void *fp;
1283 0 : void *object = NULL;
1284 : int max_objects;
1285 :
1286 0 : fp = slab->freelist;
1287 0 : while (fp && nr <= slab->objects) {
1288 0 : if (fp == search)
1289 : return 1;
1290 0 : if (!check_valid_pointer(s, slab, fp)) {
1291 0 : if (object) {
1292 0 : object_err(s, slab, object,
1293 : "Freechain corrupt");
1294 0 : set_freepointer(s, object, NULL);
1295 : } else {
1296 0 : slab_err(s, slab, "Freepointer corrupt");
1297 0 : slab->freelist = NULL;
1298 0 : slab->inuse = slab->objects;
1299 0 : slab_fix(s, "Freelist cleared");
1300 0 : return 0;
1301 : }
1302 : break;
1303 : }
1304 0 : object = fp;
1305 0 : fp = get_freepointer(s, object);
1306 0 : nr++;
1307 : }
1308 :
1309 0 : max_objects = order_objects(slab_order(slab), s->size);
1310 0 : if (max_objects > MAX_OBJS_PER_PAGE)
1311 0 : max_objects = MAX_OBJS_PER_PAGE;
1312 :
1313 0 : if (slab->objects != max_objects) {
1314 0 : slab_err(s, slab, "Wrong number of objects. Found %d but should be %d",
1315 : slab->objects, max_objects);
1316 0 : slab->objects = max_objects;
1317 0 : slab_fix(s, "Number of objects adjusted");
1318 : }
1319 0 : if (slab->inuse != slab->objects - nr) {
1320 0 : slab_err(s, slab, "Wrong object count. Counter is %d but counted were %d",
1321 : slab->inuse, slab->objects - nr);
1322 0 : slab->inuse = slab->objects - nr;
1323 0 : slab_fix(s, "Object count adjusted");
1324 : }
1325 0 : return search == NULL;
1326 : }
1327 :
1328 0 : static void trace(struct kmem_cache *s, struct slab *slab, void *object,
1329 : int alloc)
1330 : {
1331 0 : if (s->flags & SLAB_TRACE) {
1332 0 : pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
1333 : s->name,
1334 : alloc ? "alloc" : "free",
1335 : object, slab->inuse,
1336 : slab->freelist);
1337 :
1338 0 : if (!alloc)
1339 0 : print_section(KERN_INFO, "Object ", (void *)object,
1340 : s->object_size);
1341 :
1342 0 : dump_stack();
1343 : }
1344 0 : }
1345 :
1346 : /*
1347 : * Tracking of fully allocated slabs for debugging purposes.
1348 : */
1349 : static void add_full(struct kmem_cache *s,
1350 : struct kmem_cache_node *n, struct slab *slab)
1351 : {
1352 0 : if (!(s->flags & SLAB_STORE_USER))
1353 : return;
1354 :
1355 : lockdep_assert_held(&n->list_lock);
1356 0 : list_add(&slab->slab_list, &n->full);
1357 : }
1358 :
1359 : static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct slab *slab)
1360 : {
1361 35 : if (!(s->flags & SLAB_STORE_USER))
1362 : return;
1363 :
1364 : lockdep_assert_held(&n->list_lock);
1365 0 : list_del(&slab->slab_list);
1366 : }
1367 :
1368 : static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1369 : {
1370 0 : return atomic_long_read(&n->nr_slabs);
1371 : }
1372 :
1373 : static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
1374 : {
1375 436 : struct kmem_cache_node *n = get_node(s, node);
1376 :
1377 : /*
1378 : * May be called early in order to allocate a slab for the
1379 : * kmem_cache_node structure. Solve the chicken-egg
1380 : * dilemma by deferring the increment of the count during
1381 : * bootstrap (see early_kmem_cache_node_alloc).
1382 : */
1383 436 : if (likely(n)) {
1384 870 : atomic_long_inc(&n->nr_slabs);
1385 435 : atomic_long_add(objects, &n->total_objects);
1386 : }
1387 : }
1388 : static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
1389 : {
1390 0 : struct kmem_cache_node *n = get_node(s, node);
1391 :
1392 0 : atomic_long_dec(&n->nr_slabs);
1393 0 : atomic_long_sub(objects, &n->total_objects);
1394 : }
1395 :
1396 : /* Object debug checks for alloc/free paths */
1397 14462 : static void setup_object_debug(struct kmem_cache *s, void *object)
1398 : {
1399 28924 : if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))
1400 : return;
1401 :
1402 0 : init_object(s, object, SLUB_RED_INACTIVE);
1403 0 : init_tracking(s, object);
1404 : }
1405 :
1406 : static
1407 435 : void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr)
1408 : {
1409 870 : if (!kmem_cache_debug_flags(s, SLAB_POISON))
1410 : return;
1411 :
1412 0 : metadata_access_enable();
1413 0 : memset(kasan_reset_tag(addr), POISON_INUSE, slab_size(slab));
1414 : metadata_access_disable();
1415 : }
1416 :
1417 0 : static inline int alloc_consistency_checks(struct kmem_cache *s,
1418 : struct slab *slab, void *object)
1419 : {
1420 0 : if (!check_slab(s, slab))
1421 : return 0;
1422 :
1423 0 : if (!check_valid_pointer(s, slab, object)) {
1424 0 : object_err(s, slab, object, "Freelist Pointer check fails");
1425 0 : return 0;
1426 : }
1427 :
1428 0 : if (!check_object(s, slab, object, SLUB_RED_INACTIVE))
1429 : return 0;
1430 :
1431 0 : return 1;
1432 : }
1433 :
1434 0 : static noinline bool alloc_debug_processing(struct kmem_cache *s,
1435 : struct slab *slab, void *object, int orig_size)
1436 : {
1437 0 : if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1438 0 : if (!alloc_consistency_checks(s, slab, object))
1439 : goto bad;
1440 : }
1441 :
1442 : /* Success. Perform special debug activities for allocs */
1443 0 : trace(s, slab, object, 1);
1444 0 : set_orig_size(s, object, orig_size);
1445 0 : init_object(s, object, SLUB_RED_ACTIVE);
1446 0 : return true;
1447 :
1448 : bad:
1449 0 : if (folio_test_slab(slab_folio(slab))) {
1450 : /*
1451 : * If this is a slab page then lets do the best we can
1452 : * to avoid issues in the future. Marking all objects
1453 : * as used avoids touching the remaining objects.
1454 : */
1455 0 : slab_fix(s, "Marking all objects used");
1456 0 : slab->inuse = slab->objects;
1457 0 : slab->freelist = NULL;
1458 : }
1459 : return false;
1460 : }
1461 :
1462 0 : static inline int free_consistency_checks(struct kmem_cache *s,
1463 : struct slab *slab, void *object, unsigned long addr)
1464 : {
1465 0 : if (!check_valid_pointer(s, slab, object)) {
1466 0 : slab_err(s, slab, "Invalid object pointer 0x%p", object);
1467 : return 0;
1468 : }
1469 :
1470 0 : if (on_freelist(s, slab, object)) {
1471 0 : object_err(s, slab, object, "Object already free");
1472 : return 0;
1473 : }
1474 :
1475 0 : if (!check_object(s, slab, object, SLUB_RED_ACTIVE))
1476 : return 0;
1477 :
1478 0 : if (unlikely(s != slab->slab_cache)) {
1479 0 : if (!folio_test_slab(slab_folio(slab))) {
1480 0 : slab_err(s, slab, "Attempt to free object(0x%p) outside of slab",
1481 : object);
1482 0 : } else if (!slab->slab_cache) {
1483 0 : pr_err("SLUB <none>: no slab for object 0x%p.\n",
1484 : object);
1485 0 : dump_stack();
1486 : } else
1487 0 : object_err(s, slab, object,
1488 : "page slab pointer corrupt.");
1489 : return 0;
1490 : }
1491 : return 1;
1492 : }
1493 :
1494 : /*
1495 : * Parse a block of slub_debug options. Blocks are delimited by ';'
1496 : *
1497 : * @str: start of block
1498 : * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified
1499 : * @slabs: return start of list of slabs, or NULL when there's no list
1500 : * @init: assume this is initial parsing and not per-kmem-create parsing
1501 : *
1502 : * returns the start of next block if there's any, or NULL
1503 : */
1504 : static char *
1505 0 : parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init)
1506 : {
1507 0 : bool higher_order_disable = false;
1508 :
1509 : /* Skip any completely empty blocks */
1510 0 : while (*str && *str == ';')
1511 0 : str++;
1512 :
1513 0 : if (*str == ',') {
1514 : /*
1515 : * No options but restriction on slabs. This means full
1516 : * debugging for slabs matching a pattern.
1517 : */
1518 0 : *flags = DEBUG_DEFAULT_FLAGS;
1519 0 : goto check_slabs;
1520 : }
1521 0 : *flags = 0;
1522 :
1523 : /* Determine which debug features should be switched on */
1524 0 : for (; *str && *str != ',' && *str != ';'; str++) {
1525 0 : switch (tolower(*str)) {
1526 : case '-':
1527 0 : *flags = 0;
1528 0 : break;
1529 : case 'f':
1530 0 : *flags |= SLAB_CONSISTENCY_CHECKS;
1531 0 : break;
1532 : case 'z':
1533 0 : *flags |= SLAB_RED_ZONE;
1534 0 : break;
1535 : case 'p':
1536 0 : *flags |= SLAB_POISON;
1537 0 : break;
1538 : case 'u':
1539 0 : *flags |= SLAB_STORE_USER;
1540 0 : break;
1541 : case 't':
1542 0 : *flags |= SLAB_TRACE;
1543 0 : break;
1544 : case 'a':
1545 : *flags |= SLAB_FAILSLAB;
1546 0 : break;
1547 : case 'o':
1548 : /*
1549 : * Avoid enabling debugging on caches if its minimum
1550 : * order would increase as a result.
1551 : */
1552 : higher_order_disable = true;
1553 : break;
1554 : default:
1555 0 : if (init)
1556 0 : pr_err("slub_debug option '%c' unknown. skipped\n", *str);
1557 : }
1558 : }
1559 : check_slabs:
1560 0 : if (*str == ',')
1561 0 : *slabs = ++str;
1562 : else
1563 0 : *slabs = NULL;
1564 :
1565 : /* Skip over the slab list */
1566 0 : while (*str && *str != ';')
1567 0 : str++;
1568 :
1569 : /* Skip any completely empty blocks */
1570 0 : while (*str && *str == ';')
1571 0 : str++;
1572 :
1573 0 : if (init && higher_order_disable)
1574 0 : disable_higher_order_debug = 1;
1575 :
1576 0 : if (*str)
1577 : return str;
1578 : else
1579 0 : return NULL;
1580 : }
1581 :
1582 0 : static int __init setup_slub_debug(char *str)
1583 : {
1584 : slab_flags_t flags;
1585 : slab_flags_t global_flags;
1586 : char *saved_str;
1587 : char *slab_list;
1588 0 : bool global_slub_debug_changed = false;
1589 0 : bool slab_list_specified = false;
1590 :
1591 0 : global_flags = DEBUG_DEFAULT_FLAGS;
1592 0 : if (*str++ != '=' || !*str)
1593 : /*
1594 : * No options specified. Switch on full debugging.
1595 : */
1596 : goto out;
1597 :
1598 : saved_str = str;
1599 0 : while (str) {
1600 0 : str = parse_slub_debug_flags(str, &flags, &slab_list, true);
1601 :
1602 0 : if (!slab_list) {
1603 0 : global_flags = flags;
1604 0 : global_slub_debug_changed = true;
1605 : } else {
1606 0 : slab_list_specified = true;
1607 0 : if (flags & SLAB_STORE_USER)
1608 0 : stack_depot_request_early_init();
1609 : }
1610 : }
1611 :
1612 : /*
1613 : * For backwards compatibility, a single list of flags with list of
1614 : * slabs means debugging is only changed for those slabs, so the global
1615 : * slub_debug should be unchanged (0 or DEBUG_DEFAULT_FLAGS, depending
1616 : * on CONFIG_SLUB_DEBUG_ON). We can extended that to multiple lists as
1617 : * long as there is no option specifying flags without a slab list.
1618 : */
1619 0 : if (slab_list_specified) {
1620 0 : if (!global_slub_debug_changed)
1621 0 : global_flags = slub_debug;
1622 0 : slub_debug_string = saved_str;
1623 : }
1624 : out:
1625 0 : slub_debug = global_flags;
1626 0 : if (slub_debug & SLAB_STORE_USER)
1627 0 : stack_depot_request_early_init();
1628 0 : if (slub_debug != 0 || slub_debug_string)
1629 0 : static_branch_enable(&slub_debug_enabled);
1630 : else
1631 0 : static_branch_disable(&slub_debug_enabled);
1632 0 : if ((static_branch_unlikely(&init_on_alloc) ||
1633 0 : static_branch_unlikely(&init_on_free)) &&
1634 0 : (slub_debug & SLAB_POISON))
1635 0 : pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n");
1636 0 : return 1;
1637 : }
1638 :
1639 : __setup("slub_debug", setup_slub_debug);
1640 :
1641 : /*
1642 : * kmem_cache_flags - apply debugging options to the cache
1643 : * @object_size: the size of an object without meta data
1644 : * @flags: flags to set
1645 : * @name: name of the cache
1646 : *
1647 : * Debug option(s) are applied to @flags. In addition to the debug
1648 : * option(s), if a slab name (or multiple) is specified i.e.
1649 : * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ...
1650 : * then only the select slabs will receive the debug option(s).
1651 : */
1652 103 : slab_flags_t kmem_cache_flags(unsigned int object_size,
1653 : slab_flags_t flags, const char *name)
1654 : {
1655 : char *iter;
1656 : size_t len;
1657 : char *next_block;
1658 : slab_flags_t block_flags;
1659 103 : slab_flags_t slub_debug_local = slub_debug;
1660 :
1661 103 : if (flags & SLAB_NO_USER_FLAGS)
1662 : return flags;
1663 :
1664 : /*
1665 : * If the slab cache is for debugging (e.g. kmemleak) then
1666 : * don't store user (stack trace) information by default,
1667 : * but let the user enable it via the command line below.
1668 : */
1669 103 : if (flags & SLAB_NOLEAKTRACE)
1670 0 : slub_debug_local &= ~SLAB_STORE_USER;
1671 :
1672 103 : len = strlen(name);
1673 103 : next_block = slub_debug_string;
1674 : /* Go through all blocks of debug options, see if any matches our slab's name */
1675 206 : while (next_block) {
1676 0 : next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false);
1677 0 : if (!iter)
1678 0 : continue;
1679 : /* Found a block that has a slab list, search it */
1680 0 : while (*iter) {
1681 : char *end, *glob;
1682 : size_t cmplen;
1683 :
1684 0 : end = strchrnul(iter, ',');
1685 0 : if (next_block && next_block < end)
1686 0 : end = next_block - 1;
1687 :
1688 0 : glob = strnchr(iter, end - iter, '*');
1689 0 : if (glob)
1690 0 : cmplen = glob - iter;
1691 : else
1692 0 : cmplen = max_t(size_t, len, (end - iter));
1693 :
1694 0 : if (!strncmp(name, iter, cmplen)) {
1695 0 : flags |= block_flags;
1696 0 : return flags;
1697 : }
1698 :
1699 0 : if (!*end || *end == ';')
1700 : break;
1701 0 : iter = end + 1;
1702 : }
1703 : }
1704 :
1705 103 : return flags | slub_debug_local;
1706 : }
1707 : #else /* !CONFIG_SLUB_DEBUG */
1708 : static inline void setup_object_debug(struct kmem_cache *s, void *object) {}
1709 : static inline
1710 : void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr) {}
1711 :
1712 : static inline bool alloc_debug_processing(struct kmem_cache *s,
1713 : struct slab *slab, void *object, int orig_size) { return true; }
1714 :
1715 : static inline bool free_debug_processing(struct kmem_cache *s,
1716 : struct slab *slab, void *head, void *tail, int *bulk_cnt,
1717 : unsigned long addr, depot_stack_handle_t handle) { return true; }
1718 :
1719 : static inline void slab_pad_check(struct kmem_cache *s, struct slab *slab) {}
1720 : static inline int check_object(struct kmem_cache *s, struct slab *slab,
1721 : void *object, u8 val) { return 1; }
1722 : static inline depot_stack_handle_t set_track_prepare(void) { return 0; }
1723 : static inline void set_track(struct kmem_cache *s, void *object,
1724 : enum track_item alloc, unsigned long addr) {}
1725 : static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1726 : struct slab *slab) {}
1727 : static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1728 : struct slab *slab) {}
1729 : slab_flags_t kmem_cache_flags(unsigned int object_size,
1730 : slab_flags_t flags, const char *name)
1731 : {
1732 : return flags;
1733 : }
1734 : #define slub_debug 0
1735 :
1736 : #define disable_higher_order_debug 0
1737 :
1738 : static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1739 : { return 0; }
1740 : static inline void inc_slabs_node(struct kmem_cache *s, int node,
1741 : int objects) {}
1742 : static inline void dec_slabs_node(struct kmem_cache *s, int node,
1743 : int objects) {}
1744 :
1745 : #ifndef CONFIG_SLUB_TINY
1746 : static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab,
1747 : void **freelist, void *nextfree)
1748 : {
1749 : return false;
1750 : }
1751 : #endif
1752 : #endif /* CONFIG_SLUB_DEBUG */
1753 :
1754 : /*
1755 : * Hooks for other subsystems that check memory allocations. In a typical
1756 : * production configuration these hooks all should produce no code at all.
1757 : */
1758 : static __always_inline bool slab_free_hook(struct kmem_cache *s,
1759 : void *x, bool init)
1760 : {
1761 4423 : kmemleak_free_recursive(x, s->flags);
1762 4423 : kmsan_slab_free(s, x);
1763 :
1764 4423 : debug_check_no_locks_freed(x, s->object_size);
1765 :
1766 : if (!(s->flags & SLAB_DEBUG_OBJECTS))
1767 4423 : debug_check_no_obj_freed(x, s->object_size);
1768 :
1769 : /* Use KCSAN to help debug racy use-after-free. */
1770 : if (!(s->flags & SLAB_TYPESAFE_BY_RCU))
1771 : __kcsan_check_access(x, s->object_size,
1772 : KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT);
1773 :
1774 : /*
1775 : * As memory initialization might be integrated into KASAN,
1776 : * kasan_slab_free and initialization memset's must be
1777 : * kept together to avoid discrepancies in behavior.
1778 : *
1779 : * The initialization memset's clear the object and the metadata,
1780 : * but don't touch the SLAB redzone.
1781 : */
1782 4423 : if (init) {
1783 : int rsize;
1784 :
1785 : if (!kasan_has_integrated_init())
1786 0 : memset(kasan_reset_tag(x), 0, s->object_size);
1787 0 : rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0;
1788 0 : memset((char *)kasan_reset_tag(x) + s->inuse, 0,
1789 : s->size - s->inuse - rsize);
1790 : }
1791 : /* KASAN might put x into memory quarantine, delaying its reuse. */
1792 4423 : return kasan_slab_free(s, x, init);
1793 : }
1794 :
1795 4423 : static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1796 : void **head, void **tail,
1797 : int *cnt)
1798 : {
1799 :
1800 : void *object;
1801 4423 : void *next = *head;
1802 4423 : void *old_tail = *tail ? *tail : *head;
1803 :
1804 4423 : if (is_kfence_address(next)) {
1805 : slab_free_hook(s, next, false);
1806 : return true;
1807 : }
1808 :
1809 : /* Head and tail of the reconstructed freelist */
1810 4423 : *head = NULL;
1811 4423 : *tail = NULL;
1812 :
1813 : do {
1814 4423 : object = next;
1815 8846 : next = get_freepointer(s, object);
1816 :
1817 : /* If object's reuse doesn't have to be delayed */
1818 13269 : if (!slab_free_hook(s, object, slab_want_init_on_free(s))) {
1819 : /* Move object to the new freelist */
1820 8846 : set_freepointer(s, object, *head);
1821 4423 : *head = object;
1822 4423 : if (!*tail)
1823 4423 : *tail = object;
1824 : } else {
1825 : /*
1826 : * Adjust the reconstructed freelist depth
1827 : * accordingly if object's reuse is delayed.
1828 : */
1829 : --(*cnt);
1830 : }
1831 4423 : } while (object != old_tail);
1832 :
1833 4423 : if (*head == *tail)
1834 4423 : *tail = NULL;
1835 :
1836 4423 : return *head != NULL;
1837 : }
1838 :
1839 : static void *setup_object(struct kmem_cache *s, void *object)
1840 : {
1841 14462 : setup_object_debug(s, object);
1842 14462 : object = kasan_init_slab_obj(s, object);
1843 14462 : if (unlikely(s->ctor)) {
1844 260 : kasan_unpoison_object_data(s, object);
1845 260 : s->ctor(object);
1846 260 : kasan_poison_object_data(s, object);
1847 : }
1848 : return object;
1849 : }
1850 :
1851 : /*
1852 : * Slab allocation and freeing
1853 : */
1854 435 : static inline struct slab *alloc_slab_page(gfp_t flags, int node,
1855 : struct kmem_cache_order_objects oo)
1856 : {
1857 : struct folio *folio;
1858 : struct slab *slab;
1859 435 : unsigned int order = oo_order(oo);
1860 :
1861 435 : if (node == NUMA_NO_NODE)
1862 434 : folio = (struct folio *)alloc_pages(flags, order);
1863 : else
1864 1 : folio = (struct folio *)__alloc_pages_node(node, flags, order);
1865 :
1866 435 : if (!folio)
1867 : return NULL;
1868 :
1869 435 : slab = folio_slab(folio);
1870 435 : __folio_set_slab(folio);
1871 : /* Make the flag visible before any changes to folio->mapping */
1872 435 : smp_wmb();
1873 870 : if (folio_is_pfmemalloc(folio))
1874 : slab_set_pfmemalloc(slab);
1875 :
1876 : return slab;
1877 : }
1878 :
1879 : #ifdef CONFIG_SLAB_FREELIST_RANDOM
1880 : /* Pre-initialize the random sequence cache */
1881 : static int init_cache_random_seq(struct kmem_cache *s)
1882 : {
1883 : unsigned int count = oo_objects(s->oo);
1884 : int err;
1885 :
1886 : /* Bailout if already initialised */
1887 : if (s->random_seq)
1888 : return 0;
1889 :
1890 : err = cache_random_seq_create(s, count, GFP_KERNEL);
1891 : if (err) {
1892 : pr_err("SLUB: Unable to initialize free list for %s\n",
1893 : s->name);
1894 : return err;
1895 : }
1896 :
1897 : /* Transform to an offset on the set of pages */
1898 : if (s->random_seq) {
1899 : unsigned int i;
1900 :
1901 : for (i = 0; i < count; i++)
1902 : s->random_seq[i] *= s->size;
1903 : }
1904 : return 0;
1905 : }
1906 :
1907 : /* Initialize each random sequence freelist per cache */
1908 : static void __init init_freelist_randomization(void)
1909 : {
1910 : struct kmem_cache *s;
1911 :
1912 : mutex_lock(&slab_mutex);
1913 :
1914 : list_for_each_entry(s, &slab_caches, list)
1915 : init_cache_random_seq(s);
1916 :
1917 : mutex_unlock(&slab_mutex);
1918 : }
1919 :
1920 : /* Get the next entry on the pre-computed freelist randomized */
1921 : static void *next_freelist_entry(struct kmem_cache *s, struct slab *slab,
1922 : unsigned long *pos, void *start,
1923 : unsigned long page_limit,
1924 : unsigned long freelist_count)
1925 : {
1926 : unsigned int idx;
1927 :
1928 : /*
1929 : * If the target page allocation failed, the number of objects on the
1930 : * page might be smaller than the usual size defined by the cache.
