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> /* struct reclaim_state */
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 124756 : return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS);
215 : }
216 :
217 : static inline bool slub_debug_orig_size(struct kmem_cache *s)
218 : {
219 330 : 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 4992 : 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 : #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U)
296 :
297 : /*
298 : * Tracking user of a slab.
299 : */
300 : #define TRACK_ADDRS_COUNT 16
301 : struct track {
302 : unsigned long addr; /* Called from address */
303 : #ifdef CONFIG_STACKDEPOT
304 : depot_stack_handle_t handle;
305 : #endif
306 : int cpu; /* Was running on cpu */
307 : int pid; /* Pid context */
308 : unsigned long when; /* When did the operation occur */
309 : };
310 :
311 : enum track_item { TRACK_ALLOC, TRACK_FREE };
312 :
313 : #ifdef SLAB_SUPPORTS_SYSFS
314 : static int sysfs_slab_add(struct kmem_cache *);
315 : static int sysfs_slab_alias(struct kmem_cache *, const char *);
316 : #else
317 : static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
318 : static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
319 : { return 0; }
320 : #endif
321 :
322 : #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
323 : static void debugfs_slab_add(struct kmem_cache *);
324 : #else
325 : static inline void debugfs_slab_add(struct kmem_cache *s) { }
326 : #endif
327 :
328 : static inline void stat(const struct kmem_cache *s, enum stat_item si)
329 : {
330 : #ifdef CONFIG_SLUB_STATS
331 : /*
332 : * The rmw is racy on a preemptible kernel but this is acceptable, so
333 : * avoid this_cpu_add()'s irq-disable overhead.
334 : */
335 : raw_cpu_inc(s->cpu_slab->stat[si]);
336 : #endif
337 : }
338 :
339 : /*
340 : * Tracks for which NUMA nodes we have kmem_cache_nodes allocated.
341 : * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily
342 : * differ during memory hotplug/hotremove operations.
343 : * Protected by slab_mutex.
344 : */
345 : static nodemask_t slab_nodes;
346 :
347 : #ifndef CONFIG_SLUB_TINY
348 : /*
349 : * Workqueue used for flush_cpu_slab().
350 : */
351 : static struct workqueue_struct *flushwq;
352 : #endif
353 :
354 : /********************************************************************
355 : * Core slab cache functions
356 : *******************************************************************/
357 :
358 : /*
359 : * Returns freelist pointer (ptr). With hardening, this is obfuscated
360 : * with an XOR of the address where the pointer is held and a per-cache
361 : * random number.
362 : */
363 : static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
364 : unsigned long ptr_addr)
365 : {
366 : #ifdef CONFIG_SLAB_FREELIST_HARDENED
367 : /*
368 : * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged.
369 : * Normally, this doesn't cause any issues, as both set_freepointer()
370 : * and get_freepointer() are called with a pointer with the same tag.
371 : * However, there are some issues with CONFIG_SLUB_DEBUG code. For
372 : * example, when __free_slub() iterates over objects in a cache, it
373 : * passes untagged pointers to check_object(). check_object() in turns
374 : * calls get_freepointer() with an untagged pointer, which causes the
375 : * freepointer to be restored incorrectly.
376 : */
377 : return (void *)((unsigned long)ptr ^ s->random ^
378 : swab((unsigned long)kasan_reset_tag((void *)ptr_addr)));
379 : #else
380 : return ptr;
381 : #endif
382 : }
383 :
384 : /* Returns the freelist pointer recorded at location ptr_addr. */
385 : static inline void *freelist_dereference(const struct kmem_cache *s,
386 : void *ptr_addr)
387 : {
388 143717 : return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr),
389 : (unsigned long)ptr_addr);
390 : }
391 :
392 : static inline void *get_freepointer(struct kmem_cache *s, void *object)
393 : {
394 143717 : object = kasan_reset_tag(object);
395 287434 : return freelist_dereference(s, object + s->offset);
396 : }
397 :
398 : #ifndef CONFIG_SLUB_TINY
399 : static void prefetch_freepointer(const struct kmem_cache *s, void *object)
400 : {
401 75500 : prefetchw(object + s->offset);
402 : }
403 : #endif
404 :
405 : /*
406 : * When running under KMSAN, get_freepointer_safe() may return an uninitialized
407 : * pointer value in the case the current thread loses the race for the next
408 : * memory chunk in the freelist. In that case this_cpu_cmpxchg_double() in
409 : * slab_alloc_node() will fail, so the uninitialized value won't be used, but
410 : * KMSAN will still check all arguments of cmpxchg because of imperfect
411 : * handling of inline assembly.
412 : * To work around this problem, we apply __no_kmsan_checks to ensure that
413 : * get_freepointer_safe() returns initialized memory.
414 : */
415 : __no_kmsan_checks
416 : static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
417 : {
418 : unsigned long freepointer_addr;
419 : void *p;
420 :
421 : if (!debug_pagealloc_enabled_static())
422 151000 : return get_freepointer(s, object);
423 :
424 : object = kasan_reset_tag(object);
425 : freepointer_addr = (unsigned long)object + s->offset;
426 : copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p));
427 : return freelist_ptr(s, p, freepointer_addr);
428 : }
429 :
430 : static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
431 : {
432 199323 : unsigned long freeptr_addr = (unsigned long)object + s->offset;
433 :
434 : #ifdef CONFIG_SLAB_FREELIST_HARDENED
435 : BUG_ON(object == fp); /* naive detection of double free or corruption */
436 : #endif
437 :
438 199323 : freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr);
439 199323 : *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
440 : }
441 :
442 : /* Loop over all objects in a slab */
443 : #define for_each_object(__p, __s, __addr, __objects) \
444 : for (__p = fixup_red_left(__s, __addr); \
445 : __p < (__addr) + (__objects) * (__s)->size; \
446 : __p += (__s)->size)
447 :
448 : static inline unsigned int order_objects(unsigned int order, unsigned int size)
449 : {
450 209 : return ((unsigned int)PAGE_SIZE << order) / size;
451 : }
452 :
453 : static inline struct kmem_cache_order_objects oo_make(unsigned int order,
454 : unsigned int size)
455 : {
456 104 : struct kmem_cache_order_objects x = {
457 208 : (order << OO_SHIFT) + order_objects(order, size)
458 : };
459 :
460 : return x;
461 : }
462 :
463 : static inline unsigned int oo_order(struct kmem_cache_order_objects x)
464 : {
465 7450 : return x.x >> OO_SHIFT;
466 : }
467 :
468 : static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
469 : {
470 52 : return x.x & OO_MASK;
471 : }
472 :
473 : #ifdef CONFIG_SLUB_CPU_PARTIAL
474 : static void slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects)
475 : {
476 : unsigned int nr_slabs;
477 :
478 : s->cpu_partial = nr_objects;
479 :
480 : /*
481 : * We take the number of objects but actually limit the number of
482 : * slabs on the per cpu partial list, in order to limit excessive
483 : * growth of the list. For simplicity we assume that the slabs will
484 : * be half-full.
485 : */
486 : nr_slabs = DIV_ROUND_UP(nr_objects * 2, oo_objects(s->oo));
487 : s->cpu_partial_slabs = nr_slabs;
488 : }
489 : #else
490 : static inline void
491 : slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects)
492 : {
493 : }
494 : #endif /* CONFIG_SLUB_CPU_PARTIAL */
495 :
496 : /*
497 : * Per slab locking using the pagelock
498 : */
499 : static __always_inline void slab_lock(struct slab *slab)
500 : {
501 59855 : struct page *page = slab_page(slab);
502 :
503 : VM_BUG_ON_PAGE(PageTail(page), page);
504 59855 : bit_spin_lock(PG_locked, &page->flags);
505 : }
506 :
507 : static __always_inline void slab_unlock(struct slab *slab)
508 : {
509 59855 : struct page *page = slab_page(slab);
510 :
511 : VM_BUG_ON_PAGE(PageTail(page), page);
512 59855 : __bit_spin_unlock(PG_locked, &page->flags);
513 : }
514 :
515 : /*
516 : * Interrupts must be disabled (for the fallback code to work right), typically
517 : * by an _irqsave() lock variant. On PREEMPT_RT the preempt_disable(), which is
518 : * part of bit_spin_lock(), is sufficient because the policy is not to allow any
519 : * allocation/ free operation in hardirq context. Therefore nothing can
520 : * interrupt the operation.
521 : */
522 : static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct slab *slab,
523 : void *freelist_old, unsigned long counters_old,
524 : void *freelist_new, unsigned long counters_new,
525 : const char *n)
526 : {
527 : if (USE_LOCKLESS_FAST_PATH())
528 : lockdep_assert_irqs_disabled();
529 : #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
530 : defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
531 : if (s->flags & __CMPXCHG_DOUBLE) {
532 : if (cmpxchg_double(&slab->freelist, &slab->counters,
533 : freelist_old, counters_old,
534 : freelist_new, counters_new))
535 : return true;
536 : } else
537 : #endif
538 : {
539 2653 : slab_lock(slab);
540 5306 : if (slab->freelist == freelist_old &&
541 2653 : slab->counters == counters_old) {
542 2653 : slab->freelist = freelist_new;
543 2653 : slab->counters = counters_new;
544 2653 : slab_unlock(slab);
545 : return true;
546 : }
547 0 : slab_unlock(slab);
548 : }
549 :
550 : cpu_relax();
551 0 : stat(s, CMPXCHG_DOUBLE_FAIL);
552 :
553 : #ifdef SLUB_DEBUG_CMPXCHG
554 : pr_info("%s %s: cmpxchg double redo ", n, s->name);
555 : #endif
556 :
557 : return false;
558 : }
559 :
560 57202 : static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct slab *slab,
561 : void *freelist_old, unsigned long counters_old,
562 : void *freelist_new, unsigned long counters_new,
563 : const char *n)
564 : {
565 : #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
566 : defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
567 : if (s->flags & __CMPXCHG_DOUBLE) {
568 : if (cmpxchg_double(&slab->freelist, &slab->counters,
569 : freelist_old, counters_old,
570 : freelist_new, counters_new))
571 : return true;
572 : } else
573 : #endif
574 : {
575 : unsigned long flags;
576 :
577 57202 : local_irq_save(flags);
578 57202 : slab_lock(slab);
579 114404 : if (slab->freelist == freelist_old &&
580 57202 : slab->counters == counters_old) {
581 57202 : slab->freelist = freelist_new;
582 57202 : slab->counters = counters_new;
583 57202 : slab_unlock(slab);
584 114404 : local_irq_restore(flags);
585 : return true;
586 : }
587 0 : slab_unlock(slab);
588 0 : local_irq_restore(flags);
589 : }
590 :
591 : cpu_relax();
592 0 : stat(s, CMPXCHG_DOUBLE_FAIL);
593 :
594 : #ifdef SLUB_DEBUG_CMPXCHG
595 : pr_info("%s %s: cmpxchg double redo ", n, s->name);
596 : #endif
597 :
598 : return false;
599 : }
600 :
601 : #ifdef CONFIG_SLUB_DEBUG
602 : static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)];
603 : static DEFINE_SPINLOCK(object_map_lock);
604 :
605 0 : static void __fill_map(unsigned long *obj_map, struct kmem_cache *s,
606 : struct slab *slab)
607 : {
608 0 : void *addr = slab_address(slab);
609 : void *p;
610 :
611 0 : bitmap_zero(obj_map, slab->objects);
612 :
613 0 : for (p = slab->freelist; p; p = get_freepointer(s, p))
614 0 : set_bit(__obj_to_index(s, addr, p), obj_map);
615 0 : }
616 :
617 : #if IS_ENABLED(CONFIG_KUNIT)
618 0 : static bool slab_add_kunit_errors(void)
619 : {
620 : struct kunit_resource *resource;
621 :
622 0 : if (!kunit_get_current_test())
623 : return false;
624 :
625 0 : resource = kunit_find_named_resource(current->kunit_test, "slab_errors");
626 0 : if (!resource)
627 : return false;
628 :
629 0 : (*(int *)resource->data)++;
630 0 : kunit_put_resource(resource);
631 0 : return true;
632 : }
633 : #else
634 : static inline bool slab_add_kunit_errors(void) { return false; }
635 : #endif
636 :
637 : static inline unsigned int size_from_object(struct kmem_cache *s)
638 : {
639 0 : if (s->flags & SLAB_RED_ZONE)
640 0 : return s->size - s->red_left_pad;
641 :
642 : return s->size;
643 : }
644 :
645 : static inline void *restore_red_left(struct kmem_cache *s, void *p)
646 : {
647 0 : if (s->flags & SLAB_RED_ZONE)
648 0 : p -= s->red_left_pad;
649 :
650 : return p;
651 : }
652 :
653 : /*
654 : * Debug settings:
655 : */
656 : #if defined(CONFIG_SLUB_DEBUG_ON)
657 : static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
658 : #else
659 : static slab_flags_t slub_debug;
660 : #endif
661 :
662 : static char *slub_debug_string;
663 : static int disable_higher_order_debug;
664 :
665 : /*
666 : * slub is about to manipulate internal object metadata. This memory lies
667 : * outside the range of the allocated object, so accessing it would normally
668 : * be reported by kasan as a bounds error. metadata_access_enable() is used
669 : * to tell kasan that these accesses are OK.
670 : */
671 : static inline void metadata_access_enable(void)
672 : {
673 : kasan_disable_current();
674 : }
675 :
676 : static inline void metadata_access_disable(void)
677 : {
678 : kasan_enable_current();
679 : }
680 :
681 : /*
682 : * Object debugging
683 : */
684 :
685 : /* Verify that a pointer has an address that is valid within a slab page */
686 0 : static inline int check_valid_pointer(struct kmem_cache *s,
687 : struct slab *slab, void *object)
688 : {
689 : void *base;
690 :
691 0 : if (!object)
692 : return 1;
693 :
694 0 : base = slab_address(slab);
695 0 : object = kasan_reset_tag(object);
696 0 : object = restore_red_left(s, object);
697 0 : if (object < base || object >= base + slab->objects * s->size ||
698 0 : (object - base) % s->size) {
699 : return 0;
700 : }
701 :
702 0 : return 1;
703 : }
704 :
705 : static void print_section(char *level, char *text, u8 *addr,
706 : unsigned int length)
707 : {
708 : metadata_access_enable();
709 0 : print_hex_dump(level, text, DUMP_PREFIX_ADDRESS,
710 0 : 16, 1, kasan_reset_tag((void *)addr), length, 1);
711 : metadata_access_disable();
712 : }
713 :
714 : /*
715 : * See comment in calculate_sizes().
716 : */
717 : static inline bool freeptr_outside_object(struct kmem_cache *s)
718 : {
719 : return s->offset >= s->inuse;
720 : }
721 :
722 : /*
723 : * Return offset of the end of info block which is inuse + free pointer if
724 : * not overlapping with object.
725 : */
726 : static inline unsigned int get_info_end(struct kmem_cache *s)
727 : {
728 0 : if (freeptr_outside_object(s))
729 0 : return s->inuse + sizeof(void *);
730 : else
731 : return s->inuse;
732 : }
733 :
734 : static struct track *get_track(struct kmem_cache *s, void *object,
735 : enum track_item alloc)
736 : {
737 : struct track *p;
738 :
739 0 : p = object + get_info_end(s);
740 :
741 0 : return kasan_reset_tag(p + alloc);
742 : }
743 :
744 : #ifdef CONFIG_STACKDEPOT
745 0 : static noinline depot_stack_handle_t set_track_prepare(void)
746 : {
747 : depot_stack_handle_t handle;
748 : unsigned long entries[TRACK_ADDRS_COUNT];
749 : unsigned int nr_entries;
750 :
751 0 : nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 3);
752 0 : handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT);
753 :
754 0 : return handle;
755 : }
756 : #else
757 : static inline depot_stack_handle_t set_track_prepare(void)
758 : {
759 : return 0;
760 : }
761 : #endif
762 :
763 : static void set_track_update(struct kmem_cache *s, void *object,
764 : enum track_item alloc, unsigned long addr,
765 : depot_stack_handle_t handle)
766 : {
767 0 : struct track *p = get_track(s, object, alloc);
768 :
769 : #ifdef CONFIG_STACKDEPOT
770 0 : p->handle = handle;
771 : #endif
772 0 : p->addr = addr;
773 0 : p->cpu = smp_processor_id();
774 0 : p->pid = current->pid;
775 0 : p->when = jiffies;
776 : }
777 :
778 : static __always_inline void set_track(struct kmem_cache *s, void *object,
779 : enum track_item alloc, unsigned long addr)
780 : {
781 0 : depot_stack_handle_t handle = set_track_prepare();
782 :
783 : set_track_update(s, object, alloc, addr, handle);
784 : }
785 :
786 1 : static void init_tracking(struct kmem_cache *s, void *object)
787 : {
788 : struct track *p;
789 :
790 1 : if (!(s->flags & SLAB_STORE_USER))
791 : return;
792 :
793 0 : p = get_track(s, object, TRACK_ALLOC);
794 0 : memset(p, 0, 2*sizeof(struct track));
795 : }
796 :
797 0 : static void print_track(const char *s, struct track *t, unsigned long pr_time)
798 : {
799 : depot_stack_handle_t handle __maybe_unused;
800 :
801 0 : if (!t->addr)
802 : return;
803 :
804 0 : pr_err("%s in %pS age=%lu cpu=%u pid=%d\n",
805 : s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
806 : #ifdef CONFIG_STACKDEPOT
807 0 : handle = READ_ONCE(t->handle);
808 0 : if (handle)
809 0 : stack_depot_print(handle);
810 : else
811 0 : pr_err("object allocation/free stack trace missing\n");
812 : #endif
813 : }
814 :
815 0 : void print_tracking(struct kmem_cache *s, void *object)
816 : {
817 0 : unsigned long pr_time = jiffies;
818 0 : if (!(s->flags & SLAB_STORE_USER))
819 : return;
820 :
821 0 : print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
822 0 : print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
823 : }
824 :
825 : static void print_slab_info(const struct slab *slab)
826 : {
827 0 : struct folio *folio = (struct folio *)slab_folio(slab);
828 :
829 0 : pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%pGp\n",
830 : slab, slab->objects, slab->inuse, slab->freelist,
831 : folio_flags(folio, 0));
832 : }
833 :
834 : /*
835 : * kmalloc caches has fixed sizes (mostly power of 2), and kmalloc() API
836 : * family will round up the real request size to these fixed ones, so
837 : * there could be an extra area than what is requested. Save the original
838 : * request size in the meta data area, for better debug and sanity check.
839 : */
840 113 : static inline void set_orig_size(struct kmem_cache *s,
841 : void *object, unsigned int orig_size)
842 : {
843 113 : void *p = kasan_reset_tag(object);
844 :
845 113 : if (!slub_debug_orig_size(s))
846 : return;
847 :
848 : #ifdef CONFIG_KASAN_GENERIC
849 : /*
850 : * KASAN could save its free meta data in object's data area at
851 : * offset 0, if the size is larger than 'orig_size', it will
852 : * overlap the data redzone in [orig_size+1, object_size], and
853 : * the check should be skipped.
854 : */
855 : if (kasan_metadata_size(s, true) > orig_size)
856 : orig_size = s->object_size;
857 : #endif
858 :
859 0 : p += get_info_end(s);
860 0 : p += sizeof(struct track) * 2;
861 :
862 0 : *(unsigned int *)p = orig_size;
863 : }
864 :
865 0 : static inline unsigned int get_orig_size(struct kmem_cache *s, void *object)
866 : {
867 0 : void *p = kasan_reset_tag(object);
868 :
869 0 : if (!slub_debug_orig_size(s))
870 0 : return s->object_size;
871 :
872 0 : p += get_info_end(s);
873 0 : p += sizeof(struct track) * 2;
874 :
875 0 : return *(unsigned int *)p;
876 : }
877 :
878 113 : void skip_orig_size_check(struct kmem_cache *s, const void *object)
879 : {
880 113 : set_orig_size(s, (void *)object, s->object_size);
881 113 : }
882 :
883 0 : static void slab_bug(struct kmem_cache *s, char *fmt, ...)
884 : {
885 : struct va_format vaf;
886 : va_list args;
887 :
888 0 : va_start(args, fmt);
889 0 : vaf.fmt = fmt;
890 0 : vaf.va = &args;
891 0 : pr_err("=============================================================================\n");
892 0 : pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
893 0 : pr_err("-----------------------------------------------------------------------------\n\n");
894 0 : va_end(args);
895 0 : }
896 :
897 : __printf(2, 3)
898 0 : static void slab_fix(struct kmem_cache *s, char *fmt, ...)
899 : {
900 : struct va_format vaf;
901 : va_list args;
902 :
903 0 : if (slab_add_kunit_errors())
904 0 : return;
905 :
906 0 : va_start(args, fmt);
907 0 : vaf.fmt = fmt;
908 0 : vaf.va = &args;
909 0 : pr_err("FIX %s: %pV\n", s->name, &vaf);
910 0 : va_end(args);
911 : }
912 :
913 0 : static void print_trailer(struct kmem_cache *s, struct slab *slab, u8 *p)
914 : {
915 : unsigned int off; /* Offset of last byte */
916 0 : u8 *addr = slab_address(slab);
917 :
918 0 : print_tracking(s, p);
919 :
920 0 : print_slab_info(slab);
921 :
922 0 : pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n",
923 : p, p - addr, get_freepointer(s, p));
924 :
925 0 : if (s->flags & SLAB_RED_ZONE)
926 0 : print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
927 : s->red_left_pad);
928 0 : else if (p > addr + 16)
929 0 : print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
930 :
931 0 : print_section(KERN_ERR, "Object ", p,
932 0 : min_t(unsigned int, s->object_size, PAGE_SIZE));
933 0 : if (s->flags & SLAB_RED_ZONE)
934 0 : print_section(KERN_ERR, "Redzone ", p + s->object_size,
935 0 : s->inuse - s->object_size);
936 :
937 0 : off = get_info_end(s);
938 :
939 0 : if (s->flags & SLAB_STORE_USER)
940 0 : off += 2 * sizeof(struct track);
941 :
942 0 : if (slub_debug_orig_size(s))
943 0 : off += sizeof(unsigned int);
944 :
945 0 : off += kasan_metadata_size(s, false);
946 :
947 0 : if (off != size_from_object(s))
948 : /* Beginning of the filler is the free pointer */
949 0 : print_section(KERN_ERR, "Padding ", p + off,
950 0 : size_from_object(s) - off);
951 :
952 0 : dump_stack();
953 0 : }
954 :
955 0 : static void object_err(struct kmem_cache *s, struct slab *slab,
956 : u8 *object, char *reason)
957 : {
958 0 : if (slab_add_kunit_errors())
959 : return;
960 :
961 0 : slab_bug(s, "%s", reason);
962 0 : print_trailer(s, slab, object);
963 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
964 : }
965 :
966 82 : static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab,
967 : void **freelist, void *nextfree)
968 : {
969 82 : if ((s->flags & SLAB_CONSISTENCY_CHECKS) &&
970 0 : !check_valid_pointer(s, slab, nextfree) && freelist) {
971 0 : object_err(s, slab, *freelist, "Freechain corrupt");
972 0 : *freelist = NULL;
973 0 : slab_fix(s, "Isolate corrupted freechain");
974 0 : return true;
975 : }
976 :
977 : return false;
978 : }
979 :
980 0 : static __printf(3, 4) void slab_err(struct kmem_cache *s, struct slab *slab,
981 : const char *fmt, ...)
982 : {
983 : va_list args;
984 : char buf[100];
985 :
986 0 : if (slab_add_kunit_errors())
987 0 : return;
988 :
989 0 : va_start(args, fmt);
990 0 : vsnprintf(buf, sizeof(buf), fmt, args);
991 0 : va_end(args);
992 0 : slab_bug(s, "%s", buf);
993 0 : print_slab_info(slab);
994 0 : dump_stack();
995 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
996 : }
997 :
998 1 : static void init_object(struct kmem_cache *s, void *object, u8 val)
999 : {
1000 1 : u8 *p = kasan_reset_tag(object);
1001 1 : unsigned int poison_size = s->object_size;
1002 :
1003 1 : if (s->flags & SLAB_RED_ZONE) {
1004 0 : memset(p - s->red_left_pad, val, s->red_left_pad);
1005 :
1006 0 : if (slub_debug_orig_size(s) && val == SLUB_RED_ACTIVE) {
1007 : /*
1008 : * Redzone the extra allocated space by kmalloc than
1009 : * requested, and the poison size will be limited to
1010 : * the original request size accordingly.
1011 : */
1012 0 : poison_size = get_orig_size(s, object);
1013 : }
1014 : }
1015 :
1016 1 : if (s->flags & __OBJECT_POISON) {
1017 0 : memset(p, POISON_FREE, poison_size - 1);
1018 0 : p[poison_size - 1] = POISON_END;
1019 : }
1020 :
1021 1 : if (s->flags & SLAB_RED_ZONE)
1022 0 : memset(p + poison_size, val, s->inuse - poison_size);
1023 1 : }
1024 :
1025 0 : static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
1026 : void *from, void *to)
1027 : {
1028 0 : slab_fix(s, "Restoring %s 0x%p-0x%p=0x%x", message, from, to - 1, data);
1029 0 : memset(from, data, to - from);
1030 0 : }
1031 :
1032 0 : static int check_bytes_and_report(struct kmem_cache *s, struct slab *slab,
1033 : u8 *object, char *what,
1034 : u8 *start, unsigned int value, unsigned int bytes)
1035 : {
1036 : u8 *fault;
1037 : u8 *end;
1038 0 : u8 *addr = slab_address(slab);
1039 :
1040 : metadata_access_enable();
1041 0 : fault = memchr_inv(kasan_reset_tag(start), value, bytes);
1042 : metadata_access_disable();
1043 0 : if (!fault)
1044 : return 1;
1045 :
1046 0 : end = start + bytes;
1047 0 : while (end > fault && end[-1] == value)
1048 0 : end--;
1049 :
1050 0 : if (slab_add_kunit_errors())
1051 : goto skip_bug_print;
1052 :
1053 0 : slab_bug(s, "%s overwritten", what);
1054 0 : pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n",
1055 : fault, end - 1, fault - addr,
1056 : fault[0], value);
1057 0 : print_trailer(s, slab, object);
1058 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
1059 :
1060 : skip_bug_print:
1061 0 : restore_bytes(s, what, value, fault, end);
1062 0 : return 0;
1063 : }
1064 :
1065 : /*
1066 : * Object layout:
1067 : *
1068 : * object address
1069 : * Bytes of the object to be managed.
1070 : * If the freepointer may overlay the object then the free
1071 : * pointer is at the middle of the object.
1072 : *
1073 : * Poisoning uses 0x6b (POISON_FREE) and the last byte is
1074 : * 0xa5 (POISON_END)
1075 : *
1076 : * object + s->object_size
1077 : * Padding to reach word boundary. This is also used for Redzoning.
1078 : * Padding is extended by another word if Redzoning is enabled and
1079 : * object_size == inuse.
1080 : *
1081 : * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
1082 : * 0xcc (RED_ACTIVE) for objects in use.
1083 : *
1084 : * object + s->inuse
1085 : * Meta data starts here.
1086 : *
1087 : * A. Free pointer (if we cannot overwrite object on free)
1088 : * B. Tracking data for SLAB_STORE_USER
1089 : * C. Original request size for kmalloc object (SLAB_STORE_USER enabled)
1090 : * D. Padding to reach required alignment boundary or at minimum
1091 : * one word if debugging is on to be able to detect writes
1092 : * before the word boundary.
1093 : *
1094 : * Padding is done using 0x5a (POISON_INUSE)
1095 : *
1096 : * object + s->size
1097 : * Nothing is used beyond s->size.
1098 : *
1099 : * If slabcaches are merged then the object_size and inuse boundaries are mostly
1100 : * ignored. And therefore no slab options that rely on these boundaries
1101 : * may be used with merged slabcaches.
1102 : */
1103 :
1104 0 : static int check_pad_bytes(struct kmem_cache *s, struct slab *slab, u8 *p)
1105 : {
1106 0 : unsigned long off = get_info_end(s); /* The end of info */
1107 :
1108 0 : if (s->flags & SLAB_STORE_USER) {
1109 : /* We also have user information there */
1110 0 : off += 2 * sizeof(struct track);
1111 :
1112 0 : if (s->flags & SLAB_KMALLOC)
1113 0 : off += sizeof(unsigned int);
1114 : }
1115 :
1116 0 : off += kasan_metadata_size(s, false);
1117 :
1118 0 : if (size_from_object(s) == off)
1119 : return 1;
1120 :
1121 0 : return check_bytes_and_report(s, slab, p, "Object padding",
1122 0 : p + off, POISON_INUSE, size_from_object(s) - off);
1123 : }
1124 :
1125 : /* Check the pad bytes at the end of a slab page */
1126 0 : static void slab_pad_check(struct kmem_cache *s, struct slab *slab)
1127 : {
1128 : u8 *start;
1129 : u8 *fault;
1130 : u8 *end;
1131 : u8 *pad;
1132 : int length;
1133 : int remainder;
1134 :
1135 0 : if (!(s->flags & SLAB_POISON))
1136 : return;
1137 :
1138 0 : start = slab_address(slab);
1139 0 : length = slab_size(slab);
1140 0 : end = start + length;
1141 0 : remainder = length % s->size;
1142 0 : if (!remainder)
1143 : return;
1144 :
1145 0 : pad = end - remainder;
1146 : metadata_access_enable();
1147 0 : fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder);
1148 : metadata_access_disable();
1149 0 : if (!fault)
1150 : return;
1151 0 : while (end > fault && end[-1] == POISON_INUSE)
1152 0 : end--;
1153 :
1154 0 : slab_err(s, slab, "Padding overwritten. 0x%p-0x%p @offset=%tu",
1155 : fault, end - 1, fault - start);
1156 0 : print_section(KERN_ERR, "Padding ", pad, remainder);
1157 :
1158 0 : restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
1159 : }
1160 :
1161 0 : static int check_object(struct kmem_cache *s, struct slab *slab,
1162 : void *object, u8 val)
1163 : {
1164 0 : u8 *p = object;
1165 0 : u8 *endobject = object + s->object_size;
1166 : unsigned int orig_size;
1167 :
1168 0 : if (s->flags & SLAB_RED_ZONE) {
1169 0 : if (!check_bytes_and_report(s, slab, object, "Left Redzone",
1170 0 : object - s->red_left_pad, val, s->red_left_pad))
1171 : return 0;
1172 :
1173 0 : if (!check_bytes_and_report(s, slab, object, "Right Redzone",
1174 0 : endobject, val, s->inuse - s->object_size))
1175 : return 0;
1176 :
1177 0 : if (slub_debug_orig_size(s) && val == SLUB_RED_ACTIVE) {
1178 0 : orig_size = get_orig_size(s, object);
1179 :
1180 0 : if (s->object_size > orig_size &&
1181 0 : !check_bytes_and_report(s, slab, object,
1182 : "kmalloc Redzone", p + orig_size,
1183 : val, s->object_size - orig_size)) {
1184 : return 0;
1185 : }
1186 : }
1187 : } else {
1188 0 : if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
1189 0 : check_bytes_and_report(s, slab, p, "Alignment padding",
1190 : endobject, POISON_INUSE,
1191 : s->inuse - s->object_size);
1192 : }
1193 : }
1194 :
1195 0 : if (s->flags & SLAB_POISON) {
1196 0 : if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
1197 0 : (!check_bytes_and_report(s, slab, p, "Poison", p,
1198 0 : POISON_FREE, s->object_size - 1) ||
1199 0 : !check_bytes_and_report(s, slab, p, "End Poison",
1200 0 : p + s->object_size - 1, POISON_END, 1)))
1201 : return 0;
1202 : /*
1203 : * check_pad_bytes cleans up on its own.
1204 : */
1205 0 : check_pad_bytes(s, slab, p);
1206 : }
1207 :
1208 0 : if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE)
1209 : /*
1210 : * Object and freepointer overlap. Cannot check
1211 : * freepointer while object is allocated.
1212 : */
1213 : return 1;
1214 :
1215 : /* Check free pointer validity */
1216 0 : if (!check_valid_pointer(s, slab, get_freepointer(s, p))) {
1217 0 : object_err(s, slab, p, "Freepointer corrupt");
1218 : /*
1219 : * No choice but to zap it and thus lose the remainder
1220 : * of the free objects in this slab. May cause
1221 : * another error because the object count is now wrong.
1222 : */
1223 0 : set_freepointer(s, p, NULL);
1224 0 : return 0;
1225 : }
1226 : return 1;
1227 : }
1228 :
1229 0 : static int check_slab(struct kmem_cache *s, struct slab *slab)
1230 : {
1231 : int maxobj;
1232 :
1233 0 : if (!folio_test_slab(slab_folio(slab))) {
1234 0 : slab_err(s, slab, "Not a valid slab page");
1235 0 : return 0;
1236 : }
1237 :
1238 0 : maxobj = order_objects(slab_order(slab), s->size);
1239 0 : if (slab->objects > maxobj) {
1240 0 : slab_err(s, slab, "objects %u > max %u",
1241 : slab->objects, maxobj);
1242 0 : return 0;
1243 : }
1244 0 : if (slab->inuse > slab->objects) {
1245 0 : slab_err(s, slab, "inuse %u > max %u",
1246 : slab->inuse, slab->objects);
1247 0 : return 0;
1248 : }
1249 : /* Slab_pad_check fixes things up after itself */
1250 0 : slab_pad_check(s, slab);
1251 0 : return 1;
1252 : }
1253 :
1254 : /*
1255 : * Determine if a certain object in a slab is on the freelist. Must hold the
1256 : * slab lock to guarantee that the chains are in a consistent state.
1257 : */
1258 0 : static int on_freelist(struct kmem_cache *s, struct slab *slab, void *search)
1259 : {
1260 0 : int nr = 0;
1261 : void *fp;
1262 0 : void *object = NULL;
1263 : int max_objects;
1264 :
1265 0 : fp = slab->freelist;
1266 0 : while (fp && nr <= slab->objects) {
1267 0 : if (fp == search)
1268 : return 1;
1269 0 : if (!check_valid_pointer(s, slab, fp)) {
1270 0 : if (object) {
1271 0 : object_err(s, slab, object,
1272 : "Freechain corrupt");
1273 0 : set_freepointer(s, object, NULL);
1274 : } else {
1275 0 : slab_err(s, slab, "Freepointer corrupt");
1276 0 : slab->freelist = NULL;
1277 0 : slab->inuse = slab->objects;
1278 0 : slab_fix(s, "Freelist cleared");
1279 0 : return 0;
1280 : }
1281 : break;
1282 : }
1283 0 : object = fp;
1284 0 : fp = get_freepointer(s, object);
1285 0 : nr++;
1286 : }
1287 :
1288 0 : max_objects = order_objects(slab_order(slab), s->size);
1289 0 : if (max_objects > MAX_OBJS_PER_PAGE)
1290 0 : max_objects = MAX_OBJS_PER_PAGE;
1291 :
1292 0 : if (slab->objects != max_objects) {
1293 0 : slab_err(s, slab, "Wrong number of objects. Found %d but should be %d",
1294 : slab->objects, max_objects);
1295 0 : slab->objects = max_objects;
1296 0 : slab_fix(s, "Number of objects adjusted");
1297 : }
1298 0 : if (slab->inuse != slab->objects - nr) {
1299 0 : slab_err(s, slab, "Wrong object count. Counter is %d but counted were %d",
1300 : slab->inuse, slab->objects - nr);
1301 0 : slab->inuse = slab->objects - nr;
1302 0 : slab_fix(s, "Object count adjusted");
1303 : }
1304 0 : return search == NULL;
1305 : }
1306 :
1307 0 : static void trace(struct kmem_cache *s, struct slab *slab, void *object,
1308 : int alloc)
1309 : {
1310 0 : if (s->flags & SLAB_TRACE) {
1311 0 : pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
1312 : s->name,
1313 : alloc ? "alloc" : "free",
1314 : object, slab->inuse,
1315 : slab->freelist);
1316 :
1317 0 : if (!alloc)
1318 0 : print_section(KERN_INFO, "Object ", (void *)object,
1319 : s->object_size);
1320 :
1321 0 : dump_stack();
1322 : }
1323 0 : }
1324 :
1325 : /*
1326 : * Tracking of fully allocated slabs for debugging purposes.
1327 : */
1328 : static void add_full(struct kmem_cache *s,
1329 : struct kmem_cache_node *n, struct slab *slab)
1330 : {
1331 0 : if (!(s->flags & SLAB_STORE_USER))
1332 : return;
1333 :
1334 : lockdep_assert_held(&n->list_lock);
1335 0 : list_add(&slab->slab_list, &n->full);
1336 : }
1337 :
1338 : static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct slab *slab)
1339 : {
1340 2163 : if (!(s->flags & SLAB_STORE_USER))
1341 : return;
1342 :
1343 : lockdep_assert_held(&n->list_lock);
1344 0 : list_del(&slab->slab_list);
1345 : }
1346 :
1347 : /* Tracking of the number of slabs for debugging purposes */
1348 : static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1349 : {
1350 0 : struct kmem_cache_node *n = get_node(s, node);
1351 :
1352 0 : return atomic_long_read(&n->nr_slabs);
1353 : }
1354 :
1355 : static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1356 : {
1357 0 : return atomic_long_read(&n->nr_slabs);
1358 : }
1359 :
1360 : static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
1361 : {
1362 2497 : struct kmem_cache_node *n = get_node(s, node);
1363 :
1364 : /*
1365 : * May be called early in order to allocate a slab for the
1366 : * kmem_cache_node structure. Solve the chicken-egg
1367 : * dilemma by deferring the increment of the count during
1368 : * bootstrap (see early_kmem_cache_node_alloc).
1369 : */
1370 2497 : if (likely(n)) {
1371 4992 : atomic_long_inc(&n->nr_slabs);
1372 2496 : atomic_long_add(objects, &n->total_objects);
1373 : }
1374 : }
1375 : static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
1376 : {
1377 2056 : struct kmem_cache_node *n = get_node(s, node);
1378 :
1379 4112 : atomic_long_dec(&n->nr_slabs);
1380 4112 : atomic_long_sub(objects, &n->total_objects);
1381 : }
1382 :
1383 : /* Object debug checks for alloc/free paths */
1384 68231 : static void setup_object_debug(struct kmem_cache *s, void *object)
1385 : {
1386 136462 : if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))
1387 : return;
1388 :
1389 0 : init_object(s, object, SLUB_RED_INACTIVE);
1390 0 : init_tracking(s, object);
1391 : }
1392 :
1393 : static
1394 2496 : void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr)
1395 : {
1396 4992 : if (!kmem_cache_debug_flags(s, SLAB_POISON))
1397 : return;
1398 :
1399 0 : metadata_access_enable();
1400 0 : memset(kasan_reset_tag(addr), POISON_INUSE, slab_size(slab));
1401 : metadata_access_disable();
1402 : }
1403 :
1404 0 : static inline int alloc_consistency_checks(struct kmem_cache *s,
1405 : struct slab *slab, void *object)
1406 : {
1407 0 : if (!check_slab(s, slab))
1408 : return 0;
1409 :
1410 0 : if (!check_valid_pointer(s, slab, object)) {
1411 0 : object_err(s, slab, object, "Freelist Pointer check fails");
1412 0 : return 0;
1413 : }
1414 :
1415 0 : if (!check_object(s, slab, object, SLUB_RED_INACTIVE))
1416 : return 0;
1417 :
1418 0 : return 1;
1419 : }
1420 :
1421 0 : static noinline bool alloc_debug_processing(struct kmem_cache *s,
1422 : struct slab *slab, void *object, int orig_size)
1423 : {
1424 0 : if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1425 0 : if (!alloc_consistency_checks(s, slab, object))
1426 : goto bad;
1427 : }
1428 :
1429 : /* Success. Perform special debug activities for allocs */
1430 0 : trace(s, slab, object, 1);
1431 0 : set_orig_size(s, object, orig_size);
1432 0 : init_object(s, object, SLUB_RED_ACTIVE);
1433 0 : return true;
1434 :
1435 : bad:
1436 0 : if (folio_test_slab(slab_folio(slab))) {
1437 : /*
1438 : * If this is a slab page then lets do the best we can
1439 : * to avoid issues in the future. Marking all objects
1440 : * as used avoids touching the remaining objects.
1441 : */
1442 0 : slab_fix(s, "Marking all objects used");
1443 0 : slab->inuse = slab->objects;
1444 0 : slab->freelist = NULL;
1445 : }
1446 : return false;
1447 : }
1448 :
1449 0 : static inline int free_consistency_checks(struct kmem_cache *s,
1450 : struct slab *slab, void *object, unsigned long addr)
1451 : {
1452 0 : if (!check_valid_pointer(s, slab, object)) {
1453 0 : slab_err(s, slab, "Invalid object pointer 0x%p", object);
1454 : return 0;
1455 : }
1456 :
1457 0 : if (on_freelist(s, slab, object)) {
1458 0 : object_err(s, slab, object, "Object already free");
1459 : return 0;
1460 : }
1461 :
1462 0 : if (!check_object(s, slab, object, SLUB_RED_ACTIVE))
1463 : return 0;
1464 :
1465 0 : if (unlikely(s != slab->slab_cache)) {
1466 0 : if (!folio_test_slab(slab_folio(slab))) {
1467 0 : slab_err(s, slab, "Attempt to free object(0x%p) outside of slab",
1468 : object);
1469 0 : } else if (!slab->slab_cache) {
1470 0 : pr_err("SLUB <none>: no slab for object 0x%p.\n",
1471 : object);
1472 0 : dump_stack();
1473 : } else
1474 0 : object_err(s, slab, object,
1475 : "page slab pointer corrupt.");
1476 : return 0;
1477 : }
1478 : return 1;
1479 : }
1480 :
1481 : /*
1482 : * Parse a block of slub_debug options. Blocks are delimited by ';'
1483 : *
1484 : * @str: start of block
1485 : * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified
1486 : * @slabs: return start of list of slabs, or NULL when there's no list
1487 : * @init: assume this is initial parsing and not per-kmem-create parsing
1488 : *
1489 : * returns the start of next block if there's any, or NULL
1490 : */
1491 : static char *
1492 0 : parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init)
1493 : {
1494 0 : bool higher_order_disable = false;
1495 :
1496 : /* Skip any completely empty blocks */
1497 0 : while (*str && *str == ';')
1498 0 : str++;
1499 :
1500 0 : if (*str == ',') {
1501 : /*
1502 : * No options but restriction on slabs. This means full
1503 : * debugging for slabs matching a pattern.
1504 : */
1505 0 : *flags = DEBUG_DEFAULT_FLAGS;
1506 0 : goto check_slabs;
1507 : }
1508 0 : *flags = 0;
1509 :
1510 : /* Determine which debug features should be switched on */
1511 0 : for (; *str && *str != ',' && *str != ';'; str++) {
1512 0 : switch (tolower(*str)) {
1513 : case '-':
1514 0 : *flags = 0;
1515 0 : break;
1516 : case 'f':
1517 0 : *flags |= SLAB_CONSISTENCY_CHECKS;
1518 0 : break;
1519 : case 'z':
1520 0 : *flags |= SLAB_RED_ZONE;
1521 0 : break;
1522 : case 'p':
1523 0 : *flags |= SLAB_POISON;
1524 0 : break;
1525 : case 'u':
1526 0 : *flags |= SLAB_STORE_USER;
1527 0 : break;
1528 : case 't':
1529 0 : *flags |= SLAB_TRACE;
1530 0 : break;
1531 : case 'a':
1532 : *flags |= SLAB_FAILSLAB;
1533 0 : break;
1534 : case 'o':
1535 : /*
1536 : * Avoid enabling debugging on caches if its minimum
1537 : * order would increase as a result.
1538 : */
1539 : higher_order_disable = true;
1540 : break;
1541 : default:
1542 0 : if (init)
1543 0 : pr_err("slub_debug option '%c' unknown. skipped\n", *str);
1544 : }
1545 : }
1546 : check_slabs:
1547 0 : if (*str == ',')
1548 0 : *slabs = ++str;
1549 : else
1550 0 : *slabs = NULL;
1551 :
1552 : /* Skip over the slab list */
1553 0 : while (*str && *str != ';')
1554 0 : str++;
1555 :
1556 : /* Skip any completely empty blocks */
1557 0 : while (*str && *str == ';')
1558 0 : str++;
1559 :
1560 0 : if (init && higher_order_disable)
1561 0 : disable_higher_order_debug = 1;
1562 :
1563 0 : if (*str)
1564 : return str;
1565 : else
1566 0 : return NULL;
1567 : }
1568 :
1569 0 : static int __init setup_slub_debug(char *str)
1570 : {
1571 : slab_flags_t flags;
1572 : slab_flags_t global_flags;
1573 : char *saved_str;
1574 : char *slab_list;
1575 0 : bool global_slub_debug_changed = false;
1576 0 : bool slab_list_specified = false;
1577 :
1578 0 : global_flags = DEBUG_DEFAULT_FLAGS;
1579 0 : if (*str++ != '=' || !*str)
1580 : /*
1581 : * No options specified. Switch on full debugging.
1582 : */
1583 : goto out;
1584 :
1585 : saved_str = str;
1586 0 : while (str) {
1587 0 : str = parse_slub_debug_flags(str, &flags, &slab_list, true);
1588 :
1589 0 : if (!slab_list) {
1590 0 : global_flags = flags;
1591 0 : global_slub_debug_changed = true;
1592 : } else {
1593 0 : slab_list_specified = true;
1594 0 : if (flags & SLAB_STORE_USER)
1595 0 : stack_depot_request_early_init();
1596 : }
1597 : }
1598 :
1599 : /*
1600 : * For backwards compatibility, a single list of flags with list of
1601 : * slabs means debugging is only changed for those slabs, so the global
1602 : * slub_debug should be unchanged (0 or DEBUG_DEFAULT_FLAGS, depending
1603 : * on CONFIG_SLUB_DEBUG_ON). We can extended that to multiple lists as
1604 : * long as there is no option specifying flags without a slab list.
1605 : */
1606 0 : if (slab_list_specified) {
1607 0 : if (!global_slub_debug_changed)
1608 0 : global_flags = slub_debug;
1609 0 : slub_debug_string = saved_str;
1610 : }
1611 : out:
1612 0 : slub_debug = global_flags;
1613 0 : if (slub_debug & SLAB_STORE_USER)
1614 0 : stack_depot_request_early_init();
1615 0 : if (slub_debug != 0 || slub_debug_string)
1616 0 : static_branch_enable(&slub_debug_enabled);
1617 : else
1618 0 : static_branch_disable(&slub_debug_enabled);
1619 0 : if ((static_branch_unlikely(&init_on_alloc) ||
1620 0 : static_branch_unlikely(&init_on_free)) &&
1621 0 : (slub_debug & SLAB_POISON))
1622 0 : pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n");
1623 0 : return 1;
1624 : }
1625 :
1626 : __setup("slub_debug", setup_slub_debug);
1627 :
1628 : /*
1629 : * kmem_cache_flags - apply debugging options to the cache
1630 : * @object_size: the size of an object without meta data
1631 : * @flags: flags to set
1632 : * @name: name of the cache
1633 : *
1634 : * Debug option(s) are applied to @flags. In addition to the debug
1635 : * option(s), if a slab name (or multiple) is specified i.e.
1636 : * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ...
1637 : * then only the select slabs will receive the debug option(s).
1638 : */
1639 102 : slab_flags_t kmem_cache_flags(unsigned int object_size,
1640 : slab_flags_t flags, const char *name)
1641 : {
1642 : char *iter;
1643 : size_t len;
1644 : char *next_block;
1645 : slab_flags_t block_flags;
1646 102 : slab_flags_t slub_debug_local = slub_debug;
1647 :
1648 102 : if (flags & SLAB_NO_USER_FLAGS)
1649 : return flags;
1650 :
1651 : /*
1652 : * If the slab cache is for debugging (e.g. kmemleak) then
1653 : * don't store user (stack trace) information by default,
1654 : * but let the user enable it via the command line below.
1655 : */
1656 102 : if (flags & SLAB_NOLEAKTRACE)
1657 0 : slub_debug_local &= ~SLAB_STORE_USER;
1658 :
1659 102 : len = strlen(name);
1660 102 : next_block = slub_debug_string;
1661 : /* Go through all blocks of debug options, see if any matches our slab's name */
1662 204 : while (next_block) {
1663 0 : next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false);
1664 0 : if (!iter)
1665 0 : continue;
1666 : /* Found a block that has a slab list, search it */
1667 0 : while (*iter) {
1668 : char *end, *glob;
1669 : size_t cmplen;
1670 :
1671 0 : end = strchrnul(iter, ',');
1672 0 : if (next_block && next_block < end)
1673 0 : end = next_block - 1;
1674 :
1675 0 : glob = strnchr(iter, end - iter, '*');
1676 0 : if (glob)
1677 0 : cmplen = glob - iter;
1678 : else
1679 0 : cmplen = max_t(size_t, len, (end - iter));
1680 :
1681 0 : if (!strncmp(name, iter, cmplen)) {
1682 0 : flags |= block_flags;
1683 0 : return flags;
1684 : }
1685 :
1686 0 : if (!*end || *end == ';')
1687 : break;
1688 0 : iter = end + 1;
1689 : }
1690 : }
1691 :
1692 102 : return flags | slub_debug_local;
1693 : }
1694 : #else /* !CONFIG_SLUB_DEBUG */
1695 : static inline void setup_object_debug(struct kmem_cache *s, void *object) {}
1696 : static inline
1697 : void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr) {}
1698 :
1699 : static inline bool alloc_debug_processing(struct kmem_cache *s,
1700 : struct slab *slab, void *object, int orig_size) { return true; }
1701 :
1702 : static inline bool free_debug_processing(struct kmem_cache *s,
1703 : struct slab *slab, void *head, void *tail, int *bulk_cnt,
1704 : unsigned long addr, depot_stack_handle_t handle) { return true; }
1705 :
1706 : static inline void slab_pad_check(struct kmem_cache *s, struct slab *slab) {}
1707 : static inline int check_object(struct kmem_cache *s, struct slab *slab,
1708 : void *object, u8 val) { return 1; }
1709 : static inline depot_stack_handle_t set_track_prepare(void) { return 0; }
1710 : static inline void set_track(struct kmem_cache *s, void *object,
1711 : enum track_item alloc, unsigned long addr) {}
1712 : static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1713 : struct slab *slab) {}
1714 : static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1715 : struct slab *slab) {}
1716 : slab_flags_t kmem_cache_flags(unsigned int object_size,
1717 : slab_flags_t flags, const char *name)
1718 : {
1719 : return flags;
1720 : }
1721 : #define slub_debug 0
1722 :
1723 : #define disable_higher_order_debug 0
1724 :
1725 : static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1726 : { return 0; }
1727 : static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1728 : { return 0; }
1729 : static inline void inc_slabs_node(struct kmem_cache *s, int node,
1730 : int objects) {}
1731 : static inline void dec_slabs_node(struct kmem_cache *s, int node,
1732 : int objects) {}
1733 :
1734 : #ifndef CONFIG_SLUB_TINY
1735 : static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab,
1736 : void **freelist, void *nextfree)
1737 : {
1738 : return false;
1739 : }
1740 : #endif
1741 : #endif /* CONFIG_SLUB_DEBUG */
1742 :
1743 : /*
1744 : * Hooks for other subsystems that check memory allocations. In a typical
1745 : * production configuration these hooks all should produce no code at all.