1931 : */
1932 : do {
1933 : idx = s->random_seq[*pos];
1934 : *pos += 1;
1935 : if (*pos >= freelist_count)
1936 : *pos = 0;
1937 : } while (unlikely(idx >= page_limit));
1938 :
1939 : return (char *)start + idx;
1940 : }
1941 :
1942 : /* Shuffle the single linked freelist based on a random pre-computed sequence */
1943 : static bool shuffle_freelist(struct kmem_cache *s, struct slab *slab)
1944 : {
1945 : void *start;
1946 : void *cur;
1947 : void *next;
1948 : unsigned long idx, pos, page_limit, freelist_count;
1949 :
1950 : if (slab->objects < 2 || !s->random_seq)
1951 : return false;
1952 :
1953 : freelist_count = oo_objects(s->oo);
1954 : pos = get_random_u32_below(freelist_count);
1955 :
1956 : page_limit = slab->objects * s->size;
1957 : start = fixup_red_left(s, slab_address(slab));
1958 :
1959 : /* First entry is used as the base of the freelist */
1960 : cur = next_freelist_entry(s, slab, &pos, start, page_limit,
1961 : freelist_count);
1962 : cur = setup_object(s, cur);
1963 : slab->freelist = cur;
1964 :
1965 : for (idx = 1; idx < slab->objects; idx++) {
1966 : next = next_freelist_entry(s, slab, &pos, start, page_limit,
1967 : freelist_count);
1968 : next = setup_object(s, next);
1969 : set_freepointer(s, cur, next);
1970 : cur = next;
1971 : }
1972 : set_freepointer(s, cur, NULL);
1973 :
1974 : return true;
1975 : }
1976 : #else
1977 : static inline int init_cache_random_seq(struct kmem_cache *s)
1978 : {
1979 : return 0;
1980 : }
1981 : static inline void init_freelist_randomization(void) { }
1982 : static inline bool shuffle_freelist(struct kmem_cache *s, struct slab *slab)
1983 : {
1984 : return false;
1985 : }
1986 : #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1987 :
1988 435 : static struct slab *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1989 : {
1990 : struct slab *slab;
1991 435 : struct kmem_cache_order_objects oo = s->oo;
1992 : gfp_t alloc_gfp;
1993 : void *start, *p, *next;
1994 : int idx;
1995 : bool shuffle;
1996 :
1997 435 : flags &= gfp_allowed_mask;
1998 :
1999 435 : flags |= s->allocflags;
2000 :
2001 : /*
2002 : * Let the initial higher-order allocation fail under memory pressure
2003 : * so we fall-back to the minimum order allocation.
2004 : */
2005 435 : alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
2006 832 : if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
2007 76 : alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_RECLAIM;
2008 :
2009 435 : slab = alloc_slab_page(alloc_gfp, node, oo);
2010 435 : if (unlikely(!slab)) {
2011 0 : oo = s->min;
2012 0 : alloc_gfp = flags;
2013 : /*
2014 : * Allocation may have failed due to fragmentation.
2015 : * Try a lower order alloc if possible
2016 : */
2017 0 : slab = alloc_slab_page(alloc_gfp, node, oo);
2018 0 : if (unlikely(!slab))
2019 : return NULL;
2020 : stat(s, ORDER_FALLBACK);
2021 : }
2022 :
2023 435 : slab->objects = oo_objects(oo);
2024 435 : slab->inuse = 0;
2025 435 : slab->frozen = 0;
2026 :
2027 870 : account_slab(slab, oo_order(oo), s, flags);
2028 :
2029 435 : slab->slab_cache = s;
2030 :
2031 435 : kasan_poison_slab(slab);
2032 :
2033 435 : start = slab_address(slab);
2034 :
2035 435 : setup_slab_debug(s, slab, start);
2036 :
2037 435 : shuffle = shuffle_freelist(s, slab);
2038 :
2039 : if (!shuffle) {
2040 435 : start = fixup_red_left(s, start);
2041 435 : start = setup_object(s, start);
2042 435 : slab->freelist = start;
2043 14462 : for (idx = 0, p = start; idx < slab->objects - 1; idx++) {
2044 14027 : next = p + s->size;
2045 14027 : next = setup_object(s, next);
2046 28054 : set_freepointer(s, p, next);
2047 14027 : p = next;
2048 : }
2049 435 : set_freepointer(s, p, NULL);
2050 : }
2051 :
2052 435 : return slab;
2053 : }
2054 :
2055 435 : static struct slab *new_slab(struct kmem_cache *s, gfp_t flags, int node)
2056 : {
2057 435 : if (unlikely(flags & GFP_SLAB_BUG_MASK))
2058 0 : flags = kmalloc_fix_flags(flags);
2059 :
2060 435 : WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2061 :
2062 435 : return allocate_slab(s,
2063 : flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
2064 : }
2065 :
2066 0 : static void __free_slab(struct kmem_cache *s, struct slab *slab)
2067 : {
2068 0 : struct folio *folio = slab_folio(slab);
2069 0 : int order = folio_order(folio);
2070 0 : int pages = 1 << order;
2071 :
2072 0 : __slab_clear_pfmemalloc(slab);
2073 0 : folio->mapping = NULL;
2074 : /* Make the mapping reset visible before clearing the flag */
2075 0 : smp_wmb();
2076 0 : __folio_clear_slab(folio);
2077 0 : mm_account_reclaimed_pages(pages);
2078 0 : unaccount_slab(slab, order, s);
2079 0 : __free_pages(&folio->page, order);
2080 0 : }
2081 :
2082 0 : static void rcu_free_slab(struct rcu_head *h)
2083 : {
2084 0 : struct slab *slab = container_of(h, struct slab, rcu_head);
2085 :
2086 0 : __free_slab(slab->slab_cache, slab);
2087 0 : }
2088 :
2089 0 : static void free_slab(struct kmem_cache *s, struct slab *slab)
2090 : {
2091 0 : if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) {
2092 : void *p;
2093 :
2094 0 : slab_pad_check(s, slab);
2095 0 : for_each_object(p, s, slab_address(slab), slab->objects)
2096 0 : check_object(s, slab, p, SLUB_RED_INACTIVE);
2097 : }
2098 :
2099 0 : if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU))
2100 0 : call_rcu(&slab->rcu_head, rcu_free_slab);
2101 : else
2102 0 : __free_slab(s, slab);
2103 0 : }
2104 :
2105 : static void discard_slab(struct kmem_cache *s, struct slab *slab)
2106 : {
2107 0 : dec_slabs_node(s, slab_nid(slab), slab->objects);
2108 0 : free_slab(s, slab);
2109 : }
2110 :
2111 : /*
2112 : * Management of partially allocated slabs.
2113 : */
2114 : static inline void
2115 : __add_partial(struct kmem_cache_node *n, struct slab *slab, int tail)
2116 : {
2117 38 : n->nr_partial++;
2118 2 : if (tail == DEACTIVATE_TO_TAIL)
2119 35 : list_add_tail(&slab->slab_list, &n->partial);
2120 : else
2121 3 : list_add(&slab->slab_list, &n->partial);
2122 : }
2123 :
2124 : static inline void add_partial(struct kmem_cache_node *n,
2125 : struct slab *slab, int tail)
2126 : {
2127 : lockdep_assert_held(&n->list_lock);
2128 2 : __add_partial(n, slab, tail);
2129 : }
2130 :
2131 : static inline void remove_partial(struct kmem_cache_node *n,
2132 : struct slab *slab)
2133 : {
2134 : lockdep_assert_held(&n->list_lock);
2135 56 : list_del(&slab->slab_list);
2136 28 : n->nr_partial--;
2137 : }
2138 :
2139 : /*
2140 : * Called only for kmem_cache_debug() caches instead of acquire_slab(), with a
2141 : * slab from the n->partial list. Remove only a single object from the slab, do
2142 : * the alloc_debug_processing() checks and leave the slab on the list, or move
2143 : * it to full list if it was the last free object.
2144 : */
2145 0 : static void *alloc_single_from_partial(struct kmem_cache *s,
2146 : struct kmem_cache_node *n, struct slab *slab, int orig_size)
2147 : {
2148 : void *object;
2149 :
2150 : lockdep_assert_held(&n->list_lock);
2151 :
2152 0 : object = slab->freelist;
2153 0 : slab->freelist = get_freepointer(s, object);
2154 0 : slab->inuse++;
2155 :
2156 0 : if (!alloc_debug_processing(s, slab, object, orig_size)) {
2157 0 : remove_partial(n, slab);
2158 0 : return NULL;
2159 : }
2160 :
2161 0 : if (slab->inuse == slab->objects) {
2162 0 : remove_partial(n, slab);
2163 0 : add_full(s, n, slab);
2164 : }
2165 :
2166 : return object;
2167 : }
2168 :
2169 : /*
2170 : * Called only for kmem_cache_debug() caches to allocate from a freshly
2171 : * allocated slab. Allocate a single object instead of whole freelist
2172 : * and put the slab to the partial (or full) list.
2173 : */
2174 0 : static void *alloc_single_from_new_slab(struct kmem_cache *s,
2175 : struct slab *slab, int orig_size)
2176 : {
2177 0 : int nid = slab_nid(slab);
2178 0 : struct kmem_cache_node *n = get_node(s, nid);
2179 : unsigned long flags;
2180 : void *object;
2181 :
2182 :
2183 0 : object = slab->freelist;
2184 0 : slab->freelist = get_freepointer(s, object);
2185 0 : slab->inuse = 1;
2186 :
2187 0 : if (!alloc_debug_processing(s, slab, object, orig_size))
2188 : /*
2189 : * It's not really expected that this would fail on a
2190 : * freshly allocated slab, but a concurrent memory
2191 : * corruption in theory could cause that.
2192 : */
2193 : return NULL;
2194 :
2195 0 : spin_lock_irqsave(&n->list_lock, flags);
2196 :
2197 0 : if (slab->inuse == slab->objects)
2198 0 : add_full(s, n, slab);
2199 : else
2200 : add_partial(n, slab, DEACTIVATE_TO_HEAD);
2201 :
2202 0 : inc_slabs_node(s, nid, slab->objects);
2203 0 : spin_unlock_irqrestore(&n->list_lock, flags);
2204 :
2205 0 : return object;
2206 : }
2207 :
2208 : /*
2209 : * Remove slab from the partial list, freeze it and
2210 : * return the pointer to the freelist.
2211 : *
2212 : * Returns a list of objects or NULL if it fails.
2213 : */
2214 28 : static inline void *acquire_slab(struct kmem_cache *s,
2215 : struct kmem_cache_node *n, struct slab *slab,
2216 : int mode)
2217 : {
2218 : void *freelist;
2219 : unsigned long counters;
2220 : struct slab new;
2221 :
2222 : lockdep_assert_held(&n->list_lock);
2223 :
2224 : /*
2225 : * Zap the freelist and set the frozen bit.
2226 : * The old freelist is the list of objects for the
2227 : * per cpu allocation list.
2228 : */
2229 28 : freelist = slab->freelist;
2230 28 : counters = slab->counters;
2231 28 : new.counters = counters;
2232 28 : if (mode) {
2233 28 : new.inuse = slab->objects;
2234 28 : new.freelist = NULL;
2235 : } else {
2236 : new.freelist = freelist;
2237 : }
2238 :
2239 : VM_BUG_ON(new.frozen);
2240 28 : new.frozen = 1;
2241 :
2242 56 : if (!__slab_update_freelist(s, slab,
2243 : freelist, counters,
2244 : new.freelist, new.counters,
2245 : "acquire_slab"))
2246 : return NULL;
2247 :
2248 56 : remove_partial(n, slab);
2249 28 : WARN_ON(!freelist);
2250 : return freelist;
2251 : }
2252 :
2253 : #ifdef CONFIG_SLUB_CPU_PARTIAL
2254 : static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain);
2255 : #else
2256 : static inline void put_cpu_partial(struct kmem_cache *s, struct slab *slab,
2257 : int drain) { }
2258 : #endif
2259 : static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags);
2260 :
2261 : /*
2262 : * Try to allocate a partial slab from a specific node.
2263 : */
2264 462 : static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
2265 : struct partial_context *pc)
2266 : {
2267 : struct slab *slab, *slab2;
2268 462 : void *object = NULL;
2269 : unsigned long flags;
2270 462 : unsigned int partial_slabs = 0;
2271 :
2272 : /*
2273 : * Racy check. If we mistakenly see no partial slabs then we
2274 : * just allocate an empty slab. If we mistakenly try to get a
2275 : * partial slab and there is none available then get_partial()
2276 : * will return NULL.
2277 : */
2278 462 : if (!n || !n->nr_partial)
2279 : return NULL;
2280 :
2281 28 : spin_lock_irqsave(&n->list_lock, flags);
2282 28 : list_for_each_entry_safe(slab, slab2, &n->partial, slab_list) {
2283 : void *t;
2284 :
2285 56 : if (!pfmemalloc_match(slab, pc->flags))
2286 0 : continue;
2287 :
2288 28 : if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) {
2289 0 : object = alloc_single_from_partial(s, n, slab,
2290 0 : pc->orig_size);
2291 0 : if (object)
2292 : break;
2293 0 : continue;
2294 : }
2295 :
2296 28 : t = acquire_slab(s, n, slab, object == NULL);
2297 28 : if (!t)
2298 : break;
2299 :
2300 28 : if (!object) {
2301 28 : *pc->slab = slab;
2302 28 : stat(s, ALLOC_FROM_PARTIAL);
2303 28 : object = t;
2304 : } else {
2305 : put_cpu_partial(s, slab, 0);
2306 : stat(s, CPU_PARTIAL_NODE);
2307 : partial_slabs++;
2308 : }
2309 : #ifdef CONFIG_SLUB_CPU_PARTIAL
2310 : if (!kmem_cache_has_cpu_partial(s)
2311 : || partial_slabs > s->cpu_partial_slabs / 2)
2312 : break;
2313 : #else
2314 : break;
2315 : #endif
2316 :
2317 : }
2318 56 : spin_unlock_irqrestore(&n->list_lock, flags);
2319 28 : return object;
2320 : }
2321 :
2322 : /*
2323 : * Get a slab from somewhere. Search in increasing NUMA distances.
2324 : */
2325 : static void *get_any_partial(struct kmem_cache *s, struct partial_context *pc)
2326 : {
2327 : #ifdef CONFIG_NUMA
2328 : struct zonelist *zonelist;
2329 : struct zoneref *z;
2330 : struct zone *zone;
2331 : enum zone_type highest_zoneidx = gfp_zone(pc->flags);
2332 : void *object;
2333 : unsigned int cpuset_mems_cookie;
2334 :
2335 : /*
2336 : * The defrag ratio allows a configuration of the tradeoffs between
2337 : * inter node defragmentation and node local allocations. A lower
2338 : * defrag_ratio increases the tendency to do local allocations
2339 : * instead of attempting to obtain partial slabs from other nodes.
2340 : *
2341 : * If the defrag_ratio is set to 0 then kmalloc() always
2342 : * returns node local objects. If the ratio is higher then kmalloc()
2343 : * may return off node objects because partial slabs are obtained
2344 : * from other nodes and filled up.
2345 : *
2346 : * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
2347 : * (which makes defrag_ratio = 1000) then every (well almost)
2348 : * allocation will first attempt to defrag slab caches on other nodes.
2349 : * This means scanning over all nodes to look for partial slabs which
2350 : * may be expensive if we do it every time we are trying to find a slab
2351 : * with available objects.
2352 : */
2353 : if (!s->remote_node_defrag_ratio ||
2354 : get_cycles() % 1024 > s->remote_node_defrag_ratio)
2355 : return NULL;
2356 :
2357 : do {
2358 : cpuset_mems_cookie = read_mems_allowed_begin();
2359 : zonelist = node_zonelist(mempolicy_slab_node(), pc->flags);
2360 : for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) {
2361 : struct kmem_cache_node *n;
2362 :
2363 : n = get_node(s, zone_to_nid(zone));
2364 :
2365 : if (n && cpuset_zone_allowed(zone, pc->flags) &&
2366 : n->nr_partial > s->min_partial) {
2367 : object = get_partial_node(s, n, pc);
2368 : if (object) {
2369 : /*
2370 : * Don't check read_mems_allowed_retry()
2371 : * here - if mems_allowed was updated in
2372 : * parallel, that was a harmless race
2373 : * between allocation and the cpuset
2374 : * update
2375 : */
2376 : return object;
2377 : }
2378 : }
2379 : }
2380 : } while (read_mems_allowed_retry(cpuset_mems_cookie));
2381 : #endif /* CONFIG_NUMA */
2382 : return NULL;
2383 : }
2384 :
2385 : /*
2386 : * Get a partial slab, lock it and return it.
2387 : */
2388 462 : static void *get_partial(struct kmem_cache *s, int node, struct partial_context *pc)
2389 : {
2390 : void *object;
2391 462 : int searchnode = node;
2392 :
2393 462 : if (node == NUMA_NO_NODE)
2394 460 : searchnode = numa_mem_id();
2395 :
2396 462 : object = get_partial_node(s, get_node(s, searchnode), pc);
2397 462 : if (object || node != NUMA_NO_NODE)
2398 : return object;
2399 :
2400 434 : return get_any_partial(s, pc);
2401 : }
2402 :
2403 : #ifndef CONFIG_SLUB_TINY
2404 :
2405 : #ifdef CONFIG_PREEMPTION
2406 : /*
2407 : * Calculate the next globally unique transaction for disambiguation
2408 : * during cmpxchg. The transactions start with the cpu number and are then
2409 : * incremented by CONFIG_NR_CPUS.
2410 : */
2411 : #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
2412 : #else
2413 : /*
2414 : * No preemption supported therefore also no need to check for
2415 : * different cpus.
2416 : */
2417 : #define TID_STEP 1
2418 : #endif /* CONFIG_PREEMPTION */
2419 :
2420 : static inline unsigned long next_tid(unsigned long tid)
2421 : {
2422 21587 : return tid + TID_STEP;
2423 : }
2424 :
2425 : #ifdef SLUB_DEBUG_CMPXCHG
2426 : static inline unsigned int tid_to_cpu(unsigned long tid)
2427 : {
2428 : return tid % TID_STEP;
2429 : }
2430 :
2431 : static inline unsigned long tid_to_event(unsigned long tid)
2432 : {
2433 : return tid / TID_STEP;
2434 : }
2435 : #endif
2436 :
2437 : static inline unsigned int init_tid(int cpu)
2438 : {
2439 53 : return cpu;
2440 : }
2441 :
2442 : static inline void note_cmpxchg_failure(const char *n,
2443 : const struct kmem_cache *s, unsigned long tid)
2444 : {
2445 : #ifdef SLUB_DEBUG_CMPXCHG
2446 : unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
2447 :
2448 : pr_info("%s %s: cmpxchg redo ", n, s->name);
2449 :
2450 : #ifdef CONFIG_PREEMPTION
2451 : if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
2452 : pr_warn("due to cpu change %d -> %d\n",
2453 : tid_to_cpu(tid), tid_to_cpu(actual_tid));
2454 : else
2455 : #endif
2456 : if (tid_to_event(tid) != tid_to_event(actual_tid))
2457 : pr_warn("due to cpu running other code. Event %ld->%ld\n",
2458 : tid_to_event(tid), tid_to_event(actual_tid));
2459 : else
2460 : pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
2461 : actual_tid, tid, next_tid(tid));
2462 : #endif
2463 : stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
2464 : }
2465 :
2466 : static void init_kmem_cache_cpus(struct kmem_cache *s)
2467 : {
2468 : int cpu;
2469 : struct kmem_cache_cpu *c;
2470 :
2471 53 : for_each_possible_cpu(cpu) {
2472 53 : c = per_cpu_ptr(s->cpu_slab, cpu);
2473 53 : local_lock_init(&c->lock);
2474 53 : c->tid = init_tid(cpu);
2475 : }
2476 : }
2477 :
2478 : /*
2479 : * Finishes removing the cpu slab. Merges cpu's freelist with slab's freelist,
2480 : * unfreezes the slabs and puts it on the proper list.
2481 : * Assumes the slab has been already safely taken away from kmem_cache_cpu
2482 : * by the caller.
2483 : */
2484 2 : static void deactivate_slab(struct kmem_cache *s, struct slab *slab,
2485 : void *freelist)
2486 : {
2487 : enum slab_modes { M_NONE, M_PARTIAL, M_FREE, M_FULL_NOLIST };
2488 6 : struct kmem_cache_node *n = get_node(s, slab_nid(slab));
2489 2 : int free_delta = 0;
2490 2 : enum slab_modes mode = M_NONE;
2491 : void *nextfree, *freelist_iter, *freelist_tail;
2492 2 : int tail = DEACTIVATE_TO_HEAD;
2493 2 : unsigned long flags = 0;
2494 : struct slab new;
2495 : struct slab old;
2496 :
2497 2 : if (slab->freelist) {
2498 0 : stat(s, DEACTIVATE_REMOTE_FREES);
2499 0 : tail = DEACTIVATE_TO_TAIL;
2500 : }
2501 :
2502 : /*
2503 : * Stage one: Count the objects on cpu's freelist as free_delta and
2504 : * remember the last object in freelist_tail for later splicing.
2505 : */
2506 2 : freelist_tail = NULL;
2507 2 : freelist_iter = freelist;
2508 86 : while (freelist_iter) {
2509 164 : nextfree = get_freepointer(s, freelist_iter);
2510 :
2511 : /*
2512 : * If 'nextfree' is invalid, it is possible that the object at
2513 : * 'freelist_iter' is already corrupted. So isolate all objects
2514 : * starting at 'freelist_iter' by skipping them.
2515 : */
2516 82 : if (freelist_corrupted(s, slab, &freelist_iter, nextfree))
2517 : break;
2518 :
2519 82 : freelist_tail = freelist_iter;
2520 82 : free_delta++;
2521 :
2522 82 : freelist_iter = nextfree;
2523 : }
2524 :
2525 : /*
2526 : * Stage two: Unfreeze the slab while splicing the per-cpu
2527 : * freelist to the head of slab's freelist.
2528 : *
2529 : * Ensure that the slab is unfrozen while the list presence
2530 : * reflects the actual number of objects during unfreeze.
2531 : *
2532 : * We first perform cmpxchg holding lock and insert to list
2533 : * when it succeed. If there is mismatch then the slab is not
2534 : * unfrozen and number of objects in the slab may have changed.
2535 : * Then release lock and retry cmpxchg again.
2536 : */
2537 : redo:
2538 :
2539 2 : old.freelist = READ_ONCE(slab->freelist);
2540 2 : old.counters = READ_ONCE(slab->counters);
2541 : VM_BUG_ON(!old.frozen);
2542 :
2543 : /* Determine target state of the slab */
2544 2 : new.counters = old.counters;
2545 2 : if (freelist_tail) {
2546 2 : new.inuse -= free_delta;
2547 4 : set_freepointer(s, freelist_tail, old.freelist);
2548 2 : new.freelist = freelist;
2549 : } else
2550 : new.freelist = old.freelist;
2551 :
2552 2 : new.frozen = 0;
2553 :
2554 2 : if (!new.inuse && n->nr_partial >= s->min_partial) {
2555 : mode = M_FREE;
2556 2 : } else if (new.freelist) {
2557 2 : mode = M_PARTIAL;
2558 : /*
2559 : * Taking the spinlock removes the possibility that
2560 : * acquire_slab() will see a slab that is frozen
2561 : */
2562 2 : spin_lock_irqsave(&n->list_lock, flags);
2563 : } else {
2564 : mode = M_FULL_NOLIST;
2565 : }
2566 :
2567 :
2568 2 : if (!slab_update_freelist(s, slab,
2569 : old.freelist, old.counters,
2570 : new.freelist, new.counters,
2571 : "unfreezing slab")) {
2572 0 : if (mode == M_PARTIAL)
2573 0 : spin_unlock_irqrestore(&n->list_lock, flags);
2574 : goto redo;
2575 : }
2576 :
2577 :
2578 2 : if (mode == M_PARTIAL) {
2579 2 : add_partial(n, slab, tail);
2580 4 : spin_unlock_irqrestore(&n->list_lock, flags);
2581 2 : stat(s, tail);
2582 0 : } else if (mode == M_FREE) {
2583 0 : stat(s, DEACTIVATE_EMPTY);
2584 : discard_slab(s, slab);
2585 : stat(s, FREE_SLAB);
2586 : } else if (mode == M_FULL_NOLIST) {
2587 : stat(s, DEACTIVATE_FULL);
2588 : }
2589 2 : }
2590 :
2591 : #ifdef CONFIG_SLUB_CPU_PARTIAL
2592 : static void __unfreeze_partials(struct kmem_cache *s, struct slab *partial_slab)
2593 : {
2594 : struct kmem_cache_node *n = NULL, *n2 = NULL;
2595 : struct slab *slab, *slab_to_discard = NULL;
2596 : unsigned long flags = 0;
2597 :
2598 : while (partial_slab) {
2599 : struct slab new;
2600 : struct slab old;
2601 :
2602 : slab = partial_slab;
2603 : partial_slab = slab->next;
2604 :
2605 : n2 = get_node(s, slab_nid(slab));
2606 : if (n != n2) {
2607 : if (n)
2608 : spin_unlock_irqrestore(&n->list_lock, flags);
2609 :
2610 : n = n2;
2611 : spin_lock_irqsave(&n->list_lock, flags);
2612 : }
2613 :
2614 : do {
2615 :
2616 : old.freelist = slab->freelist;
2617 : old.counters = slab->counters;
2618 : VM_BUG_ON(!old.frozen);
2619 :
2620 : new.counters = old.counters;
2621 : new.freelist = old.freelist;
2622 :
2623 : new.frozen = 0;
2624 :
2625 : } while (!__slab_update_freelist(s, slab,
2626 : old.freelist, old.counters,
2627 : new.freelist, new.counters,
2628 : "unfreezing slab"));
2629 :
2630 : if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
2631 : slab->next = slab_to_discard;
2632 : slab_to_discard = slab;
2633 : } else {
2634 : add_partial(n, slab, DEACTIVATE_TO_TAIL);
2635 : stat(s, FREE_ADD_PARTIAL);
2636 : }
2637 : }
2638 :
2639 : if (n)
2640 : spin_unlock_irqrestore(&n->list_lock, flags);
2641 :
2642 : while (slab_to_discard) {
2643 : slab = slab_to_discard;
2644 : slab_to_discard = slab_to_discard->next;
2645 :
2646 : stat(s, DEACTIVATE_EMPTY);
2647 : discard_slab(s, slab);
2648 : stat(s, FREE_SLAB);
2649 : }
2650 : }
2651 :
2652 : /*
2653 : * Unfreeze all the cpu partial slabs.
2654 : */
2655 : static void unfreeze_partials(struct kmem_cache *s)
2656 : {
2657 : struct slab *partial_slab;
2658 : unsigned long flags;
2659 :
2660 : local_lock_irqsave(&s->cpu_slab->lock, flags);
2661 : partial_slab = this_cpu_read(s->cpu_slab->partial);
2662 : this_cpu_write(s->cpu_slab->partial, NULL);
2663 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2664 :
2665 : if (partial_slab)
2666 : __unfreeze_partials(s, partial_slab);
2667 : }
2668 :
2669 : static void unfreeze_partials_cpu(struct kmem_cache *s,
2670 : struct kmem_cache_cpu *c)
2671 : {
2672 : struct slab *partial_slab;
2673 :
2674 : partial_slab = slub_percpu_partial(c);
2675 : c->partial = NULL;
2676 :
2677 : if (partial_slab)
2678 : __unfreeze_partials(s, partial_slab);
2679 : }
2680 :
2681 : /*
2682 : * Put a slab that was just frozen (in __slab_free|get_partial_node) into a
2683 : * partial slab slot if available.
2684 : *
2685 : * If we did not find a slot then simply move all the partials to the
2686 : * per node partial list.
2687 : */
2688 : static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain)
2689 : {
2690 : struct slab *oldslab;
2691 : struct slab *slab_to_unfreeze = NULL;
2692 : unsigned long flags;
2693 : int slabs = 0;
2694 :
2695 : local_lock_irqsave(&s->cpu_slab->lock, flags);
2696 :
2697 : oldslab = this_cpu_read(s->cpu_slab->partial);
2698 :
2699 : if (oldslab) {
2700 : if (drain && oldslab->slabs >= s->cpu_partial_slabs) {
2701 : /*
2702 : * Partial array is full. Move the existing set to the
2703 : * per node partial list. Postpone the actual unfreezing
2704 : * outside of the critical section.
2705 : */
2706 : slab_to_unfreeze = oldslab;
2707 : oldslab = NULL;
2708 : } else {
2709 : slabs = oldslab->slabs;
2710 : }
2711 : }
2712 :
2713 : slabs++;
2714 :
2715 : slab->slabs = slabs;
2716 : slab->next = oldslab;
2717 :
2718 : this_cpu_write(s->cpu_slab->partial, slab);
2719 :
2720 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2721 :
2722 : if (slab_to_unfreeze) {
2723 : __unfreeze_partials(s, slab_to_unfreeze);
2724 : stat(s, CPU_PARTIAL_DRAIN);
2725 : }
2726 : }
2727 :
2728 : #else /* CONFIG_SLUB_CPU_PARTIAL */
2729 :
2730 : static inline void unfreeze_partials(struct kmem_cache *s) { }
2731 : static inline void unfreeze_partials_cpu(struct kmem_cache *s,
2732 : struct kmem_cache_cpu *c) { }
2733 :
2734 : #endif /* CONFIG_SLUB_CPU_PARTIAL */
2735 :
2736 0 : static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
2737 : {
2738 : unsigned long flags;
2739 : struct slab *slab;
2740 : void *freelist;
2741 :
2742 0 : local_lock_irqsave(&s->cpu_slab->lock, flags);
2743 :
2744 0 : slab = c->slab;
2745 0 : freelist = c->freelist;
2746 :
2747 0 : c->slab = NULL;
2748 0 : c->freelist = NULL;
2749 0 : c->tid = next_tid(c->tid);
2750 :
2751 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2752 :
2753 0 : if (slab) {
2754 0 : deactivate_slab(s, slab, freelist);
2755 0 : stat(s, CPUSLAB_FLUSH);
2756 : }
2757 0 : }
2758 :
2759 2 : static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
2760 : {
2761 2 : struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2762 2 : void *freelist = c->freelist;
2763 2 : struct slab *slab = c->slab;
2764 :
2765 2 : c->slab = NULL;
2766 2 : c->freelist = NULL;
2767 4 : c->tid = next_tid(c->tid);
2768 :
2769 2 : if (slab) {
2770 2 : deactivate_slab(s, slab, freelist);
2771 2 : stat(s, CPUSLAB_FLUSH);
2772 : }
2773 :
2774 2 : unfreeze_partials_cpu(s, c);
2775 2 : }
2776 :
2777 : struct slub_flush_work {
2778 : struct work_struct work;
2779 : struct kmem_cache *s;
2780 : bool skip;
2781 : };
2782 :
2783 : /*
2784 : * Flush cpu slab.
2785 : *
2786 : * Called from CPU work handler with migration disabled.
2787 : */
2788 0 : static void flush_cpu_slab(struct work_struct *w)
2789 : {
2790 : struct kmem_cache *s;
2791 : struct kmem_cache_cpu *c;
2792 : struct slub_flush_work *sfw;
2793 :
2794 0 : sfw = container_of(w, struct slub_flush_work, work);
2795 :
2796 0 : s = sfw->s;
2797 0 : c = this_cpu_ptr(s->cpu_slab);
2798 :
2799 0 : if (c->slab)
2800 0 : flush_slab(s, c);
2801 :
2802 0 : unfreeze_partials(s);
2803 0 : }
2804 :
2805 : static bool has_cpu_slab(int cpu, struct kmem_cache *s)
2806 : {
2807 0 : struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2808 :
2809 0 : return c->slab || slub_percpu_partial(c);
2810 : }
2811 :
2812 : static DEFINE_MUTEX(flush_lock);
2813 : static DEFINE_PER_CPU(struct slub_flush_work, slub_flush);
2814 :
2815 0 : static void flush_all_cpus_locked(struct kmem_cache *s)
2816 : {
2817 : struct slub_flush_work *sfw;
2818 : unsigned int cpu;
2819 :
2820 : lockdep_assert_cpus_held();
2821 0 : mutex_lock(&flush_lock);
2822 :
2823 0 : for_each_online_cpu(cpu) {
2824 0 : sfw = &per_cpu(slub_flush, cpu);
2825 0 : if (!has_cpu_slab(cpu, s)) {
2826 0 : sfw->skip = true;
2827 0 : continue;
2828 : }
2829 0 : INIT_WORK(&sfw->work, flush_cpu_slab);
2830 0 : sfw->skip = false;
2831 0 : sfw->s = s;
2832 0 : queue_work_on(cpu, flushwq, &sfw->work);
2833 : }
2834 :
2835 0 : for_each_online_cpu(cpu) {
2836 0 : sfw = &per_cpu(slub_flush, cpu);
2837 0 : if (sfw->skip)
2838 0 : continue;
2839 0 : flush_work(&sfw->work);
2840 : }
2841 :
2842 0 : mutex_unlock(&flush_lock);
2843 0 : }
2844 :
2845 : static void flush_all(struct kmem_cache *s)
2846 : {
2847 : cpus_read_lock();
2848 0 : flush_all_cpus_locked(s);
2849 : cpus_read_unlock();
2850 : }
2851 :
2852 : /*
2853 : * Use the cpu notifier to insure that the cpu slabs are flushed when
2854 : * necessary.
2855 : */
2856 0 : static int slub_cpu_dead(unsigned int cpu)
2857 : {
2858 : struct kmem_cache *s;
2859 :
2860 0 : mutex_lock(&slab_mutex);
2861 0 : list_for_each_entry(s, &slab_caches, list)
2862 0 : __flush_cpu_slab(s, cpu);
2863 0 : mutex_unlock(&slab_mutex);
2864 0 : return 0;
2865 : }
2866 :
2867 : #else /* CONFIG_SLUB_TINY */
2868 : static inline void flush_all_cpus_locked(struct kmem_cache *s) { }
2869 : static inline void flush_all(struct kmem_cache *s) { }
2870 : static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) { }
2871 : static inline int slub_cpu_dead(unsigned int cpu) { return 0; }
2872 : #endif /* CONFIG_SLUB_TINY */
2873 :
2874 : /*
2875 : * Check if the objects in a per cpu structure fit numa
2876 : * locality expectations.
2877 : */
2878 : static inline int node_match(struct slab *slab, int node)
2879 : {
2880 : #ifdef CONFIG_NUMA
2881 : if (node != NUMA_NO_NODE && slab_nid(slab) != node)
2882 : return 0;
2883 : #endif
2884 : return 1;
2885 : }
2886 :
2887 : #ifdef CONFIG_SLUB_DEBUG
2888 0 : static int count_free(struct slab *slab)
2889 : {
2890 0 : return slab->objects - slab->inuse;
2891 : }
2892 :
2893 : static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2894 : {
2895 0 : return atomic_long_read(&n->total_objects);
2896 : }
2897 :
2898 : /* Supports checking bulk free of a constructed freelist */
2899 0 : static inline bool free_debug_processing(struct kmem_cache *s,
2900 : struct slab *slab, void *head, void *tail, int *bulk_cnt,
2901 : unsigned long addr, depot_stack_handle_t handle)
2902 : {
2903 0 : bool checks_ok = false;
2904 0 : void *object = head;
2905 0 : int cnt = 0;
2906 :
2907 0 : if (s->flags & SLAB_CONSISTENCY_CHECKS) {
2908 0 : if (!check_slab(s, slab))
2909 : goto out;
2910 : }
2911 :
2912 0 : if (slab->inuse < *bulk_cnt) {
2913 0 : slab_err(s, slab, "Slab has %d allocated objects but %d are to be freed\n",
2914 : slab->inuse, *bulk_cnt);
2915 0 : goto out;
2916 : }
2917 :
2918 : next_object:
2919 :
2920 0 : if (++cnt > *bulk_cnt)
2921 : goto out_cnt;
2922 :
2923 0 : if (s->flags & SLAB_CONSISTENCY_CHECKS) {
2924 0 : if (!free_consistency_checks(s, slab, object, addr))
2925 : goto out;
2926 : }
2927 :
2928 0 : if (s->flags & SLAB_STORE_USER)
2929 : set_track_update(s, object, TRACK_FREE, addr, handle);
2930 0 : trace(s, slab, object, 0);
2931 : /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
2932 0 : init_object(s, object, SLUB_RED_INACTIVE);
2933 :
2934 : /* Reached end of constructed freelist yet? */
2935 0 : if (object != tail) {
2936 0 : object = get_freepointer(s, object);
2937 0 : goto next_object;
2938 : }
2939 : checks_ok = true;
2940 :
2941 : out_cnt:
2942 0 : if (cnt != *bulk_cnt) {
2943 0 : slab_err(s, slab, "Bulk free expected %d objects but found %d\n",
2944 : *bulk_cnt, cnt);
2945 0 : *bulk_cnt = cnt;
2946 : }
2947 :
2948 : out:
2949 :
2950 0 : if (!checks_ok)
2951 0 : slab_fix(s, "Object at 0x%p not freed", object);
2952 :
2953 0 : return checks_ok;
2954 : }
2955 : #endif /* CONFIG_SLUB_DEBUG */
2956 :
2957 : #if defined(CONFIG_SLUB_DEBUG) || defined(SLAB_SUPPORTS_SYSFS)
2958 0 : static unsigned long count_partial(struct kmem_cache_node *n,
2959 : int (*get_count)(struct slab *))
2960 : {
2961 : unsigned long flags;
2962 0 : unsigned long x = 0;
2963 : struct slab *slab;
2964 :
2965 0 : spin_lock_irqsave(&n->list_lock, flags);
2966 0 : list_for_each_entry(slab, &n->partial, slab_list)
2967 0 : x += get_count(slab);
2968 0 : spin_unlock_irqrestore(&n->list_lock, flags);
2969 0 : return x;
2970 : }
2971 : #endif /* CONFIG_SLUB_DEBUG || SLAB_SUPPORTS_SYSFS */
2972 :
2973 : #ifdef CONFIG_SLUB_DEBUG
2974 : static noinline void
2975 0 : slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2976 : {
2977 : static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2978 : DEFAULT_RATELIMIT_BURST);
2979 : int node;
2980 : struct kmem_cache_node *n;
2981 :
2982 0 : if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2983 : return;
2984 :
2985 0 : pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2986 : nid, gfpflags, &gfpflags);
2987 0 : pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
2988 : s->name, s->object_size, s->size, oo_order(s->oo),
2989 : oo_order(s->min));
2990 :
2991 0 : if (oo_order(s->min) > get_order(s->object_size))
2992 0 : pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2993 : s->name);
2994 :
2995 0 : for_each_kmem_cache_node(s, node, n) {
2996 : unsigned long nr_slabs;
2997 : unsigned long nr_objs;
2998 : unsigned long nr_free;
2999 :
3000 0 : nr_free = count_partial(n, count_free);
3001 0 : nr_slabs = node_nr_slabs(n);
3002 0 : nr_objs = node_nr_objs(n);
3003 :
3004 0 : pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
3005 : node, nr_slabs, nr_objs, nr_free);
3006 : }
3007 : }
3008 : #else /* CONFIG_SLUB_DEBUG */
3009 : static inline void
3010 : slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) { }
3011 : #endif
3012 :
3013 : static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags)
3014 : {
3015 922 : if (unlikely(slab_test_pfmemalloc(slab)))
3016 0 : return gfp_pfmemalloc_allowed(gfpflags);
3017 :
3018 : return true;
3019 : }
3020 :
3021 : #ifndef CONFIG_SLUB_TINY
3022 : static inline bool
3023 20683 : __update_cpu_freelist_fast(struct kmem_cache *s,
3024 : void *freelist_old, void *freelist_new,
3025 : unsigned long tid)
3026 : {
3027 20683 : freelist_aba_t old = { .freelist = freelist_old, .counter = tid };
3028 41366 : freelist_aba_t new = { .freelist = freelist_new, .counter = next_tid(tid) };
3029 :
3030 41366 : return this_cpu_try_cmpxchg_freelist(s->cpu_slab->freelist_tid.full,
3031 : &old.full, new.full);
3032 : }
3033 :
3034 : /*
3035 : * Check the slab->freelist and either transfer the freelist to the
3036 : * per cpu freelist or deactivate the slab.
3037 : *
3038 : * The slab is still frozen if the return value is not NULL.
3039 : *
3040 : * If this function returns NULL then the slab has been unfrozen.
3041 : */
3042 432 : static inline void *get_freelist(struct kmem_cache *s, struct slab *slab)
3043 : {
3044 : struct slab new;
3045 : unsigned long counters;
3046 : void *freelist;
3047 :
3048 : lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock));
3049 :
3050 : do {
3051 432 : freelist = slab->freelist;
3052 432 : counters = slab->counters;
3053 :
3054 432 : new.counters = counters;
3055 : VM_BUG_ON(!new.frozen);
3056 :
3057 432 : new.inuse = slab->objects;
3058 432 : new.frozen = freelist != NULL;
3059 :
3060 864 : } while (!__slab_update_freelist(s, slab,
3061 : freelist, counters,
3062 : NULL, new.counters,
3063 432 : "get_freelist"));
3064 :
3065 432 : return freelist;
3066 : }
3067 :
3068 : /*
3069 : * Slow path. The lockless freelist is empty or we need to perform
3070 : * debugging duties.
3071 : *
3072 : * Processing is still very fast if new objects have been freed to the
3073 : * regular freelist. In that case we simply take over the regular freelist
3074 : * as the lockless freelist and zap the regular freelist.
3075 : *
3076 : * If that is not working then we fall back to the partial lists. We take the
3077 : * first element of the freelist as the object to allocate now and move the
3078 : * rest of the freelist to the lockless freelist.
3079 : *
3080 : * And if we were unable to get a new slab from the partial slab lists then
3081 : * we need to allocate a new slab. This is the slowest path since it involves
3082 : * a call to the page allocator and the setup of a new slab.
3083 : *
3084 : * Version of __slab_alloc to use when we know that preemption is
3085 : * already disabled (which is the case for bulk allocation).
3086 : */
3087 462 : static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
3088 : unsigned long addr, struct kmem_cache_cpu *c, unsigned int orig_size)
3089 : {
3090 : void *freelist;
3091 : struct slab *slab;
3092 : unsigned long flags;
3093 : struct partial_context pc;
3094 :
3095 462 : stat(s, ALLOC_SLOWPATH);
3096 :
3097 : reread_slab:
3098 :
3099 462 : slab = READ_ONCE(c->slab);
3100 462 : if (!slab) {
3101 : /*
3102 : * if the node is not online or has no normal memory, just
3103 : * ignore the node constraint
3104 : */
3105 32 : if (unlikely(node != NUMA_NO_NODE &&
3106 : !node_isset(node, slab_nodes)))
3107 0 : node = NUMA_NO_NODE;
3108 : goto new_slab;
3109 : }
3110 : redo:
3111 :
3112 432 : if (unlikely(!node_match(slab, node))) {
3113 : /*
3114 : * same as above but node_match() being false already
3115 : * implies node != NUMA_NO_NODE
3116 : */
3117 : if (!node_isset(node, slab_nodes)) {
3118 : node = NUMA_NO_NODE;
3119 : } else {
3120 : stat(s, ALLOC_NODE_MISMATCH);
3121 : goto deactivate_slab;
3122 : }
3123 : }
3124 :
3125 : /*
3126 : * By rights, we should be searching for a slab page that was
3127 : * PFMEMALLOC but right now, we are losing the pfmemalloc
3128 : * information when the page leaves the per-cpu allocator
3129 : */
3130 864 : if (unlikely(!pfmemalloc_match(slab, gfpflags)))
3131 : goto deactivate_slab;
3132 :
3133 : /* must check again c->slab in case we got preempted and it changed */
3134 432 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3135 432 : if (unlikely(slab != c->slab)) {
3136 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3137 : goto reread_slab;
3138 : }
3139 432 : freelist = c->freelist;
3140 432 : if (freelist)
3141 : goto load_freelist;
3142 :
3143 432 : freelist = get_freelist(s, slab);
3144 :
3145 432 : if (!freelist) {
3146 432 : c->slab = NULL;
3147 864 : c->tid = next_tid(c->tid);
3148 432 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3149 : stat(s, DEACTIVATE_BYPASS);
3150 : goto new_slab;
3151 : }
3152 :
3153 : stat(s, ALLOC_REFILL);
3154 :
3155 : load_freelist:
3156 :
3157 462 : lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock));
3158 :
3159 : /*
3160 : * freelist is pointing to the list of objects to be used.
3161 : * slab is pointing to the slab from which the objects are obtained.
3162 : * That slab must be frozen for per cpu allocations to work.
3163 : */
3164 : VM_BUG_ON(!c->slab->frozen);
3165 924 : c->freelist = get_freepointer(s, freelist);
3166 924 : c->tid = next_tid(c->tid);
3167 924 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3168 462 : return freelist;
3169 :
3170 : deactivate_slab:
3171 :
3172 0 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3173 0 : if (slab != c->slab) {
3174 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3175 : goto reread_slab;
3176 : }
3177 0 : freelist = c->freelist;
3178 0 : c->slab = NULL;
3179 0 : c->freelist = NULL;
3180 0 : c->tid = next_tid(c->tid);
3181 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3182 0 : deactivate_slab(s, slab, freelist);
3183 :
3184 : new_slab:
3185 :
3186 : if (slub_percpu_partial(c)) {
3187 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3188 : if (unlikely(c->slab)) {
3189 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3190 : goto reread_slab;
3191 : }
3192 : if (unlikely(!slub_percpu_partial(c))) {
3193 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3194 : /* we were preempted and partial list got empty */
3195 : goto new_objects;
3196 : }
3197 :
3198 : slab = c->slab = slub_percpu_partial(c);
3199 : slub_set_percpu_partial(c, slab);
3200 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3201 : stat(s, CPU_PARTIAL_ALLOC);
3202 : goto redo;
3203 : }
3204 :
3205 : new_objects:
3206 :
3207 462 : pc.flags = gfpflags;
3208 462 : pc.slab = &slab;
3209 462 : pc.orig_size = orig_size;
3210 462 : freelist = get_partial(s, node, &pc);
3211 462 : if (freelist)
3212 : goto check_new_slab;
3213 :
3214 434 : slub_put_cpu_ptr(s->cpu_slab);
3215 434 : slab = new_slab(s, gfpflags, node);
3216 434 : c = slub_get_cpu_ptr(s->cpu_slab);
3217 :
3218 434 : if (unlikely(!slab)) {
3219 0 : slab_out_of_memory(s, gfpflags, node);
3220 0 : return NULL;
3221 : }
3222 :
3223 434 : stat(s, ALLOC_SLAB);
3224 :
3225 434 : if (kmem_cache_debug(s)) {
3226 0 : freelist = alloc_single_from_new_slab(s, slab, orig_size);
3227 :
3228 0 : if (unlikely(!freelist))
3229 : goto new_objects;
3230 :
3231 0 : if (s->flags & SLAB_STORE_USER)
3232 : set_track(s, freelist, TRACK_ALLOC, addr);
3233 :
3234 : return freelist;
3235 : }
3236 :
3237 : /*
3238 : * No other reference to the slab yet so we can
3239 : * muck around with it freely without cmpxchg
3240 : */
3241 434 : freelist = slab->freelist;
3242 434 : slab->freelist = NULL;
3243 434 : slab->inuse = slab->objects;
3244 434 : slab->frozen = 1;
3245 :
3246 868 : inc_slabs_node(s, slab_nid(slab), slab->objects);
3247 :
3248 : check_new_slab:
3249 :
3250 462 : if (kmem_cache_debug(s)) {
3251 : /*
3252 : * For debug caches here we had to go through
3253 : * alloc_single_from_partial() so just store the tracking info
3254 : * and return the object
3255 : */
3256 0 : if (s->flags & SLAB_STORE_USER)
3257 : set_track(s, freelist, TRACK_ALLOC, addr);
3258 :
3259 : return freelist;
3260 : }
3261 :
3262 924 : if (unlikely(!pfmemalloc_match(slab, gfpflags))) {
3263 : /*
3264 : * For !pfmemalloc_match() case we don't load freelist so that
3265 : * we don't make further mismatched allocations easier.