1746 : */
1747 : static __always_inline bool slab_free_hook(struct kmem_cache *s,
1748 : void *x, bool init)
1749 : {
1750 65545 : kmemleak_free_recursive(x, s->flags);
1751 65545 : kmsan_slab_free(s, x);
1752 :
1753 65545 : debug_check_no_locks_freed(x, s->object_size);
1754 :
1755 : if (!(s->flags & SLAB_DEBUG_OBJECTS))
1756 65545 : debug_check_no_obj_freed(x, s->object_size);
1757 :
1758 : /* Use KCSAN to help debug racy use-after-free. */
1759 : if (!(s->flags & SLAB_TYPESAFE_BY_RCU))
1760 : __kcsan_check_access(x, s->object_size,
1761 : KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT);
1762 :
1763 : /*
1764 : * As memory initialization might be integrated into KASAN,
1765 : * kasan_slab_free and initialization memset's must be
1766 : * kept together to avoid discrepancies in behavior.
1767 : *
1768 : * The initialization memset's clear the object and the metadata,
1769 : * but don't touch the SLAB redzone.
1770 : */
1771 65545 : if (init) {
1772 : int rsize;
1773 :
1774 : if (!kasan_has_integrated_init())
1775 0 : memset(kasan_reset_tag(x), 0, s->object_size);
1776 0 : rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0;
1777 0 : memset((char *)kasan_reset_tag(x) + s->inuse, 0,
1778 0 : s->size - s->inuse - rsize);
1779 : }
1780 : /* KASAN might put x into memory quarantine, delaying its reuse. */
1781 65545 : return kasan_slab_free(s, x, init);
1782 : }
1783 :
1784 65545 : static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1785 : void **head, void **tail,
1786 : int *cnt)
1787 : {
1788 :
1789 : void *object;
1790 65545 : void *next = *head;
1791 65545 : void *old_tail = *tail ? *tail : *head;
1792 :
1793 65545 : if (is_kfence_address(next)) {
1794 : slab_free_hook(s, next, false);
1795 : return true;
1796 : }
1797 :
1798 : /* Head and tail of the reconstructed freelist */
1799 65545 : *head = NULL;
1800 65545 : *tail = NULL;
1801 :
1802 : do {
1803 65545 : object = next;
1804 131090 : next = get_freepointer(s, object);
1805 :
1806 : /* If object's reuse doesn't have to be delayed */
1807 196635 : if (!slab_free_hook(s, object, slab_want_init_on_free(s))) {
1808 : /* Move object to the new freelist */
1809 131090 : set_freepointer(s, object, *head);
1810 65545 : *head = object;
1811 65545 : if (!*tail)
1812 65545 : *tail = object;
1813 : } else {
1814 : /*
1815 : * Adjust the reconstructed freelist depth
1816 : * accordingly if object's reuse is delayed.
1817 : */
1818 : --(*cnt);
1819 : }
1820 65545 : } while (object != old_tail);
1821 :
1822 65545 : if (*head == *tail)
1823 65545 : *tail = NULL;
1824 :
1825 65545 : return *head != NULL;
1826 : }
1827 :
1828 : static void *setup_object(struct kmem_cache *s, void *object)
1829 : {
1830 68231 : setup_object_debug(s, object);
1831 68231 : object = kasan_init_slab_obj(s, object);
1832 68231 : if (unlikely(s->ctor)) {
1833 260 : kasan_unpoison_object_data(s, object);
1834 260 : s->ctor(object);
1835 260 : kasan_poison_object_data(s, object);
1836 : }
1837 : return object;
1838 : }
1839 :
1840 : /*
1841 : * Slab allocation and freeing
1842 : */
1843 2496 : static inline struct slab *alloc_slab_page(gfp_t flags, int node,
1844 : struct kmem_cache_order_objects oo)
1845 : {
1846 : struct folio *folio;
1847 : struct slab *slab;
1848 2496 : unsigned int order = oo_order(oo);
1849 :
1850 2496 : if (node == NUMA_NO_NODE)
1851 2495 : folio = (struct folio *)alloc_pages(flags, order);
1852 : else
1853 1 : folio = (struct folio *)__alloc_pages_node(node, flags, order);
1854 :
1855 2496 : if (!folio)
1856 : return NULL;
1857 :
1858 2496 : slab = folio_slab(folio);
1859 2496 : __folio_set_slab(folio);
1860 : /* Make the flag visible before any changes to folio->mapping */
1861 2496 : smp_wmb();
1862 4992 : if (folio_is_pfmemalloc(folio))
1863 : slab_set_pfmemalloc(slab);
1864 :
1865 : return slab;
1866 : }
1867 :
1868 : #ifdef CONFIG_SLAB_FREELIST_RANDOM
1869 : /* Pre-initialize the random sequence cache */
1870 : static int init_cache_random_seq(struct kmem_cache *s)
1871 : {
1872 : unsigned int count = oo_objects(s->oo);
1873 : int err;
1874 :
1875 : /* Bailout if already initialised */
1876 : if (s->random_seq)
1877 : return 0;
1878 :
1879 : err = cache_random_seq_create(s, count, GFP_KERNEL);
1880 : if (err) {
1881 : pr_err("SLUB: Unable to initialize free list for %s\n",
1882 : s->name);
1883 : return err;
1884 : }
1885 :
1886 : /* Transform to an offset on the set of pages */
1887 : if (s->random_seq) {
1888 : unsigned int i;
1889 :
1890 : for (i = 0; i < count; i++)
1891 : s->random_seq[i] *= s->size;
1892 : }
1893 : return 0;
1894 : }
1895 :
1896 : /* Initialize each random sequence freelist per cache */
1897 : static void __init init_freelist_randomization(void)
1898 : {
1899 : struct kmem_cache *s;
1900 :
1901 : mutex_lock(&slab_mutex);
1902 :
1903 : list_for_each_entry(s, &slab_caches, list)
1904 : init_cache_random_seq(s);
1905 :
1906 : mutex_unlock(&slab_mutex);
1907 : }
1908 :
1909 : /* Get the next entry on the pre-computed freelist randomized */
1910 : static void *next_freelist_entry(struct kmem_cache *s, struct slab *slab,
1911 : unsigned long *pos, void *start,
1912 : unsigned long page_limit,
1913 : unsigned long freelist_count)
1914 : {
1915 : unsigned int idx;
1916 :
1917 : /*
1918 : * If the target page allocation failed, the number of objects on the
1919 : * page might be smaller than the usual size defined by the cache.
1920 : */
1921 : do {
1922 : idx = s->random_seq[*pos];
1923 : *pos += 1;
1924 : if (*pos >= freelist_count)
1925 : *pos = 0;
1926 : } while (unlikely(idx >= page_limit));
1927 :
1928 : return (char *)start + idx;
1929 : }
1930 :
1931 : /* Shuffle the single linked freelist based on a random pre-computed sequence */
1932 : static bool shuffle_freelist(struct kmem_cache *s, struct slab *slab)
1933 : {
1934 : void *start;
1935 : void *cur;
1936 : void *next;
1937 : unsigned long idx, pos, page_limit, freelist_count;
1938 :
1939 : if (slab->objects < 2 || !s->random_seq)
1940 : return false;
1941 :
1942 : freelist_count = oo_objects(s->oo);
1943 : pos = get_random_u32_below(freelist_count);
1944 :
1945 : page_limit = slab->objects * s->size;
1946 : start = fixup_red_left(s, slab_address(slab));
1947 :
1948 : /* First entry is used as the base of the freelist */
1949 : cur = next_freelist_entry(s, slab, &pos, start, page_limit,
1950 : freelist_count);
1951 : cur = setup_object(s, cur);
1952 : slab->freelist = cur;
1953 :
1954 : for (idx = 1; idx < slab->objects; idx++) {
1955 : next = next_freelist_entry(s, slab, &pos, start, page_limit,
1956 : freelist_count);
1957 : next = setup_object(s, next);
1958 : set_freepointer(s, cur, next);
1959 : cur = next;
1960 : }
1961 : set_freepointer(s, cur, NULL);
1962 :
1963 : return true;
1964 : }
1965 : #else
1966 : static inline int init_cache_random_seq(struct kmem_cache *s)
1967 : {
1968 : return 0;
1969 : }
1970 : static inline void init_freelist_randomization(void) { }
1971 : static inline bool shuffle_freelist(struct kmem_cache *s, struct slab *slab)
1972 : {
1973 : return false;
1974 : }
1975 : #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1976 :
1977 2496 : static struct slab *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1978 : {
1979 : struct slab *slab;
1980 2496 : struct kmem_cache_order_objects oo = s->oo;
1981 : gfp_t alloc_gfp;
1982 : void *start, *p, *next;
1983 : int idx;
1984 : bool shuffle;
1985 :
1986 2496 : flags &= gfp_allowed_mask;
1987 :
1988 2496 : flags |= s->allocflags;
1989 :
1990 : /*
1991 : * Let the initial higher-order allocation fail under memory pressure
1992 : * so we fall-back to the minimum order allocation.
1993 : */
1994 2496 : alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1995 4954 : if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
1996 75 : alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_RECLAIM;
1997 :
1998 2496 : slab = alloc_slab_page(alloc_gfp, node, oo);
1999 2496 : if (unlikely(!slab)) {
2000 0 : oo = s->min;
2001 0 : alloc_gfp = flags;
2002 : /*
2003 : * Allocation may have failed due to fragmentation.
2004 : * Try a lower order alloc if possible
2005 : */
2006 0 : slab = alloc_slab_page(alloc_gfp, node, oo);
2007 0 : if (unlikely(!slab))
2008 : return NULL;
2009 : stat(s, ORDER_FALLBACK);
2010 : }
2011 :
2012 2496 : slab->objects = oo_objects(oo);
2013 2496 : slab->inuse = 0;
2014 2496 : slab->frozen = 0;
2015 :
2016 4992 : account_slab(slab, oo_order(oo), s, flags);
2017 :
2018 2496 : slab->slab_cache = s;
2019 :
2020 2496 : kasan_poison_slab(slab);
2021 :
2022 2496 : start = slab_address(slab);
2023 :
2024 2496 : setup_slab_debug(s, slab, start);
2025 :
2026 2496 : shuffle = shuffle_freelist(s, slab);
2027 :
2028 : if (!shuffle) {
2029 2496 : start = fixup_red_left(s, start);
2030 2496 : start = setup_object(s, start);
2031 2496 : slab->freelist = start;
2032 68231 : for (idx = 0, p = start; idx < slab->objects - 1; idx++) {
2033 65735 : next = p + s->size;
2034 65735 : next = setup_object(s, next);
2035 131470 : set_freepointer(s, p, next);
2036 65735 : p = next;
2037 : }
2038 2496 : set_freepointer(s, p, NULL);
2039 : }
2040 :
2041 2496 : return slab;
2042 : }
2043 :
2044 2496 : static struct slab *new_slab(struct kmem_cache *s, gfp_t flags, int node)
2045 : {
2046 2496 : if (unlikely(flags & GFP_SLAB_BUG_MASK))
2047 0 : flags = kmalloc_fix_flags(flags);
2048 :
2049 2496 : WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2050 :
2051 2496 : return allocate_slab(s,
2052 : flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
2053 : }
2054 :
2055 2056 : static void __free_slab(struct kmem_cache *s, struct slab *slab)
2056 : {
2057 2056 : struct folio *folio = slab_folio(slab);
2058 2056 : int order = folio_order(folio);
2059 2056 : int pages = 1 << order;
2060 :
2061 2056 : __slab_clear_pfmemalloc(slab);
2062 2056 : folio->mapping = NULL;
2063 : /* Make the mapping reset visible before clearing the flag */
2064 2056 : smp_wmb();
2065 2056 : __folio_clear_slab(folio);
2066 2056 : if (current->reclaim_state)
2067 0 : current->reclaim_state->reclaimed_slab += pages;
2068 2056 : unaccount_slab(slab, order, s);
2069 2056 : __free_pages(&folio->page, order);
2070 2056 : }
2071 :
2072 0 : static void rcu_free_slab(struct rcu_head *h)
2073 : {
2074 0 : struct slab *slab = container_of(h, struct slab, rcu_head);
2075 :
2076 0 : __free_slab(slab->slab_cache, slab);
2077 0 : }
2078 :
2079 2056 : static void free_slab(struct kmem_cache *s, struct slab *slab)
2080 : {
2081 4112 : if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) {
2082 : void *p;
2083 :
2084 0 : slab_pad_check(s, slab);
2085 0 : for_each_object(p, s, slab_address(slab), slab->objects)
2086 0 : check_object(s, slab, p, SLUB_RED_INACTIVE);
2087 : }
2088 :
2089 2056 : if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU))
2090 0 : call_rcu(&slab->rcu_head, rcu_free_slab);
2091 : else
2092 2056 : __free_slab(s, slab);
2093 2056 : }
2094 :
2095 : static void discard_slab(struct kmem_cache *s, struct slab *slab)
2096 : {
2097 6168 : dec_slabs_node(s, slab_nid(slab), slab->objects);
2098 2056 : free_slab(s, slab);
2099 : }
2100 :
2101 : /*
2102 : * Management of partially allocated slabs.
2103 : */
2104 : static inline void
2105 : __add_partial(struct kmem_cache_node *n, struct slab *slab, int tail)
2106 : {
2107 2166 : n->nr_partial++;
2108 2 : if (tail == DEACTIVATE_TO_TAIL)
2109 2163 : list_add_tail(&slab->slab_list, &n->partial);
2110 : else
2111 3 : list_add(&slab->slab_list, &n->partial);
2112 : }
2113 :
2114 : static inline void add_partial(struct kmem_cache_node *n,
2115 : struct slab *slab, int tail)
2116 : {
2117 : lockdep_assert_held(&n->list_lock);
2118 2 : __add_partial(n, slab, tail);
2119 : }
2120 :
2121 : static inline void remove_partial(struct kmem_cache_node *n,
2122 : struct slab *slab)
2123 : {
2124 : lockdep_assert_held(&n->list_lock);
2125 4300 : list_del(&slab->slab_list);
2126 2150 : n->nr_partial--;
2127 : }
2128 :
2129 : /*
2130 : * Called only for kmem_cache_debug() caches instead of acquire_slab(), with a
2131 : * slab from the n->partial list. Remove only a single object from the slab, do
2132 : * the alloc_debug_processing() checks and leave the slab on the list, or move
2133 : * it to full list if it was the last free object.
2134 : */
2135 0 : static void *alloc_single_from_partial(struct kmem_cache *s,
2136 : struct kmem_cache_node *n, struct slab *slab, int orig_size)
2137 : {
2138 : void *object;
2139 :
2140 : lockdep_assert_held(&n->list_lock);
2141 :
2142 0 : object = slab->freelist;
2143 0 : slab->freelist = get_freepointer(s, object);
2144 0 : slab->inuse++;
2145 :
2146 0 : if (!alloc_debug_processing(s, slab, object, orig_size)) {
2147 0 : remove_partial(n, slab);
2148 0 : return NULL;
2149 : }
2150 :
2151 0 : if (slab->inuse == slab->objects) {
2152 0 : remove_partial(n, slab);
2153 0 : add_full(s, n, slab);
2154 : }
2155 :
2156 : return object;
2157 : }
2158 :
2159 : /*
2160 : * Called only for kmem_cache_debug() caches to allocate from a freshly
2161 : * allocated slab. Allocate a single object instead of whole freelist
2162 : * and put the slab to the partial (or full) list.
2163 : */
2164 0 : static void *alloc_single_from_new_slab(struct kmem_cache *s,
2165 : struct slab *slab, int orig_size)
2166 : {
2167 0 : int nid = slab_nid(slab);
2168 0 : struct kmem_cache_node *n = get_node(s, nid);
2169 : unsigned long flags;
2170 : void *object;
2171 :
2172 :
2173 0 : object = slab->freelist;
2174 0 : slab->freelist = get_freepointer(s, object);
2175 0 : slab->inuse = 1;
2176 :
2177 0 : if (!alloc_debug_processing(s, slab, object, orig_size))
2178 : /*
2179 : * It's not really expected that this would fail on a
2180 : * freshly allocated slab, but a concurrent memory
2181 : * corruption in theory could cause that.
2182 : */
2183 : return NULL;
2184 :
2185 0 : spin_lock_irqsave(&n->list_lock, flags);
2186 :
2187 0 : if (slab->inuse == slab->objects)
2188 0 : add_full(s, n, slab);
2189 : else
2190 : add_partial(n, slab, DEACTIVATE_TO_HEAD);
2191 :
2192 0 : inc_slabs_node(s, nid, slab->objects);
2193 0 : spin_unlock_irqrestore(&n->list_lock, flags);
2194 :
2195 0 : return object;
2196 : }
2197 :
2198 : /*
2199 : * Remove slab from the partial list, freeze it and
2200 : * return the pointer to the freelist.
2201 : *
2202 : * Returns a list of objects or NULL if it fails.
2203 : */
2204 94 : static inline void *acquire_slab(struct kmem_cache *s,
2205 : struct kmem_cache_node *n, struct slab *slab,
2206 : int mode)
2207 : {
2208 : void *freelist;
2209 : unsigned long counters;
2210 : struct slab new;
2211 :
2212 : lockdep_assert_held(&n->list_lock);
2213 :
2214 : /*
2215 : * Zap the freelist and set the frozen bit.
2216 : * The old freelist is the list of objects for the
2217 : * per cpu allocation list.
2218 : */
2219 94 : freelist = slab->freelist;
2220 94 : counters = slab->counters;
2221 94 : new.counters = counters;
2222 94 : if (mode) {
2223 94 : new.inuse = slab->objects;
2224 94 : new.freelist = NULL;
2225 : } else {
2226 : new.freelist = freelist;
2227 : }
2228 :
2229 : VM_BUG_ON(new.frozen);
2230 94 : new.frozen = 1;
2231 :
2232 188 : if (!__cmpxchg_double_slab(s, slab,
2233 : freelist, counters,
2234 : new.freelist, new.counters,
2235 : "acquire_slab"))
2236 : return NULL;
2237 :
2238 188 : remove_partial(n, slab);
2239 94 : WARN_ON(!freelist);
2240 : return freelist;
2241 : }
2242 :
2243 : #ifdef CONFIG_SLUB_CPU_PARTIAL
2244 : static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain);
2245 : #else
2246 : static inline void put_cpu_partial(struct kmem_cache *s, struct slab *slab,
2247 : int drain) { }
2248 : #endif
2249 : static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags);
2250 :
2251 : /*
2252 : * Try to allocate a partial slab from a specific node.
2253 : */
2254 2589 : static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
2255 : struct partial_context *pc)
2256 : {
2257 : struct slab *slab, *slab2;
2258 2589 : void *object = NULL;
2259 : unsigned long flags;
2260 2589 : unsigned int partial_slabs = 0;
2261 :
2262 : /*
2263 : * Racy check. If we mistakenly see no partial slabs then we
2264 : * just allocate an empty slab. If we mistakenly try to get a
2265 : * partial slab and there is none available then get_partial()
2266 : * will return NULL.
2267 : */
2268 2589 : if (!n || !n->nr_partial)
2269 : return NULL;
2270 :
2271 94 : spin_lock_irqsave(&n->list_lock, flags);
2272 94 : list_for_each_entry_safe(slab, slab2, &n->partial, slab_list) {
2273 : void *t;
2274 :
2275 188 : if (!pfmemalloc_match(slab, pc->flags))
2276 0 : continue;
2277 :
2278 94 : if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) {
2279 0 : object = alloc_single_from_partial(s, n, slab,
2280 0 : pc->orig_size);
2281 0 : if (object)
2282 : break;
2283 0 : continue;
2284 : }
2285 :
2286 94 : t = acquire_slab(s, n, slab, object == NULL);
2287 94 : if (!t)
2288 : break;
2289 :
2290 94 : if (!object) {
2291 94 : *pc->slab = slab;
2292 94 : stat(s, ALLOC_FROM_PARTIAL);
2293 94 : object = t;
2294 : } else {
2295 : put_cpu_partial(s, slab, 0);
2296 : stat(s, CPU_PARTIAL_NODE);
2297 : partial_slabs++;
2298 : }
2299 : #ifdef CONFIG_SLUB_CPU_PARTIAL
2300 : if (!kmem_cache_has_cpu_partial(s)
2301 : || partial_slabs > s->cpu_partial_slabs / 2)
2302 : break;
2303 : #else
2304 : break;
2305 : #endif
2306 :
2307 : }
2308 188 : spin_unlock_irqrestore(&n->list_lock, flags);
2309 94 : return object;
2310 : }
2311 :
2312 : /*
2313 : * Get a slab from somewhere. Search in increasing NUMA distances.
2314 : */
2315 : static void *get_any_partial(struct kmem_cache *s, struct partial_context *pc)
2316 : {
2317 : #ifdef CONFIG_NUMA
2318 : struct zonelist *zonelist;
2319 : struct zoneref *z;
2320 : struct zone *zone;
2321 : enum zone_type highest_zoneidx = gfp_zone(pc->flags);
2322 : void *object;
2323 : unsigned int cpuset_mems_cookie;
2324 :
2325 : /*
2326 : * The defrag ratio allows a configuration of the tradeoffs between
2327 : * inter node defragmentation and node local allocations. A lower
2328 : * defrag_ratio increases the tendency to do local allocations
2329 : * instead of attempting to obtain partial slabs from other nodes.
2330 : *
2331 : * If the defrag_ratio is set to 0 then kmalloc() always
2332 : * returns node local objects. If the ratio is higher then kmalloc()
2333 : * may return off node objects because partial slabs are obtained
2334 : * from other nodes and filled up.
2335 : *
2336 : * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
2337 : * (which makes defrag_ratio = 1000) then every (well almost)
2338 : * allocation will first attempt to defrag slab caches on other nodes.
2339 : * This means scanning over all nodes to look for partial slabs which
2340 : * may be expensive if we do it every time we are trying to find a slab
2341 : * with available objects.
2342 : */
2343 : if (!s->remote_node_defrag_ratio ||
2344 : get_cycles() % 1024 > s->remote_node_defrag_ratio)
2345 : return NULL;
2346 :
2347 : do {
2348 : cpuset_mems_cookie = read_mems_allowed_begin();
2349 : zonelist = node_zonelist(mempolicy_slab_node(), pc->flags);
2350 : for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) {
2351 : struct kmem_cache_node *n;
2352 :
2353 : n = get_node(s, zone_to_nid(zone));
2354 :
2355 : if (n && cpuset_zone_allowed(zone, pc->flags) &&
2356 : n->nr_partial > s->min_partial) {
2357 : object = get_partial_node(s, n, pc);
2358 : if (object) {
2359 : /*
2360 : * Don't check read_mems_allowed_retry()
2361 : * here - if mems_allowed was updated in
2362 : * parallel, that was a harmless race
2363 : * between allocation and the cpuset
2364 : * update
2365 : */
2366 : return object;
2367 : }
2368 : }
2369 : }
2370 : } while (read_mems_allowed_retry(cpuset_mems_cookie));
2371 : #endif /* CONFIG_NUMA */
2372 : return NULL;
2373 : }
2374 :
2375 : /*
2376 : * Get a partial slab, lock it and return it.
2377 : */
2378 2589 : static void *get_partial(struct kmem_cache *s, int node, struct partial_context *pc)
2379 : {
2380 : void *object;
2381 2589 : int searchnode = node;
2382 :
2383 2589 : if (node == NUMA_NO_NODE)
2384 2587 : searchnode = numa_mem_id();
2385 :
2386 2589 : object = get_partial_node(s, get_node(s, searchnode), pc);
2387 2589 : if (object || node != NUMA_NO_NODE)
2388 : return object;
2389 :
2390 2495 : return get_any_partial(s, pc);
2391 : }
2392 :
2393 : #ifndef CONFIG_SLUB_TINY
2394 :
2395 : #ifdef CONFIG_PREEMPTION
2396 : /*
2397 : * Calculate the next globally unique transaction for disambiguation
2398 : * during cmpxchg. The transactions start with the cpu number and are then
2399 : * incremented by CONFIG_NR_CPUS.
2400 : */
2401 : #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
2402 : #else
2403 : /*
2404 : * No preemption supported therefore also no need to check for
2405 : * different cpus.