3266 : */
3267 0 : deactivate_slab(s, slab, get_freepointer(s, freelist));
3268 0 : return freelist;
3269 : }
3270 :
3271 : retry_load_slab:
3272 :
3273 462 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3274 462 : if (unlikely(c->slab)) {
3275 0 : void *flush_freelist = c->freelist;
3276 0 : struct slab *flush_slab = c->slab;
3277 :
3278 0 : c->slab = NULL;
3279 0 : c->freelist = NULL;
3280 0 : c->tid = next_tid(c->tid);
3281 :
3282 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3283 :
3284 0 : deactivate_slab(s, flush_slab, flush_freelist);
3285 :
3286 0 : stat(s, CPUSLAB_FLUSH);
3287 :
3288 : goto retry_load_slab;
3289 : }
3290 462 : c->slab = slab;
3291 :
3292 462 : goto load_freelist;
3293 : }
3294 :
3295 : /*
3296 : * A wrapper for ___slab_alloc() for contexts where preemption is not yet
3297 : * disabled. Compensates for possible cpu changes by refetching the per cpu area
3298 : * pointer.
3299 : */
3300 : static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
3301 : unsigned long addr, struct kmem_cache_cpu *c, unsigned int orig_size)
3302 : {
3303 : void *p;
3304 :
3305 : #ifdef CONFIG_PREEMPT_COUNT
3306 : /*
3307 : * We may have been preempted and rescheduled on a different
3308 : * cpu before disabling preemption. Need to reload cpu area
3309 : * pointer.
3310 : */
3311 : c = slub_get_cpu_ptr(s->cpu_slab);
3312 : #endif
3313 :
3314 461 : p = ___slab_alloc(s, gfpflags, node, addr, c, orig_size);
3315 : #ifdef CONFIG_PREEMPT_COUNT
3316 : slub_put_cpu_ptr(s->cpu_slab);
3317 : #endif
3318 : return p;
3319 : }
3320 :
3321 : static __always_inline void *__slab_alloc_node(struct kmem_cache *s,
3322 : gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
3323 : {
3324 : struct kmem_cache_cpu *c;
3325 : struct slab *slab;
3326 : unsigned long tid;
3327 : void *object;
3328 :
3329 : redo:
3330 : /*
3331 : * Must read kmem_cache cpu data via this cpu ptr. Preemption is
3332 : * enabled. We may switch back and forth between cpus while
3333 : * reading from one cpu area. That does not matter as long
3334 : * as we end up on the original cpu again when doing the cmpxchg.
3335 : *
3336 : * We must guarantee that tid and kmem_cache_cpu are retrieved on the
3337 : * same cpu. We read first the kmem_cache_cpu pointer and use it to read
3338 : * the tid. If we are preempted and switched to another cpu between the
3339 : * two reads, it's OK as the two are still associated with the same cpu
3340 : * and cmpxchg later will validate the cpu.
3341 : */
3342 16949 : c = raw_cpu_ptr(s->cpu_slab);
3343 16949 : tid = READ_ONCE(c->tid);
3344 :
3345 : /*
3346 : * Irqless object alloc/free algorithm used here depends on sequence
3347 : * of fetching cpu_slab's data. tid should be fetched before anything
3348 : * on c to guarantee that object and slab associated with previous tid
3349 : * won't be used with current tid. If we fetch tid first, object and
3350 : * slab could be one associated with next tid and our alloc/free
3351 : * request will be failed. In this case, we will retry. So, no problem.
3352 : */
3353 16949 : barrier();
3354 :
3355 : /*
3356 : * The transaction ids are globally unique per cpu and per operation on
3357 : * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
3358 : * occurs on the right processor and that there was no operation on the
3359 : * linked list in between.
3360 : */
3361 :
3362 16949 : object = c->freelist;
3363 16949 : slab = c->slab;
3364 :
3365 16949 : if (!USE_LOCKLESS_FAST_PATH() ||
3366 33437 : unlikely(!object || !slab || !node_match(slab, node))) {
3367 622 : object = __slab_alloc(s, gfpflags, node, addr, c, orig_size);
3368 : } else {
3369 16488 : void *next_object = get_freepointer_safe(s, object);
3370 :
3371 : /*
3372 : * The cmpxchg will only match if there was no additional
3373 : * operation and if we are on the right processor.
3374 : *
3375 : * The cmpxchg does the following atomically (without lock
3376 : * semantics!)
3377 : * 1. Relocate first pointer to the current per cpu area.
3378 : * 2. Verify that tid and freelist have not been changed
3379 : * 3. If they were not changed replace tid and freelist
3380 : *
3381 : * Since this is without lock semantics the protection is only
3382 : * against code executing on this cpu *not* from access by
3383 : * other cpus.
3384 : */
3385 16488 : if (unlikely(!__update_cpu_freelist_fast(s, object, next_object, tid))) {
3386 : note_cmpxchg_failure("slab_alloc", s, tid);
3387 : goto redo;
3388 : }
3389 16488 : prefetch_freepointer(s, next_object);
3390 : stat(s, ALLOC_FASTPATH);
3391 : }
3392 :
3393 : return object;
3394 : }
3395 : #else /* CONFIG_SLUB_TINY */
3396 : static void *__slab_alloc_node(struct kmem_cache *s,
3397 : gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
3398 : {
3399 : struct partial_context pc;
3400 : struct slab *slab;
3401 : void *object;
3402 :
3403 : pc.flags = gfpflags;
3404 : pc.slab = &slab;
3405 : pc.orig_size = orig_size;
3406 : object = get_partial(s, node, &pc);
3407 :
3408 : if (object)
3409 : return object;
3410 :
3411 : slab = new_slab(s, gfpflags, node);
3412 : if (unlikely(!slab)) {
3413 : slab_out_of_memory(s, gfpflags, node);
3414 : return NULL;
3415 : }
3416 :
3417 : object = alloc_single_from_new_slab(s, slab, orig_size);
3418 :
3419 : return object;
3420 : }
3421 : #endif /* CONFIG_SLUB_TINY */
3422 :
3423 : /*
3424 : * If the object has been wiped upon free, make sure it's fully initialized by
3425 : * zeroing out freelist pointer.
3426 : */
3427 : static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s,
3428 : void *obj)
3429 : {
3430 17001 : if (unlikely(slab_want_init_on_free(s)) && obj)
3431 0 : memset((void *)((char *)kasan_reset_tag(obj) + s->offset),
3432 : 0, sizeof(void *));
3433 : }
3434 :
3435 : /*
3436 : * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
3437 : * have the fastpath folded into their functions. So no function call
3438 : * overhead for requests that can be satisfied on the fastpath.
3439 : *
3440 : * The fastpath works by first checking if the lockless freelist can be used.
3441 : * If not then __slab_alloc is called for slow processing.
3442 : *
3443 : * Otherwise we can simply pick the next object from the lockless free list.
3444 : */
3445 : static __fastpath_inline void *slab_alloc_node(struct kmem_cache *s, struct list_lru *lru,
3446 : gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
3447 : {
3448 : void *object;
3449 16949 : struct obj_cgroup *objcg = NULL;
3450 16949 : bool init = false;
3451 :
3452 33898 : s = slab_pre_alloc_hook(s, lru, &objcg, 1, gfpflags);
3453 16949 : if (!s)
3454 : return NULL;
3455 :
3456 16949 : object = kfence_alloc(s, orig_size, gfpflags);
3457 : if (unlikely(object))
3458 : goto out;
3459 :
3460 16949 : object = __slab_alloc_node(s, gfpflags, node, addr, orig_size);
3461 :
3462 33898 : maybe_wipe_obj_freeptr(s, object);
3463 33898 : init = slab_want_init_on_alloc(gfpflags, s);
3464 :
3465 : out:
3466 : /*
3467 : * When init equals 'true', like for kzalloc() family, only
3468 : * @orig_size bytes might be zeroed instead of s->object_size
3469 : */
3470 16949 : slab_post_alloc_hook(s, objcg, gfpflags, 1, &object, init, orig_size);
3471 :
3472 16949 : return object;
3473 : }
3474 :
3475 : static __fastpath_inline void *slab_alloc(struct kmem_cache *s, struct list_lru *lru,
3476 : gfp_t gfpflags, unsigned long addr, size_t orig_size)
3477 : {
3478 9849 : return slab_alloc_node(s, lru, gfpflags, NUMA_NO_NODE, addr, orig_size);
3479 : }
3480 :
3481 : static __fastpath_inline
3482 : void *__kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
3483 : gfp_t gfpflags)
3484 : {
3485 19698 : void *ret = slab_alloc(s, lru, gfpflags, _RET_IP_, s->object_size);
3486 :
3487 9849 : trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, NUMA_NO_NODE);
3488 :
3489 : return ret;
3490 : }
3491 :
3492 9800 : void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
3493 : {
3494 9800 : return __kmem_cache_alloc_lru(s, NULL, gfpflags);
3495 : }
3496 : EXPORT_SYMBOL(kmem_cache_alloc);
3497 :
3498 49 : void *kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
3499 : gfp_t gfpflags)
3500 : {
3501 49 : return __kmem_cache_alloc_lru(s, lru, gfpflags);
3502 : }
3503 : EXPORT_SYMBOL(kmem_cache_alloc_lru);
3504 :
3505 6838 : void *__kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags,
3506 : int node, size_t orig_size,
3507 : unsigned long caller)
3508 : {
3509 6838 : return slab_alloc_node(s, NULL, gfpflags, node,
3510 : caller, orig_size);
3511 : }
3512 :
3513 262 : void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
3514 : {
3515 524 : void *ret = slab_alloc_node(s, NULL, gfpflags, node, _RET_IP_, s->object_size);
3516 :
3517 262 : trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, node);
3518 :
3519 262 : return ret;
3520 : }
3521 : EXPORT_SYMBOL(kmem_cache_alloc_node);
3522 :
3523 0 : static noinline void free_to_partial_list(
3524 : struct kmem_cache *s, struct slab *slab,
3525 : void *head, void *tail, int bulk_cnt,
3526 : unsigned long addr)
3527 : {
3528 0 : struct kmem_cache_node *n = get_node(s, slab_nid(slab));
3529 0 : struct slab *slab_free = NULL;
3530 0 : int cnt = bulk_cnt;
3531 : unsigned long flags;
3532 0 : depot_stack_handle_t handle = 0;
3533 :
3534 0 : if (s->flags & SLAB_STORE_USER)
3535 0 : handle = set_track_prepare();
3536 :
3537 0 : spin_lock_irqsave(&n->list_lock, flags);
3538 :
3539 0 : if (free_debug_processing(s, slab, head, tail, &cnt, addr, handle)) {
3540 0 : void *prior = slab->freelist;
3541 :
3542 : /* Perform the actual freeing while we still hold the locks */
3543 0 : slab->inuse -= cnt;
3544 0 : set_freepointer(s, tail, prior);
3545 0 : slab->freelist = head;
3546 :
3547 : /*
3548 : * If the slab is empty, and node's partial list is full,
3549 : * it should be discarded anyway no matter it's on full or
3550 : * partial list.
3551 : */
3552 0 : if (slab->inuse == 0 && n->nr_partial >= s->min_partial)
3553 0 : slab_free = slab;
3554 :
3555 0 : if (!prior) {
3556 : /* was on full list */
3557 0 : remove_full(s, n, slab);
3558 0 : if (!slab_free) {
3559 : add_partial(n, slab, DEACTIVATE_TO_TAIL);
3560 : stat(s, FREE_ADD_PARTIAL);
3561 : }
3562 0 : } else if (slab_free) {
3563 0 : remove_partial(n, slab);
3564 : stat(s, FREE_REMOVE_PARTIAL);
3565 : }
3566 : }
3567 :
3568 0 : if (slab_free) {
3569 : /*
3570 : * Update the counters while still holding n->list_lock to
3571 : * prevent spurious validation warnings
3572 : */
3573 0 : dec_slabs_node(s, slab_nid(slab_free), slab_free->objects);
3574 : }
3575 :
3576 0 : spin_unlock_irqrestore(&n->list_lock, flags);
3577 :
3578 0 : if (slab_free) {
3579 0 : stat(s, FREE_SLAB);
3580 0 : free_slab(s, slab_free);
3581 : }
3582 0 : }
3583 :
3584 : /*
3585 : * Slow path handling. This may still be called frequently since objects
3586 : * have a longer lifetime than the cpu slabs in most processing loads.
3587 : *
3588 : * So we still attempt to reduce cache line usage. Just take the slab
3589 : * lock and free the item. If there is no additional partial slab
3590 : * handling required then we can return immediately.
3591 : */
3592 228 : static void __slab_free(struct kmem_cache *s, struct slab *slab,
3593 : void *head, void *tail, int cnt,
3594 : unsigned long addr)
3595 :
3596 : {
3597 : void *prior;
3598 : int was_frozen;
3599 : struct slab new;
3600 : unsigned long counters;
3601 228 : struct kmem_cache_node *n = NULL;
3602 : unsigned long flags;
3603 :
3604 228 : stat(s, FREE_SLOWPATH);
3605 :
3606 228 : if (kfence_free(head))
3607 228 : return;
3608 :
3609 228 : if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) {
3610 0 : free_to_partial_list(s, slab, head, tail, cnt, addr);
3611 0 : return;
3612 : }
3613 :
3614 : do {
3615 228 : if (unlikely(n)) {
3616 0 : spin_unlock_irqrestore(&n->list_lock, flags);
3617 0 : n = NULL;
3618 : }
3619 228 : prior = slab->freelist;
3620 228 : counters = slab->counters;
3621 456 : set_freepointer(s, tail, prior);
3622 228 : new.counters = counters;
3623 228 : was_frozen = new.frozen;
3624 228 : new.inuse -= cnt;
3625 228 : if ((!new.inuse || !prior) && !was_frozen) {
3626 :
3627 37 : if (kmem_cache_has_cpu_partial(s) && !prior) {
3628 :
3629 : /*
3630 : * Slab was on no list before and will be
3631 : * partially empty
3632 : * We can defer the list move and instead
3633 : * freeze it.
3634 : */
3635 : new.frozen = 1;
3636 :
3637 : } else { /* Needs to be taken off a list */
3638 :
3639 111 : n = get_node(s, slab_nid(slab));
3640 : /*
3641 : * Speculatively acquire the list_lock.
3642 : * If the cmpxchg does not succeed then we may
3643 : * drop the list_lock without any processing.
3644 : *
3645 : * Otherwise the list_lock will synchronize with
3646 : * other processors updating the list of slabs.
3647 : */
3648 37 : spin_lock_irqsave(&n->list_lock, flags);
3649 :
3650 : }
3651 : }
3652 :
3653 228 : } while (!slab_update_freelist(s, slab,
3654 : prior, counters,
3655 : head, new.counters,
3656 228 : "__slab_free"));
3657 :
3658 228 : if (likely(!n)) {
3659 :
3660 : if (likely(was_frozen)) {
3661 : /*
3662 : * The list lock was not taken therefore no list
3663 : * activity can be necessary.
3664 : */
3665 : stat(s, FREE_FROZEN);
3666 : } else if (new.frozen) {
3667 : /*
3668 : * If we just froze the slab then put it onto the
3669 : * per cpu partial list.
3670 : */
3671 : put_cpu_partial(s, slab, 1);
3672 : stat(s, CPU_PARTIAL_FREE);
3673 : }
3674 :
3675 : return;
3676 : }
3677 :
3678 37 : if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
3679 : goto slab_empty;
3680 :
3681 : /*
3682 : * Objects left in the slab. If it was not on the partial list before
3683 : * then add it.
3684 : */
3685 37 : if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
3686 70 : remove_full(s, n, slab);
3687 : add_partial(n, slab, DEACTIVATE_TO_TAIL);
3688 : stat(s, FREE_ADD_PARTIAL);
3689 : }
3690 37 : spin_unlock_irqrestore(&n->list_lock, flags);
3691 : return;
3692 :
3693 : slab_empty:
3694 0 : if (prior) {
3695 : /*
3696 : * Slab on the partial list.
3697 : */
3698 0 : remove_partial(n, slab);
3699 : stat(s, FREE_REMOVE_PARTIAL);
3700 : } else {
3701 : /* Slab must be on the full list */
3702 0 : remove_full(s, n, slab);
3703 : }
3704 :
3705 0 : spin_unlock_irqrestore(&n->list_lock, flags);
3706 0 : stat(s, FREE_SLAB);
3707 0 : discard_slab(s, slab);
3708 : }
3709 :
3710 : #ifndef CONFIG_SLUB_TINY
3711 : /*
3712 : * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
3713 : * can perform fastpath freeing without additional function calls.
3714 : *
3715 : * The fastpath is only possible if we are freeing to the current cpu slab
3716 : * of this processor. This typically the case if we have just allocated
3717 : * the item before.
3718 : *
3719 : * If fastpath is not possible then fall back to __slab_free where we deal
3720 : * with all sorts of special processing.
3721 : *
3722 : * Bulk free of a freelist with several objects (all pointing to the
3723 : * same slab) possible by specifying head and tail ptr, plus objects
3724 : * count (cnt). Bulk free indicated by tail pointer being set.
3725 : */
3726 : static __always_inline void do_slab_free(struct kmem_cache *s,
3727 : struct slab *slab, void *head, void *tail,
3728 : int cnt, unsigned long addr)
3729 : {
3730 4423 : void *tail_obj = tail ? : head;
3731 : struct kmem_cache_cpu *c;
3732 : unsigned long tid;
3733 : void **freelist;
3734 :
3735 : redo:
3736 : /*
3737 : * Determine the currently cpus per cpu slab.
3738 : * The cpu may change afterward. However that does not matter since
3739 : * data is retrieved via this pointer. If we are on the same cpu
3740 : * during the cmpxchg then the free will succeed.
3741 : */
3742 4423 : c = raw_cpu_ptr(s->cpu_slab);
3743 4423 : tid = READ_ONCE(c->tid);
3744 :
3745 : /* Same with comment on barrier() in slab_alloc_node() */
3746 4423 : barrier();
3747 :
3748 4423 : if (unlikely(slab != c->slab)) {
3749 228 : __slab_free(s, slab, head, tail_obj, cnt, addr);
3750 : return;
3751 : }
3752 :
3753 : if (USE_LOCKLESS_FAST_PATH()) {
3754 4195 : freelist = READ_ONCE(c->freelist);
3755 :
3756 8390 : set_freepointer(s, tail_obj, freelist);
3757 :
3758 4195 : if (unlikely(!__update_cpu_freelist_fast(s, freelist, head, tid))) {
3759 : note_cmpxchg_failure("slab_free", s, tid);
3760 : goto redo;
3761 : }
3762 : } else {
3763 : /* Update the free list under the local lock */
3764 : local_lock(&s->cpu_slab->lock);
3765 : c = this_cpu_ptr(s->cpu_slab);
3766 : if (unlikely(slab != c->slab)) {
3767 : local_unlock(&s->cpu_slab->lock);
3768 : goto redo;
3769 : }
3770 : tid = c->tid;
3771 : freelist = c->freelist;
3772 :
3773 : set_freepointer(s, tail_obj, freelist);
3774 : c->freelist = head;
3775 : c->tid = next_tid(tid);
3776 :
3777 : local_unlock(&s->cpu_slab->lock);
3778 : }
3779 : stat(s, FREE_FASTPATH);
3780 : }
3781 : #else /* CONFIG_SLUB_TINY */
3782 : static void do_slab_free(struct kmem_cache *s,
3783 : struct slab *slab, void *head, void *tail,
3784 : int cnt, unsigned long addr)
3785 : {
3786 : void *tail_obj = tail ? : head;
3787 :
3788 : __slab_free(s, slab, head, tail_obj, cnt, addr);
3789 : }
3790 : #endif /* CONFIG_SLUB_TINY */
3791 :
3792 : static __fastpath_inline void slab_free(struct kmem_cache *s, struct slab *slab,
3793 : void *head, void *tail, void **p, int cnt,
3794 : unsigned long addr)
3795 : {
3796 4423 : memcg_slab_free_hook(s, slab, p, cnt);
3797 : /*
3798 : * With KASAN enabled slab_free_freelist_hook modifies the freelist
3799 : * to remove objects, whose reuse must be delayed.
3800 : */
3801 4423 : if (slab_free_freelist_hook(s, &head, &tail, &cnt))
3802 4423 : do_slab_free(s, slab, head, tail, cnt, addr);
3803 : }
3804 :
3805 : #ifdef CONFIG_KASAN_GENERIC
3806 : void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
3807 : {
3808 : do_slab_free(cache, virt_to_slab(x), x, NULL, 1, addr);
3809 : }
3810 : #endif
3811 :
3812 3127 : void __kmem_cache_free(struct kmem_cache *s, void *x, unsigned long caller)
3813 : {
3814 9381 : slab_free(s, virt_to_slab(x), x, NULL, &x, 1, caller);
3815 3127 : }
3816 :
3817 1296 : void kmem_cache_free(struct kmem_cache *s, void *x)
3818 : {
3819 1296 : s = cache_from_obj(s, x);
3820 1296 : if (!s)
3821 : return;
3822 1296 : trace_kmem_cache_free(_RET_IP_, x, s);
3823 2592 : slab_free(s, virt_to_slab(x), x, NULL, &x, 1, _RET_IP_);
3824 : }
3825 : EXPORT_SYMBOL(kmem_cache_free);
3826 :
3827 : struct detached_freelist {
3828 : struct slab *slab;
3829 : void *tail;
3830 : void *freelist;
3831 : int cnt;
3832 : struct kmem_cache *s;
3833 : };
3834 :
3835 : /*
3836 : * This function progressively scans the array with free objects (with
3837 : * a limited look ahead) and extract objects belonging to the same
3838 : * slab. It builds a detached freelist directly within the given
3839 : * slab/objects. This can happen without any need for
3840 : * synchronization, because the objects are owned by running process.
3841 : * The freelist is build up as a single linked list in the objects.
3842 : * The idea is, that this detached freelist can then be bulk
3843 : * transferred to the real freelist(s), but only requiring a single
3844 : * synchronization primitive. Look ahead in the array is limited due
3845 : * to performance reasons.