2406 : */
2407 : #define TID_STEP 1
2408 : #endif /* CONFIG_PREEMPTION */
2409 :
2410 : static inline unsigned long next_tid(unsigned long tid)
2411 : {
2412 88995 : return tid + TID_STEP;
2413 : }
2414 :
2415 : #ifdef SLUB_DEBUG_CMPXCHG
2416 : static inline unsigned int tid_to_cpu(unsigned long tid)
2417 : {
2418 : return tid % TID_STEP;
2419 : }
2420 :
2421 : static inline unsigned long tid_to_event(unsigned long tid)
2422 : {
2423 : return tid / TID_STEP;
2424 : }
2425 : #endif
2426 :
2427 : static inline unsigned int init_tid(int cpu)
2428 : {
2429 52 : return cpu;
2430 : }
2431 :
2432 : static inline void note_cmpxchg_failure(const char *n,
2433 : const struct kmem_cache *s, unsigned long tid)
2434 : {
2435 : #ifdef SLUB_DEBUG_CMPXCHG
2436 : unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
2437 :
2438 : pr_info("%s %s: cmpxchg redo ", n, s->name);
2439 :
2440 : #ifdef CONFIG_PREEMPTION
2441 : if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
2442 : pr_warn("due to cpu change %d -> %d\n",
2443 : tid_to_cpu(tid), tid_to_cpu(actual_tid));
2444 : else
2445 : #endif
2446 : if (tid_to_event(tid) != tid_to_event(actual_tid))
2447 : pr_warn("due to cpu running other code. Event %ld->%ld\n",
2448 : tid_to_event(tid), tid_to_event(actual_tid));
2449 : else
2450 : pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
2451 : actual_tid, tid, next_tid(tid));
2452 : #endif
2453 : stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
2454 : }
2455 :
2456 : static void init_kmem_cache_cpus(struct kmem_cache *s)
2457 : {
2458 : int cpu;
2459 : struct kmem_cache_cpu *c;
2460 :
2461 52 : for_each_possible_cpu(cpu) {
2462 52 : c = per_cpu_ptr(s->cpu_slab, cpu);
2463 52 : local_lock_init(&c->lock);
2464 52 : c->tid = init_tid(cpu);
2465 : }
2466 : }
2467 :
2468 : /*
2469 : * Finishes removing the cpu slab. Merges cpu's freelist with slab's freelist,
2470 : * unfreezes the slabs and puts it on the proper list.
2471 : * Assumes the slab has been already safely taken away from kmem_cache_cpu
2472 : * by the caller.
2473 : */
2474 2 : static void deactivate_slab(struct kmem_cache *s, struct slab *slab,
2475 : void *freelist)
2476 : {
2477 : enum slab_modes { M_NONE, M_PARTIAL, M_FREE, M_FULL_NOLIST };
2478 6 : struct kmem_cache_node *n = get_node(s, slab_nid(slab));
2479 2 : int free_delta = 0;
2480 2 : enum slab_modes mode = M_NONE;
2481 : void *nextfree, *freelist_iter, *freelist_tail;
2482 2 : int tail = DEACTIVATE_TO_HEAD;
2483 2 : unsigned long flags = 0;
2484 : struct slab new;
2485 : struct slab old;
2486 :
2487 2 : if (slab->freelist) {
2488 0 : stat(s, DEACTIVATE_REMOTE_FREES);
2489 0 : tail = DEACTIVATE_TO_TAIL;
2490 : }
2491 :
2492 : /*
2493 : * Stage one: Count the objects on cpu's freelist as free_delta and
2494 : * remember the last object in freelist_tail for later splicing.
2495 : */
2496 2 : freelist_tail = NULL;
2497 2 : freelist_iter = freelist;
2498 86 : while (freelist_iter) {
2499 164 : nextfree = get_freepointer(s, freelist_iter);
2500 :
2501 : /*
2502 : * If 'nextfree' is invalid, it is possible that the object at
2503 : * 'freelist_iter' is already corrupted. So isolate all objects
2504 : * starting at 'freelist_iter' by skipping them.
2505 : */
2506 82 : if (freelist_corrupted(s, slab, &freelist_iter, nextfree))
2507 : break;
2508 :
2509 82 : freelist_tail = freelist_iter;
2510 82 : free_delta++;
2511 :
2512 82 : freelist_iter = nextfree;
2513 : }
2514 :
2515 : /*
2516 : * Stage two: Unfreeze the slab while splicing the per-cpu
2517 : * freelist to the head of slab's freelist.
2518 : *
2519 : * Ensure that the slab is unfrozen while the list presence
2520 : * reflects the actual number of objects during unfreeze.
2521 : *
2522 : * We first perform cmpxchg holding lock and insert to list
2523 : * when it succeed. If there is mismatch then the slab is not
2524 : * unfrozen and number of objects in the slab may have changed.
2525 : * Then release lock and retry cmpxchg again.
2526 : */
2527 : redo:
2528 :
2529 2 : old.freelist = READ_ONCE(slab->freelist);
2530 2 : old.counters = READ_ONCE(slab->counters);
2531 : VM_BUG_ON(!old.frozen);
2532 :
2533 : /* Determine target state of the slab */
2534 2 : new.counters = old.counters;
2535 2 : if (freelist_tail) {
2536 2 : new.inuse -= free_delta;
2537 4 : set_freepointer(s, freelist_tail, old.freelist);
2538 2 : new.freelist = freelist;
2539 : } else
2540 : new.freelist = old.freelist;
2541 :
2542 2 : new.frozen = 0;
2543 :
2544 2 : if (!new.inuse && n->nr_partial >= s->min_partial) {
2545 : mode = M_FREE;
2546 2 : } else if (new.freelist) {
2547 2 : mode = M_PARTIAL;
2548 : /*
2549 : * Taking the spinlock removes the possibility that
2550 : * acquire_slab() will see a slab that is frozen
2551 : */
2552 2 : spin_lock_irqsave(&n->list_lock, flags);
2553 : } else {
2554 : mode = M_FULL_NOLIST;
2555 : }
2556 :
2557 :
2558 2 : if (!cmpxchg_double_slab(s, slab,
2559 : old.freelist, old.counters,
2560 : new.freelist, new.counters,
2561 : "unfreezing slab")) {
2562 0 : if (mode == M_PARTIAL)
2563 0 : spin_unlock_irqrestore(&n->list_lock, flags);
2564 : goto redo;
2565 : }
2566 :
2567 :
2568 2 : if (mode == M_PARTIAL) {
2569 2 : add_partial(n, slab, tail);
2570 4 : spin_unlock_irqrestore(&n->list_lock, flags);
2571 2 : stat(s, tail);
2572 0 : } else if (mode == M_FREE) {
2573 0 : stat(s, DEACTIVATE_EMPTY);
2574 : discard_slab(s, slab);
2575 : stat(s, FREE_SLAB);
2576 : } else if (mode == M_FULL_NOLIST) {
2577 : stat(s, DEACTIVATE_FULL);
2578 : }
2579 2 : }
2580 :
2581 : #ifdef CONFIG_SLUB_CPU_PARTIAL
2582 : static void __unfreeze_partials(struct kmem_cache *s, struct slab *partial_slab)
2583 : {
2584 : struct kmem_cache_node *n = NULL, *n2 = NULL;
2585 : struct slab *slab, *slab_to_discard = NULL;
2586 : unsigned long flags = 0;
2587 :
2588 : while (partial_slab) {
2589 : struct slab new;
2590 : struct slab old;
2591 :
2592 : slab = partial_slab;
2593 : partial_slab = slab->next;
2594 :
2595 : n2 = get_node(s, slab_nid(slab));
2596 : if (n != n2) {
2597 : if (n)
2598 : spin_unlock_irqrestore(&n->list_lock, flags);
2599 :
2600 : n = n2;
2601 : spin_lock_irqsave(&n->list_lock, flags);
2602 : }
2603 :
2604 : do {
2605 :
2606 : old.freelist = slab->freelist;
2607 : old.counters = slab->counters;
2608 : VM_BUG_ON(!old.frozen);
2609 :
2610 : new.counters = old.counters;
2611 : new.freelist = old.freelist;
2612 :
2613 : new.frozen = 0;
2614 :
2615 : } while (!__cmpxchg_double_slab(s, slab,
2616 : old.freelist, old.counters,
2617 : new.freelist, new.counters,
2618 : "unfreezing slab"));
2619 :
2620 : if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
2621 : slab->next = slab_to_discard;
2622 : slab_to_discard = slab;
2623 : } else {
2624 : add_partial(n, slab, DEACTIVATE_TO_TAIL);
2625 : stat(s, FREE_ADD_PARTIAL);
2626 : }
2627 : }
2628 :
2629 : if (n)
2630 : spin_unlock_irqrestore(&n->list_lock, flags);
2631 :
2632 : while (slab_to_discard) {
2633 : slab = slab_to_discard;
2634 : slab_to_discard = slab_to_discard->next;
2635 :
2636 : stat(s, DEACTIVATE_EMPTY);
2637 : discard_slab(s, slab);
2638 : stat(s, FREE_SLAB);
2639 : }
2640 : }
2641 :
2642 : /*
2643 : * Unfreeze all the cpu partial slabs.
2644 : */
2645 : static void unfreeze_partials(struct kmem_cache *s)
2646 : {
2647 : struct slab *partial_slab;
2648 : unsigned long flags;
2649 :
2650 : local_lock_irqsave(&s->cpu_slab->lock, flags);
2651 : partial_slab = this_cpu_read(s->cpu_slab->partial);
2652 : this_cpu_write(s->cpu_slab->partial, NULL);
2653 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2654 :
2655 : if (partial_slab)
2656 : __unfreeze_partials(s, partial_slab);
2657 : }
2658 :
2659 : static void unfreeze_partials_cpu(struct kmem_cache *s,
2660 : struct kmem_cache_cpu *c)
2661 : {
2662 : struct slab *partial_slab;
2663 :
2664 : partial_slab = slub_percpu_partial(c);
2665 : c->partial = NULL;
2666 :
2667 : if (partial_slab)
2668 : __unfreeze_partials(s, partial_slab);
2669 : }
2670 :
2671 : /*
2672 : * Put a slab that was just frozen (in __slab_free|get_partial_node) into a
2673 : * partial slab slot if available.
2674 : *
2675 : * If we did not find a slot then simply move all the partials to the
2676 : * per node partial list.
2677 : */
2678 : static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain)
2679 : {
2680 : struct slab *oldslab;
2681 : struct slab *slab_to_unfreeze = NULL;
2682 : unsigned long flags;
2683 : int slabs = 0;
2684 :
2685 : local_lock_irqsave(&s->cpu_slab->lock, flags);
2686 :
2687 : oldslab = this_cpu_read(s->cpu_slab->partial);
2688 :
2689 : if (oldslab) {
2690 : if (drain && oldslab->slabs >= s->cpu_partial_slabs) {
2691 : /*
2692 : * Partial array is full. Move the existing set to the
2693 : * per node partial list. Postpone the actual unfreezing
2694 : * outside of the critical section.
2695 : */
2696 : slab_to_unfreeze = oldslab;
2697 : oldslab = NULL;
2698 : } else {
2699 : slabs = oldslab->slabs;
2700 : }
2701 : }
2702 :
2703 : slabs++;
2704 :
2705 : slab->slabs = slabs;
2706 : slab->next = oldslab;
2707 :
2708 : this_cpu_write(s->cpu_slab->partial, slab);
2709 :
2710 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2711 :
2712 : if (slab_to_unfreeze) {
2713 : __unfreeze_partials(s, slab_to_unfreeze);
2714 : stat(s, CPU_PARTIAL_DRAIN);
2715 : }
2716 : }
2717 :
2718 : #else /* CONFIG_SLUB_CPU_PARTIAL */
2719 :
2720 : static inline void unfreeze_partials(struct kmem_cache *s) { }
2721 : static inline void unfreeze_partials_cpu(struct kmem_cache *s,
2722 : struct kmem_cache_cpu *c) { }
2723 :
2724 : #endif /* CONFIG_SLUB_CPU_PARTIAL */
2725 :
2726 0 : static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
2727 : {
2728 : unsigned long flags;
2729 : struct slab *slab;
2730 : void *freelist;
2731 :
2732 0 : local_lock_irqsave(&s->cpu_slab->lock, flags);
2733 :
2734 0 : slab = c->slab;
2735 0 : freelist = c->freelist;
2736 :
2737 0 : c->slab = NULL;
2738 0 : c->freelist = NULL;
2739 0 : c->tid = next_tid(c->tid);
2740 :
2741 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2742 :
2743 0 : if (slab) {
2744 0 : deactivate_slab(s, slab, freelist);
2745 0 : stat(s, CPUSLAB_FLUSH);
2746 : }
2747 0 : }
2748 :
2749 2 : static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
2750 : {
2751 2 : struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2752 2 : void *freelist = c->freelist;
2753 2 : struct slab *slab = c->slab;
2754 :
2755 2 : c->slab = NULL;
2756 2 : c->freelist = NULL;
2757 4 : c->tid = next_tid(c->tid);
2758 :
2759 2 : if (slab) {
2760 2 : deactivate_slab(s, slab, freelist);
2761 2 : stat(s, CPUSLAB_FLUSH);
2762 : }
2763 :
2764 2 : unfreeze_partials_cpu(s, c);
2765 2 : }
2766 :
2767 : struct slub_flush_work {
2768 : struct work_struct work;
2769 : struct kmem_cache *s;
2770 : bool skip;
2771 : };
2772 :
2773 : /*
2774 : * Flush cpu slab.
2775 : *
2776 : * Called from CPU work handler with migration disabled.
2777 : */
2778 0 : static void flush_cpu_slab(struct work_struct *w)
2779 : {
2780 : struct kmem_cache *s;
2781 : struct kmem_cache_cpu *c;
2782 : struct slub_flush_work *sfw;
2783 :
2784 0 : sfw = container_of(w, struct slub_flush_work, work);
2785 :
2786 0 : s = sfw->s;
2787 0 : c = this_cpu_ptr(s->cpu_slab);
2788 :
2789 0 : if (c->slab)
2790 0 : flush_slab(s, c);
2791 :
2792 0 : unfreeze_partials(s);
2793 0 : }
2794 :
2795 : static bool has_cpu_slab(int cpu, struct kmem_cache *s)
2796 : {
2797 0 : struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2798 :
2799 0 : return c->slab || slub_percpu_partial(c);
2800 : }
2801 :
2802 : static DEFINE_MUTEX(flush_lock);
2803 : static DEFINE_PER_CPU(struct slub_flush_work, slub_flush);
2804 :
2805 0 : static void flush_all_cpus_locked(struct kmem_cache *s)
2806 : {
2807 : struct slub_flush_work *sfw;
2808 : unsigned int cpu;
2809 :
2810 : lockdep_assert_cpus_held();
2811 0 : mutex_lock(&flush_lock);
2812 :
2813 0 : for_each_online_cpu(cpu) {
2814 0 : sfw = &per_cpu(slub_flush, cpu);
2815 0 : if (!has_cpu_slab(cpu, s)) {
2816 0 : sfw->skip = true;
2817 0 : continue;
2818 : }
2819 0 : INIT_WORK(&sfw->work, flush_cpu_slab);
2820 0 : sfw->skip = false;
2821 0 : sfw->s = s;
2822 0 : queue_work_on(cpu, flushwq, &sfw->work);
2823 : }
2824 :
2825 0 : for_each_online_cpu(cpu) {
2826 0 : sfw = &per_cpu(slub_flush, cpu);
2827 0 : if (sfw->skip)
2828 0 : continue;
2829 0 : flush_work(&sfw->work);
2830 : }
2831 :
2832 0 : mutex_unlock(&flush_lock);
2833 0 : }
2834 :
2835 : static void flush_all(struct kmem_cache *s)
2836 : {
2837 : cpus_read_lock();
2838 0 : flush_all_cpus_locked(s);
2839 : cpus_read_unlock();
2840 : }
2841 :
2842 : /*
2843 : * Use the cpu notifier to insure that the cpu slabs are flushed when
2844 : * necessary.
2845 : */
2846 0 : static int slub_cpu_dead(unsigned int cpu)
2847 : {
2848 : struct kmem_cache *s;
2849 :
2850 0 : mutex_lock(&slab_mutex);
2851 0 : list_for_each_entry(s, &slab_caches, list)
2852 0 : __flush_cpu_slab(s, cpu);
2853 0 : mutex_unlock(&slab_mutex);
2854 0 : return 0;
2855 : }
2856 :
2857 : #else /* CONFIG_SLUB_TINY */
2858 : static inline void flush_all_cpus_locked(struct kmem_cache *s) { }
2859 : static inline void flush_all(struct kmem_cache *s) { }
2860 : static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) { }
2861 : static inline int slub_cpu_dead(unsigned int cpu) { return 0; }
2862 : #endif /* CONFIG_SLUB_TINY */
2863 :
2864 : /*
2865 : * Check if the objects in a per cpu structure fit numa
2866 : * locality expectations.
2867 : */
2868 : static inline int node_match(struct slab *slab, int node)
2869 : {
2870 : #ifdef CONFIG_NUMA
2871 : if (node != NUMA_NO_NODE && slab_nid(slab) != node)
2872 : return 0;
2873 : #endif
2874 : return 1;
2875 : }
2876 :
2877 : #ifdef CONFIG_SLUB_DEBUG
2878 0 : static int count_free(struct slab *slab)
2879 : {
2880 0 : return slab->objects - slab->inuse;
2881 : }
2882 :
2883 : static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2884 : {
2885 0 : return atomic_long_read(&n->total_objects);
2886 : }
2887 :
2888 : /* Supports checking bulk free of a constructed freelist */
2889 0 : static inline bool free_debug_processing(struct kmem_cache *s,
2890 : struct slab *slab, void *head, void *tail, int *bulk_cnt,
2891 : unsigned long addr, depot_stack_handle_t handle)
2892 : {
2893 0 : bool checks_ok = false;
2894 0 : void *object = head;
2895 0 : int cnt = 0;
2896 :
2897 0 : if (s->flags & SLAB_CONSISTENCY_CHECKS) {
2898 0 : if (!check_slab(s, slab))
2899 : goto out;
2900 : }
2901 :
2902 0 : if (slab->inuse < *bulk_cnt) {
2903 0 : slab_err(s, slab, "Slab has %d allocated objects but %d are to be freed\n",
2904 : slab->inuse, *bulk_cnt);
2905 0 : goto out;
2906 : }
2907 :
2908 : next_object:
2909 :
2910 0 : if (++cnt > *bulk_cnt)
2911 : goto out_cnt;
2912 :
2913 0 : if (s->flags & SLAB_CONSISTENCY_CHECKS) {
2914 0 : if (!free_consistency_checks(s, slab, object, addr))
2915 : goto out;
2916 : }
2917 :
2918 0 : if (s->flags & SLAB_STORE_USER)
2919 : set_track_update(s, object, TRACK_FREE, addr, handle);
2920 0 : trace(s, slab, object, 0);
2921 : /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
2922 0 : init_object(s, object, SLUB_RED_INACTIVE);
2923 :
2924 : /* Reached end of constructed freelist yet? */
2925 0 : if (object != tail) {
2926 0 : object = get_freepointer(s, object);
2927 0 : goto next_object;
2928 : }
2929 : checks_ok = true;
2930 :
2931 : out_cnt:
2932 0 : if (cnt != *bulk_cnt) {
2933 0 : slab_err(s, slab, "Bulk free expected %d objects but found %d\n",
2934 : *bulk_cnt, cnt);
2935 0 : *bulk_cnt = cnt;
2936 : }
2937 :
2938 : out:
2939 :
2940 0 : if (!checks_ok)
2941 0 : slab_fix(s, "Object at 0x%p not freed", object);
2942 :
2943 0 : return checks_ok;
2944 : }
2945 : #endif /* CONFIG_SLUB_DEBUG */
2946 :
2947 : #if defined(CONFIG_SLUB_DEBUG) || defined(SLAB_SUPPORTS_SYSFS)
2948 0 : static unsigned long count_partial(struct kmem_cache_node *n,
2949 : int (*get_count)(struct slab *))
2950 : {
2951 : unsigned long flags;
2952 0 : unsigned long x = 0;
2953 : struct slab *slab;
2954 :
2955 0 : spin_lock_irqsave(&n->list_lock, flags);
2956 0 : list_for_each_entry(slab, &n->partial, slab_list)
2957 0 : x += get_count(slab);
2958 0 : spin_unlock_irqrestore(&n->list_lock, flags);
2959 0 : return x;
2960 : }
2961 : #endif /* CONFIG_SLUB_DEBUG || SLAB_SUPPORTS_SYSFS */
2962 :
2963 : #ifdef CONFIG_SLUB_DEBUG
2964 : static noinline void
2965 0 : slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2966 : {
2967 : static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2968 : DEFAULT_RATELIMIT_BURST);
2969 : int node;
2970 : struct kmem_cache_node *n;
2971 :
2972 0 : if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2973 : return;
2974 :
2975 0 : pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2976 : nid, gfpflags, &gfpflags);
2977 0 : pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
2978 : s->name, s->object_size, s->size, oo_order(s->oo),
2979 : oo_order(s->min));
2980 :
2981 0 : if (oo_order(s->min) > get_order(s->object_size))
2982 0 : pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2983 : s->name);
2984 :
2985 0 : for_each_kmem_cache_node(s, node, n) {
2986 : unsigned long nr_slabs;
2987 : unsigned long nr_objs;
2988 : unsigned long nr_free;
2989 :
2990 0 : nr_free = count_partial(n, count_free);
2991 0 : nr_slabs = node_nr_slabs(n);
2992 0 : nr_objs = node_nr_objs(n);
2993 :
2994 0 : pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
2995 : node, nr_slabs, nr_objs, nr_free);
2996 : }
2997 : }
2998 : #else /* CONFIG_SLUB_DEBUG */
2999 : static inline void
3000 : slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) { }
3001 : #endif
3002 :
3003 : static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags)
3004 : {
3005 5242 : if (unlikely(slab_test_pfmemalloc(slab)))
3006 0 : return gfp_pfmemalloc_allowed(gfpflags);
3007 :
3008 : return true;
3009 : }
3010 :
3011 : #ifndef CONFIG_SLUB_TINY
3012 : /*
3013 : * Check the slab->freelist and either transfer the freelist to the
3014 : * per cpu freelist or deactivate the slab.
3015 : *
3016 : * The slab is still frozen if the return value is not NULL.
3017 : *
3018 : * If this function returns NULL then the slab has been unfrozen.
3019 : */
3020 : static inline void *get_freelist(struct kmem_cache *s, struct slab *slab)
3021 : {
3022 : struct slab new;
3023 : unsigned long counters;
3024 : void *freelist;
3025 :
3026 : lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock));
3027 :
3028 : do {
3029 2559 : freelist = slab->freelist;
3030 2559 : counters = slab->counters;
3031 :
3032 2559 : new.counters = counters;
3033 : VM_BUG_ON(!new.frozen);
3034 :
3035 2559 : new.inuse = slab->objects;
3036 2559 : new.frozen = freelist != NULL;
3037 :
3038 5118 : } while (!__cmpxchg_double_slab(s, slab,
3039 : freelist, counters,
3040 : NULL, new.counters,
3041 2559 : "get_freelist"));
3042 :
3043 : return freelist;
3044 : }
3045 :
3046 : /*
3047 : * Slow path. The lockless freelist is empty or we need to perform
3048 : * debugging duties.
3049 : *
3050 : * Processing is still very fast if new objects have been freed to the
3051 : * regular freelist. In that case we simply take over the regular freelist
3052 : * as the lockless freelist and zap the regular freelist.
3053 : *
3054 : * If that is not working then we fall back to the partial lists. We take the
3055 : * first element of the freelist as the object to allocate now and move the
3056 : * rest of the freelist to the lockless freelist.
3057 : *
3058 : * And if we were unable to get a new slab from the partial slab lists then
3059 : * we need to allocate a new slab. This is the slowest path since it involves
3060 : * a call to the page allocator and the setup of a new slab.
3061 : *
3062 : * Version of __slab_alloc to use when we know that preemption is
3063 : * already disabled (which is the case for bulk allocation).
3064 : */
3065 2589 : static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
3066 : unsigned long addr, struct kmem_cache_cpu *c, unsigned int orig_size)
3067 : {
3068 : void *freelist;
3069 : struct slab *slab;
3070 : unsigned long flags;
3071 : struct partial_context pc;
3072 :
3073 2589 : stat(s, ALLOC_SLOWPATH);
3074 :
3075 : reread_slab:
3076 :
3077 2589 : slab = READ_ONCE(c->slab);
3078 2589 : if (!slab) {
3079 : /*
3080 : * if the node is not online or has no normal memory, just
3081 : * ignore the node constraint
3082 : */
3083 32 : if (unlikely(node != NUMA_NO_NODE &&
3084 : !node_isset(node, slab_nodes)))
3085 0 : node = NUMA_NO_NODE;
3086 : goto new_slab;
3087 : }
3088 : redo:
3089 :
3090 2559 : if (unlikely(!node_match(slab, node))) {
3091 : /*
3092 : * same as above but node_match() being false already
3093 : * implies node != NUMA_NO_NODE
3094 : */
3095 : if (!node_isset(node, slab_nodes)) {
3096 : node = NUMA_NO_NODE;
3097 : } else {
3098 : stat(s, ALLOC_NODE_MISMATCH);
3099 : goto deactivate_slab;
3100 : }
3101 : }
3102 :
3103 : /*
3104 : * By rights, we should be searching for a slab page that was
3105 : * PFMEMALLOC but right now, we are losing the pfmemalloc
3106 : * information when the page leaves the per-cpu allocator
3107 : */
3108 5118 : if (unlikely(!pfmemalloc_match(slab, gfpflags)))
3109 : goto deactivate_slab;
3110 :
3111 : /* must check again c->slab in case we got preempted and it changed */
3112 2559 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3113 2559 : if (unlikely(slab != c->slab)) {
3114 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3115 : goto reread_slab;
3116 : }
3117 2559 : freelist = c->freelist;
3118 2559 : if (freelist)
3119 : goto load_freelist;
3120 :
3121 2559 : freelist = get_freelist(s, slab);
3122 :
3123 2559 : if (!freelist) {
3124 2559 : c->slab = NULL;
3125 5118 : c->tid = next_tid(c->tid);
3126 2559 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3127 : stat(s, DEACTIVATE_BYPASS);
3128 : goto new_slab;
3129 : }
3130 :
3131 : stat(s, ALLOC_REFILL);
3132 :
3133 : load_freelist:
3134 :
3135 2589 : lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock));
3136 :
3137 : /*
3138 : * freelist is pointing to the list of objects to be used.