3846 : */
3847 : static inline
3848 0 : int build_detached_freelist(struct kmem_cache *s, size_t size,
3849 : void **p, struct detached_freelist *df)
3850 : {
3851 0 : int lookahead = 3;
3852 : void *object;
3853 : struct folio *folio;
3854 : size_t same;
3855 :
3856 0 : object = p[--size];
3857 0 : folio = virt_to_folio(object);
3858 0 : if (!s) {
3859 : /* Handle kalloc'ed objects */
3860 0 : if (unlikely(!folio_test_slab(folio))) {
3861 0 : free_large_kmalloc(folio, object);
3862 0 : df->slab = NULL;
3863 0 : return size;
3864 : }
3865 : /* Derive kmem_cache from object */
3866 0 : df->slab = folio_slab(folio);
3867 0 : df->s = df->slab->slab_cache;
3868 : } else {
3869 0 : df->slab = folio_slab(folio);
3870 0 : df->s = cache_from_obj(s, object); /* Support for memcg */
3871 : }
3872 :
3873 : /* Start new detached freelist */
3874 0 : df->tail = object;
3875 0 : df->freelist = object;
3876 0 : df->cnt = 1;
3877 :
3878 0 : if (is_kfence_address(object))
3879 : return size;
3880 :
3881 0 : set_freepointer(df->s, object, NULL);
3882 :
3883 0 : same = size;
3884 0 : while (size) {
3885 0 : object = p[--size];
3886 : /* df->slab is always set at this point */
3887 0 : if (df->slab == virt_to_slab(object)) {
3888 : /* Opportunity build freelist */
3889 0 : set_freepointer(df->s, object, df->freelist);
3890 0 : df->freelist = object;
3891 0 : df->cnt++;
3892 0 : same--;
3893 0 : if (size != same)
3894 0 : swap(p[size], p[same]);
3895 0 : continue;
3896 : }
3897 :
3898 : /* Limit look ahead search */
3899 0 : if (!--lookahead)
3900 : break;
3901 : }
3902 :
3903 0 : return same;
3904 : }
3905 :
3906 : /* Note that interrupts must be enabled when calling this function. */
3907 0 : void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
3908 : {
3909 0 : if (!size)
3910 : return;
3911 :
3912 : do {
3913 : struct detached_freelist df;
3914 :
3915 0 : size = build_detached_freelist(s, size, p, &df);
3916 0 : if (!df.slab)
3917 0 : continue;
3918 :
3919 0 : slab_free(df.s, df.slab, df.freelist, df.tail, &p[size], df.cnt,
3920 0 : _RET_IP_);
3921 0 : } while (likely(size));
3922 : }
3923 : EXPORT_SYMBOL(kmem_cache_free_bulk);
3924 :
3925 : #ifndef CONFIG_SLUB_TINY
3926 7 : static inline int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags,
3927 : size_t size, void **p, struct obj_cgroup *objcg)
3928 : {
3929 : struct kmem_cache_cpu *c;
3930 : unsigned long irqflags;
3931 : int i;
3932 :
3933 : /*
3934 : * Drain objects in the per cpu slab, while disabling local
3935 : * IRQs, which protects against PREEMPT and interrupts
3936 : * handlers invoking normal fastpath.
3937 : */
3938 7 : c = slub_get_cpu_ptr(s->cpu_slab);
3939 7 : local_lock_irqsave(&s->cpu_slab->lock, irqflags);
3940 :
3941 33 : for (i = 0; i < size; i++) {
3942 26 : void *object = kfence_alloc(s, s->object_size, flags);
3943 :
3944 : if (unlikely(object)) {
3945 : p[i] = object;
3946 : continue;
3947 : }
3948 :
3949 26 : object = c->freelist;
3950 26 : if (unlikely(!object)) {
3951 : /*
3952 : * We may have removed an object from c->freelist using
3953 : * the fastpath in the previous iteration; in that case,
3954 : * c->tid has not been bumped yet.
3955 : * Since ___slab_alloc() may reenable interrupts while
3956 : * allocating memory, we should bump c->tid now.
3957 : */
3958 2 : c->tid = next_tid(c->tid);
3959 :
3960 2 : local_unlock_irqrestore(&s->cpu_slab->lock, irqflags);
3961 :
3962 : /*
3963 : * Invoking slow path likely have side-effect
3964 : * of re-populating per CPU c->freelist
3965 : */
3966 2 : p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
3967 1 : _RET_IP_, c, s->object_size);
3968 1 : if (unlikely(!p[i]))
3969 : goto error;
3970 :
3971 1 : c = this_cpu_ptr(s->cpu_slab);
3972 2 : maybe_wipe_obj_freeptr(s, p[i]);
3973 :
3974 1 : local_lock_irqsave(&s->cpu_slab->lock, irqflags);
3975 :
3976 1 : continue; /* goto for-loop */
3977 : }
3978 50 : c->freelist = get_freepointer(s, object);
3979 25 : p[i] = object;
3980 25 : maybe_wipe_obj_freeptr(s, p[i]);
3981 : }
3982 14 : c->tid = next_tid(c->tid);
3983 14 : local_unlock_irqrestore(&s->cpu_slab->lock, irqflags);
3984 7 : slub_put_cpu_ptr(s->cpu_slab);
3985 :
3986 7 : return i;
3987 :
3988 : error:
3989 0 : slub_put_cpu_ptr(s->cpu_slab);
3990 0 : slab_post_alloc_hook(s, objcg, flags, i, p, false, s->object_size);
3991 0 : kmem_cache_free_bulk(s, i, p);
3992 0 : return 0;
3993 :
3994 : }
3995 : #else /* CONFIG_SLUB_TINY */
3996 : static int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags,
3997 : size_t size, void **p, struct obj_cgroup *objcg)
3998 : {
3999 : int i;
4000 :
4001 : for (i = 0; i < size; i++) {
4002 : void *object = kfence_alloc(s, s->object_size, flags);
4003 :
4004 : if (unlikely(object)) {
4005 : p[i] = object;
4006 : continue;
4007 : }
4008 :
4009 : p[i] = __slab_alloc_node(s, flags, NUMA_NO_NODE,
4010 : _RET_IP_, s->object_size);
4011 : if (unlikely(!p[i]))
4012 : goto error;
4013 :
4014 : maybe_wipe_obj_freeptr(s, p[i]);
4015 : }
4016 :
4017 : return i;
4018 :
4019 : error:
4020 : slab_post_alloc_hook(s, objcg, flags, i, p, false, s->object_size);
4021 : kmem_cache_free_bulk(s, i, p);
4022 : return 0;
4023 : }
4024 : #endif /* CONFIG_SLUB_TINY */
4025 :
4026 : /* Note that interrupts must be enabled when calling this function. */
4027 7 : int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
4028 : void **p)
4029 : {
4030 : int i;
4031 7 : struct obj_cgroup *objcg = NULL;
4032 :
4033 7 : if (!size)
4034 : return 0;
4035 :
4036 : /* memcg and kmem_cache debug support */
4037 14 : s = slab_pre_alloc_hook(s, NULL, &objcg, size, flags);
4038 7 : if (unlikely(!s))
4039 : return 0;
4040 :
4041 7 : i = __kmem_cache_alloc_bulk(s, flags, size, p, objcg);
4042 :
4043 : /*
4044 : * memcg and kmem_cache debug support and memory initialization.
4045 : * Done outside of the IRQ disabled fastpath loop.
4046 : */
4047 7 : if (i != 0)
4048 14 : slab_post_alloc_hook(s, objcg, flags, size, p,
4049 14 : slab_want_init_on_alloc(flags, s), s->object_size);
4050 : return i;
4051 : }
4052 : EXPORT_SYMBOL(kmem_cache_alloc_bulk);
4053 :
4054 :
4055 : /*
4056 : * Object placement in a slab is made very easy because we always start at
4057 : * offset 0. If we tune the size of the object to the alignment then we can
4058 : * get the required alignment by putting one properly sized object after
4059 : * another.
4060 : *
4061 : * Notice that the allocation order determines the sizes of the per cpu
4062 : * caches. Each processor has always one slab available for allocations.
4063 : * Increasing the allocation order reduces the number of times that slabs
4064 : * must be moved on and off the partial lists and is therefore a factor in
4065 : * locking overhead.
4066 : */
4067 :
4068 : /*
4069 : * Minimum / Maximum order of slab pages. This influences locking overhead
4070 : * and slab fragmentation. A higher order reduces the number of partial slabs
4071 : * and increases the number of allocations possible without having to
4072 : * take the list_lock.
4073 : */
4074 : static unsigned int slub_min_order;
4075 : static unsigned int slub_max_order =
4076 : IS_ENABLED(CONFIG_SLUB_TINY) ? 1 : PAGE_ALLOC_COSTLY_ORDER;
4077 : static unsigned int slub_min_objects;
4078 :
4079 : /*
4080 : * Calculate the order of allocation given an slab object size.
4081 : *
4082 : * The order of allocation has significant impact on performance and other
4083 : * system components. Generally order 0 allocations should be preferred since
4084 : * order 0 does not cause fragmentation in the page allocator. Larger objects
4085 : * be problematic to put into order 0 slabs because there may be too much
4086 : * unused space left. We go to a higher order if more than 1/16th of the slab
4087 : * would be wasted.
4088 : *
4089 : * In order to reach satisfactory performance we must ensure that a minimum
4090 : * number of objects is in one slab. Otherwise we may generate too much
4091 : * activity on the partial lists which requires taking the list_lock. This is
4092 : * less a concern for large slabs though which are rarely used.
4093 : *
4094 : * slub_max_order specifies the order where we begin to stop considering the
4095 : * number of objects in a slab as critical. If we reach slub_max_order then
4096 : * we try to keep the page order as low as possible. So we accept more waste
4097 : * of space in favor of a small page order.
4098 : *
4099 : * Higher order allocations also allow the placement of more objects in a
4100 : * slab and thereby reduce object handling overhead. If the user has
4101 : * requested a higher minimum order then we start with that one instead of
4102 : * the smallest order which will fit the object.
4103 : */
4104 55 : static inline unsigned int calc_slab_order(unsigned int size,
4105 : unsigned int min_objects, unsigned int max_order,
4106 : unsigned int fract_leftover)
4107 : {
4108 55 : unsigned int min_order = slub_min_order;
4109 : unsigned int order;
4110 :
4111 55 : if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
4112 0 : return get_order(size * MAX_OBJS_PER_PAGE) - 1;
4113 :
4114 168 : for (order = max(min_order, (unsigned int)get_order(min_objects * size));
4115 3 : order <= max_order; order++) {
4116 :
4117 56 : unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
4118 : unsigned int rem;
4119 :
4120 56 : rem = slab_size % size;
4121 :
4122 56 : if (rem <= slab_size / fract_leftover)
4123 : break;
4124 : }
4125 :
4126 : return order;
4127 : }
4128 :
4129 53 : static inline int calculate_order(unsigned int size)
4130 : {
4131 : unsigned int order;
4132 : unsigned int min_objects;
4133 : unsigned int max_objects;
4134 : unsigned int nr_cpus;
4135 :
4136 : /*
4137 : * Attempt to find best configuration for a slab. This
4138 : * works by first attempting to generate a layout with
4139 : * the best configuration and backing off gradually.
4140 : *
4141 : * First we increase the acceptable waste in a slab. Then
4142 : * we reduce the minimum objects required in a slab.
4143 : */
4144 53 : min_objects = slub_min_objects;
4145 53 : if (!min_objects) {
4146 : /*
4147 : * Some architectures will only update present cpus when
4148 : * onlining them, so don't trust the number if it's just 1. But
4149 : * we also don't want to use nr_cpu_ids always, as on some other
4150 : * architectures, there can be many possible cpus, but never
4151 : * onlined. Here we compromise between trying to avoid too high
4152 : * order on systems that appear larger than they are, and too
4153 : * low order on systems that appear smaller than they are.
4154 : */
4155 53 : nr_cpus = num_present_cpus();
4156 : if (nr_cpus <= 1)
4157 53 : nr_cpus = nr_cpu_ids;
4158 53 : min_objects = 4 * (fls(nr_cpus) + 1);
4159 : }
4160 106 : max_objects = order_objects(slub_max_order, size);
4161 53 : min_objects = min(min_objects, max_objects);
4162 :
4163 106 : while (min_objects > 1) {
4164 : unsigned int fraction;
4165 :
4166 : fraction = 16;
4167 55 : while (fraction >= 4) {
4168 55 : order = calc_slab_order(size, min_objects,
4169 : slub_max_order, fraction);
4170 55 : if (order <= slub_max_order)
4171 53 : return order;
4172 2 : fraction /= 2;
4173 : }
4174 0 : min_objects--;
4175 : }
4176 :
4177 : /*
4178 : * We were unable to place multiple objects in a slab. Now
4179 : * lets see if we can place a single object there.
4180 : */
4181 0 : order = calc_slab_order(size, 1, slub_max_order, 1);
4182 0 : if (order <= slub_max_order)
4183 0 : return order;
4184 :
4185 : /*
4186 : * Doh this slab cannot be placed using slub_max_order.
4187 : */
4188 0 : order = calc_slab_order(size, 1, MAX_ORDER, 1);
4189 0 : if (order <= MAX_ORDER)
4190 0 : return order;
4191 : return -ENOSYS;
4192 : }
4193 :
4194 : static void
4195 : init_kmem_cache_node(struct kmem_cache_node *n)
4196 : {
4197 53 : n->nr_partial = 0;
4198 53 : spin_lock_init(&n->list_lock);
4199 106 : INIT_LIST_HEAD(&n->partial);
4200 : #ifdef CONFIG_SLUB_DEBUG
4201 106 : atomic_long_set(&n->nr_slabs, 0);
4202 106 : atomic_long_set(&n->total_objects, 0);
4203 106 : INIT_LIST_HEAD(&n->full);
4204 : #endif
4205 : }
4206 :
4207 : #ifndef CONFIG_SLUB_TINY
4208 53 : static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
4209 : {
4210 : BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
4211 : NR_KMALLOC_TYPES * KMALLOC_SHIFT_HIGH *
4212 : sizeof(struct kmem_cache_cpu));
4213 :
4214 : /*
4215 : * Must align to double word boundary for the double cmpxchg
4216 : * instructions to work; see __pcpu_double_call_return_bool().
4217 : */
4218 53 : s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
4219 : 2 * sizeof(void *));
4220 :
4221 53 : if (!s->cpu_slab)
4222 : return 0;
4223 :
4224 : init_kmem_cache_cpus(s);
4225 :
4226 : return 1;
4227 : }
4228 : #else
4229 : static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
4230 : {
4231 : return 1;
4232 : }
4233 : #endif /* CONFIG_SLUB_TINY */
4234 :
4235 : static struct kmem_cache *kmem_cache_node;
4236 :
4237 : /*
4238 : * No kmalloc_node yet so do it by hand. We know that this is the first
4239 : * slab on the node for this slabcache. There are no concurrent accesses
4240 : * possible.
4241 : *
4242 : * Note that this function only works on the kmem_cache_node
4243 : * when allocating for the kmem_cache_node. This is used for bootstrapping
4244 : * memory on a fresh node that has no slab structures yet.
4245 : */
4246 1 : static void early_kmem_cache_node_alloc(int node)
4247 : {
4248 : struct slab *slab;
4249 : struct kmem_cache_node *n;
4250 :
4251 1 : BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
4252 :
4253 1 : slab = new_slab(kmem_cache_node, GFP_NOWAIT, node);
4254 :
4255 1 : BUG_ON(!slab);
4256 3 : inc_slabs_node(kmem_cache_node, slab_nid(slab), slab->objects);
4257 2 : if (slab_nid(slab) != node) {
4258 0 : pr_err("SLUB: Unable to allocate memory from node %d\n", node);
4259 0 : pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
4260 : }
4261 :
4262 1 : n = slab->freelist;
4263 1 : BUG_ON(!n);
4264 : #ifdef CONFIG_SLUB_DEBUG
4265 1 : init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
4266 1 : init_tracking(kmem_cache_node, n);
4267 : #endif
4268 1 : n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false);
4269 2 : slab->freelist = get_freepointer(kmem_cache_node, n);
4270 1 : slab->inuse = 1;
4271 1 : kmem_cache_node->node[node] = n;
4272 1 : init_kmem_cache_node(n);
4273 2 : inc_slabs_node(kmem_cache_node, node, slab->objects);
4274 :
4275 : /*
4276 : * No locks need to be taken here as it has just been
4277 : * initialized and there is no concurrent access.
4278 : */
4279 1 : __add_partial(n, slab, DEACTIVATE_TO_HEAD);
4280 1 : }
4281 :
4282 0 : static void free_kmem_cache_nodes(struct kmem_cache *s)
4283 : {
4284 : int node;
4285 : struct kmem_cache_node *n;
4286 :
4287 0 : for_each_kmem_cache_node(s, node, n) {
4288 0 : s->node[node] = NULL;
4289 0 : kmem_cache_free(kmem_cache_node, n);
4290 : }
4291 0 : }
4292 :
4293 0 : void __kmem_cache_release(struct kmem_cache *s)
4294 : {
4295 0 : cache_random_seq_destroy(s);
4296 : #ifndef CONFIG_SLUB_TINY
4297 0 : free_percpu(s->cpu_slab);
4298 : #endif
4299 0 : free_kmem_cache_nodes(s);
4300 0 : }
4301 :
4302 53 : static int init_kmem_cache_nodes(struct kmem_cache *s)
4303 : {
4304 : int node;
4305 :
4306 159 : for_each_node_mask(node, slab_nodes) {
4307 : struct kmem_cache_node *n;
4308 :
4309 53 : if (slab_state == DOWN) {
4310 1 : early_kmem_cache_node_alloc(node);
4311 1 : continue;
4312 : }
4313 52 : n = kmem_cache_alloc_node(kmem_cache_node,
4314 : GFP_KERNEL, node);
4315 :
4316 52 : if (!n) {
4317 0 : free_kmem_cache_nodes(s);
4318 0 : return 0;
4319 : }
4320 :
4321 52 : init_kmem_cache_node(n);
4322 52 : s->node[node] = n;
4323 : }
4324 : return 1;
4325 : }
4326 :
4327 : static void set_cpu_partial(struct kmem_cache *s)
4328 : {
4329 : #ifdef CONFIG_SLUB_CPU_PARTIAL
4330 : unsigned int nr_objects;
4331 :
4332 : /*
4333 : * cpu_partial determined the maximum number of objects kept in the
4334 : * per cpu partial lists of a processor.
4335 : *
4336 : * Per cpu partial lists mainly contain slabs that just have one
4337 : * object freed. If they are used for allocation then they can be
4338 : * filled up again with minimal effort. The slab will never hit the
4339 : * per node partial lists and therefore no locking will be required.
4340 : *
4341 : * For backwards compatibility reasons, this is determined as number
4342 : * of objects, even though we now limit maximum number of pages, see
4343 : * slub_set_cpu_partial()
4344 : */
4345 : if (!kmem_cache_has_cpu_partial(s))
4346 : nr_objects = 0;
4347 : else if (s->size >= PAGE_SIZE)
4348 : nr_objects = 6;
4349 : else if (s->size >= 1024)
4350 : nr_objects = 24;
4351 : else if (s->size >= 256)
4352 : nr_objects = 52;
4353 : else
4354 : nr_objects = 120;
4355 :
4356 : slub_set_cpu_partial(s, nr_objects);
4357 : #endif
4358 : }
4359 :
4360 : /*
4361 : * calculate_sizes() determines the order and the distribution of data within
4362 : * a slab object.
4363 : */
4364 53 : static int calculate_sizes(struct kmem_cache *s)
4365 : {
4366 53 : slab_flags_t flags = s->flags;
4367 53 : unsigned int size = s->object_size;
4368 : unsigned int order;
4369 :
4370 : /*
4371 : * Round up object size to the next word boundary. We can only
4372 : * place the free pointer at word boundaries and this determines
4373 : * the possible location of the free pointer.
4374 : */
4375 53 : size = ALIGN(size, sizeof(void *));
4376 :
4377 : #ifdef CONFIG_SLUB_DEBUG
4378 : /*
4379 : * Determine if we can poison the object itself. If the user of
4380 : * the slab may touch the object after free or before allocation
4381 : * then we should never poison the object itself.
4382 : */
4383 53 : if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
4384 0 : !s->ctor)
4385 0 : s->flags |= __OBJECT_POISON;
4386 : else
4387 53 : s->flags &= ~__OBJECT_POISON;
4388 :
4389 :
4390 : /*
4391 : * If we are Redzoning then check if there is some space between the
4392 : * end of the object and the free pointer. If not then add an
4393 : * additional word to have some bytes to store Redzone information.
4394 : */
4395 53 : if ((flags & SLAB_RED_ZONE) && size == s->object_size)
4396 0 : size += sizeof(void *);
4397 : #endif
4398 :
4399 : /*
4400 : * With that we have determined the number of bytes in actual use
4401 : * by the object and redzoning.
4402 : */
4403 53 : s->inuse = size;
4404 :
4405 53 : if (slub_debug_orig_size(s) ||
4406 48 : (flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
4407 48 : ((flags & SLAB_RED_ZONE) && s->object_size < sizeof(void *)) ||
4408 48 : s->ctor) {
4409 : /*
4410 : * Relocate free pointer after the object if it is not
4411 : * permitted to overwrite the first word of the object on
4412 : * kmem_cache_free.
4413 : *
4414 : * This is the case if we do RCU, have a constructor or
4415 : * destructor, are poisoning the objects, or are
4416 : * redzoning an object smaller than sizeof(void *).
4417 : *
4418 : * The assumption that s->offset >= s->inuse means free
4419 : * pointer is outside of the object is used in the
4420 : * freeptr_outside_object() function. If that is no
4421 : * longer true, the function needs to be modified.
4422 : */
4423 10 : s->offset = size;
4424 10 : size += sizeof(void *);
4425 : } else {
4426 : /*
4427 : * Store freelist pointer near middle of object to keep
4428 : * it away from the edges of the object to avoid small
4429 : * sized over/underflows from neighboring allocations.
4430 : */
4431 43 : s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *));
4432 : }
4433 :
4434 : #ifdef CONFIG_SLUB_DEBUG
4435 53 : if (flags & SLAB_STORE_USER) {
4436 : /*
4437 : * Need to store information about allocs and frees after
4438 : * the object.
4439 : */
4440 0 : size += 2 * sizeof(struct track);
4441 :
4442 : /* Save the original kmalloc request size */
4443 0 : if (flags & SLAB_KMALLOC)
4444 0 : size += sizeof(unsigned int);
4445 : }
4446 : #endif
4447 :
4448 53 : kasan_cache_create(s, &size, &s->flags);
4449 : #ifdef CONFIG_SLUB_DEBUG
4450 53 : if (flags & SLAB_RED_ZONE) {
4451 : /*
4452 : * Add some empty padding so that we can catch
4453 : * overwrites from earlier objects rather than let
4454 : * tracking information or the free pointer be
4455 : * corrupted if a user writes before the start
4456 : * of the object.
4457 : */
4458 0 : size += sizeof(void *);
4459 :
4460 : s->red_left_pad = sizeof(void *);
4461 0 : s->red_left_pad = ALIGN(s->red_left_pad, s->align);
4462 0 : size += s->red_left_pad;
4463 : }
4464 : #endif
4465 :
4466 : /*
4467 : * SLUB stores one object immediately after another beginning from
4468 : * offset 0. In order to align the objects we have to simply size
4469 : * each object to conform to the alignment.