3139 : * slab is pointing to the slab from which the objects are obtained.
3140 : * That slab must be frozen for per cpu allocations to work.
3141 : */
3142 : VM_BUG_ON(!c->slab->frozen);
3143 5178 : c->freelist = get_freepointer(s, freelist);
3144 5178 : c->tid = next_tid(c->tid);
3145 5178 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3146 2589 : return freelist;
3147 :
3148 : deactivate_slab:
3149 :
3150 0 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3151 0 : if (slab != c->slab) {
3152 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3153 : goto reread_slab;
3154 : }
3155 0 : freelist = c->freelist;
3156 0 : c->slab = NULL;
3157 0 : c->freelist = NULL;
3158 0 : c->tid = next_tid(c->tid);
3159 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3160 0 : deactivate_slab(s, slab, freelist);
3161 :
3162 : new_slab:
3163 :
3164 : if (slub_percpu_partial(c)) {
3165 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3166 : if (unlikely(c->slab)) {
3167 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3168 : goto reread_slab;
3169 : }
3170 : if (unlikely(!slub_percpu_partial(c))) {
3171 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3172 : /* we were preempted and partial list got empty */
3173 : goto new_objects;
3174 : }
3175 :
3176 : slab = c->slab = slub_percpu_partial(c);
3177 : slub_set_percpu_partial(c, slab);
3178 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3179 : stat(s, CPU_PARTIAL_ALLOC);
3180 : goto redo;
3181 : }
3182 :
3183 : new_objects:
3184 :
3185 2589 : pc.flags = gfpflags;
3186 2589 : pc.slab = &slab;
3187 2589 : pc.orig_size = orig_size;
3188 2589 : freelist = get_partial(s, node, &pc);
3189 2589 : if (freelist)
3190 : goto check_new_slab;
3191 :
3192 2495 : slub_put_cpu_ptr(s->cpu_slab);
3193 2495 : slab = new_slab(s, gfpflags, node);
3194 2495 : c = slub_get_cpu_ptr(s->cpu_slab);
3195 :
3196 2495 : if (unlikely(!slab)) {
3197 0 : slab_out_of_memory(s, gfpflags, node);
3198 0 : return NULL;
3199 : }
3200 :
3201 2495 : stat(s, ALLOC_SLAB);
3202 :
3203 2495 : if (kmem_cache_debug(s)) {
3204 0 : freelist = alloc_single_from_new_slab(s, slab, orig_size);
3205 :
3206 0 : if (unlikely(!freelist))
3207 : goto new_objects;
3208 :
3209 0 : if (s->flags & SLAB_STORE_USER)
3210 : set_track(s, freelist, TRACK_ALLOC, addr);
3211 :
3212 : return freelist;
3213 : }
3214 :
3215 : /*
3216 : * No other reference to the slab yet so we can
3217 : * muck around with it freely without cmpxchg
3218 : */
3219 2495 : freelist = slab->freelist;
3220 2495 : slab->freelist = NULL;
3221 2495 : slab->inuse = slab->objects;
3222 2495 : slab->frozen = 1;
3223 :
3224 4990 : inc_slabs_node(s, slab_nid(slab), slab->objects);
3225 :
3226 : check_new_slab:
3227 :
3228 2589 : if (kmem_cache_debug(s)) {
3229 : /*
3230 : * For debug caches here we had to go through
3231 : * alloc_single_from_partial() so just store the tracking info
3232 : * and return the object
3233 : */
3234 0 : if (s->flags & SLAB_STORE_USER)
3235 : set_track(s, freelist, TRACK_ALLOC, addr);
3236 :
3237 : return freelist;
3238 : }
3239 :
3240 5178 : if (unlikely(!pfmemalloc_match(slab, gfpflags))) {
3241 : /*
3242 : * For !pfmemalloc_match() case we don't load freelist so that
3243 : * we don't make further mismatched allocations easier.
3244 : */
3245 0 : deactivate_slab(s, slab, get_freepointer(s, freelist));
3246 0 : return freelist;
3247 : }
3248 :
3249 : retry_load_slab:
3250 :
3251 2589 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3252 2589 : if (unlikely(c->slab)) {
3253 0 : void *flush_freelist = c->freelist;
3254 0 : struct slab *flush_slab = c->slab;
3255 :
3256 0 : c->slab = NULL;
3257 0 : c->freelist = NULL;
3258 0 : c->tid = next_tid(c->tid);
3259 :
3260 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3261 :
3262 0 : deactivate_slab(s, flush_slab, flush_freelist);
3263 :
3264 0 : stat(s, CPUSLAB_FLUSH);
3265 :
3266 : goto retry_load_slab;
3267 : }
3268 2589 : c->slab = slab;
3269 :
3270 2589 : goto load_freelist;
3271 : }
3272 :
3273 : /*
3274 : * A wrapper for ___slab_alloc() for contexts where preemption is not yet
3275 : * disabled. Compensates for possible cpu changes by refetching the per cpu area
3276 : * pointer.
3277 : */
3278 : static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
3279 : unsigned long addr, struct kmem_cache_cpu *c, unsigned int orig_size)
3280 : {
3281 : void *p;
3282 :
3283 : #ifdef CONFIG_PREEMPT_COUNT
3284 : /*
3285 : * We may have been preempted and rescheduled on a different
3286 : * cpu before disabling preemption. Need to reload cpu area
3287 : * pointer.
3288 : */
3289 : c = slub_get_cpu_ptr(s->cpu_slab);
3290 : #endif
3291 :
3292 2589 : p = ___slab_alloc(s, gfpflags, node, addr, c, orig_size);
3293 : #ifdef CONFIG_PREEMPT_COUNT
3294 : slub_put_cpu_ptr(s->cpu_slab);
3295 : #endif
3296 : return p;
3297 : }
3298 :
3299 : static __always_inline void *__slab_alloc_node(struct kmem_cache *s,
3300 : gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
3301 : {
3302 : struct kmem_cache_cpu *c;
3303 : struct slab *slab;
3304 : unsigned long tid;
3305 : void *object;
3306 :
3307 : redo:
3308 : /*
3309 : * Must read kmem_cache cpu data via this cpu ptr. Preemption is
3310 : * enabled. We may switch back and forth between cpus while
3311 : * reading from one cpu area. That does not matter as long
3312 : * as we end up on the original cpu again when doing the cmpxchg.
3313 : *
3314 : * We must guarantee that tid and kmem_cache_cpu are retrieved on the
3315 : * same cpu. We read first the kmem_cache_cpu pointer and use it to read
3316 : * the tid. If we are preempted and switched to another cpu between the
3317 : * two reads, it's OK as the two are still associated with the same cpu
3318 : * and cmpxchg later will validate the cpu.
3319 : */
3320 78089 : c = raw_cpu_ptr(s->cpu_slab);
3321 78089 : tid = READ_ONCE(c->tid);
3322 :
3323 : /*
3324 : * Irqless object alloc/free algorithm used here depends on sequence
3325 : * of fetching cpu_slab's data. tid should be fetched before anything
3326 : * on c to guarantee that object and slab associated with previous tid
3327 : * won't be used with current tid. If we fetch tid first, object and
3328 : * slab could be one associated with next tid and our alloc/free
3329 : * request will be failed. In this case, we will retry. So, no problem.
3330 : */
3331 78089 : barrier();
3332 :
3333 : /*
3334 : * The transaction ids are globally unique per cpu and per operation on
3335 : * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
3336 : * occurs on the right processor and that there was no operation on the
3337 : * linked list in between.
3338 : */
3339 :
3340 78089 : object = c->freelist;
3341 78089 : slab = c->slab;
3342 :
3343 78089 : if (!USE_LOCKLESS_FAST_PATH() ||
3344 153589 : unlikely(!object || !slab || !node_match(slab, node))) {
3345 4526 : object = __slab_alloc(s, gfpflags, node, addr, c, orig_size);
3346 : } else {
3347 75500 : void *next_object = get_freepointer_safe(s, object);
3348 :
3349 : /*
3350 : * The cmpxchg will only match if there was no additional
3351 : * operation and if we are on the right processor.
3352 : *
3353 : * The cmpxchg does the following atomically (without lock
3354 : * semantics!)
3355 : * 1. Relocate first pointer to the current per cpu area.
3356 : * 2. Verify that tid and freelist have not been changed
3357 : * 3. If they were not changed replace tid and freelist
3358 : *
3359 : * Since this is without lock semantics the protection is only
3360 : * against code executing on this cpu *not* from access by
3361 : * other cpus.
3362 : */
3363 302000 : if (unlikely(!this_cpu_cmpxchg_double(
3364 : s->cpu_slab->freelist, s->cpu_slab->tid,
3365 : object, tid,
3366 : next_object, next_tid(tid)))) {
3367 :
3368 : note_cmpxchg_failure("slab_alloc", s, tid);
3369 : goto redo;
3370 : }
3371 75500 : prefetch_freepointer(s, next_object);
3372 : stat(s, ALLOC_FASTPATH);
3373 : }
3374 :
3375 : return object;
3376 : }
3377 : #else /* CONFIG_SLUB_TINY */
3378 : static void *__slab_alloc_node(struct kmem_cache *s,
3379 : gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
3380 : {
3381 : struct partial_context pc;
3382 : struct slab *slab;
3383 : void *object;
3384 :
3385 : pc.flags = gfpflags;
3386 : pc.slab = &slab;
3387 : pc.orig_size = orig_size;
3388 : object = get_partial(s, node, &pc);
3389 :
3390 : if (object)
3391 : return object;
3392 :
3393 : slab = new_slab(s, gfpflags, node);
3394 : if (unlikely(!slab)) {
3395 : slab_out_of_memory(s, gfpflags, node);
3396 : return NULL;
3397 : }
3398 :
3399 : object = alloc_single_from_new_slab(s, slab, orig_size);
3400 :
3401 : return object;
3402 : }
3403 : #endif /* CONFIG_SLUB_TINY */
3404 :
3405 : /*
3406 : * If the object has been wiped upon free, make sure it's fully initialized by
3407 : * zeroing out freelist pointer.
3408 : */
3409 : static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s,
3410 : void *obj)
3411 : {
3412 78089 : if (unlikely(slab_want_init_on_free(s)) && obj)
3413 0 : memset((void *)((char *)kasan_reset_tag(obj) + s->offset),
3414 : 0, sizeof(void *));
3415 : }
3416 :
3417 : /*
3418 : * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
3419 : * have the fastpath folded into their functions. So no function call
3420 : * overhead for requests that can be satisfied on the fastpath.
3421 : *
3422 : * The fastpath works by first checking if the lockless freelist can be used.
3423 : * If not then __slab_alloc is called for slow processing.
3424 : *
3425 : * Otherwise we can simply pick the next object from the lockless free list.
3426 : */
3427 : static __fastpath_inline void *slab_alloc_node(struct kmem_cache *s, struct list_lru *lru,
3428 : gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
3429 : {
3430 : void *object;
3431 78089 : struct obj_cgroup *objcg = NULL;
3432 78089 : bool init = false;
3433 :
3434 156178 : s = slab_pre_alloc_hook(s, lru, &objcg, 1, gfpflags);
3435 78089 : if (!s)
3436 : return NULL;
3437 :
3438 78089 : object = kfence_alloc(s, orig_size, gfpflags);
3439 : if (unlikely(object))
3440 : goto out;
3441 :
3442 78089 : object = __slab_alloc_node(s, gfpflags, node, addr, orig_size);
3443 :
3444 156178 : maybe_wipe_obj_freeptr(s, object);
3445 156178 : init = slab_want_init_on_alloc(gfpflags, s);
3446 :
3447 : out:
3448 : /*
3449 : * When init equals 'true', like for kzalloc() family, only
3450 : * @orig_size bytes might be zeroed instead of s->object_size
3451 : */
3452 78089 : slab_post_alloc_hook(s, objcg, gfpflags, 1, &object, init, orig_size);
3453 :
3454 78089 : return object;
3455 : }
3456 :
3457 : static __fastpath_inline void *slab_alloc(struct kmem_cache *s, struct list_lru *lru,
3458 : gfp_t gfpflags, unsigned long addr, size_t orig_size)
3459 : {
3460 30851 : return slab_alloc_node(s, lru, gfpflags, NUMA_NO_NODE, addr, orig_size);
3461 : }
3462 :
3463 : static __fastpath_inline
3464 : void *__kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
3465 : gfp_t gfpflags)
3466 : {
3467 61702 : void *ret = slab_alloc(s, lru, gfpflags, _RET_IP_, s->object_size);
3468 :
3469 30851 : trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, NUMA_NO_NODE);
3470 :
3471 : return ret;
3472 : }
3473 :
3474 30766 : void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
3475 : {
3476 30766 : return __kmem_cache_alloc_lru(s, NULL, gfpflags);
3477 : }
3478 : EXPORT_SYMBOL(kmem_cache_alloc);
3479 :
3480 85 : void *kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
3481 : gfp_t gfpflags)
3482 : {
3483 85 : return __kmem_cache_alloc_lru(s, lru, gfpflags);
3484 : }
3485 : EXPORT_SYMBOL(kmem_cache_alloc_lru);
3486 :
3487 46553 : void *__kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags,
3488 : int node, size_t orig_size,
3489 : unsigned long caller)
3490 : {
3491 46553 : return slab_alloc_node(s, NULL, gfpflags, node,
3492 : caller, orig_size);
3493 : }
3494 :
3495 685 : void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
3496 : {
3497 1370 : void *ret = slab_alloc_node(s, NULL, gfpflags, node, _RET_IP_, s->object_size);
3498 :
3499 685 : trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, node);
3500 :
3501 685 : return ret;
3502 : }
3503 : EXPORT_SYMBOL(kmem_cache_alloc_node);
3504 :
3505 0 : static noinline void free_to_partial_list(
3506 : struct kmem_cache *s, struct slab *slab,
3507 : void *head, void *tail, int bulk_cnt,
3508 : unsigned long addr)
3509 : {
3510 0 : struct kmem_cache_node *n = get_node(s, slab_nid(slab));
3511 0 : struct slab *slab_free = NULL;
3512 0 : int cnt = bulk_cnt;
3513 : unsigned long flags;
3514 0 : depot_stack_handle_t handle = 0;
3515 :
3516 0 : if (s->flags & SLAB_STORE_USER)
3517 0 : handle = set_track_prepare();
3518 :
3519 0 : spin_lock_irqsave(&n->list_lock, flags);
3520 :
3521 0 : if (free_debug_processing(s, slab, head, tail, &cnt, addr, handle)) {
3522 0 : void *prior = slab->freelist;
3523 :
3524 : /* Perform the actual freeing while we still hold the locks */
3525 0 : slab->inuse -= cnt;
3526 0 : set_freepointer(s, tail, prior);
3527 0 : slab->freelist = head;
3528 :
3529 : /*
3530 : * If the slab is empty, and node's partial list is full,
3531 : * it should be discarded anyway no matter it's on full or
3532 : * partial list.
3533 : */
3534 0 : if (slab->inuse == 0 && n->nr_partial >= s->min_partial)
3535 0 : slab_free = slab;
3536 :
3537 0 : if (!prior) {
3538 : /* was on full list */
3539 0 : remove_full(s, n, slab);
3540 0 : if (!slab_free) {
3541 : add_partial(n, slab, DEACTIVATE_TO_TAIL);
3542 : stat(s, FREE_ADD_PARTIAL);
3543 : }
3544 0 : } else if (slab_free) {
3545 0 : remove_partial(n, slab);
3546 : stat(s, FREE_REMOVE_PARTIAL);
3547 : }
3548 : }
3549 :
3550 0 : if (slab_free) {
3551 : /*
3552 : * Update the counters while still holding n->list_lock to
3553 : * prevent spurious validation warnings
3554 : */
3555 0 : dec_slabs_node(s, slab_nid(slab_free), slab_free->objects);
3556 : }
3557 :
3558 0 : spin_unlock_irqrestore(&n->list_lock, flags);
3559 :
3560 0 : if (slab_free) {
3561 0 : stat(s, FREE_SLAB);
3562 0 : free_slab(s, slab_free);
3563 : }
3564 0 : }
3565 :
3566 : /*
3567 : * Slow path handling. This may still be called frequently since objects
3568 : * have a longer lifetime than the cpu slabs in most processing loads.
3569 : *
3570 : * So we still attempt to reduce cache line usage. Just take the slab
3571 : * lock and free the item. If there is no additional partial slab
3572 : * handling required then we can return immediately.
3573 : */
3574 57200 : static void __slab_free(struct kmem_cache *s, struct slab *slab,
3575 : void *head, void *tail, int cnt,
3576 : unsigned long addr)
3577 :
3578 : {
3579 : void *prior;
3580 : int was_frozen;
3581 : struct slab new;
3582 : unsigned long counters;
3583 57200 : struct kmem_cache_node *n = NULL;
3584 : unsigned long flags;
3585 :
3586 57200 : stat(s, FREE_SLOWPATH);
3587 :
3588 57200 : if (kfence_free(head))
3589 55144 : return;
3590 :
3591 57200 : if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) {
3592 0 : free_to_partial_list(s, slab, head, tail, cnt, addr);
3593 0 : return;
3594 : }
3595 :
3596 : do {
3597 57200 : if (unlikely(n)) {
3598 0 : spin_unlock_irqrestore(&n->list_lock, flags);
3599 0 : n = NULL;
3600 : }
3601 57200 : prior = slab->freelist;
3602 57200 : counters = slab->counters;
3603 114400 : set_freepointer(s, tail, prior);
3604 57200 : new.counters = counters;
3605 57200 : was_frozen = new.frozen;
3606 57200 : new.inuse -= cnt;
3607 57200 : if ((!new.inuse || !prior) && !was_frozen) {
3608 :
3609 4261 : if (kmem_cache_has_cpu_partial(s) && !prior) {
3610 :
3611 : /*
3612 : * Slab was on no list before and will be
3613 : * partially empty
3614 : * We can defer the list move and instead
3615 : * freeze it.
3616 : */
3617 : new.frozen = 1;
3618 :
3619 : } else { /* Needs to be taken off a list */
3620 :
3621 12783 : n = get_node(s, slab_nid(slab));
3622 : /*
3623 : * Speculatively acquire the list_lock.
3624 : * If the cmpxchg does not succeed then we may
3625 : * drop the list_lock without any processing.
3626 : *
3627 : * Otherwise the list_lock will synchronize with
3628 : * other processors updating the list of slabs.
3629 : */
3630 4261 : spin_lock_irqsave(&n->list_lock, flags);
3631 :
3632 : }
3633 : }
3634 :
3635 57200 : } while (!cmpxchg_double_slab(s, slab,
3636 : prior, counters,
3637 : head, new.counters,
3638 57200 : "__slab_free"));
3639 :
3640 57200 : if (likely(!n)) {
3641 :
3642 : if (likely(was_frozen)) {
3643 : /*
3644 : * The list lock was not taken therefore no list
3645 : * activity can be necessary.
3646 : */
3647 : stat(s, FREE_FROZEN);
3648 : } else if (new.frozen) {
3649 : /*
3650 : * If we just froze the slab then put it onto the
3651 : * per cpu partial list.
3652 : */
3653 : put_cpu_partial(s, slab, 1);
3654 : stat(s, CPU_PARTIAL_FREE);
3655 : }
3656 :
3657 : return;
3658 : }
3659 :
3660 4261 : if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
3661 : goto slab_empty;
3662 :
3663 : /*
3664 : * Objects left in the slab. If it was not on the partial list before
3665 : * then add it.
3666 : */
3667 2205 : if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
3668 4326 : remove_full(s, n, slab);
3669 : add_partial(n, slab, DEACTIVATE_TO_TAIL);
3670 : stat(s, FREE_ADD_PARTIAL);
3671 : }
3672 2205 : spin_unlock_irqrestore(&n->list_lock, flags);
3673 : return;
3674 :
3675 : slab_empty:
3676 2056 : if (prior) {
3677 : /*
3678 : * Slab on the partial list.
3679 : */
3680 2056 : remove_partial(n, slab);
3681 : stat(s, FREE_REMOVE_PARTIAL);
3682 : } else {
3683 : /* Slab must be on the full list */
3684 0 : remove_full(s, n, slab);
3685 : }
3686 :
3687 4112 : spin_unlock_irqrestore(&n->list_lock, flags);
3688 2056 : stat(s, FREE_SLAB);
3689 2056 : discard_slab(s, slab);
3690 : }
3691 :
3692 : #ifndef CONFIG_SLUB_TINY
3693 : /*
3694 : * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
3695 : * can perform fastpath freeing without additional function calls.
3696 : *
3697 : * The fastpath is only possible if we are freeing to the current cpu slab
3698 : * of this processor. This typically the case if we have just allocated
3699 : * the item before.
3700 : *
3701 : * If fastpath is not possible then fall back to __slab_free where we deal
3702 : * with all sorts of special processing.
3703 : *
3704 : * Bulk free of a freelist with several objects (all pointing to the
3705 : * same slab) possible by specifying head and tail ptr, plus objects
3706 : * count (cnt). Bulk free indicated by tail pointer being set.
3707 : */
3708 : static __always_inline void do_slab_free(struct kmem_cache *s,
3709 : struct slab *slab, void *head, void *tail,
3710 : int cnt, unsigned long addr)
3711 : {
3712 65545 : void *tail_obj = tail ? : head;
3713 : struct kmem_cache_cpu *c;
3714 : unsigned long tid;
3715 : void **freelist;
3716 :
3717 : redo:
3718 : /*
3719 : * Determine the currently cpus per cpu slab.
3720 : * The cpu may change afterward. However that does not matter since
3721 : * data is retrieved via this pointer. If we are on the same cpu
3722 : * during the cmpxchg then the free will succeed.
3723 : */
3724 65545 : c = raw_cpu_ptr(s->cpu_slab);
3725 65545 : tid = READ_ONCE(c->tid);
3726 :
3727 : /* Same with comment on barrier() in slab_alloc_node() */
3728 65545 : barrier();
3729 :
3730 65545 : if (unlikely(slab != c->slab)) {
3731 57200 : __slab_free(s, slab, head, tail_obj, cnt, addr);
3732 : return;
3733 : }
3734 :
3735 : if (USE_LOCKLESS_FAST_PATH()) {
3736 8345 : freelist = READ_ONCE(c->freelist);
3737 :
3738 16690 : set_freepointer(s, tail_obj, freelist);
3739 :
3740 33380 : if (unlikely(!this_cpu_cmpxchg_double(
3741 : s->cpu_slab->freelist, s->cpu_slab->tid,
3742 : freelist, tid,
3743 : head, next_tid(tid)))) {
3744 :
3745 : note_cmpxchg_failure("slab_free", s, tid);
3746 : goto redo;
3747 : }
3748 : } else {
3749 : /* Update the free list under the local lock */
3750 : local_lock(&s->cpu_slab->lock);
3751 : c = this_cpu_ptr(s->cpu_slab);
3752 : if (unlikely(slab != c->slab)) {
3753 : local_unlock(&s->cpu_slab->lock);
3754 : goto redo;
3755 : }
3756 : tid = c->tid;
3757 : freelist = c->freelist;
3758 :
3759 : set_freepointer(s, tail_obj, freelist);
3760 : c->freelist = head;
3761 : c->tid = next_tid(tid);
3762 :
3763 : local_unlock(&s->cpu_slab->lock);
3764 : }
3765 : stat(s, FREE_FASTPATH);
3766 : }
3767 : #else /* CONFIG_SLUB_TINY */
3768 : static void do_slab_free(struct kmem_cache *s,
3769 : struct slab *slab, void *head, void *tail,
3770 : int cnt, unsigned long addr)
3771 : {
3772 : void *tail_obj = tail ? : head;
3773 :
3774 : __slab_free(s, slab, head, tail_obj, cnt, addr);
3775 : }
3776 : #endif /* CONFIG_SLUB_TINY */
3777 :
3778 : static __fastpath_inline void slab_free(struct kmem_cache *s, struct slab *slab,
3779 : void *head, void *tail, void **p, int cnt,
3780 : unsigned long addr)
3781 : {
3782 65545 : memcg_slab_free_hook(s, slab, p, cnt);
3783 : /*
3784 : * With KASAN enabled slab_free_freelist_hook modifies the freelist
3785 : * to remove objects, whose reuse must be delayed.