4470 : */
4471 53 : size = ALIGN(size, s->align);
4472 53 : s->size = size;
4473 53 : s->reciprocal_size = reciprocal_value(size);
4474 53 : order = calculate_order(size);
4475 :
4476 53 : if ((int)order < 0)
4477 : return 0;
4478 :
4479 53 : s->allocflags = 0;
4480 53 : if (order)
4481 18 : s->allocflags |= __GFP_COMP;
4482 :
4483 53 : if (s->flags & SLAB_CACHE_DMA)
4484 0 : s->allocflags |= GFP_DMA;
4485 :
4486 53 : if (s->flags & SLAB_CACHE_DMA32)
4487 0 : s->allocflags |= GFP_DMA32;
4488 :
4489 53 : if (s->flags & SLAB_RECLAIM_ACCOUNT)
4490 18 : s->allocflags |= __GFP_RECLAIMABLE;
4491 :
4492 : /*
4493 : * Determine the number of objects per slab
4494 : */
4495 106 : s->oo = oo_make(order, size);
4496 159 : s->min = oo_make(get_order(size), size);
4497 :
4498 53 : return !!oo_objects(s->oo);
4499 : }
4500 :
4501 53 : static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
4502 : {
4503 53 : s->flags = kmem_cache_flags(s->size, flags, s->name);
4504 : #ifdef CONFIG_SLAB_FREELIST_HARDENED
4505 : s->random = get_random_long();
4506 : #endif
4507 :
4508 53 : if (!calculate_sizes(s))
4509 : goto error;
4510 53 : if (disable_higher_order_debug) {
4511 : /*
4512 : * Disable debugging flags that store metadata if the min slab
4513 : * order increased.
4514 : */
4515 0 : if (get_order(s->size) > get_order(s->object_size)) {
4516 0 : s->flags &= ~DEBUG_METADATA_FLAGS;
4517 0 : s->offset = 0;
4518 0 : if (!calculate_sizes(s))
4519 : goto error;
4520 : }
4521 : }
4522 :
4523 : #ifdef system_has_freelist_aba
4524 : if (system_has_freelist_aba() && !(s->flags & SLAB_NO_CMPXCHG)) {
4525 : /* Enable fast mode */
4526 : s->flags |= __CMPXCHG_DOUBLE;
4527 : }
4528 : #endif
4529 :
4530 : /*
4531 : * The larger the object size is, the more slabs we want on the partial
4532 : * list to avoid pounding the page allocator excessively.
4533 : */
4534 106 : s->min_partial = min_t(unsigned long, MAX_PARTIAL, ilog2(s->size) / 2);
4535 53 : s->min_partial = max_t(unsigned long, MIN_PARTIAL, s->min_partial);
4536 :
4537 53 : set_cpu_partial(s);
4538 :
4539 : #ifdef CONFIG_NUMA
4540 : s->remote_node_defrag_ratio = 1000;
4541 : #endif
4542 :
4543 : /* Initialize the pre-computed randomized freelist if slab is up */
4544 : if (slab_state >= UP) {
4545 : if (init_cache_random_seq(s))
4546 : goto error;
4547 : }
4548 :
4549 53 : if (!init_kmem_cache_nodes(s))
4550 : goto error;
4551 :
4552 53 : if (alloc_kmem_cache_cpus(s))
4553 : return 0;
4554 :
4555 : error:
4556 0 : __kmem_cache_release(s);
4557 0 : return -EINVAL;
4558 : }
4559 :
4560 0 : static void list_slab_objects(struct kmem_cache *s, struct slab *slab,
4561 : const char *text)
4562 : {
4563 : #ifdef CONFIG_SLUB_DEBUG
4564 0 : void *addr = slab_address(slab);
4565 : void *p;
4566 :
4567 0 : slab_err(s, slab, text, s->name);
4568 :
4569 0 : spin_lock(&object_map_lock);
4570 0 : __fill_map(object_map, s, slab);
4571 :
4572 0 : for_each_object(p, s, addr, slab->objects) {
4573 :
4574 0 : if (!test_bit(__obj_to_index(s, addr, p), object_map)) {
4575 0 : pr_err("Object 0x%p @offset=%tu\n", p, p - addr);
4576 0 : print_tracking(s, p);
4577 : }
4578 : }
4579 0 : spin_unlock(&object_map_lock);
4580 : #endif
4581 0 : }
4582 :
4583 : /*
4584 : * Attempt to free all partial slabs on a node.
4585 : * This is called from __kmem_cache_shutdown(). We must take list_lock
4586 : * because sysfs file might still access partial list after the shutdowning.
4587 : */
4588 0 : static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
4589 : {
4590 0 : LIST_HEAD(discard);
4591 : struct slab *slab, *h;
4592 :
4593 0 : BUG_ON(irqs_disabled());
4594 0 : spin_lock_irq(&n->list_lock);
4595 0 : list_for_each_entry_safe(slab, h, &n->partial, slab_list) {
4596 0 : if (!slab->inuse) {
4597 0 : remove_partial(n, slab);
4598 0 : list_add(&slab->slab_list, &discard);
4599 : } else {
4600 0 : list_slab_objects(s, slab,
4601 : "Objects remaining in %s on __kmem_cache_shutdown()");
4602 : }
4603 : }
4604 0 : spin_unlock_irq(&n->list_lock);
4605 :
4606 0 : list_for_each_entry_safe(slab, h, &discard, slab_list)
4607 0 : discard_slab(s, slab);
4608 0 : }
4609 :
4610 0 : bool __kmem_cache_empty(struct kmem_cache *s)
4611 : {
4612 : int node;
4613 : struct kmem_cache_node *n;
4614 :
4615 0 : for_each_kmem_cache_node(s, node, n)
4616 0 : if (n->nr_partial || node_nr_slabs(n))
4617 : return false;
4618 : return true;
4619 : }
4620 :
4621 : /*
4622 : * Release all resources used by a slab cache.
4623 : */
4624 0 : int __kmem_cache_shutdown(struct kmem_cache *s)
4625 : {
4626 : int node;
4627 : struct kmem_cache_node *n;
4628 :
4629 0 : flush_all_cpus_locked(s);
4630 : /* Attempt to free all objects */
4631 0 : for_each_kmem_cache_node(s, node, n) {
4632 0 : free_partial(s, n);
4633 0 : if (n->nr_partial || node_nr_slabs(n))
4634 : return 1;
4635 : }
4636 : return 0;
4637 : }
4638 :
4639 : #ifdef CONFIG_PRINTK
4640 0 : void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
4641 : {
4642 : void *base;
4643 : int __maybe_unused i;
4644 : unsigned int objnr;
4645 : void *objp;
4646 : void *objp0;
4647 0 : struct kmem_cache *s = slab->slab_cache;
4648 : struct track __maybe_unused *trackp;
4649 :
4650 0 : kpp->kp_ptr = object;
4651 0 : kpp->kp_slab = slab;
4652 0 : kpp->kp_slab_cache = s;
4653 0 : base = slab_address(slab);
4654 0 : objp0 = kasan_reset_tag(object);
4655 : #ifdef CONFIG_SLUB_DEBUG
4656 0 : objp = restore_red_left(s, objp0);
4657 : #else
4658 : objp = objp0;
4659 : #endif
4660 0 : objnr = obj_to_index(s, slab, objp);
4661 0 : kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp);
4662 0 : objp = base + s->size * objnr;
4663 0 : kpp->kp_objp = objp;
4664 0 : if (WARN_ON_ONCE(objp < base || objp >= base + slab->objects * s->size
4665 0 : || (objp - base) % s->size) ||
4666 0 : !(s->flags & SLAB_STORE_USER))
4667 : return;
4668 : #ifdef CONFIG_SLUB_DEBUG
4669 0 : objp = fixup_red_left(s, objp);
4670 0 : trackp = get_track(s, objp, TRACK_ALLOC);
4671 0 : kpp->kp_ret = (void *)trackp->addr;
4672 : #ifdef CONFIG_STACKDEPOT
4673 : {
4674 : depot_stack_handle_t handle;
4675 : unsigned long *entries;
4676 : unsigned int nr_entries;
4677 :
4678 0 : handle = READ_ONCE(trackp->handle);
4679 0 : if (handle) {
4680 0 : nr_entries = stack_depot_fetch(handle, &entries);
4681 0 : for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
4682 0 : kpp->kp_stack[i] = (void *)entries[i];
4683 : }
4684 :
4685 0 : trackp = get_track(s, objp, TRACK_FREE);
4686 0 : handle = READ_ONCE(trackp->handle);
4687 0 : if (handle) {
4688 0 : nr_entries = stack_depot_fetch(handle, &entries);
4689 0 : for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
4690 0 : kpp->kp_free_stack[i] = (void *)entries[i];
4691 : }
4692 : }
4693 : #endif
4694 : #endif
4695 : }
4696 : #endif
4697 :
4698 : /********************************************************************
4699 : * Kmalloc subsystem
4700 : *******************************************************************/
4701 :
4702 0 : static int __init setup_slub_min_order(char *str)
4703 : {
4704 0 : get_option(&str, (int *)&slub_min_order);
4705 :
4706 0 : return 1;
4707 : }
4708 :
4709 : __setup("slub_min_order=", setup_slub_min_order);
4710 :
4711 0 : static int __init setup_slub_max_order(char *str)
4712 : {
4713 0 : get_option(&str, (int *)&slub_max_order);
4714 0 : slub_max_order = min_t(unsigned int, slub_max_order, MAX_ORDER);
4715 :
4716 0 : return 1;
4717 : }
4718 :
4719 : __setup("slub_max_order=", setup_slub_max_order);
4720 :
4721 0 : static int __init setup_slub_min_objects(char *str)
4722 : {
4723 0 : get_option(&str, (int *)&slub_min_objects);
4724 :
4725 0 : return 1;
4726 : }
4727 :
4728 : __setup("slub_min_objects=", setup_slub_min_objects);
4729 :
4730 : #ifdef CONFIG_HARDENED_USERCOPY
4731 : /*
4732 : * Rejects incorrectly sized objects and objects that are to be copied
4733 : * to/from userspace but do not fall entirely within the containing slab
4734 : * cache's usercopy region.
4735 : *
4736 : * Returns NULL if check passes, otherwise const char * to name of cache
4737 : * to indicate an error.
4738 : */
4739 : void __check_heap_object(const void *ptr, unsigned long n,
4740 : const struct slab *slab, bool to_user)
4741 : {
4742 : struct kmem_cache *s;
4743 : unsigned int offset;
4744 : bool is_kfence = is_kfence_address(ptr);
4745 :
4746 : ptr = kasan_reset_tag(ptr);
4747 :
4748 : /* Find object and usable object size. */
4749 : s = slab->slab_cache;
4750 :
4751 : /* Reject impossible pointers. */
4752 : if (ptr < slab_address(slab))
4753 : usercopy_abort("SLUB object not in SLUB page?!", NULL,
4754 : to_user, 0, n);
4755 :
4756 : /* Find offset within object. */
4757 : if (is_kfence)
4758 : offset = ptr - kfence_object_start(ptr);
4759 : else
4760 : offset = (ptr - slab_address(slab)) % s->size;
4761 :
4762 : /* Adjust for redzone and reject if within the redzone. */
4763 : if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) {
4764 : if (offset < s->red_left_pad)
4765 : usercopy_abort("SLUB object in left red zone",
4766 : s->name, to_user, offset, n);
4767 : offset -= s->red_left_pad;
4768 : }
4769 :
4770 : /* Allow address range falling entirely within usercopy region. */
4771 : if (offset >= s->useroffset &&
4772 : offset - s->useroffset <= s->usersize &&
4773 : n <= s->useroffset - offset + s->usersize)
4774 : return;
4775 :
4776 : usercopy_abort("SLUB object", s->name, to_user, offset, n);
4777 : }
4778 : #endif /* CONFIG_HARDENED_USERCOPY */
4779 :
4780 : #define SHRINK_PROMOTE_MAX 32
4781 :
4782 : /*
4783 : * kmem_cache_shrink discards empty slabs and promotes the slabs filled
4784 : * up most to the head of the partial lists. New allocations will then
4785 : * fill those up and thus they can be removed from the partial lists.
4786 : *
4787 : * The slabs with the least items are placed last. This results in them
4788 : * being allocated from last increasing the chance that the last objects
4789 : * are freed in them.
4790 : */
4791 0 : static int __kmem_cache_do_shrink(struct kmem_cache *s)
4792 : {
4793 : int node;
4794 : int i;
4795 : struct kmem_cache_node *n;
4796 : struct slab *slab;
4797 : struct slab *t;
4798 : struct list_head discard;
4799 : struct list_head promote[SHRINK_PROMOTE_MAX];
4800 : unsigned long flags;
4801 0 : int ret = 0;
4802 :
4803 0 : for_each_kmem_cache_node(s, node, n) {
4804 0 : INIT_LIST_HEAD(&discard);
4805 0 : for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
4806 0 : INIT_LIST_HEAD(promote + i);
4807 :
4808 0 : spin_lock_irqsave(&n->list_lock, flags);
4809 :
4810 : /*
4811 : * Build lists of slabs to discard or promote.
4812 : *
4813 : * Note that concurrent frees may occur while we hold the
4814 : * list_lock. slab->inuse here is the upper limit.
4815 : */
4816 0 : list_for_each_entry_safe(slab, t, &n->partial, slab_list) {
4817 0 : int free = slab->objects - slab->inuse;
4818 :
4819 : /* Do not reread slab->inuse */
4820 0 : barrier();
4821 :
4822 : /* We do not keep full slabs on the list */
4823 0 : BUG_ON(free <= 0);
4824 :
4825 0 : if (free == slab->objects) {
4826 0 : list_move(&slab->slab_list, &discard);
4827 0 : n->nr_partial--;
4828 0 : dec_slabs_node(s, node, slab->objects);
4829 0 : } else if (free <= SHRINK_PROMOTE_MAX)
4830 0 : list_move(&slab->slab_list, promote + free - 1);
4831 : }
4832 :
4833 : /*
4834 : * Promote the slabs filled up most to the head of the
4835 : * partial list.
4836 : */
4837 0 : for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
4838 0 : list_splice(promote + i, &n->partial);
4839 :
4840 0 : spin_unlock_irqrestore(&n->list_lock, flags);
4841 :
4842 : /* Release empty slabs */
4843 0 : list_for_each_entry_safe(slab, t, &discard, slab_list)
4844 0 : free_slab(s, slab);
4845 :
4846 0 : if (node_nr_slabs(n))
4847 0 : ret = 1;
4848 : }
4849 :
4850 0 : return ret;
4851 : }
4852 :
4853 0 : int __kmem_cache_shrink(struct kmem_cache *s)
4854 : {
4855 0 : flush_all(s);
4856 0 : return __kmem_cache_do_shrink(s);
4857 : }
4858 :
4859 : static int slab_mem_going_offline_callback(void *arg)
4860 : {
4861 : struct kmem_cache *s;
4862 :
4863 : mutex_lock(&slab_mutex);
4864 : list_for_each_entry(s, &slab_caches, list) {
4865 : flush_all_cpus_locked(s);
4866 : __kmem_cache_do_shrink(s);
4867 : }
4868 : mutex_unlock(&slab_mutex);
4869 :
4870 : return 0;
4871 : }
4872 :
4873 : static void slab_mem_offline_callback(void *arg)
4874 : {
4875 : struct memory_notify *marg = arg;
4876 : int offline_node;
4877 :
4878 : offline_node = marg->status_change_nid_normal;
4879 :
4880 : /*
4881 : * If the node still has available memory. we need kmem_cache_node
4882 : * for it yet.
4883 : */
4884 : if (offline_node < 0)
4885 : return;
4886 :
4887 : mutex_lock(&slab_mutex);
4888 : node_clear(offline_node, slab_nodes);
4889 : /*
4890 : * We no longer free kmem_cache_node structures here, as it would be
4891 : * racy with all get_node() users, and infeasible to protect them with
4892 : * slab_mutex.
4893 : */
4894 : mutex_unlock(&slab_mutex);
4895 : }
4896 :
4897 : static int slab_mem_going_online_callback(void *arg)
4898 : {
4899 : struct kmem_cache_node *n;
4900 : struct kmem_cache *s;
4901 : struct memory_notify *marg = arg;
4902 : int nid = marg->status_change_nid_normal;
4903 : int ret = 0;
4904 :
4905 : /*
4906 : * If the node's memory is already available, then kmem_cache_node is
4907 : * already created. Nothing to do.
4908 : */
4909 : if (nid < 0)
4910 : return 0;
4911 :
4912 : /*
4913 : * We are bringing a node online. No memory is available yet. We must
4914 : * allocate a kmem_cache_node structure in order to bring the node
4915 : * online.
4916 : */
4917 : mutex_lock(&slab_mutex);
4918 : list_for_each_entry(s, &slab_caches, list) {
4919 : /*
4920 : * The structure may already exist if the node was previously
4921 : * onlined and offlined.
4922 : */
4923 : if (get_node(s, nid))
4924 : continue;
4925 : /*
4926 : * XXX: kmem_cache_alloc_node will fallback to other nodes
4927 : * since memory is not yet available from the node that
4928 : * is brought up.
4929 : */
4930 : n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
4931 : if (!n) {
4932 : ret = -ENOMEM;
4933 : goto out;
4934 : }
4935 : init_kmem_cache_node(n);
4936 : s->node[nid] = n;
4937 : }
4938 : /*
4939 : * Any cache created after this point will also have kmem_cache_node
4940 : * initialized for the new node.
4941 : */
4942 : node_set(nid, slab_nodes);
4943 : out:
4944 : mutex_unlock(&slab_mutex);
4945 : return ret;
4946 : }
4947 :
4948 : static int slab_memory_callback(struct notifier_block *self,
4949 : unsigned long action, void *arg)
4950 : {
4951 : int ret = 0;
4952 :
4953 : switch (action) {
4954 : case MEM_GOING_ONLINE:
4955 : ret = slab_mem_going_online_callback(arg);
4956 : break;
4957 : case MEM_GOING_OFFLINE:
4958 : ret = slab_mem_going_offline_callback(arg);
4959 : break;
4960 : case MEM_OFFLINE:
4961 : case MEM_CANCEL_ONLINE:
4962 : slab_mem_offline_callback(arg);
4963 : break;
4964 : case MEM_ONLINE:
4965 : case MEM_CANCEL_OFFLINE:
4966 : break;
4967 : }
4968 : if (ret)
4969 : ret = notifier_from_errno(ret);
4970 : else
4971 : ret = NOTIFY_OK;
4972 : return ret;
4973 : }
4974 :
4975 : /********************************************************************
4976 : * Basic setup of slabs
4977 : *******************************************************************/
4978 :
4979 : /*
4980 : * Used for early kmem_cache structures that were allocated using
4981 : * the page allocator. Allocate them properly then fix up the pointers
4982 : * that may be pointing to the wrong kmem_cache structure.
4983 : */
4984 :
4985 2 : static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
4986 : {
4987 : int node;
4988 4 : struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
4989 : struct kmem_cache_node *n;
4990 :
4991 4 : memcpy(s, static_cache, kmem_cache->object_size);
4992 :
4993 : /*
4994 : * This runs very early, and only the boot processor is supposed to be
4995 : * up. Even if it weren't true, IRQs are not up so we couldn't fire
4996 : * IPIs around.
4997 : */
4998 2 : __flush_cpu_slab(s, smp_processor_id());
4999 6 : for_each_kmem_cache_node(s, node, n) {
5000 : struct slab *p;
5001 :
5002 4 : list_for_each_entry(p, &n->partial, slab_list)
5003 2 : p->slab_cache = s;
5004 :
5005 : #ifdef CONFIG_SLUB_DEBUG
5006 2 : list_for_each_entry(p, &n->full, slab_list)
5007 0 : p->slab_cache = s;
5008 : #endif
5009 : }
5010 4 : list_add(&s->list, &slab_caches);
5011 2 : return s;
5012 : }
5013 :
5014 1 : void __init kmem_cache_init(void)
5015 : {
5016 : static __initdata struct kmem_cache boot_kmem_cache,
5017 : boot_kmem_cache_node;
5018 : int node;
5019 :
5020 : if (debug_guardpage_minorder())
5021 : slub_max_order = 0;
5022 :
5023 : /* Print slub debugging pointers without hashing */
5024 1 : if (__slub_debug_enabled())
5025 0 : no_hash_pointers_enable(NULL);
5026 :
5027 1 : kmem_cache_node = &boot_kmem_cache_node;
5028 1 : kmem_cache = &boot_kmem_cache;
5029 :
5030 : /*
5031 : * Initialize the nodemask for which we will allocate per node
5032 : * structures. Here we don't need taking slab_mutex yet.
5033 : */
5034 3 : for_each_node_state(node, N_NORMAL_MEMORY)
5035 : node_set(node, slab_nodes);
5036 :
5037 1 : create_boot_cache(kmem_cache_node, "kmem_cache_node",
5038 : sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
5039 :
5040 1 : hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
5041 :
5042 : /* Able to allocate the per node structures */
5043 1 : slab_state = PARTIAL;
5044 :
5045 1 : create_boot_cache(kmem_cache, "kmem_cache",
5046 : offsetof(struct kmem_cache, node) +
5047 : nr_node_ids * sizeof(struct kmem_cache_node *),
5048 : SLAB_HWCACHE_ALIGN, 0, 0);
5049 :
5050 1 : kmem_cache = bootstrap(&boot_kmem_cache);
5051 1 : kmem_cache_node = bootstrap(&boot_kmem_cache_node);
5052 :
5053 : /* Now we can use the kmem_cache to allocate kmalloc slabs */
5054 1 : setup_kmalloc_cache_index_table();
5055 1 : create_kmalloc_caches(0);
5056 :
5057 : /* Setup random freelists for each cache */
5058 1 : init_freelist_randomization();
5059 :
5060 1 : cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
5061 : slub_cpu_dead);
5062 :
5063 1 : pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n",
5064 : cache_line_size(),
5065 : slub_min_order, slub_max_order, slub_min_objects,
5066 : nr_cpu_ids, nr_node_ids);
5067 1 : }
5068 :
5069 1 : void __init kmem_cache_init_late(void)
5070 : {
5071 : #ifndef CONFIG_SLUB_TINY
5072 1 : flushwq = alloc_workqueue("slub_flushwq", WQ_MEM_RECLAIM, 0);
5073 1 : WARN_ON(!flushwq);
5074 : #endif
5075 1 : }
5076 :
5077 : struct kmem_cache *
5078 57 : __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
5079 : slab_flags_t flags, void (*ctor)(void *))
5080 : {
5081 : struct kmem_cache *s;
5082 :
5083 57 : s = find_mergeable(size, align, flags, name, ctor);
5084 57 : if (s) {
5085 32 : if (sysfs_slab_alias(s, name))
5086 : return NULL;
5087 :
5088 32 : s->refcount++;
5089 :
5090 : /*
5091 : * Adjust the object sizes so that we clear
5092 : * the complete object on kzalloc.
5093 : */
5094 32 : s->object_size = max(s->object_size, size);
5095 32 : s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
5096 : }
5097 :
5098 : return s;
5099 : }
5100 :
5101 53 : int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
5102 : {
5103 : int err;
5104 :
5105 53 : err = kmem_cache_open(s, flags);
5106 53 : if (err)
5107 : return err;
5108 :
5109 : /* Mutex is not taken during early boot */
5110 53 : if (slab_state <= UP)
5111 : return 0;
5112 :
5113 0 : err = sysfs_slab_add(s);
5114 0 : if (err) {
5115 0 : __kmem_cache_release(s);
5116 0 : return err;
5117 : }
5118 :
5119 : if (s->flags & SLAB_STORE_USER)
5120 : debugfs_slab_add(s);
5121 :
5122 : return 0;
5123 : }
5124 :
5125 : #ifdef SLAB_SUPPORTS_SYSFS
5126 0 : static int count_inuse(struct slab *slab)
5127 : {
5128 0 : return slab->inuse;
5129 : }
5130 :
5131 0 : static int count_total(struct slab *slab)
5132 : {
5133 0 : return slab->objects;
5134 : }
5135 : #endif
5136 :
5137 : #ifdef CONFIG_SLUB_DEBUG
5138 0 : static void validate_slab(struct kmem_cache *s, struct slab *slab,
5139 : unsigned long *obj_map)
5140 : {
5141 : void *p;
5142 0 : void *addr = slab_address(slab);
5143 :
5144 0 : if (!check_slab(s, slab) || !on_freelist(s, slab, NULL))
5145 : return;
5146 :
5147 : /* Now we know that a valid freelist exists */
5148 0 : __fill_map(obj_map, s, slab);
5149 0 : for_each_object(p, s, addr, slab->objects) {
5150 0 : u8 val = test_bit(__obj_to_index(s, addr, p), obj_map) ?
5151 : SLUB_RED_INACTIVE : SLUB_RED_ACTIVE;
5152 :
5153 0 : if (!check_object(s, slab, p, val))
5154 : break;
5155 : }
5156 : }
5157 :
5158 0 : static int validate_slab_node(struct kmem_cache *s,
5159 : struct kmem_cache_node *n, unsigned long *obj_map)
5160 : {
5161 0 : unsigned long count = 0;
5162 : struct slab *slab;
5163 : unsigned long flags;
5164 :
5165 0 : spin_lock_irqsave(&n->list_lock, flags);
5166 :
5167 0 : list_for_each_entry(slab, &n->partial, slab_list) {
5168 0 : validate_slab(s, slab, obj_map);
5169 0 : count++;
5170 : }
5171 0 : if (count != n->nr_partial) {
5172 0 : pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
5173 : s->name, count, n->nr_partial);
5174 0 : slab_add_kunit_errors();
5175 : }
5176 :
5177 0 : if (!(s->flags & SLAB_STORE_USER))
5178 : goto out;
5179 :
5180 0 : list_for_each_entry(slab, &n->full, slab_list) {
5181 0 : validate_slab(s, slab, obj_map);
5182 0 : count++;
5183 : }
5184 0 : if (count != node_nr_slabs(n)) {
5185 0 : pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
5186 : s->name, count, node_nr_slabs(n));
5187 0 : slab_add_kunit_errors();
5188 : }
5189 :
5190 : out:
5191 0 : spin_unlock_irqrestore(&n->list_lock, flags);
5192 0 : return count;
5193 : }
5194 :
5195 0 : long validate_slab_cache(struct kmem_cache *s)
5196 : {
5197 : int node;
5198 0 : unsigned long count = 0;
5199 : struct kmem_cache_node *n;
5200 : unsigned long *obj_map;
5201 :
5202 0 : obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL);
5203 0 : if (!obj_map)
5204 : return -ENOMEM;
5205 :
5206 0 : flush_all(s);
5207 0 : for_each_kmem_cache_node(s, node, n)
5208 0 : count += validate_slab_node(s, n, obj_map);
5209 :
5210 0 : bitmap_free(obj_map);
5211 :
5212 0 : return count;
5213 : }
5214 : EXPORT_SYMBOL(validate_slab_cache);
5215 :
5216 : #ifdef CONFIG_DEBUG_FS
5217 : /*
5218 : * Generate lists of code addresses where slabcache objects are allocated
5219 : * and freed.
5220 : */
5221 :
5222 : struct location {
5223 : depot_stack_handle_t handle;
5224 : unsigned long count;
5225 : unsigned long addr;
5226 : unsigned long waste;
5227 : long long sum_time;
5228 : long min_time;
5229 : long max_time;
5230 : long min_pid;
5231 : long max_pid;
5232 : DECLARE_BITMAP(cpus, NR_CPUS);
5233 : nodemask_t nodes;
5234 : };
5235 :
5236 : struct loc_track {
5237 : unsigned long max;
5238 : unsigned long count;
5239 : struct location *loc;
5240 : loff_t idx;
5241 : };
5242 :
5243 : static struct dentry *slab_debugfs_root;
5244 :
5245 : static void free_loc_track(struct loc_track *t)
5246 : {
5247 : if (t->max)
5248 : free_pages((unsigned long)t->loc,
5249 : get_order(sizeof(struct location) * t->max));
5250 : }
5251 :
5252 : static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
5253 : {
5254 : struct location *l;
5255 : int order;
5256 :
5257 : order = get_order(sizeof(struct location) * max);
5258 :
5259 : l = (void *)__get_free_pages(flags, order);
5260 : if (!l)
5261 : return 0;
5262 :
5263 : if (t->count) {
5264 : memcpy(l, t->loc, sizeof(struct location) * t->count);
5265 : free_loc_track(t);
5266 : }
5267 : t->max = max;
5268 : t->loc = l;
5269 : return 1;
5270 : }
5271 :
5272 : static int add_location(struct loc_track *t, struct kmem_cache *s,
5273 : const struct track *track,
5274 : unsigned int orig_size)
5275 : {
5276 : long start, end, pos;
5277 : struct location *l;
5278 : unsigned long caddr, chandle, cwaste;
5279 : unsigned long age = jiffies - track->when;
5280 : depot_stack_handle_t handle = 0;
5281 : unsigned int waste = s->object_size - orig_size;
5282 :
5283 : #ifdef CONFIG_STACKDEPOT
5284 : handle = READ_ONCE(track->handle);
5285 : #endif
5286 : start = -1;
5287 : end = t->count;
5288 :
5289 : for ( ; ; ) {
5290 : pos = start + (end - start + 1) / 2;
5291 :
5292 : /*
5293 : * There is nothing at "end". If we end up there
5294 : * we need to add something to before end.
5295 : */
5296 : if (pos == end)
5297 : break;
5298 :
5299 : l = &t->loc[pos];
5300 : caddr = l->addr;
5301 : chandle = l->handle;
5302 : cwaste = l->waste;
5303 : if ((track->addr == caddr) && (handle == chandle) &&
5304 : (waste == cwaste)) {
5305 :
5306 : l->count++;
5307 : if (track->when) {
5308 : l->sum_time += age;
5309 : if (age < l->min_time)
5310 : l->min_time = age;
5311 : if (age > l->max_time)
5312 : l->max_time = age;
5313 :
5314 : if (track->pid < l->min_pid)
5315 : l->min_pid = track->pid;
5316 : if (track->pid > l->max_pid)
5317 : l->max_pid = track->pid;
5318 :
5319 : cpumask_set_cpu(track->cpu,
5320 : to_cpumask(l->cpus));
5321 : }
5322 : node_set(page_to_nid(virt_to_page(track)), l->nodes);
5323 : return 1;
5324 : }
5325 :
5326 : if (track->addr < caddr)
5327 : end = pos;
5328 : else if (track->addr == caddr && handle < chandle)
5329 : end = pos;
5330 : else if (track->addr == caddr && handle == chandle &&
5331 : waste < cwaste)
5332 : end = pos;
5333 : else
5334 : start = pos;
5335 : }
5336 :
5337 : /*
5338 : * Not found. Insert new tracking element.
5339 : */
5340 : if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
5341 : return 0;
5342 :
5343 : l = t->loc + pos;
5344 : if (pos < t->count)
5345 : memmove(l + 1, l,
5346 : (t->count - pos) * sizeof(struct location));
5347 : t->count++;
5348 : l->count = 1;
5349 : l->addr = track->addr;
5350 : l->sum_time = age;
5351 : l->min_time = age;
5352 : l->max_time = age;
5353 : l->min_pid = track->pid;
5354 : l->max_pid = track->pid;
5355 : l->handle = handle;
5356 : l->waste = waste;
5357 : cpumask_clear(to_cpumask(l->cpus));
5358 : cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
5359 : nodes_clear(l->nodes);
5360 : node_set(page_to_nid(virt_to_page(track)), l->nodes);
5361 : return 1;
5362 : }
5363 :
5364 : static void process_slab(struct loc_track *t, struct kmem_cache *s,
5365 : struct slab *slab, enum track_item alloc,
5366 : unsigned long *obj_map)
5367 : {
5368 : void *addr = slab_address(slab);
5369 : bool is_alloc = (alloc == TRACK_ALLOC);
5370 : void *p;
5371 :
5372 : __fill_map(obj_map, s, slab);
5373 :
5374 : for_each_object(p, s, addr, slab->objects)
5375 : if (!test_bit(__obj_to_index(s, addr, p), obj_map))
5376 : add_location(t, s, get_track(s, p, alloc),
5377 : is_alloc ? get_orig_size(s, p) :
5378 : s->object_size);
5379 : }
5380 : #endif /* CONFIG_DEBUG_FS */
5381 : #endif /* CONFIG_SLUB_DEBUG */
5382 :
5383 : #ifdef SLAB_SUPPORTS_SYSFS
5384 : enum slab_stat_type {
5385 : SL_ALL, /* All slabs */
5386 : SL_PARTIAL, /* Only partially allocated slabs */
5387 : SL_CPU, /* Only slabs used for cpu caches */
5388 : SL_OBJECTS, /* Determine allocated objects not slabs */
5389 : SL_TOTAL /* Determine object capacity not slabs */
5390 : };
5391 :
5392 : #define SO_ALL (1 << SL_ALL)
5393 : #define SO_PARTIAL (1 << SL_PARTIAL)
5394 : #define SO_CPU (1 << SL_CPU)
5395 : #define SO_OBJECTS (1 << SL_OBJECTS)
5396 : #define SO_TOTAL (1 << SL_TOTAL)
5397 :
5398 0 : static ssize_t show_slab_objects(struct kmem_cache *s,
5399 : char *buf, unsigned long flags)
5400 : {
5401 0 : unsigned long total = 0;
5402 : int node;
5403 : int x;
5404 : unsigned long *nodes;
5405 0 : int len = 0;
5406 :
5407 0 : nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
5408 0 : if (!nodes)
5409 : return -ENOMEM;
5410 :
5411 0 : if (flags & SO_CPU) {
5412 : int cpu;
5413 :
5414 0 : for_each_possible_cpu(cpu) {
5415 0 : struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
5416 : cpu);
5417 : int node;
5418 : struct slab *slab;
5419 :
5420 0 : slab = READ_ONCE(c->slab);
5421 0 : if (!slab)
5422 0 : continue;
5423 :
5424 0 : node = slab_nid(slab);
5425 0 : if (flags & SO_TOTAL)
5426 0 : x = slab->objects;
5427 0 : else if (flags & SO_OBJECTS)
5428 0 : x = slab->inuse;
5429 : else
5430 : x = 1;
5431 :
5432 0 : total += x;
5433 0 : nodes[node] += x;
5434 :
5435 : #ifdef CONFIG_SLUB_CPU_PARTIAL
5436 : slab = slub_percpu_partial_read_once(c);
5437 : if (slab) {
5438 : node = slab_nid(slab);
5439 : if (flags & SO_TOTAL)
5440 : WARN_ON_ONCE(1);
5441 : else if (flags & SO_OBJECTS)
5442 : WARN_ON_ONCE(1);
5443 : else
5444 : x = slab->slabs;
5445 : total += x;
5446 : nodes[node] += x;
5447 : }
5448 : #endif
5449 : }
5450 : }
5451 :
5452 : /*
5453 : * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex"
5454 : * already held which will conflict with an existing lock order:
5455 : *
5456 : * mem_hotplug_lock->slab_mutex->kernfs_mutex
5457 : *
5458 : * We don't really need mem_hotplug_lock (to hold off
5459 : * slab_mem_going_offline_callback) here because slab's memory hot
5460 : * unplug code doesn't destroy the kmem_cache->node[] data.
5461 : */
5462 :
5463 : #ifdef CONFIG_SLUB_DEBUG
5464 0 : if (flags & SO_ALL) {
5465 : struct kmem_cache_node *n;
5466 :
5467 0 : for_each_kmem_cache_node(s, node, n) {
5468 :
5469 0 : if (flags & SO_TOTAL)
5470 0 : x = node_nr_objs(n);
5471 0 : else if (flags & SO_OBJECTS)
5472 0 : x = node_nr_objs(n) - count_partial(n, count_free);
5473 : else
5474 0 : x = node_nr_slabs(n);
5475 0 : total += x;
5476 0 : nodes[node] += x;
5477 : }
5478 :
5479 : } else
5480 : #endif
5481 0 : if (flags & SO_PARTIAL) {
5482 : struct kmem_cache_node *n;
5483 :
5484 0 : for_each_kmem_cache_node(s, node, n) {
5485 0 : if (flags & SO_TOTAL)
5486 0 : x = count_partial(n, count_total);
5487 0 : else if (flags & SO_OBJECTS)
5488 0 : x = count_partial(n, count_inuse);
5489 : else
5490 0 : x = n->nr_partial;
5491 0 : total += x;
5492 0 : nodes[node] += x;
5493 : }
5494 : }
5495 :
5496 0 : len += sysfs_emit_at(buf, len, "%lu", total);
5497 : #ifdef CONFIG_NUMA
5498 : for (node = 0; node < nr_node_ids; node++) {
5499 : if (nodes[node])
5500 : len += sysfs_emit_at(buf, len, " N%d=%lu",
5501 : node, nodes[node]);
5502 : }
5503 : #endif
5504 0 : len += sysfs_emit_at(buf, len, "\n");
5505 0 : kfree(nodes);
5506 :
5507 0 : return len;
5508 : }
5509 :
5510 : #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
5511 : #define to_slab(n) container_of(n, struct kmem_cache, kobj)
5512 :
5513 : struct slab_attribute {
5514 : struct attribute attr;
5515 : ssize_t (*show)(struct kmem_cache *s, char *buf);
5516 : ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
5517 : };
5518 :
5519 : #define SLAB_ATTR_RO(_name) \
5520 : static struct slab_attribute _name##_attr = __ATTR_RO_MODE(_name, 0400)
5521 :
5522 : #define SLAB_ATTR(_name) \
5523 : static struct slab_attribute _name##_attr = __ATTR_RW_MODE(_name, 0600)
5524 :
5525 0 : static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
5526 : {
5527 0 : return sysfs_emit(buf, "%u\n", s->size);
5528 : }
5529 : SLAB_ATTR_RO(slab_size);
5530 :
5531 0 : static ssize_t align_show(struct kmem_cache *s, char *buf)
5532 : {
5533 0 : return sysfs_emit(buf, "%u\n", s->align);
5534 : }
5535 : SLAB_ATTR_RO(align);
5536 :
5537 0 : static ssize_t object_size_show(struct kmem_cache *s, char *buf)
5538 : {
5539 0 : return sysfs_emit(buf, "%u\n", s->object_size);
5540 : }
5541 : SLAB_ATTR_RO(object_size);
5542 :
5543 0 : static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
5544 : {
5545 0 : return sysfs_emit(buf, "%u\n", oo_objects(s->oo));
5546 : }
5547 : SLAB_ATTR_RO(objs_per_slab);
5548 :
5549 0 : static ssize_t order_show(struct kmem_cache *s, char *buf)
5550 : {
5551 0 : return sysfs_emit(buf, "%u\n", oo_order(s->oo));
5552 : }
5553 : SLAB_ATTR_RO(order);
5554 :
5555 0 : static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
5556 : {
5557 0 : return sysfs_emit(buf, "%lu\n", s->min_partial);
5558 : }
5559 :
5560 0 : static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
5561 : size_t length)
5562 : {
5563 : unsigned long min;
5564 : int err;
5565 :
5566 0 : err = kstrtoul(buf, 10, &min);
5567 0 : if (err)
5568 0 : return err;
5569 :
5570 0 : s->min_partial = min;
5571 0 : return length;
5572 : }
5573 : SLAB_ATTR(min_partial);
5574 :
5575 0 : static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
5576 : {
5577 0 : unsigned int nr_partial = 0;
5578 : #ifdef CONFIG_SLUB_CPU_PARTIAL
5579 : nr_partial = s->cpu_partial;
5580 : #endif
5581 :
5582 0 : return sysfs_emit(buf, "%u\n", nr_partial);
5583 : }
5584 :
5585 0 : static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
5586 : size_t length)
5587 : {
5588 : unsigned int objects;
5589 : int err;
5590 :
5591 0 : err = kstrtouint(buf, 10, &objects);
5592 0 : if (err)
5593 0 : return err;
5594 0 : if (objects && !kmem_cache_has_cpu_partial(s))
5595 : return -EINVAL;
5596 :
5597 0 : slub_set_cpu_partial(s, objects);
5598 0 : flush_all(s);
5599 0 : return length;
5600 : }
5601 : SLAB_ATTR(cpu_partial);
5602 :
5603 0 : static ssize_t ctor_show(struct kmem_cache *s, char *buf)
5604 : {
5605 0 : if (!s->ctor)
5606 : return 0;
5607 0 : return sysfs_emit(buf, "%pS\n", s->ctor);
5608 : }
5609 : SLAB_ATTR_RO(ctor);
5610 :
5611 0 : static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5612 : {
5613 0 : return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
5614 : }
5615 : SLAB_ATTR_RO(aliases);
5616 :
5617 0 : static ssize_t partial_show(struct kmem_cache *s, char *buf)
5618 : {
5619 0 : return show_slab_objects(s, buf, SO_PARTIAL);
5620 : }
5621 : SLAB_ATTR_RO(partial);
5622 :
5623 0 : static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5624 : {
5625 0 : return show_slab_objects(s, buf, SO_CPU);
5626 : }
5627 : SLAB_ATTR_RO(cpu_slabs);
5628 :
5629 0 : static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5630 : {
5631 0 : return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5632 : }
5633 : SLAB_ATTR_RO(objects_partial);
5634 :
5635 0 : static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5636 : {
5637 0 : int objects = 0;
5638 0 : int slabs = 0;
5639 : int cpu __maybe_unused;
5640 0 : int len = 0;
5641 :
5642 : #ifdef CONFIG_SLUB_CPU_PARTIAL
5643 : for_each_online_cpu(cpu) {
5644 : struct slab *slab;
5645 :
5646 : slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5647 :
5648 : if (slab)
5649 : slabs += slab->slabs;
5650 : }
5651 : #endif
5652 :
5653 : /* Approximate half-full slabs, see slub_set_cpu_partial() */
5654 0 : objects = (slabs * oo_objects(s->oo)) / 2;
5655 0 : len += sysfs_emit_at(buf, len, "%d(%d)", objects, slabs);
5656 :
5657 : #ifdef CONFIG_SLUB_CPU_PARTIAL
5658 : for_each_online_cpu(cpu) {
5659 : struct slab *slab;
5660 :
5661 : slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5662 : if (slab) {
5663 : slabs = READ_ONCE(slab->slabs);
5664 : objects = (slabs * oo_objects(s->oo)) / 2;
5665 : len += sysfs_emit_at(buf, len, " C%d=%d(%d)",
5666 : cpu, objects, slabs);
5667 : }
5668 : }
5669 : #endif
5670 0 : len += sysfs_emit_at(buf, len, "\n");
5671 :
5672 0 : return len;
5673 : }
5674 : SLAB_ATTR_RO(slabs_cpu_partial);
5675 :
5676 0 : static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5677 : {
5678 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5679 : }
5680 : SLAB_ATTR_RO(reclaim_account);
5681 :
5682 0 : static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5683 : {
5684 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
5685 : }
5686 : SLAB_ATTR_RO(hwcache_align);
5687 :
5688 : #ifdef CONFIG_ZONE_DMA
5689 : static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5690 : {
5691 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5692 : }
5693 : SLAB_ATTR_RO(cache_dma);
5694 : #endif
5695 :
5696 : #ifdef CONFIG_HARDENED_USERCOPY
5697 : static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5698 : {
5699 : return sysfs_emit(buf, "%u\n", s->usersize);
5700 : }
5701 : SLAB_ATTR_RO(usersize);
5702 : #endif
5703 :
5704 0 : static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5705 : {
5706 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
5707 : }
5708 : SLAB_ATTR_RO(destroy_by_rcu);
5709 :
5710 : #ifdef CONFIG_SLUB_DEBUG
5711 0 : static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5712 : {
5713 0 : return show_slab_objects(s, buf, SO_ALL);
5714 : }
5715 : SLAB_ATTR_RO(slabs);
5716 :
5717 0 : static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5718 : {
5719 0 : return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5720 : }
5721 : SLAB_ATTR_RO(total_objects);
5722 :
5723 0 : static ssize_t objects_show(struct kmem_cache *s, char *buf)
5724 : {
5725 0 : return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
5726 : }
5727 : SLAB_ATTR_RO(objects);
5728 :
5729 0 : static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5730 : {
5731 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
5732 : }
5733 : SLAB_ATTR_RO(sanity_checks);
5734 :
5735 0 : static ssize_t trace_show(struct kmem_cache *s, char *buf)
5736 : {
5737 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5738 : }
5739 : SLAB_ATTR_RO(trace);
5740 :
5741 0 : static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5742 : {
5743 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5744 : }
5745 :
5746 : SLAB_ATTR_RO(red_zone);
5747 :
5748 0 : static ssize_t poison_show(struct kmem_cache *s, char *buf)
5749 : {
5750 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON));
5751 : }
5752 :
5753 : SLAB_ATTR_RO(poison);
5754 :
5755 0 : static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5756 : {
5757 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5758 : }
5759 :
5760 : SLAB_ATTR_RO(store_user);
5761 :
5762 0 : static ssize_t validate_show(struct kmem_cache *s, char *buf)
5763 : {
5764 0 : return 0;
5765 : }
5766 :
5767 0 : static ssize_t validate_store(struct kmem_cache *s,
5768 : const char *buf, size_t length)
5769 : {
5770 0 : int ret = -EINVAL;
5771 :
5772 0 : if (buf[0] == '1' && kmem_cache_debug(s)) {
5773 0 : ret = validate_slab_cache(s);
5774 0 : if (ret >= 0)
5775 0 : ret = length;
5776 : }
5777 0 : return ret;
5778 : }
5779 : SLAB_ATTR(validate);
5780 :
5781 : #endif /* CONFIG_SLUB_DEBUG */
5782 :
5783 : #ifdef CONFIG_FAILSLAB
5784 : static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5785 : {
5786 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5787 : }
5788 :
5789 : static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5790 : size_t length)
5791 : {
5792 : if (s->refcount > 1)
5793 : return -EINVAL;
5794 :
5795 : if (buf[0] == '1')
5796 : WRITE_ONCE(s->flags, s->flags | SLAB_FAILSLAB);
5797 : else
5798 : WRITE_ONCE(s->flags, s->flags & ~SLAB_FAILSLAB);
5799 :
5800 : return length;
5801 : }
5802 : SLAB_ATTR(failslab);
5803 : #endif
5804 :
5805 0 : static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5806 : {
5807 0 : return 0;
5808 : }
5809 :
5810 0 : static ssize_t shrink_store(struct kmem_cache *s,
5811 : const char *buf, size_t length)
5812 : {
5813 0 : if (buf[0] == '1')
5814 0 : kmem_cache_shrink(s);
5815 : else
5816 : return -EINVAL;
5817 0 : return length;
5818 : }
5819 : SLAB_ATTR(shrink);
5820 :
5821 : #ifdef CONFIG_NUMA
5822 : static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
5823 : {
5824 : return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10);
5825 : }
5826 :
5827 : static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
5828 : const char *buf, size_t length)
5829 : {
5830 : unsigned int ratio;
5831 : int err;
5832 :
5833 : err = kstrtouint(buf, 10, &ratio);
5834 : if (err)
5835 : return err;
5836 : if (ratio > 100)
5837 : return -ERANGE;
5838 :
5839 : s->remote_node_defrag_ratio = ratio * 10;
5840 :
5841 : return length;
5842 : }
5843 : SLAB_ATTR(remote_node_defrag_ratio);
5844 : #endif
5845 :
5846 : #ifdef CONFIG_SLUB_STATS
5847 : static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5848 : {
5849 : unsigned long sum = 0;
5850 : int cpu;
5851 : int len = 0;