3786 : */
3787 65545 : if (slab_free_freelist_hook(s, &head, &tail, &cnt))
3788 65545 : do_slab_free(s, slab, head, tail, cnt, addr);
3789 : }
3790 :
3791 : #ifdef CONFIG_KASAN_GENERIC
3792 : void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
3793 : {
3794 : do_slab_free(cache, virt_to_slab(x), x, NULL, 1, addr);
3795 : }
3796 : #endif
3797 :
3798 42825 : void __kmem_cache_free(struct kmem_cache *s, void *x, unsigned long caller)
3799 : {
3800 128475 : slab_free(s, virt_to_slab(x), x, NULL, &x, 1, caller);
3801 42825 : }
3802 :
3803 22720 : void kmem_cache_free(struct kmem_cache *s, void *x)
3804 : {
3805 22720 : s = cache_from_obj(s, x);
3806 22720 : if (!s)
3807 : return;
3808 22720 : trace_kmem_cache_free(_RET_IP_, x, s);
3809 45440 : slab_free(s, virt_to_slab(x), x, NULL, &x, 1, _RET_IP_);
3810 : }
3811 : EXPORT_SYMBOL(kmem_cache_free);
3812 :
3813 : struct detached_freelist {
3814 : struct slab *slab;
3815 : void *tail;
3816 : void *freelist;
3817 : int cnt;
3818 : struct kmem_cache *s;
3819 : };
3820 :
3821 : /*
3822 : * This function progressively scans the array with free objects (with
3823 : * a limited look ahead) and extract objects belonging to the same
3824 : * slab. It builds a detached freelist directly within the given
3825 : * slab/objects. This can happen without any need for
3826 : * synchronization, because the objects are owned by running process.
3827 : * The freelist is build up as a single linked list in the objects.
3828 : * The idea is, that this detached freelist can then be bulk
3829 : * transferred to the real freelist(s), but only requiring a single
3830 : * synchronization primitive. Look ahead in the array is limited due
3831 : * to performance reasons.
3832 : */
3833 : static inline
3834 0 : int build_detached_freelist(struct kmem_cache *s, size_t size,
3835 : void **p, struct detached_freelist *df)
3836 : {
3837 0 : int lookahead = 3;
3838 : void *object;
3839 : struct folio *folio;
3840 : size_t same;
3841 :
3842 0 : object = p[--size];
3843 0 : folio = virt_to_folio(object);
3844 0 : if (!s) {
3845 : /* Handle kalloc'ed objects */
3846 0 : if (unlikely(!folio_test_slab(folio))) {
3847 0 : free_large_kmalloc(folio, object);
3848 0 : df->slab = NULL;
3849 0 : return size;
3850 : }
3851 : /* Derive kmem_cache from object */
3852 0 : df->slab = folio_slab(folio);
3853 0 : df->s = df->slab->slab_cache;
3854 : } else {
3855 0 : df->slab = folio_slab(folio);
3856 0 : df->s = cache_from_obj(s, object); /* Support for memcg */
3857 : }
3858 :
3859 : /* Start new detached freelist */
3860 0 : df->tail = object;
3861 0 : df->freelist = object;
3862 0 : df->cnt = 1;
3863 :
3864 0 : if (is_kfence_address(object))
3865 : return size;
3866 :
3867 0 : set_freepointer(df->s, object, NULL);
3868 :
3869 0 : same = size;
3870 0 : while (size) {
3871 0 : object = p[--size];
3872 : /* df->slab is always set at this point */
3873 0 : if (df->slab == virt_to_slab(object)) {
3874 : /* Opportunity build freelist */
3875 0 : set_freepointer(df->s, object, df->freelist);
3876 0 : df->freelist = object;
3877 0 : df->cnt++;
3878 0 : same--;
3879 0 : if (size != same)
3880 0 : swap(p[size], p[same]);
3881 0 : continue;
3882 : }
3883 :
3884 : /* Limit look ahead search */
3885 0 : if (!--lookahead)
3886 : break;
3887 : }
3888 :
3889 0 : return same;
3890 : }
3891 :
3892 : /* Note that interrupts must be enabled when calling this function. */
3893 0 : void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
3894 : {
3895 0 : if (!size)
3896 : return;
3897 :
3898 : do {
3899 : struct detached_freelist df;
3900 :
3901 0 : size = build_detached_freelist(s, size, p, &df);
3902 0 : if (!df.slab)
3903 0 : continue;
3904 :
3905 0 : slab_free(df.s, df.slab, df.freelist, df.tail, &p[size], df.cnt,
3906 0 : _RET_IP_);
3907 0 : } while (likely(size));
3908 : }
3909 : EXPORT_SYMBOL(kmem_cache_free_bulk);
3910 :
3911 : #ifndef CONFIG_SLUB_TINY
3912 0 : static inline int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags,
3913 : size_t size, void **p, struct obj_cgroup *objcg)
3914 : {
3915 : struct kmem_cache_cpu *c;
3916 : unsigned long irqflags;
3917 : int i;
3918 :
3919 : /*
3920 : * Drain objects in the per cpu slab, while disabling local
3921 : * IRQs, which protects against PREEMPT and interrupts
3922 : * handlers invoking normal fastpath.
3923 : */
3924 0 : c = slub_get_cpu_ptr(s->cpu_slab);
3925 0 : local_lock_irqsave(&s->cpu_slab->lock, irqflags);
3926 :
3927 0 : for (i = 0; i < size; i++) {
3928 0 : void *object = kfence_alloc(s, s->object_size, flags);
3929 :
3930 : if (unlikely(object)) {
3931 : p[i] = object;
3932 : continue;
3933 : }
3934 :
3935 0 : object = c->freelist;
3936 0 : if (unlikely(!object)) {
3937 : /*
3938 : * We may have removed an object from c->freelist using
3939 : * the fastpath in the previous iteration; in that case,
3940 : * c->tid has not been bumped yet.
3941 : * Since ___slab_alloc() may reenable interrupts while
3942 : * allocating memory, we should bump c->tid now.
3943 : */
3944 0 : c->tid = next_tid(c->tid);
3945 :
3946 0 : local_unlock_irqrestore(&s->cpu_slab->lock, irqflags);
3947 :
3948 : /*
3949 : * Invoking slow path likely have side-effect
3950 : * of re-populating per CPU c->freelist
3951 : */
3952 0 : p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
3953 0 : _RET_IP_, c, s->object_size);
3954 0 : if (unlikely(!p[i]))
3955 : goto error;
3956 :
3957 0 : c = this_cpu_ptr(s->cpu_slab);
3958 0 : maybe_wipe_obj_freeptr(s, p[i]);
3959 :
3960 0 : local_lock_irqsave(&s->cpu_slab->lock, irqflags);
3961 :
3962 0 : continue; /* goto for-loop */
3963 : }
3964 0 : c->freelist = get_freepointer(s, object);
3965 0 : p[i] = object;
3966 0 : maybe_wipe_obj_freeptr(s, p[i]);
3967 : }
3968 0 : c->tid = next_tid(c->tid);
3969 0 : local_unlock_irqrestore(&s->cpu_slab->lock, irqflags);
3970 0 : slub_put_cpu_ptr(s->cpu_slab);
3971 :
3972 0 : return i;
3973 :
3974 : error:
3975 0 : slub_put_cpu_ptr(s->cpu_slab);
3976 0 : slab_post_alloc_hook(s, objcg, flags, i, p, false, s->object_size);
3977 0 : kmem_cache_free_bulk(s, i, p);
3978 0 : return 0;
3979 :
3980 : }
3981 : #else /* CONFIG_SLUB_TINY */
3982 : static int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags,
3983 : size_t size, void **p, struct obj_cgroup *objcg)
3984 : {
3985 : int i;
3986 :
3987 : for (i = 0; i < size; i++) {
3988 : void *object = kfence_alloc(s, s->object_size, flags);
3989 :
3990 : if (unlikely(object)) {
3991 : p[i] = object;
3992 : continue;
3993 : }
3994 :
3995 : p[i] = __slab_alloc_node(s, flags, NUMA_NO_NODE,
3996 : _RET_IP_, s->object_size);
3997 : if (unlikely(!p[i]))
3998 : goto error;
3999 :
4000 : maybe_wipe_obj_freeptr(s, p[i]);
4001 : }
4002 :
4003 : return i;
4004 :
4005 : error:
4006 : slab_post_alloc_hook(s, objcg, flags, i, p, false, s->object_size);
4007 : kmem_cache_free_bulk(s, i, p);
4008 : return 0;
4009 : }
4010 : #endif /* CONFIG_SLUB_TINY */
4011 :
4012 : /* Note that interrupts must be enabled when calling this function. */
4013 0 : int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
4014 : void **p)
4015 : {
4016 : int i;
4017 0 : struct obj_cgroup *objcg = NULL;
4018 :
4019 0 : if (!size)
4020 : return 0;
4021 :
4022 : /* memcg and kmem_cache debug support */
4023 0 : s = slab_pre_alloc_hook(s, NULL, &objcg, size, flags);
4024 0 : if (unlikely(!s))
4025 : return 0;
4026 :
4027 0 : i = __kmem_cache_alloc_bulk(s, flags, size, p, objcg);
4028 :
4029 : /*
4030 : * memcg and kmem_cache debug support and memory initialization.
4031 : * Done outside of the IRQ disabled fastpath loop.
4032 : */
4033 0 : if (i != 0)
4034 0 : slab_post_alloc_hook(s, objcg, flags, size, p,
4035 0 : slab_want_init_on_alloc(flags, s), s->object_size);
4036 : return i;
4037 : }
4038 : EXPORT_SYMBOL(kmem_cache_alloc_bulk);
4039 :
4040 :
4041 : /*
4042 : * Object placement in a slab is made very easy because we always start at
4043 : * offset 0. If we tune the size of the object to the alignment then we can
4044 : * get the required alignment by putting one properly sized object after
4045 : * another.
4046 : *
4047 : * Notice that the allocation order determines the sizes of the per cpu
4048 : * caches. Each processor has always one slab available for allocations.
4049 : * Increasing the allocation order reduces the number of times that slabs
4050 : * must be moved on and off the partial lists and is therefore a factor in
4051 : * locking overhead.
4052 : */
4053 :
4054 : /*
4055 : * Minimum / Maximum order of slab pages. This influences locking overhead
4056 : * and slab fragmentation. A higher order reduces the number of partial slabs
4057 : * and increases the number of allocations possible without having to
4058 : * take the list_lock.
4059 : */
4060 : static unsigned int slub_min_order;
4061 : static unsigned int slub_max_order =
4062 : IS_ENABLED(CONFIG_SLUB_TINY) ? 1 : PAGE_ALLOC_COSTLY_ORDER;
4063 : static unsigned int slub_min_objects;
4064 :
4065 : /*
4066 : * Calculate the order of allocation given an slab object size.
4067 : *
4068 : * The order of allocation has significant impact on performance and other
4069 : * system components. Generally order 0 allocations should be preferred since
4070 : * order 0 does not cause fragmentation in the page allocator. Larger objects
4071 : * be problematic to put into order 0 slabs because there may be too much
4072 : * unused space left. We go to a higher order if more than 1/16th of the slab
4073 : * would be wasted.
4074 : *
4075 : * In order to reach satisfactory performance we must ensure that a minimum
4076 : * number of objects is in one slab. Otherwise we may generate too much
4077 : * activity on the partial lists which requires taking the list_lock. This is
4078 : * less a concern for large slabs though which are rarely used.
4079 : *
4080 : * slub_max_order specifies the order where we begin to stop considering the
4081 : * number of objects in a slab as critical. If we reach slub_max_order then
4082 : * we try to keep the page order as low as possible. So we accept more waste
4083 : * of space in favor of a small page order.
4084 : *
4085 : * Higher order allocations also allow the placement of more objects in a
4086 : * slab and thereby reduce object handling overhead. If the user has
4087 : * requested a higher minimum order then we start with that one instead of
4088 : * the smallest order which will fit the object.
4089 : */
4090 53 : static inline unsigned int calc_slab_order(unsigned int size,
4091 : unsigned int min_objects, unsigned int max_order,
4092 : unsigned int fract_leftover)
4093 : {
4094 53 : unsigned int min_order = slub_min_order;
4095 : unsigned int order;
4096 :
4097 53 : if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
4098 0 : return get_order(size * MAX_OBJS_PER_PAGE) - 1;
4099 :
4100 162 : for (order = max(min_order, (unsigned int)get_order(min_objects * size));
4101 3 : order <= max_order; order++) {
4102 :
4103 55 : unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
4104 : unsigned int rem;
4105 :
4106 55 : rem = slab_size % size;
4107 :
4108 55 : if (rem <= slab_size / fract_leftover)
4109 : break;
4110 : }
4111 :
4112 : return order;
4113 : }
4114 :
4115 52 : static inline int calculate_order(unsigned int size)
4116 : {
4117 : unsigned int order;
4118 : unsigned int min_objects;
4119 : unsigned int max_objects;
4120 : unsigned int nr_cpus;
4121 :
4122 : /*
4123 : * Attempt to find best configuration for a slab. This
4124 : * works by first attempting to generate a layout with
4125 : * the best configuration and backing off gradually.
4126 : *
4127 : * First we increase the acceptable waste in a slab. Then
4128 : * we reduce the minimum objects required in a slab.
4129 : */
4130 52 : min_objects = slub_min_objects;
4131 52 : if (!min_objects) {
4132 : /*
4133 : * Some architectures will only update present cpus when
4134 : * onlining them, so don't trust the number if it's just 1. But
4135 : * we also don't want to use nr_cpu_ids always, as on some other
4136 : * architectures, there can be many possible cpus, but never
4137 : * onlined. Here we compromise between trying to avoid too high
4138 : * order on systems that appear larger than they are, and too
4139 : * low order on systems that appear smaller than they are.
4140 : */
4141 52 : nr_cpus = num_present_cpus();
4142 : if (nr_cpus <= 1)
4143 52 : nr_cpus = nr_cpu_ids;
4144 52 : min_objects = 4 * (fls(nr_cpus) + 1);
4145 : }
4146 104 : max_objects = order_objects(slub_max_order, size);
4147 52 : min_objects = min(min_objects, max_objects);
4148 :
4149 104 : while (min_objects > 1) {
4150 : unsigned int fraction;
4151 :
4152 : fraction = 16;
4153 53 : while (fraction >= 4) {
4154 53 : order = calc_slab_order(size, min_objects,
4155 : slub_max_order, fraction);
4156 53 : if (order <= slub_max_order)
4157 52 : return order;
4158 1 : fraction /= 2;
4159 : }
4160 0 : min_objects--;
4161 : }
4162 :
4163 : /*
4164 : * We were unable to place multiple objects in a slab. Now
4165 : * lets see if we can place a single object there.
4166 : */
4167 0 : order = calc_slab_order(size, 1, slub_max_order, 1);
4168 0 : if (order <= slub_max_order)
4169 0 : return order;
4170 :
4171 : /*
4172 : * Doh this slab cannot be placed using slub_max_order.
4173 : */
4174 0 : order = calc_slab_order(size, 1, MAX_ORDER, 1);
4175 0 : if (order < MAX_ORDER)
4176 0 : return order;
4177 : return -ENOSYS;
4178 : }
4179 :
4180 : static void
4181 : init_kmem_cache_node(struct kmem_cache_node *n)
4182 : {
4183 52 : n->nr_partial = 0;
4184 52 : spin_lock_init(&n->list_lock);
4185 104 : INIT_LIST_HEAD(&n->partial);
4186 : #ifdef CONFIG_SLUB_DEBUG
4187 104 : atomic_long_set(&n->nr_slabs, 0);
4188 104 : atomic_long_set(&n->total_objects, 0);
4189 104 : INIT_LIST_HEAD(&n->full);
4190 : #endif
4191 : }
4192 :
4193 : #ifndef CONFIG_SLUB_TINY
4194 52 : static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
4195 : {
4196 : BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
4197 : NR_KMALLOC_TYPES * KMALLOC_SHIFT_HIGH *
4198 : sizeof(struct kmem_cache_cpu));
4199 :
4200 : /*
4201 : * Must align to double word boundary for the double cmpxchg
4202 : * instructions to work; see __pcpu_double_call_return_bool().
4203 : */
4204 52 : s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
4205 : 2 * sizeof(void *));
4206 :
4207 52 : if (!s->cpu_slab)
4208 : return 0;
4209 :
4210 : init_kmem_cache_cpus(s);
4211 :
4212 : return 1;
4213 : }
4214 : #else
4215 : static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
4216 : {
4217 : return 1;
4218 : }
4219 : #endif /* CONFIG_SLUB_TINY */
4220 :
4221 : static struct kmem_cache *kmem_cache_node;
4222 :
4223 : /*
4224 : * No kmalloc_node yet so do it by hand. We know that this is the first
4225 : * slab on the node for this slabcache. There are no concurrent accesses
4226 : * possible.
4227 : *
4228 : * Note that this function only works on the kmem_cache_node
4229 : * when allocating for the kmem_cache_node. This is used for bootstrapping
4230 : * memory on a fresh node that has no slab structures yet.
4231 : */
4232 1 : static void early_kmem_cache_node_alloc(int node)
4233 : {
4234 : struct slab *slab;
4235 : struct kmem_cache_node *n;
4236 :
4237 1 : BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
4238 :
4239 1 : slab = new_slab(kmem_cache_node, GFP_NOWAIT, node);
4240 :
4241 1 : BUG_ON(!slab);
4242 3 : inc_slabs_node(kmem_cache_node, slab_nid(slab), slab->objects);
4243 2 : if (slab_nid(slab) != node) {
4244 0 : pr_err("SLUB: Unable to allocate memory from node %d\n", node);
4245 0 : pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
4246 : }
4247 :
4248 1 : n = slab->freelist;
4249 1 : BUG_ON(!n);
4250 : #ifdef CONFIG_SLUB_DEBUG
4251 1 : init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
4252 1 : init_tracking(kmem_cache_node, n);
4253 : #endif
4254 1 : n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false);
4255 2 : slab->freelist = get_freepointer(kmem_cache_node, n);
4256 1 : slab->inuse = 1;
4257 1 : kmem_cache_node->node[node] = n;
4258 1 : init_kmem_cache_node(n);
4259 2 : inc_slabs_node(kmem_cache_node, node, slab->objects);
4260 :
4261 : /*
4262 : * No locks need to be taken here as it has just been
4263 : * initialized and there is no concurrent access.
4264 : */
4265 1 : __add_partial(n, slab, DEACTIVATE_TO_HEAD);
4266 1 : }
4267 :
4268 0 : static void free_kmem_cache_nodes(struct kmem_cache *s)
4269 : {
4270 : int node;
4271 : struct kmem_cache_node *n;
4272 :
4273 0 : for_each_kmem_cache_node(s, node, n) {
4274 0 : s->node[node] = NULL;
4275 0 : kmem_cache_free(kmem_cache_node, n);
4276 : }
4277 0 : }
4278 :
4279 0 : void __kmem_cache_release(struct kmem_cache *s)
4280 : {
4281 0 : cache_random_seq_destroy(s);
4282 : #ifndef CONFIG_SLUB_TINY
4283 0 : free_percpu(s->cpu_slab);
4284 : #endif
4285 0 : free_kmem_cache_nodes(s);
4286 0 : }
4287 :
4288 52 : static int init_kmem_cache_nodes(struct kmem_cache *s)
4289 : {
4290 : int node;
4291 :
4292 156 : for_each_node_mask(node, slab_nodes) {
4293 : struct kmem_cache_node *n;
4294 :
4295 52 : if (slab_state == DOWN) {
4296 1 : early_kmem_cache_node_alloc(node);
4297 1 : continue;
4298 : }
4299 51 : n = kmem_cache_alloc_node(kmem_cache_node,
4300 : GFP_KERNEL, node);
4301 :
4302 51 : if (!n) {
4303 0 : free_kmem_cache_nodes(s);
4304 0 : return 0;
4305 : }
4306 :
4307 51 : init_kmem_cache_node(n);
4308 51 : s->node[node] = n;
4309 : }
4310 : return 1;
4311 : }
4312 :
4313 : static void set_cpu_partial(struct kmem_cache *s)
4314 : {
4315 : #ifdef CONFIG_SLUB_CPU_PARTIAL
4316 : unsigned int nr_objects;
4317 :
4318 : /*
4319 : * cpu_partial determined the maximum number of objects kept in the
4320 : * per cpu partial lists of a processor.
4321 : *
4322 : * Per cpu partial lists mainly contain slabs that just have one
4323 : * object freed. If they are used for allocation then they can be
4324 : * filled up again with minimal effort. The slab will never hit the
4325 : * per node partial lists and therefore no locking will be required.
4326 : *
4327 : * For backwards compatibility reasons, this is determined as number
4328 : * of objects, even though we now limit maximum number of pages, see
4329 : * slub_set_cpu_partial()
4330 : */
4331 : if (!kmem_cache_has_cpu_partial(s))
4332 : nr_objects = 0;
4333 : else if (s->size >= PAGE_SIZE)
4334 : nr_objects = 6;
4335 : else if (s->size >= 1024)
4336 : nr_objects = 24;
4337 : else if (s->size >= 256)
4338 : nr_objects = 52;
4339 : else
4340 : nr_objects = 120;
4341 :
4342 : slub_set_cpu_partial(s, nr_objects);
4343 : #endif
4344 : }
4345 :
4346 : /*
4347 : * calculate_sizes() determines the order and the distribution of data within
4348 : * a slab object.
4349 : */
4350 52 : static int calculate_sizes(struct kmem_cache *s)
4351 : {
4352 52 : slab_flags_t flags = s->flags;
4353 52 : unsigned int size = s->object_size;
4354 : unsigned int order;
4355 :
4356 : /*
4357 : * Round up object size to the next word boundary. We can only
4358 : * place the free pointer at word boundaries and this determines
4359 : * the possible location of the free pointer.
4360 : */
4361 52 : size = ALIGN(size, sizeof(void *));
4362 :
4363 : #ifdef CONFIG_SLUB_DEBUG
4364 : /*
4365 : * Determine if we can poison the object itself. If the user of
4366 : * the slab may touch the object after free or before allocation
4367 : * then we should never poison the object itself.
4368 : */
4369 52 : if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
4370 0 : !s->ctor)
4371 0 : s->flags |= __OBJECT_POISON;
4372 : else
4373 52 : s->flags &= ~__OBJECT_POISON;
4374 :
4375 :
4376 : /*
4377 : * If we are Redzoning then check if there is some space between the
4378 : * end of the object and the free pointer. If not then add an
4379 : * additional word to have some bytes to store Redzone information.
4380 : */
4381 52 : if ((flags & SLAB_RED_ZONE) && size == s->object_size)
4382 0 : size += sizeof(void *);
4383 : #endif
4384 :
4385 : /*
4386 : * With that we have determined the number of bytes in actual use
4387 : * by the object and redzoning.
4388 : */
4389 52 : s->inuse = size;
4390 :
4391 52 : if (slub_debug_orig_size(s) ||
4392 48 : (flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
4393 48 : ((flags & SLAB_RED_ZONE) && s->object_size < sizeof(void *)) ||
4394 48 : s->ctor) {
4395 : /*
4396 : * Relocate free pointer after the object if it is not
4397 : * permitted to overwrite the first word of the object on
4398 : * kmem_cache_free.
4399 : *
4400 : * This is the case if we do RCU, have a constructor or
4401 : * destructor, are poisoning the objects, or are
4402 : * redzoning an object smaller than sizeof(void *).
4403 : *
4404 : * The assumption that s->offset >= s->inuse means free
4405 : * pointer is outside of the object is used in the
4406 : * freeptr_outside_object() function. If that is no
4407 : * longer true, the function needs to be modified.
4408 : */
4409 9 : s->offset = size;
4410 9 : size += sizeof(void *);
4411 : } else {
4412 : /*
4413 : * Store freelist pointer near middle of object to keep
4414 : * it away from the edges of the object to avoid small
4415 : * sized over/underflows from neighboring allocations.
4416 : */
4417 43 : s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *));
4418 : }
4419 :
4420 : #ifdef CONFIG_SLUB_DEBUG
4421 52 : if (flags & SLAB_STORE_USER) {
4422 : /*
4423 : * Need to store information about allocs and frees after
4424 : * the object.
4425 : */
4426 0 : size += 2 * sizeof(struct track);
4427 :
4428 : /* Save the original kmalloc request size */
4429 0 : if (flags & SLAB_KMALLOC)
4430 0 : size += sizeof(unsigned int);
4431 : }
4432 : #endif
4433 :
4434 52 : kasan_cache_create(s, &size, &s->flags);
4435 : #ifdef CONFIG_SLUB_DEBUG
4436 52 : if (flags & SLAB_RED_ZONE) {
4437 : /*
4438 : * Add some empty padding so that we can catch
4439 : * overwrites from earlier objects rather than let
4440 : * tracking information or the free pointer be
4441 : * corrupted if a user writes before the start
4442 : * of the object.
4443 : */
4444 0 : size += sizeof(void *);
4445 :
4446 : s->red_left_pad = sizeof(void *);
4447 0 : s->red_left_pad = ALIGN(s->red_left_pad, s->align);
4448 0 : size += s->red_left_pad;
4449 : }
4450 : #endif
4451 :
4452 : /*
4453 : * SLUB stores one object immediately after another beginning from
4454 : * offset 0. In order to align the objects we have to simply size
4455 : * each object to conform to the alignment.
4456 : */
4457 52 : size = ALIGN(size, s->align);
4458 52 : s->size = size;
4459 52 : s->reciprocal_size = reciprocal_value(size);
4460 52 : order = calculate_order(size);
4461 :
4462 52 : if ((int)order < 0)
4463 : return 0;
4464 :
4465 52 : s->allocflags = 0;
4466 52 : if (order)
4467 18 : s->allocflags |= __GFP_COMP;
4468 :
4469 52 : if (s->flags & SLAB_CACHE_DMA)
4470 0 : s->allocflags |= GFP_DMA;
4471 :
4472 52 : if (s->flags & SLAB_CACHE_DMA32)
4473 0 : s->allocflags |= GFP_DMA32;
4474 :
4475 52 : if (s->flags & SLAB_RECLAIM_ACCOUNT)
4476 18 : s->allocflags |= __GFP_RECLAIMABLE;
4477 :
4478 : /*
4479 : * Determine the number of objects per slab
4480 : */
4481 104 : s->oo = oo_make(order, size);
4482 156 : s->min = oo_make(get_order(size), size);
4483 :
4484 52 : return !!oo_objects(s->oo);
4485 : }
4486 :
4487 52 : static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
4488 : {
4489 52 : s->flags = kmem_cache_flags(s->size, flags, s->name);
4490 : #ifdef CONFIG_SLAB_FREELIST_HARDENED
4491 : s->random = get_random_long();
4492 : #endif
4493 :
4494 52 : if (!calculate_sizes(s))
4495 : goto error;
4496 52 : if (disable_higher_order_debug) {
4497 : /*
4498 : * Disable debugging flags that store metadata if the min slab
4499 : * order increased.
4500 : */
4501 0 : if (get_order(s->size) > get_order(s->object_size)) {
4502 0 : s->flags &= ~DEBUG_METADATA_FLAGS;
4503 0 : s->offset = 0;
4504 0 : if (!calculate_sizes(s))
4505 : goto error;
4506 : }
4507 : }
4508 :
4509 : #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
4510 : defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
4511 : if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
4512 : /* Enable fast mode */
4513 : s->flags |= __CMPXCHG_DOUBLE;
4514 : #endif
4515 :
4516 : /*
4517 : * The larger the object size is, the more slabs we want on the partial
4518 : * list to avoid pounding the page allocator excessively.
4519 : */
4520 104 : s->min_partial = min_t(unsigned long, MAX_PARTIAL, ilog2(s->size) / 2);
4521 52 : s->min_partial = max_t(unsigned long, MIN_PARTIAL, s->min_partial);
4522 :
4523 52 : set_cpu_partial(s);
4524 :
4525 : #ifdef CONFIG_NUMA
4526 : s->remote_node_defrag_ratio = 1000;
4527 : #endif
4528 :
4529 : /* Initialize the pre-computed randomized freelist if slab is up */
4530 : if (slab_state >= UP) {
4531 : if (init_cache_random_seq(s))
4532 : goto error;
4533 : }
4534 :
4535 52 : if (!init_kmem_cache_nodes(s))
4536 : goto error;
4537 :
4538 52 : if (alloc_kmem_cache_cpus(s))
4539 : return 0;
4540 :
4541 : error:
4542 0 : __kmem_cache_release(s);
4543 0 : return -EINVAL;
4544 : }
4545 :
4546 0 : static void list_slab_objects(struct kmem_cache *s, struct slab *slab,
4547 : const char *text)
4548 : {
4549 : #ifdef CONFIG_SLUB_DEBUG
4550 0 : void *addr = slab_address(slab);
4551 : void *p;
4552 :
4553 0 : slab_err(s, slab, text, s->name);
4554 :
4555 0 : spin_lock(&object_map_lock);
4556 0 : __fill_map(object_map, s, slab);
4557 :
4558 0 : for_each_object(p, s, addr, slab->objects) {
4559 :
4560 0 : if (!test_bit(__obj_to_index(s, addr, p), object_map)) {
4561 0 : pr_err("Object 0x%p @offset=%tu\n", p, p - addr);
4562 0 : print_tracking(s, p);
4563 : }
4564 : }
4565 0 : spin_unlock(&object_map_lock);
4566 : #endif
4567 0 : }
4568 :
4569 : /*
4570 : * Attempt to free all partial slabs on a node.
4571 : * This is called from __kmem_cache_shutdown(). We must take list_lock
4572 : * because sysfs file might still access partial list after the shutdowning.
4573 : */
4574 0 : static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
4575 : {
4576 0 : LIST_HEAD(discard);
4577 : struct slab *slab, *h;
4578 :
4579 0 : BUG_ON(irqs_disabled());
4580 0 : spin_lock_irq(&n->list_lock);
4581 0 : list_for_each_entry_safe(slab, h, &n->partial, slab_list) {
4582 0 : if (!slab->inuse) {
4583 0 : remove_partial(n, slab);
4584 0 : list_add(&slab->slab_list, &discard);
4585 : } else {
4586 0 : list_slab_objects(s, slab,
4587 : "Objects remaining in %s on __kmem_cache_shutdown()");
4588 : }
4589 : }
4590 0 : spin_unlock_irq(&n->list_lock);
4591 :
4592 0 : list_for_each_entry_safe(slab, h, &discard, slab_list)
4593 0 : discard_slab(s, slab);
4594 0 : }
4595 :
4596 0 : bool __kmem_cache_empty(struct kmem_cache *s)
4597 : {
4598 : int node;
4599 : struct kmem_cache_node *n;
4600 :
4601 0 : for_each_kmem_cache_node(s, node, n)
4602 0 : if (n->nr_partial || slabs_node(s, node))
4603 : return false;
4604 : return true;
4605 : }
4606 :
4607 : /*
4608 : * Release all resources used by a slab cache.
4609 : */
4610 0 : int __kmem_cache_shutdown(struct kmem_cache *s)
4611 : {
4612 : int node;
4613 : struct kmem_cache_node *n;
4614 :
4615 0 : flush_all_cpus_locked(s);
4616 : /* Attempt to free all objects */
4617 0 : for_each_kmem_cache_node(s, node, n) {
4618 0 : free_partial(s, n);
4619 0 : if (n->nr_partial || slabs_node(s, node))
4620 : return 1;
4621 : }
4622 : return 0;
4623 : }
4624 :
4625 : #ifdef CONFIG_PRINTK
4626 0 : void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
4627 : {
4628 : void *base;
4629 : int __maybe_unused i;
4630 : unsigned int objnr;
4631 : void *objp;
4632 : void *objp0;
4633 0 : struct kmem_cache *s = slab->slab_cache;
4634 : struct track __maybe_unused *trackp;
4635 :
4636 0 : kpp->kp_ptr = object;
4637 0 : kpp->kp_slab = slab;
4638 0 : kpp->kp_slab_cache = s;
4639 0 : base = slab_address(slab);
4640 0 : objp0 = kasan_reset_tag(object);
4641 : #ifdef CONFIG_SLUB_DEBUG
4642 0 : objp = restore_red_left(s, objp0);
4643 : #else
4644 : objp = objp0;
4645 : #endif
4646 0 : objnr = obj_to_index(s, slab, objp);
4647 0 : kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp);
4648 0 : objp = base + s->size * objnr;
4649 0 : kpp->kp_objp = objp;
4650 0 : if (WARN_ON_ONCE(objp < base || objp >= base + slab->objects * s->size
4651 0 : || (objp - base) % s->size) ||
4652 0 : !(s->flags & SLAB_STORE_USER))
4653 : return;
4654 : #ifdef CONFIG_SLUB_DEBUG
4655 0 : objp = fixup_red_left(s, objp);
4656 0 : trackp = get_track(s, objp, TRACK_ALLOC);
4657 0 : kpp->kp_ret = (void *)trackp->addr;
4658 : #ifdef CONFIG_STACKDEPOT
4659 : {
4660 : depot_stack_handle_t handle;
4661 : unsigned long *entries;
4662 : unsigned int nr_entries;
4663 :
4664 0 : handle = READ_ONCE(trackp->handle);
4665 0 : if (handle) {
4666 0 : nr_entries = stack_depot_fetch(handle, &entries);
4667 0 : for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
4668 0 : kpp->kp_stack[i] = (void *)entries[i];
4669 : }
4670 :
4671 0 : trackp = get_track(s, objp, TRACK_FREE);
4672 0 : handle = READ_ONCE(trackp->handle);
4673 0 : if (handle) {
4674 0 : nr_entries = stack_depot_fetch(handle, &entries);
4675 0 : for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
4676 0 : kpp->kp_free_stack[i] = (void *)entries[i];
4677 : }
4678 : }
4679 : #endif
4680 : #endif
4681 : }
4682 : #endif
4683 :
4684 : /********************************************************************
4685 : * Kmalloc subsystem
4686 : *******************************************************************/
4687 :
4688 0 : static int __init setup_slub_min_order(char *str)
4689 : {
4690 0 : get_option(&str, (int *)&slub_min_order);
4691 :
4692 0 : return 1;
4693 : }
4694 :
4695 : __setup("slub_min_order=", setup_slub_min_order);
4696 :
4697 0 : static int __init setup_slub_max_order(char *str)
4698 : {
4699 0 : get_option(&str, (int *)&slub_max_order);
4700 0 : slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1);
4701 :
4702 0 : return 1;
4703 : }
4704 :
4705 : __setup("slub_max_order=", setup_slub_max_order);
4706 :
4707 0 : static int __init setup_slub_min_objects(char *str)
4708 : {
4709 0 : get_option(&str, (int *)&slub_min_objects);
4710 :
4711 0 : return 1;
4712 : }
4713 :
4714 : __setup("slub_min_objects=", setup_slub_min_objects);
4715 :
4716 : #ifdef CONFIG_HARDENED_USERCOPY
4717 : /*
4718 : * Rejects incorrectly sized objects and objects that are to be copied
4719 : * to/from userspace but do not fall entirely within the containing slab
4720 : * cache's usercopy region.
4721 : *
4722 : * Returns NULL if check passes, otherwise const char * to name of cache
4723 : * to indicate an error.
4724 : */
4725 : void __check_heap_object(const void *ptr, unsigned long n,
4726 : const struct slab *slab, bool to_user)
4727 : {
4728 : struct kmem_cache *s;
4729 : unsigned int offset;
4730 : bool is_kfence = is_kfence_address(ptr);
4731 :
4732 : ptr = kasan_reset_tag(ptr);
4733 :
4734 : /* Find object and usable object size. */
4735 : s = slab->slab_cache;
4736 :
4737 : /* Reject impossible pointers. */
4738 : if (ptr < slab_address(slab))
4739 : usercopy_abort("SLUB object not in SLUB page?!", NULL,
4740 : to_user, 0, n);
4741 :
4742 : /* Find offset within object. */
4743 : if (is_kfence)
4744 : offset = ptr - kfence_object_start(ptr);
4745 : else
4746 : offset = (ptr - slab_address(slab)) % s->size;
4747 :
4748 : /* Adjust for redzone and reject if within the redzone. */
4749 : if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) {
4750 : if (offset < s->red_left_pad)
4751 : usercopy_abort("SLUB object in left red zone",
4752 : s->name, to_user, offset, n);
4753 : offset -= s->red_left_pad;
4754 : }
4755 :
4756 : /* Allow address range falling entirely within usercopy region. */
4757 : if (offset >= s->useroffset &&
4758 : offset - s->useroffset <= s->usersize &&
4759 : n <= s->useroffset - offset + s->usersize)
4760 : return;
4761 :
4762 : usercopy_abort("SLUB object", s->name, to_user, offset, n);
4763 : }
4764 : #endif /* CONFIG_HARDENED_USERCOPY */
4765 :
4766 : #define SHRINK_PROMOTE_MAX 32
4767 :
4768 : /*
4769 : * kmem_cache_shrink discards empty slabs and promotes the slabs filled
4770 : * up most to the head of the partial lists. New allocations will then
4771 : * fill those up and thus they can be removed from the partial lists.
4772 : *
4773 : * The slabs with the least items are placed last. This results in them
4774 : * being allocated from last increasing the chance that the last objects
4775 : * are freed in them.
4776 : */
4777 0 : static int __kmem_cache_do_shrink(struct kmem_cache *s)
4778 : {
4779 : int node;
4780 : int i;
4781 : struct kmem_cache_node *n;
4782 : struct slab *slab;
4783 : struct slab *t;
4784 : struct list_head discard;
4785 : struct list_head promote[SHRINK_PROMOTE_MAX];
4786 : unsigned long flags;
4787 0 : int ret = 0;
4788 :
4789 0 : for_each_kmem_cache_node(s, node, n) {
4790 0 : INIT_LIST_HEAD(&discard);
4791 0 : for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
4792 0 : INIT_LIST_HEAD(promote + i);
4793 :
4794 0 : spin_lock_irqsave(&n->list_lock, flags);
4795 :
4796 : /*
4797 : * Build lists of slabs to discard or promote.
4798 : *
4799 : * Note that concurrent frees may occur while we hold the
4800 : * list_lock. slab->inuse here is the upper limit.
4801 : */
4802 0 : list_for_each_entry_safe(slab, t, &n->partial, slab_list) {
4803 0 : int free = slab->objects - slab->inuse;
4804 :
4805 : /* Do not reread slab->inuse */
4806 0 : barrier();
4807 :
4808 : /* We do not keep full slabs on the list */
4809 0 : BUG_ON(free <= 0);
4810 :
4811 0 : if (free == slab->objects) {
4812 0 : list_move(&slab->slab_list, &discard);
4813 0 : n->nr_partial--;
4814 0 : dec_slabs_node(s, node, slab->objects);
4815 0 : } else if (free <= SHRINK_PROMOTE_MAX)
4816 0 : list_move(&slab->slab_list, promote + free - 1);
4817 : }
4818 :
4819 : /*
4820 : * Promote the slabs filled up most to the head of the
4821 : * partial list.
4822 : */
4823 0 : for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
4824 0 : list_splice(promote + i, &n->partial);
4825 :
4826 0 : spin_unlock_irqrestore(&n->list_lock, flags);
4827 :
4828 : /* Release empty slabs */
4829 0 : list_for_each_entry_safe(slab, t, &discard, slab_list)
4830 0 : free_slab(s, slab);
4831 :
4832 0 : if (slabs_node(s, node))
4833 0 : ret = 1;
4834 : }
4835 :
4836 0 : return ret;
4837 : }
4838 :
4839 0 : int __kmem_cache_shrink(struct kmem_cache *s)
4840 : {
4841 0 : flush_all(s);
4842 0 : return __kmem_cache_do_shrink(s);
4843 : }
4844 :
4845 : static int slab_mem_going_offline_callback(void *arg)
4846 : {
4847 : struct kmem_cache *s;
4848 :
4849 : mutex_lock(&slab_mutex);
4850 : list_for_each_entry(s, &slab_caches, list) {
4851 : flush_all_cpus_locked(s);
4852 : __kmem_cache_do_shrink(s);
4853 : }
4854 : mutex_unlock(&slab_mutex);
4855 :
4856 : return 0;
4857 : }
4858 :
4859 : static void slab_mem_offline_callback(void *arg)
4860 : {
4861 : struct memory_notify *marg = arg;
4862 : int offline_node;
4863 :
4864 : offline_node = marg->status_change_nid_normal;
4865 :
4866 : /*
4867 : * If the node still has available memory. we need kmem_cache_node
4868 : * for it yet.
4869 : */
4870 : if (offline_node < 0)
4871 : return;
4872 :
4873 : mutex_lock(&slab_mutex);
4874 : node_clear(offline_node, slab_nodes);
4875 : /*
4876 : * We no longer free kmem_cache_node structures here, as it would be
4877 : * racy with all get_node() users, and infeasible to protect them with
4878 : * slab_mutex.
4879 : */
4880 : mutex_unlock(&slab_mutex);
4881 : }
4882 :
4883 : static int slab_mem_going_online_callback(void *arg)
4884 : {
4885 : struct kmem_cache_node *n;
4886 : struct kmem_cache *s;
4887 : struct memory_notify *marg = arg;
4888 : int nid = marg->status_change_nid_normal;
4889 : int ret = 0;
4890 :
4891 : /*
4892 : * If the node's memory is already available, then kmem_cache_node is
4893 : * already created. Nothing to do.
4894 : */
4895 : if (nid < 0)
4896 : return 0;
4897 :
4898 : /*
4899 : * We are bringing a node online. No memory is available yet. We must
4900 : * allocate a kmem_cache_node structure in order to bring the node
4901 : * online.
4902 : */
4903 : mutex_lock(&slab_mutex);
4904 : list_for_each_entry(s, &slab_caches, list) {
4905 : /*
4906 : * The structure may already exist if the node was previously
4907 : * onlined and offlined.
4908 : */
4909 : if (get_node(s, nid))
4910 : continue;
4911 : /*
4912 : * XXX: kmem_cache_alloc_node will fallback to other nodes
4913 : * since memory is not yet available from the node that
4914 : * is brought up.
4915 : */
4916 : n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
4917 : if (!n) {
4918 : ret = -ENOMEM;
4919 : goto out;
4920 : }
4921 : init_kmem_cache_node(n);
4922 : s->node[nid] = n;
4923 : }
4924 : /*
4925 : * Any cache created after this point will also have kmem_cache_node
4926 : * initialized for the new node.
4927 : */
4928 : node_set(nid, slab_nodes);
4929 : out:
4930 : mutex_unlock(&slab_mutex);
4931 : return ret;
4932 : }
4933 :
4934 : static int slab_memory_callback(struct notifier_block *self,
4935 : unsigned long action, void *arg)
4936 : {
4937 : int ret = 0;
4938 :
4939 : switch (action) {
4940 : case MEM_GOING_ONLINE:
4941 : ret = slab_mem_going_online_callback(arg);
4942 : break;
4943 : case MEM_GOING_OFFLINE:
4944 : ret = slab_mem_going_offline_callback(arg);
4945 : break;
4946 : case MEM_OFFLINE:
4947 : case MEM_CANCEL_ONLINE:
4948 : slab_mem_offline_callback(arg);
4949 : break;
4950 : case MEM_ONLINE:
4951 : case MEM_CANCEL_OFFLINE:
4952 : break;
4953 : }
4954 : if (ret)
4955 : ret = notifier_from_errno(ret);
4956 : else
4957 : ret = NOTIFY_OK;
4958 : return ret;
4959 : }
4960 :
4961 : /********************************************************************
4962 : * Basic setup of slabs
4963 : *******************************************************************/
4964 :
4965 : /*
4966 : * Used for early kmem_cache structures that were allocated using
4967 : * the page allocator. Allocate them properly then fix up the pointers
4968 : * that may be pointing to the wrong kmem_cache structure.
4969 : */
4970 :
4971 2 : static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
4972 : {
4973 : int node;
4974 4 : struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
4975 : struct kmem_cache_node *n;
4976 :
4977 2 : memcpy(s, static_cache, kmem_cache->object_size);
4978 :
4979 : /*
4980 : * This runs very early, and only the boot processor is supposed to be
4981 : * up. Even if it weren't true, IRQs are not up so we couldn't fire
4982 : * IPIs around.
4983 : */
4984 2 : __flush_cpu_slab(s, smp_processor_id());
4985 6 : for_each_kmem_cache_node(s, node, n) {
4986 : struct slab *p;
4987 :
4988 4 : list_for_each_entry(p, &n->partial, slab_list)
4989 2 : p->slab_cache = s;
4990 :
4991 : #ifdef CONFIG_SLUB_DEBUG
4992 2 : list_for_each_entry(p, &n->full, slab_list)
4993 0 : p->slab_cache = s;
4994 : #endif
4995 : }
4996 4 : list_add(&s->list, &slab_caches);
4997 2 : return s;
4998 : }
4999 :
5000 1 : void __init kmem_cache_init(void)
5001 : {
5002 : static __initdata struct kmem_cache boot_kmem_cache,
5003 : boot_kmem_cache_node;
5004 : int node;
5005 :
5006 : if (debug_guardpage_minorder())
5007 : slub_max_order = 0;
5008 :
5009 : /* Print slub debugging pointers without hashing */
5010 1 : if (__slub_debug_enabled())
5011 0 : no_hash_pointers_enable(NULL);
5012 :
5013 1 : kmem_cache_node = &boot_kmem_cache_node;
5014 1 : kmem_cache = &boot_kmem_cache;
5015 :
5016 : /*
5017 : * Initialize the nodemask for which we will allocate per node
5018 : * structures. Here we don't need taking slab_mutex yet.
5019 : */
5020 3 : for_each_node_state(node, N_NORMAL_MEMORY)
5021 : node_set(node, slab_nodes);
5022 :
5023 1 : create_boot_cache(kmem_cache_node, "kmem_cache_node",
5024 : sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
5025 :
5026 1 : hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
5027 :
5028 : /* Able to allocate the per node structures */
5029 1 : slab_state = PARTIAL;
5030 :
5031 1 : create_boot_cache(kmem_cache, "kmem_cache",
5032 : offsetof(struct kmem_cache, node) +
5033 : nr_node_ids * sizeof(struct kmem_cache_node *),
5034 : SLAB_HWCACHE_ALIGN, 0, 0);
5035 :
5036 1 : kmem_cache = bootstrap(&boot_kmem_cache);
5037 1 : kmem_cache_node = bootstrap(&boot_kmem_cache_node);
5038 :
5039 : /* Now we can use the kmem_cache to allocate kmalloc slabs */
5040 1 : setup_kmalloc_cache_index_table();
5041 1 : create_kmalloc_caches(0);
5042 :
5043 : /* Setup random freelists for each cache */
5044 1 : init_freelist_randomization();
5045 :
5046 1 : cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
5047 : slub_cpu_dead);
5048 :
5049 1 : pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n",
5050 : cache_line_size(),
5051 : slub_min_order, slub_max_order, slub_min_objects,
5052 : nr_cpu_ids, nr_node_ids);
5053 1 : }
5054 :
5055 1 : void __init kmem_cache_init_late(void)
5056 : {
5057 : #ifndef CONFIG_SLUB_TINY
5058 1 : flushwq = alloc_workqueue("slub_flushwq", WQ_MEM_RECLAIM, 0);
5059 1 : WARN_ON(!flushwq);
5060 : #endif
5061 1 : }
5062 :
5063 : struct kmem_cache *
5064 57 : __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
5065 : slab_flags_t flags, void (*ctor)(void *))
5066 : {
5067 : struct kmem_cache *s;
5068 :
5069 57 : s = find_mergeable(size, align, flags, name, ctor);
5070 57 : if (s) {
5071 33 : if (sysfs_slab_alias(s, name))
5072 : return NULL;
5073 :
5074 33 : s->refcount++;
5075 :
5076 : /*
5077 : * Adjust the object sizes so that we clear
5078 : * the complete object on kzalloc.
5079 : */
5080 33 : s->object_size = max(s->object_size, size);
5081 33 : s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
5082 : }
5083 :
5084 : return s;
5085 : }
5086 :
5087 52 : int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
5088 : {
5089 : int err;
5090 :
5091 52 : err = kmem_cache_open(s, flags);
5092 52 : if (err)
5093 : return err;
5094 :
5095 : /* Mutex is not taken during early boot */
5096 52 : if (slab_state <= UP)
5097 : return 0;
5098 :
5099 0 : err = sysfs_slab_add(s);
5100 0 : if (err) {
5101 0 : __kmem_cache_release(s);
5102 0 : return err;
5103 : }
5104 :
5105 : if (s->flags & SLAB_STORE_USER)
5106 : debugfs_slab_add(s);
5107 :
5108 : return 0;
5109 : }
5110 :
5111 : #ifdef SLAB_SUPPORTS_SYSFS
5112 0 : static int count_inuse(struct slab *slab)
5113 : {
5114 0 : return slab->inuse;
5115 : }
5116 :
5117 0 : static int count_total(struct slab *slab)
5118 : {
5119 0 : return slab->objects;
5120 : }
5121 : #endif
5122 :
5123 : #ifdef CONFIG_SLUB_DEBUG
5124 0 : static void validate_slab(struct kmem_cache *s, struct slab *slab,
5125 : unsigned long *obj_map)
5126 : {
5127 : void *p;
5128 0 : void *addr = slab_address(slab);
5129 :
5130 0 : if (!check_slab(s, slab) || !on_freelist(s, slab, NULL))
5131 : return;
5132 :
5133 : /* Now we know that a valid freelist exists */
5134 0 : __fill_map(obj_map, s, slab);
5135 0 : for_each_object(p, s, addr, slab->objects) {
5136 0 : u8 val = test_bit(__obj_to_index(s, addr, p), obj_map) ?
5137 : SLUB_RED_INACTIVE : SLUB_RED_ACTIVE;
5138 :
5139 0 : if (!check_object(s, slab, p, val))
5140 : break;
5141 : }
5142 : }
5143 :
5144 0 : static int validate_slab_node(struct kmem_cache *s,
5145 : struct kmem_cache_node *n, unsigned long *obj_map)
5146 : {
5147 0 : unsigned long count = 0;
5148 : struct slab *slab;
5149 : unsigned long flags;
5150 :
5151 0 : spin_lock_irqsave(&n->list_lock, flags);
5152 :
5153 0 : list_for_each_entry(slab, &n->partial, slab_list) {
5154 0 : validate_slab(s, slab, obj_map);
5155 0 : count++;
5156 : }
5157 0 : if (count != n->nr_partial) {
5158 0 : pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
5159 : s->name, count, n->nr_partial);
5160 0 : slab_add_kunit_errors();
5161 : }
5162 :
5163 0 : if (!(s->flags & SLAB_STORE_USER))
5164 : goto out;
5165 :
5166 0 : list_for_each_entry(slab, &n->full, slab_list) {
5167 0 : validate_slab(s, slab, obj_map);
5168 0 : count++;
5169 : }
5170 0 : if (count != atomic_long_read(&n->nr_slabs)) {
5171 0 : pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
5172 : s->name, count, atomic_long_read(&n->nr_slabs));
5173 0 : slab_add_kunit_errors();
5174 : }
5175 :
5176 : out:
5177 0 : spin_unlock_irqrestore(&n->list_lock, flags);
5178 0 : return count;
5179 : }
5180 :
5181 0 : long validate_slab_cache(struct kmem_cache *s)
5182 : {
5183 : int node;
5184 0 : unsigned long count = 0;
5185 : struct kmem_cache_node *n;
5186 : unsigned long *obj_map;
5187 :
5188 0 : obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL);
5189 0 : if (!obj_map)
5190 : return -ENOMEM;
5191 :
5192 0 : flush_all(s);
5193 0 : for_each_kmem_cache_node(s, node, n)
5194 0 : count += validate_slab_node(s, n, obj_map);
5195 :
5196 0 : bitmap_free(obj_map);
5197 :
5198 0 : return count;
5199 : }
5200 : EXPORT_SYMBOL(validate_slab_cache);
5201 :
5202 : #ifdef CONFIG_DEBUG_FS
5203 : /*
5204 : * Generate lists of code addresses where slabcache objects are allocated
5205 : * and freed.