5852 : int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
5853 :
5854 : if (!data)
5855 : return -ENOMEM;
5856 :
5857 : for_each_online_cpu(cpu) {
5858 : unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
5859 :
5860 : data[cpu] = x;
5861 : sum += x;
5862 : }
5863 :
5864 : len += sysfs_emit_at(buf, len, "%lu", sum);
5865 :
5866 : #ifdef CONFIG_SMP
5867 : for_each_online_cpu(cpu) {
5868 : if (data[cpu])
5869 : len += sysfs_emit_at(buf, len, " C%d=%u",
5870 : cpu, data[cpu]);
5871 : }
5872 : #endif
5873 : kfree(data);
5874 : len += sysfs_emit_at(buf, len, "\n");
5875 :
5876 : return len;
5877 : }
5878 :
5879 : static void clear_stat(struct kmem_cache *s, enum stat_item si)
5880 : {
5881 : int cpu;
5882 :
5883 : for_each_online_cpu(cpu)
5884 : per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
5885 : }
5886 :
5887 : #define STAT_ATTR(si, text) \
5888 : static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5889 : { \
5890 : return show_stat(s, buf, si); \
5891 : } \
5892 : static ssize_t text##_store(struct kmem_cache *s, \
5893 : const char *buf, size_t length) \
5894 : { \
5895 : if (buf[0] != '0') \
5896 : return -EINVAL; \
5897 : clear_stat(s, si); \
5898 : return length; \
5899 : } \
5900 : SLAB_ATTR(text); \
5901 :
5902 : STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5903 : STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5904 : STAT_ATTR(FREE_FASTPATH, free_fastpath);
5905 : STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5906 : STAT_ATTR(FREE_FROZEN, free_frozen);
5907 : STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5908 : STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5909 : STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5910 : STAT_ATTR(ALLOC_SLAB, alloc_slab);
5911 : STAT_ATTR(ALLOC_REFILL, alloc_refill);
5912 : STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
5913 : STAT_ATTR(FREE_SLAB, free_slab);
5914 : STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5915 : STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5916 : STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5917 : STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5918 : STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5919 : STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
5920 : STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
5921 : STAT_ATTR(ORDER_FALLBACK, order_fallback);
5922 : STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5923 : STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
5924 : STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5925 : STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
5926 : STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5927 : STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
5928 : #endif /* CONFIG_SLUB_STATS */
5929 :
5930 : #ifdef CONFIG_KFENCE
5931 : static ssize_t skip_kfence_show(struct kmem_cache *s, char *buf)
5932 : {
5933 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_SKIP_KFENCE));
5934 : }
5935 :
5936 : static ssize_t skip_kfence_store(struct kmem_cache *s,
5937 : const char *buf, size_t length)
5938 : {
5939 : int ret = length;
5940 :
5941 : if (buf[0] == '0')
5942 : s->flags &= ~SLAB_SKIP_KFENCE;
5943 : else if (buf[0] == '1')
5944 : s->flags |= SLAB_SKIP_KFENCE;
5945 : else
5946 : ret = -EINVAL;
5947 :
5948 : return ret;
5949 : }
5950 : SLAB_ATTR(skip_kfence);
5951 : #endif
5952 :
5953 : static struct attribute *slab_attrs[] = {
5954 : &slab_size_attr.attr,
5955 : &object_size_attr.attr,
5956 : &objs_per_slab_attr.attr,
5957 : &order_attr.attr,
5958 : &min_partial_attr.attr,
5959 : &cpu_partial_attr.attr,
5960 : &objects_partial_attr.attr,
5961 : &partial_attr.attr,
5962 : &cpu_slabs_attr.attr,
5963 : &ctor_attr.attr,
5964 : &aliases_attr.attr,
5965 : &align_attr.attr,
5966 : &hwcache_align_attr.attr,
5967 : &reclaim_account_attr.attr,
5968 : &destroy_by_rcu_attr.attr,
5969 : &shrink_attr.attr,
5970 : &slabs_cpu_partial_attr.attr,
5971 : #ifdef CONFIG_SLUB_DEBUG
5972 : &total_objects_attr.attr,
5973 : &objects_attr.attr,
5974 : &slabs_attr.attr,
5975 : &sanity_checks_attr.attr,
5976 : &trace_attr.attr,
5977 : &red_zone_attr.attr,
5978 : &poison_attr.attr,
5979 : &store_user_attr.attr,
5980 : &validate_attr.attr,
5981 : #endif
5982 : #ifdef CONFIG_ZONE_DMA
5983 : &cache_dma_attr.attr,
5984 : #endif
5985 : #ifdef CONFIG_NUMA
5986 : &remote_node_defrag_ratio_attr.attr,
5987 : #endif
5988 : #ifdef CONFIG_SLUB_STATS
5989 : &alloc_fastpath_attr.attr,
5990 : &alloc_slowpath_attr.attr,
5991 : &free_fastpath_attr.attr,
5992 : &free_slowpath_attr.attr,
5993 : &free_frozen_attr.attr,
5994 : &free_add_partial_attr.attr,
5995 : &free_remove_partial_attr.attr,
5996 : &alloc_from_partial_attr.attr,
5997 : &alloc_slab_attr.attr,
5998 : &alloc_refill_attr.attr,
5999 : &alloc_node_mismatch_attr.attr,
6000 : &free_slab_attr.attr,
6001 : &cpuslab_flush_attr.attr,
6002 : &deactivate_full_attr.attr,
6003 : &deactivate_empty_attr.attr,
6004 : &deactivate_to_head_attr.attr,
6005 : &deactivate_to_tail_attr.attr,
6006 : &deactivate_remote_frees_attr.attr,
6007 : &deactivate_bypass_attr.attr,
6008 : &order_fallback_attr.attr,
6009 : &cmpxchg_double_fail_attr.attr,
6010 : &cmpxchg_double_cpu_fail_attr.attr,
6011 : &cpu_partial_alloc_attr.attr,
6012 : &cpu_partial_free_attr.attr,
6013 : &cpu_partial_node_attr.attr,
6014 : &cpu_partial_drain_attr.attr,
6015 : #endif
6016 : #ifdef CONFIG_FAILSLAB
6017 : &failslab_attr.attr,
6018 : #endif
6019 : #ifdef CONFIG_HARDENED_USERCOPY
6020 : &usersize_attr.attr,
6021 : #endif
6022 : #ifdef CONFIG_KFENCE
6023 : &skip_kfence_attr.attr,
6024 : #endif
6025 :
6026 : NULL
6027 : };
6028 :
6029 : static const struct attribute_group slab_attr_group = {
6030 : .attrs = slab_attrs,
6031 : };
6032 :
6033 0 : static ssize_t slab_attr_show(struct kobject *kobj,
6034 : struct attribute *attr,
6035 : char *buf)
6036 : {
6037 : struct slab_attribute *attribute;
6038 : struct kmem_cache *s;
6039 :
6040 0 : attribute = to_slab_attr(attr);
6041 0 : s = to_slab(kobj);
6042 :
6043 0 : if (!attribute->show)
6044 : return -EIO;
6045 :
6046 0 : return attribute->show(s, buf);
6047 : }
6048 :
6049 0 : static ssize_t slab_attr_store(struct kobject *kobj,
6050 : struct attribute *attr,
6051 : const char *buf, size_t len)
6052 : {
6053 : struct slab_attribute *attribute;
6054 : struct kmem_cache *s;
6055 :
6056 0 : attribute = to_slab_attr(attr);
6057 0 : s = to_slab(kobj);
6058 :
6059 0 : if (!attribute->store)
6060 : return -EIO;
6061 :
6062 0 : return attribute->store(s, buf, len);
6063 : }
6064 :
6065 0 : static void kmem_cache_release(struct kobject *k)
6066 : {
6067 0 : slab_kmem_cache_release(to_slab(k));
6068 0 : }
6069 :
6070 : static const struct sysfs_ops slab_sysfs_ops = {
6071 : .show = slab_attr_show,
6072 : .store = slab_attr_store,
6073 : };
6074 :
6075 : static const struct kobj_type slab_ktype = {
6076 : .sysfs_ops = &slab_sysfs_ops,
6077 : .release = kmem_cache_release,
6078 : };
6079 :
6080 : static struct kset *slab_kset;
6081 :
6082 : static inline struct kset *cache_kset(struct kmem_cache *s)
6083 : {
6084 53 : return slab_kset;
6085 : }
6086 :
6087 : #define ID_STR_LENGTH 32
6088 :
6089 : /* Create a unique string id for a slab cache:
6090 : *
6091 : * Format :[flags-]size
6092 : */
6093 41 : static char *create_unique_id(struct kmem_cache *s)
6094 : {
6095 41 : char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
6096 41 : char *p = name;
6097 :
6098 41 : if (!name)
6099 : return ERR_PTR(-ENOMEM);
6100 :
6101 41 : *p++ = ':';
6102 : /*
6103 : * First flags affecting slabcache operations. We will only
6104 : * get here for aliasable slabs so we do not need to support
6105 : * too many flags. The flags here must cover all flags that
6106 : * are matched during merging to guarantee that the id is
6107 : * unique.
6108 : */
6109 41 : if (s->flags & SLAB_CACHE_DMA)
6110 0 : *p++ = 'd';
6111 41 : if (s->flags & SLAB_CACHE_DMA32)
6112 0 : *p++ = 'D';
6113 41 : if (s->flags & SLAB_RECLAIM_ACCOUNT)
6114 14 : *p++ = 'a';
6115 41 : if (s->flags & SLAB_CONSISTENCY_CHECKS)
6116 0 : *p++ = 'F';
6117 : if (s->flags & SLAB_ACCOUNT)
6118 : *p++ = 'A';
6119 41 : if (p != name + 1)
6120 14 : *p++ = '-';
6121 41 : p += snprintf(p, ID_STR_LENGTH - (p - name), "%07u", s->size);
6122 :
6123 41 : if (WARN_ON(p > name + ID_STR_LENGTH - 1)) {
6124 0 : kfree(name);
6125 0 : return ERR_PTR(-EINVAL);
6126 : }
6127 : kmsan_unpoison_memory(name, p - name);
6128 : return name;
6129 : }
6130 :
6131 53 : static int sysfs_slab_add(struct kmem_cache *s)
6132 : {
6133 : int err;
6134 : const char *name;
6135 106 : struct kset *kset = cache_kset(s);
6136 53 : int unmergeable = slab_unmergeable(s);
6137 :
6138 53 : if (!unmergeable && disable_higher_order_debug &&
6139 0 : (slub_debug & DEBUG_METADATA_FLAGS))
6140 0 : unmergeable = 1;
6141 :
6142 53 : if (unmergeable) {
6143 : /*
6144 : * Slabcache can never be merged so we can use the name proper.
6145 : * This is typically the case for debug situations. In that
6146 : * case we can catch duplicate names easily.
6147 : */
6148 12 : sysfs_remove_link(&slab_kset->kobj, s->name);
6149 12 : name = s->name;
6150 : } else {
6151 : /*
6152 : * Create a unique name for the slab as a target
6153 : * for the symlinks.
6154 : */
6155 41 : name = create_unique_id(s);
6156 41 : if (IS_ERR(name))
6157 0 : return PTR_ERR(name);
6158 : }
6159 :
6160 53 : s->kobj.kset = kset;
6161 53 : err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
6162 53 : if (err)
6163 : goto out;
6164 :
6165 53 : err = sysfs_create_group(&s->kobj, &slab_attr_group);
6166 53 : if (err)
6167 : goto out_del_kobj;
6168 :
6169 53 : if (!unmergeable) {
6170 : /* Setup first alias */
6171 41 : sysfs_slab_alias(s, s->name);
6172 : }
6173 : out:
6174 53 : if (!unmergeable)
6175 41 : kfree(name);
6176 : return err;
6177 : out_del_kobj:
6178 0 : kobject_del(&s->kobj);
6179 0 : goto out;
6180 : }
6181 :
6182 0 : void sysfs_slab_unlink(struct kmem_cache *s)
6183 : {
6184 0 : if (slab_state >= FULL)
6185 0 : kobject_del(&s->kobj);
6186 0 : }
6187 :
6188 0 : void sysfs_slab_release(struct kmem_cache *s)
6189 : {
6190 0 : if (slab_state >= FULL)
6191 0 : kobject_put(&s->kobj);
6192 0 : }
6193 :
6194 : /*
6195 : * Need to buffer aliases during bootup until sysfs becomes
6196 : * available lest we lose that information.
6197 : */
6198 : struct saved_alias {
6199 : struct kmem_cache *s;
6200 : const char *name;
6201 : struct saved_alias *next;
6202 : };
6203 :
6204 : static struct saved_alias *alias_list;
6205 :
6206 105 : static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
6207 : {
6208 : struct saved_alias *al;
6209 :
6210 105 : if (slab_state == FULL) {
6211 : /*
6212 : * If we have a leftover link then remove it.
6213 : */
6214 73 : sysfs_remove_link(&slab_kset->kobj, name);
6215 73 : return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
6216 : }
6217 :
6218 32 : al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
6219 32 : if (!al)
6220 : return -ENOMEM;
6221 :
6222 32 : al->s = s;
6223 32 : al->name = name;
6224 32 : al->next = alias_list;
6225 32 : alias_list = al;
6226 32 : kmsan_unpoison_memory(al, sizeof(*al));
6227 32 : return 0;
6228 : }
6229 :
6230 1 : static int __init slab_sysfs_init(void)
6231 : {
6232 : struct kmem_cache *s;
6233 : int err;
6234 :
6235 1 : mutex_lock(&slab_mutex);
6236 :
6237 1 : slab_kset = kset_create_and_add("slab", NULL, kernel_kobj);
6238 1 : if (!slab_kset) {
6239 0 : mutex_unlock(&slab_mutex);
6240 0 : pr_err("Cannot register slab subsystem.\n");
6241 0 : return -ENOMEM;
6242 : }
6243 :
6244 1 : slab_state = FULL;
6245 :
6246 54 : list_for_each_entry(s, &slab_caches, list) {
6247 53 : err = sysfs_slab_add(s);
6248 53 : if (err)
6249 0 : pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
6250 : s->name);
6251 : }
6252 :
6253 33 : while (alias_list) {
6254 32 : struct saved_alias *al = alias_list;
6255 :
6256 32 : alias_list = alias_list->next;
6257 32 : err = sysfs_slab_alias(al->s, al->name);
6258 32 : if (err)
6259 0 : pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
6260 : al->name);
6261 32 : kfree(al);
6262 : }
6263 :
6264 1 : mutex_unlock(&slab_mutex);
6265 1 : return 0;
6266 : }
6267 : late_initcall(slab_sysfs_init);
6268 : #endif /* SLAB_SUPPORTS_SYSFS */
6269 :
6270 : #if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS)
6271 : static int slab_debugfs_show(struct seq_file *seq, void *v)
6272 : {
6273 : struct loc_track *t = seq->private;
6274 : struct location *l;
6275 : unsigned long idx;
6276 :
6277 : idx = (unsigned long) t->idx;
6278 : if (idx < t->count) {
6279 : l = &t->loc[idx];
6280 :
6281 : seq_printf(seq, "%7ld ", l->count);
6282 :
6283 : if (l->addr)
6284 : seq_printf(seq, "%pS", (void *)l->addr);
6285 : else
6286 : seq_puts(seq, "<not-available>");
6287 :
6288 : if (l->waste)
6289 : seq_printf(seq, " waste=%lu/%lu",
6290 : l->count * l->waste, l->waste);
6291 :
6292 : if (l->sum_time != l->min_time) {
6293 : seq_printf(seq, " age=%ld/%llu/%ld",
6294 : l->min_time, div_u64(l->sum_time, l->count),
6295 : l->max_time);
6296 : } else
6297 : seq_printf(seq, " age=%ld", l->min_time);
6298 :
6299 : if (l->min_pid != l->max_pid)
6300 : seq_printf(seq, " pid=%ld-%ld", l->min_pid, l->max_pid);
6301 : else
6302 : seq_printf(seq, " pid=%ld",
6303 : l->min_pid);
6304 :
6305 : if (num_online_cpus() > 1 && !cpumask_empty(to_cpumask(l->cpus)))
6306 : seq_printf(seq, " cpus=%*pbl",
6307 : cpumask_pr_args(to_cpumask(l->cpus)));
6308 :
6309 : if (nr_online_nodes > 1 && !nodes_empty(l->nodes))
6310 : seq_printf(seq, " nodes=%*pbl",
6311 : nodemask_pr_args(&l->nodes));
6312 :
6313 : #ifdef CONFIG_STACKDEPOT
6314 : {
6315 : depot_stack_handle_t handle;
6316 : unsigned long *entries;
6317 : unsigned int nr_entries, j;
6318 :
6319 : handle = READ_ONCE(l->handle);
6320 : if (handle) {
6321 : nr_entries = stack_depot_fetch(handle, &entries);
6322 : seq_puts(seq, "\n");
6323 : for (j = 0; j < nr_entries; j++)
6324 : seq_printf(seq, " %pS\n", (void *)entries[j]);
6325 : }
6326 : }
6327 : #endif
6328 : seq_puts(seq, "\n");
6329 : }
6330 :
6331 : if (!idx && !t->count)
6332 : seq_puts(seq, "No data\n");
6333 :
6334 : return 0;
6335 : }
6336 :
6337 : static void slab_debugfs_stop(struct seq_file *seq, void *v)
6338 : {
6339 : }
6340 :
6341 : static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos)
6342 : {
6343 : struct loc_track *t = seq->private;
6344 :
6345 : t->idx = ++(*ppos);
6346 : if (*ppos <= t->count)
6347 : return ppos;
6348 :
6349 : return NULL;
6350 : }
6351 :
6352 : static int cmp_loc_by_count(const void *a, const void *b, const void *data)
6353 : {
6354 : struct location *loc1 = (struct location *)a;
6355 : struct location *loc2 = (struct location *)b;
6356 :
6357 : if (loc1->count > loc2->count)
6358 : return -1;
6359 : else
6360 : return 1;
6361 : }
6362 :
6363 : static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos)
6364 : {
6365 : struct loc_track *t = seq->private;
6366 :
6367 : t->idx = *ppos;
6368 : return ppos;
6369 : }
6370 :
6371 : static const struct seq_operations slab_debugfs_sops = {
6372 : .start = slab_debugfs_start,
6373 : .next = slab_debugfs_next,
6374 : .stop = slab_debugfs_stop,
6375 : .show = slab_debugfs_show,
6376 : };
6377 :
6378 : static int slab_debug_trace_open(struct inode *inode, struct file *filep)
6379 : {
6380 :
6381 : struct kmem_cache_node *n;
6382 : enum track_item alloc;
6383 : int node;
6384 : struct loc_track *t = __seq_open_private(filep, &slab_debugfs_sops,
6385 : sizeof(struct loc_track));
6386 : struct kmem_cache *s = file_inode(filep)->i_private;
6387 : unsigned long *obj_map;
6388 :
6389 : if (!t)
6390 : return -ENOMEM;
6391 :
6392 : obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL);
6393 : if (!obj_map) {
6394 : seq_release_private(inode, filep);
6395 : return -ENOMEM;
6396 : }
6397 :
6398 : if (strcmp(filep->f_path.dentry->d_name.name, "alloc_traces") == 0)
6399 : alloc = TRACK_ALLOC;
6400 : else
6401 : alloc = TRACK_FREE;
6402 :
6403 : if (!alloc_loc_track(t, PAGE_SIZE / sizeof(struct location), GFP_KERNEL)) {
6404 : bitmap_free(obj_map);
6405 : seq_release_private(inode, filep);
6406 : return -ENOMEM;
6407 : }
6408 :
6409 : for_each_kmem_cache_node(s, node, n) {
6410 : unsigned long flags;
6411 : struct slab *slab;
6412 :
6413 : if (!node_nr_slabs(n))
6414 : continue;
6415 :
6416 : spin_lock_irqsave(&n->list_lock, flags);
6417 : list_for_each_entry(slab, &n->partial, slab_list)
6418 : process_slab(t, s, slab, alloc, obj_map);
6419 : list_for_each_entry(slab, &n->full, slab_list)
6420 : process_slab(t, s, slab, alloc, obj_map);
6421 : spin_unlock_irqrestore(&n->list_lock, flags);
6422 : }
6423 :
6424 : /* Sort locations by count */
6425 : sort_r(t->loc, t->count, sizeof(struct location),
6426 : cmp_loc_by_count, NULL, NULL);
6427 :
6428 : bitmap_free(obj_map);
6429 : return 0;
6430 : }
6431 :
6432 : static int slab_debug_trace_release(struct inode *inode, struct file *file)
6433 : {
6434 : struct seq_file *seq = file->private_data;
6435 : struct loc_track *t = seq->private;
6436 :
6437 : free_loc_track(t);
6438 : return seq_release_private(inode, file);
6439 : }
6440 :
6441 : static const struct file_operations slab_debugfs_fops = {
6442 : .open = slab_debug_trace_open,
6443 : .read = seq_read,
6444 : .llseek = seq_lseek,
6445 : .release = slab_debug_trace_release,
6446 : };
6447 :
6448 : static void debugfs_slab_add(struct kmem_cache *s)
6449 : {
6450 : struct dentry *slab_cache_dir;
6451 :
6452 : if (unlikely(!slab_debugfs_root))
6453 : return;
6454 :
6455 : slab_cache_dir = debugfs_create_dir(s->name, slab_debugfs_root);
6456 :
6457 : debugfs_create_file("alloc_traces", 0400,
6458 : slab_cache_dir, s, &slab_debugfs_fops);
6459 :
6460 : debugfs_create_file("free_traces", 0400,
6461 : slab_cache_dir, s, &slab_debugfs_fops);
6462 : }
6463 :
6464 : void debugfs_slab_release(struct kmem_cache *s)
6465 : {
6466 : debugfs_lookup_and_remove(s->name, slab_debugfs_root);
6467 : }
6468 :
6469 : static int __init slab_debugfs_init(void)
6470 : {
6471 : struct kmem_cache *s;
6472 :
6473 : slab_debugfs_root = debugfs_create_dir("slab", NULL);
6474 :
6475 : list_for_each_entry(s, &slab_caches, list)
6476 : if (s->flags & SLAB_STORE_USER)
6477 : debugfs_slab_add(s);
6478 :
6479 : return 0;
6480 :
6481 : }
6482 : __initcall(slab_debugfs_init);
6483 : #endif
6484 : /*
6485 : * The /proc/slabinfo ABI
6486 : */
6487 : #ifdef CONFIG_SLUB_DEBUG
6488 0 : void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
6489 : {
6490 0 : unsigned long nr_slabs = 0;
6491 0 : unsigned long nr_objs = 0;
6492 0 : unsigned long nr_free = 0;
6493 : int node;
6494 : struct kmem_cache_node *n;
6495 :
6496 0 : for_each_kmem_cache_node(s, node, n) {
6497 0 : nr_slabs += node_nr_slabs(n);
6498 0 : nr_objs += node_nr_objs(n);
6499 0 : nr_free += count_partial(n, count_free);
6500 : }
6501 :
6502 0 : sinfo->active_objs = nr_objs - nr_free;
6503 0 : sinfo->num_objs = nr_objs;
6504 0 : sinfo->active_slabs = nr_slabs;
6505 0 : sinfo->num_slabs = nr_slabs;
6506 0 : sinfo->objects_per_slab = oo_objects(s->oo);
6507 0 : sinfo->cache_order = oo_order(s->oo);
6508 0 : }
6509 :
6510 0 : void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
6511 : {
6512 0 : }
6513 :
6514 0 : ssize_t slabinfo_write(struct file *file, const char __user *buffer,
6515 : size_t count, loff_t *ppos)
6516 : {
6517 0 : return -EIO;
6518 : }
6519 : #endif /* CONFIG_SLUB_DEBUG */
|