5206 : */
5207 :
5208 : struct location {
5209 : depot_stack_handle_t handle;
5210 : unsigned long count;
5211 : unsigned long addr;
5212 : unsigned long waste;
5213 : long long sum_time;
5214 : long min_time;
5215 : long max_time;
5216 : long min_pid;
5217 : long max_pid;
5218 : DECLARE_BITMAP(cpus, NR_CPUS);
5219 : nodemask_t nodes;
5220 : };
5221 :
5222 : struct loc_track {
5223 : unsigned long max;
5224 : unsigned long count;
5225 : struct location *loc;
5226 : loff_t idx;
5227 : };
5228 :
5229 : static struct dentry *slab_debugfs_root;
5230 :
5231 : static void free_loc_track(struct loc_track *t)
5232 : {
5233 : if (t->max)
5234 : free_pages((unsigned long)t->loc,
5235 : get_order(sizeof(struct location) * t->max));
5236 : }
5237 :
5238 : static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
5239 : {
5240 : struct location *l;
5241 : int order;
5242 :
5243 : order = get_order(sizeof(struct location) * max);
5244 :
5245 : l = (void *)__get_free_pages(flags, order);
5246 : if (!l)
5247 : return 0;
5248 :
5249 : if (t->count) {
5250 : memcpy(l, t->loc, sizeof(struct location) * t->count);
5251 : free_loc_track(t);
5252 : }
5253 : t->max = max;
5254 : t->loc = l;
5255 : return 1;
5256 : }
5257 :
5258 : static int add_location(struct loc_track *t, struct kmem_cache *s,
5259 : const struct track *track,
5260 : unsigned int orig_size)
5261 : {
5262 : long start, end, pos;
5263 : struct location *l;
5264 : unsigned long caddr, chandle, cwaste;
5265 : unsigned long age = jiffies - track->when;
5266 : depot_stack_handle_t handle = 0;
5267 : unsigned int waste = s->object_size - orig_size;
5268 :
5269 : #ifdef CONFIG_STACKDEPOT
5270 : handle = READ_ONCE(track->handle);
5271 : #endif
5272 : start = -1;
5273 : end = t->count;
5274 :
5275 : for ( ; ; ) {
5276 : pos = start + (end - start + 1) / 2;
5277 :
5278 : /*
5279 : * There is nothing at "end". If we end up there
5280 : * we need to add something to before end.
5281 : */
5282 : if (pos == end)
5283 : break;
5284 :
5285 : l = &t->loc[pos];
5286 : caddr = l->addr;
5287 : chandle = l->handle;
5288 : cwaste = l->waste;
5289 : if ((track->addr == caddr) && (handle == chandle) &&
5290 : (waste == cwaste)) {
5291 :
5292 : l->count++;
5293 : if (track->when) {
5294 : l->sum_time += age;
5295 : if (age < l->min_time)
5296 : l->min_time = age;
5297 : if (age > l->max_time)
5298 : l->max_time = age;
5299 :
5300 : if (track->pid < l->min_pid)
5301 : l->min_pid = track->pid;
5302 : if (track->pid > l->max_pid)
5303 : l->max_pid = track->pid;
5304 :
5305 : cpumask_set_cpu(track->cpu,
5306 : to_cpumask(l->cpus));
5307 : }
5308 : node_set(page_to_nid(virt_to_page(track)), l->nodes);
5309 : return 1;
5310 : }
5311 :
5312 : if (track->addr < caddr)
5313 : end = pos;
5314 : else if (track->addr == caddr && handle < chandle)
5315 : end = pos;
5316 : else if (track->addr == caddr && handle == chandle &&
5317 : waste < cwaste)
5318 : end = pos;
5319 : else
5320 : start = pos;
5321 : }
5322 :
5323 : /*
5324 : * Not found. Insert new tracking element.
5325 : */
5326 : if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
5327 : return 0;
5328 :
5329 : l = t->loc + pos;
5330 : if (pos < t->count)
5331 : memmove(l + 1, l,
5332 : (t->count - pos) * sizeof(struct location));
5333 : t->count++;
5334 : l->count = 1;
5335 : l->addr = track->addr;
5336 : l->sum_time = age;
5337 : l->min_time = age;
5338 : l->max_time = age;
5339 : l->min_pid = track->pid;
5340 : l->max_pid = track->pid;
5341 : l->handle = handle;
5342 : l->waste = waste;
5343 : cpumask_clear(to_cpumask(l->cpus));
5344 : cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
5345 : nodes_clear(l->nodes);
5346 : node_set(page_to_nid(virt_to_page(track)), l->nodes);
5347 : return 1;
5348 : }
5349 :
5350 : static void process_slab(struct loc_track *t, struct kmem_cache *s,
5351 : struct slab *slab, enum track_item alloc,
5352 : unsigned long *obj_map)
5353 : {
5354 : void *addr = slab_address(slab);
5355 : bool is_alloc = (alloc == TRACK_ALLOC);
5356 : void *p;
5357 :
5358 : __fill_map(obj_map, s, slab);
5359 :
5360 : for_each_object(p, s, addr, slab->objects)
5361 : if (!test_bit(__obj_to_index(s, addr, p), obj_map))
5362 : add_location(t, s, get_track(s, p, alloc),
5363 : is_alloc ? get_orig_size(s, p) :
5364 : s->object_size);
5365 : }
5366 : #endif /* CONFIG_DEBUG_FS */
5367 : #endif /* CONFIG_SLUB_DEBUG */
5368 :
5369 : #ifdef SLAB_SUPPORTS_SYSFS
5370 : enum slab_stat_type {
5371 : SL_ALL, /* All slabs */
5372 : SL_PARTIAL, /* Only partially allocated slabs */
5373 : SL_CPU, /* Only slabs used for cpu caches */
5374 : SL_OBJECTS, /* Determine allocated objects not slabs */
5375 : SL_TOTAL /* Determine object capacity not slabs */
5376 : };
5377 :
5378 : #define SO_ALL (1 << SL_ALL)
5379 : #define SO_PARTIAL (1 << SL_PARTIAL)
5380 : #define SO_CPU (1 << SL_CPU)
5381 : #define SO_OBJECTS (1 << SL_OBJECTS)
5382 : #define SO_TOTAL (1 << SL_TOTAL)
5383 :
5384 0 : static ssize_t show_slab_objects(struct kmem_cache *s,
5385 : char *buf, unsigned long flags)
5386 : {
5387 0 : unsigned long total = 0;
5388 : int node;
5389 : int x;
5390 : unsigned long *nodes;
5391 0 : int len = 0;
5392 :
5393 0 : nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
5394 0 : if (!nodes)
5395 : return -ENOMEM;
5396 :
5397 0 : if (flags & SO_CPU) {
5398 : int cpu;
5399 :
5400 0 : for_each_possible_cpu(cpu) {
5401 0 : struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
5402 : cpu);
5403 : int node;
5404 : struct slab *slab;
5405 :
5406 0 : slab = READ_ONCE(c->slab);
5407 0 : if (!slab)
5408 0 : continue;
5409 :
5410 0 : node = slab_nid(slab);
5411 0 : if (flags & SO_TOTAL)
5412 0 : x = slab->objects;
5413 0 : else if (flags & SO_OBJECTS)
5414 0 : x = slab->inuse;
5415 : else
5416 : x = 1;
5417 :
5418 0 : total += x;
5419 0 : nodes[node] += x;
5420 :
5421 : #ifdef CONFIG_SLUB_CPU_PARTIAL
5422 : slab = slub_percpu_partial_read_once(c);
5423 : if (slab) {
5424 : node = slab_nid(slab);
5425 : if (flags & SO_TOTAL)
5426 : WARN_ON_ONCE(1);
5427 : else if (flags & SO_OBJECTS)
5428 : WARN_ON_ONCE(1);
5429 : else
5430 : x = slab->slabs;
5431 : total += x;
5432 : nodes[node] += x;
5433 : }
5434 : #endif
5435 : }
5436 : }
5437 :
5438 : /*
5439 : * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex"
5440 : * already held which will conflict with an existing lock order:
5441 : *
5442 : * mem_hotplug_lock->slab_mutex->kernfs_mutex
5443 : *
5444 : * We don't really need mem_hotplug_lock (to hold off
5445 : * slab_mem_going_offline_callback) here because slab's memory hot
5446 : * unplug code doesn't destroy the kmem_cache->node[] data.
5447 : */
5448 :
5449 : #ifdef CONFIG_SLUB_DEBUG
5450 0 : if (flags & SO_ALL) {
5451 : struct kmem_cache_node *n;
5452 :
5453 0 : for_each_kmem_cache_node(s, node, n) {
5454 :
5455 0 : if (flags & SO_TOTAL)
5456 0 : x = atomic_long_read(&n->total_objects);
5457 0 : else if (flags & SO_OBJECTS)
5458 0 : x = atomic_long_read(&n->total_objects) -
5459 0 : count_partial(n, count_free);
5460 : else
5461 0 : x = atomic_long_read(&n->nr_slabs);
5462 0 : total += x;
5463 0 : nodes[node] += x;
5464 : }
5465 :
5466 : } else
5467 : #endif
5468 0 : if (flags & SO_PARTIAL) {
5469 : struct kmem_cache_node *n;
5470 :
5471 0 : for_each_kmem_cache_node(s, node, n) {
5472 0 : if (flags & SO_TOTAL)
5473 0 : x = count_partial(n, count_total);
5474 0 : else if (flags & SO_OBJECTS)
5475 0 : x = count_partial(n, count_inuse);
5476 : else
5477 0 : x = n->nr_partial;
5478 0 : total += x;
5479 0 : nodes[node] += x;
5480 : }
5481 : }
5482 :
5483 0 : len += sysfs_emit_at(buf, len, "%lu", total);
5484 : #ifdef CONFIG_NUMA
5485 : for (node = 0; node < nr_node_ids; node++) {
5486 : if (nodes[node])
5487 : len += sysfs_emit_at(buf, len, " N%d=%lu",
5488 : node, nodes[node]);
5489 : }
5490 : #endif
5491 0 : len += sysfs_emit_at(buf, len, "\n");
5492 0 : kfree(nodes);
5493 :
5494 0 : return len;
5495 : }
5496 :
5497 : #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
5498 : #define to_slab(n) container_of(n, struct kmem_cache, kobj)
5499 :
5500 : struct slab_attribute {
5501 : struct attribute attr;
5502 : ssize_t (*show)(struct kmem_cache *s, char *buf);
5503 : ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
5504 : };
5505 :
5506 : #define SLAB_ATTR_RO(_name) \
5507 : static struct slab_attribute _name##_attr = __ATTR_RO_MODE(_name, 0400)
5508 :
5509 : #define SLAB_ATTR(_name) \
5510 : static struct slab_attribute _name##_attr = __ATTR_RW_MODE(_name, 0600)
5511 :
5512 0 : static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
5513 : {
5514 0 : return sysfs_emit(buf, "%u\n", s->size);
5515 : }
5516 : SLAB_ATTR_RO(slab_size);
5517 :
5518 0 : static ssize_t align_show(struct kmem_cache *s, char *buf)
5519 : {
5520 0 : return sysfs_emit(buf, "%u\n", s->align);
5521 : }
5522 : SLAB_ATTR_RO(align);
5523 :
5524 0 : static ssize_t object_size_show(struct kmem_cache *s, char *buf)
5525 : {
5526 0 : return sysfs_emit(buf, "%u\n", s->object_size);
5527 : }
5528 : SLAB_ATTR_RO(object_size);
5529 :
5530 0 : static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
5531 : {
5532 0 : return sysfs_emit(buf, "%u\n", oo_objects(s->oo));
5533 : }
5534 : SLAB_ATTR_RO(objs_per_slab);
5535 :
5536 0 : static ssize_t order_show(struct kmem_cache *s, char *buf)
5537 : {
5538 0 : return sysfs_emit(buf, "%u\n", oo_order(s->oo));
5539 : }
5540 : SLAB_ATTR_RO(order);
5541 :
5542 0 : static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
5543 : {
5544 0 : return sysfs_emit(buf, "%lu\n", s->min_partial);
5545 : }
5546 :
5547 0 : static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
5548 : size_t length)
5549 : {
5550 : unsigned long min;
5551 : int err;
5552 :
5553 0 : err = kstrtoul(buf, 10, &min);
5554 0 : if (err)
5555 0 : return err;
5556 :
5557 0 : s->min_partial = min;
5558 0 : return length;
5559 : }
5560 : SLAB_ATTR(min_partial);
5561 :
5562 0 : static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
5563 : {
5564 0 : unsigned int nr_partial = 0;
5565 : #ifdef CONFIG_SLUB_CPU_PARTIAL
5566 : nr_partial = s->cpu_partial;
5567 : #endif
5568 :
5569 0 : return sysfs_emit(buf, "%u\n", nr_partial);
5570 : }
5571 :
5572 0 : static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
5573 : size_t length)
5574 : {
5575 : unsigned int objects;
5576 : int err;
5577 :
5578 0 : err = kstrtouint(buf, 10, &objects);
5579 0 : if (err)
5580 0 : return err;
5581 0 : if (objects && !kmem_cache_has_cpu_partial(s))
5582 : return -EINVAL;
5583 :
5584 0 : slub_set_cpu_partial(s, objects);
5585 0 : flush_all(s);
5586 0 : return length;
5587 : }
5588 : SLAB_ATTR(cpu_partial);
5589 :
5590 0 : static ssize_t ctor_show(struct kmem_cache *s, char *buf)
5591 : {
5592 0 : if (!s->ctor)
5593 : return 0;
5594 0 : return sysfs_emit(buf, "%pS\n", s->ctor);
5595 : }
5596 : SLAB_ATTR_RO(ctor);
5597 :
5598 0 : static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5599 : {
5600 0 : return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
5601 : }
5602 : SLAB_ATTR_RO(aliases);
5603 :
5604 0 : static ssize_t partial_show(struct kmem_cache *s, char *buf)
5605 : {
5606 0 : return show_slab_objects(s, buf, SO_PARTIAL);
5607 : }
5608 : SLAB_ATTR_RO(partial);
5609 :
5610 0 : static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5611 : {
5612 0 : return show_slab_objects(s, buf, SO_CPU);
5613 : }
5614 : SLAB_ATTR_RO(cpu_slabs);
5615 :
5616 0 : static ssize_t objects_show(struct kmem_cache *s, char *buf)
5617 : {
5618 0 : return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
5619 : }
5620 : SLAB_ATTR_RO(objects);
5621 :
5622 0 : static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5623 : {
5624 0 : return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5625 : }
5626 : SLAB_ATTR_RO(objects_partial);
5627 :
5628 0 : static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5629 : {
5630 0 : int objects = 0;
5631 0 : int slabs = 0;
5632 : int cpu __maybe_unused;
5633 0 : int len = 0;
5634 :
5635 : #ifdef CONFIG_SLUB_CPU_PARTIAL
5636 : for_each_online_cpu(cpu) {
5637 : struct slab *slab;
5638 :
5639 : slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5640 :
5641 : if (slab)
5642 : slabs += slab->slabs;
5643 : }
5644 : #endif
5645 :
5646 : /* Approximate half-full slabs, see slub_set_cpu_partial() */
5647 0 : objects = (slabs * oo_objects(s->oo)) / 2;
5648 0 : len += sysfs_emit_at(buf, len, "%d(%d)", objects, slabs);
5649 :
5650 : #if defined(CONFIG_SLUB_CPU_PARTIAL) && defined(CONFIG_SMP)
5651 : for_each_online_cpu(cpu) {
5652 : struct slab *slab;
5653 :
5654 : slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5655 : if (slab) {
5656 : slabs = READ_ONCE(slab->slabs);
5657 : objects = (slabs * oo_objects(s->oo)) / 2;
5658 : len += sysfs_emit_at(buf, len, " C%d=%d(%d)",
5659 : cpu, objects, slabs);
5660 : }
5661 : }
5662 : #endif
5663 0 : len += sysfs_emit_at(buf, len, "\n");
5664 :
5665 0 : return len;
5666 : }
5667 : SLAB_ATTR_RO(slabs_cpu_partial);
5668 :
5669 0 : static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5670 : {
5671 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5672 : }
5673 : SLAB_ATTR_RO(reclaim_account);
5674 :
5675 0 : static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5676 : {
5677 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
5678 : }
5679 : SLAB_ATTR_RO(hwcache_align);
5680 :
5681 : #ifdef CONFIG_ZONE_DMA
5682 : static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5683 : {
5684 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5685 : }
5686 : SLAB_ATTR_RO(cache_dma);
5687 : #endif
5688 :
5689 : #ifdef CONFIG_HARDENED_USERCOPY
5690 : static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5691 : {
5692 : return sysfs_emit(buf, "%u\n", s->usersize);
5693 : }
5694 : SLAB_ATTR_RO(usersize);
5695 : #endif
5696 :
5697 0 : static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5698 : {
5699 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
5700 : }
5701 : SLAB_ATTR_RO(destroy_by_rcu);
5702 :
5703 : #ifdef CONFIG_SLUB_DEBUG
5704 0 : static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5705 : {
5706 0 : return show_slab_objects(s, buf, SO_ALL);
5707 : }
5708 : SLAB_ATTR_RO(slabs);
5709 :
5710 0 : static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5711 : {
5712 0 : return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5713 : }
5714 : SLAB_ATTR_RO(total_objects);
5715 :
5716 0 : static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5717 : {
5718 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
5719 : }
5720 : SLAB_ATTR_RO(sanity_checks);
5721 :
5722 0 : static ssize_t trace_show(struct kmem_cache *s, char *buf)
5723 : {
5724 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5725 : }
5726 : SLAB_ATTR_RO(trace);
5727 :
5728 0 : static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5729 : {
5730 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5731 : }
5732 :
5733 : SLAB_ATTR_RO(red_zone);
5734 :
5735 0 : static ssize_t poison_show(struct kmem_cache *s, char *buf)
5736 : {
5737 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON));
5738 : }
5739 :
5740 : SLAB_ATTR_RO(poison);
5741 :
5742 0 : static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5743 : {
5744 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5745 : }
5746 :
5747 : SLAB_ATTR_RO(store_user);
5748 :
5749 0 : static ssize_t validate_show(struct kmem_cache *s, char *buf)
5750 : {
5751 0 : return 0;
5752 : }
5753 :
5754 0 : static ssize_t validate_store(struct kmem_cache *s,
5755 : const char *buf, size_t length)
5756 : {
5757 0 : int ret = -EINVAL;
5758 :
5759 0 : if (buf[0] == '1' && kmem_cache_debug(s)) {
5760 0 : ret = validate_slab_cache(s);
5761 0 : if (ret >= 0)
5762 0 : ret = length;
5763 : }
5764 0 : return ret;
5765 : }
5766 : SLAB_ATTR(validate);
5767 :
5768 : #endif /* CONFIG_SLUB_DEBUG */
5769 :
5770 : #ifdef CONFIG_FAILSLAB
5771 : static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5772 : {
5773 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5774 : }
5775 :
5776 : static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5777 : size_t length)
5778 : {
5779 : if (s->refcount > 1)
5780 : return -EINVAL;
5781 :
5782 : if (buf[0] == '1')
5783 : WRITE_ONCE(s->flags, s->flags | SLAB_FAILSLAB);
5784 : else
5785 : WRITE_ONCE(s->flags, s->flags & ~SLAB_FAILSLAB);
5786 :
5787 : return length;
5788 : }
5789 : SLAB_ATTR(failslab);
5790 : #endif
5791 :
5792 0 : static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5793 : {
5794 0 : return 0;
5795 : }
5796 :
5797 0 : static ssize_t shrink_store(struct kmem_cache *s,
5798 : const char *buf, size_t length)
5799 : {
5800 0 : if (buf[0] == '1')
5801 0 : kmem_cache_shrink(s);
5802 : else
5803 : return -EINVAL;
5804 0 : return length;
5805 : }
5806 : SLAB_ATTR(shrink);
5807 :
5808 : #ifdef CONFIG_NUMA
5809 : static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
5810 : {
5811 : return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10);
5812 : }
5813 :
5814 : static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
5815 : const char *buf, size_t length)
5816 : {
5817 : unsigned int ratio;
5818 : int err;
5819 :
5820 : err = kstrtouint(buf, 10, &ratio);
5821 : if (err)
5822 : return err;
5823 : if (ratio > 100)
5824 : return -ERANGE;
5825 :
5826 : s->remote_node_defrag_ratio = ratio * 10;
5827 :
5828 : return length;
5829 : }
5830 : SLAB_ATTR(remote_node_defrag_ratio);
5831 : #endif
5832 :
5833 : #ifdef CONFIG_SLUB_STATS
5834 : static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5835 : {
5836 : unsigned long sum = 0;
5837 : int cpu;
5838 : int len = 0;
5839 : int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
5840 :
5841 : if (!data)
5842 : return -ENOMEM;
5843 :
5844 : for_each_online_cpu(cpu) {
5845 : unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
5846 :
5847 : data[cpu] = x;
5848 : sum += x;
5849 : }
5850 :
5851 : len += sysfs_emit_at(buf, len, "%lu", sum);
5852 :
5853 : #ifdef CONFIG_SMP
5854 : for_each_online_cpu(cpu) {
5855 : if (data[cpu])
5856 : len += sysfs_emit_at(buf, len, " C%d=%u",
5857 : cpu, data[cpu]);
5858 : }
5859 : #endif
5860 : kfree(data);
5861 : len += sysfs_emit_at(buf, len, "\n");
5862 :
5863 : return len;
5864 : }
5865 :
5866 : static void clear_stat(struct kmem_cache *s, enum stat_item si)
5867 : {
5868 : int cpu;
5869 :
5870 : for_each_online_cpu(cpu)
5871 : per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
5872 : }
5873 :
5874 : #define STAT_ATTR(si, text) \
5875 : static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5876 : { \
5877 : return show_stat(s, buf, si); \
5878 : } \
5879 : static ssize_t text##_store(struct kmem_cache *s, \
5880 : const char *buf, size_t length) \
5881 : { \
5882 : if (buf[0] != '0') \
5883 : return -EINVAL; \
5884 : clear_stat(s, si); \
5885 : return length; \
5886 : } \
5887 : SLAB_ATTR(text); \
5888 :
5889 : STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5890 : STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5891 : STAT_ATTR(FREE_FASTPATH, free_fastpath);
5892 : STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5893 : STAT_ATTR(FREE_FROZEN, free_frozen);
5894 : STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5895 : STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5896 : STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5897 : STAT_ATTR(ALLOC_SLAB, alloc_slab);
5898 : STAT_ATTR(ALLOC_REFILL, alloc_refill);
5899 : STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
5900 : STAT_ATTR(FREE_SLAB, free_slab);
5901 : STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5902 : STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5903 : STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5904 : STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5905 : STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5906 : STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
5907 : STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
5908 : STAT_ATTR(ORDER_FALLBACK, order_fallback);
5909 : STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5910 : STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
5911 : STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5912 : STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
5913 : STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5914 : STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
5915 : #endif /* CONFIG_SLUB_STATS */
5916 :
5917 : #ifdef CONFIG_KFENCE
5918 : static ssize_t skip_kfence_show(struct kmem_cache *s, char *buf)
5919 : {
5920 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_SKIP_KFENCE));
5921 : }
5922 :
5923 : static ssize_t skip_kfence_store(struct kmem_cache *s,
5924 : const char *buf, size_t length)
5925 : {
5926 : int ret = length;
5927 :
5928 : if (buf[0] == '0')
5929 : s->flags &= ~SLAB_SKIP_KFENCE;
5930 : else if (buf[0] == '1')
5931 : s->flags |= SLAB_SKIP_KFENCE;
5932 : else
5933 : ret = -EINVAL;
5934 :
5935 : return ret;
5936 : }
5937 : SLAB_ATTR(skip_kfence);
5938 : #endif
5939 :
5940 : static struct attribute *slab_attrs[] = {
5941 : &slab_size_attr.attr,
5942 : &object_size_attr.attr,
5943 : &objs_per_slab_attr.attr,
5944 : &order_attr.attr,
5945 : &min_partial_attr.attr,
5946 : &cpu_partial_attr.attr,
5947 : &objects_attr.attr,
5948 : &objects_partial_attr.attr,
5949 : &partial_attr.attr,
5950 : &cpu_slabs_attr.attr,
5951 : &ctor_attr.attr,
5952 : &aliases_attr.attr,
5953 : &align_attr.attr,
5954 : &hwcache_align_attr.attr,
5955 : &reclaim_account_attr.attr,
5956 : &destroy_by_rcu_attr.attr,
5957 : &shrink_attr.attr,
5958 : &slabs_cpu_partial_attr.attr,
5959 : #ifdef CONFIG_SLUB_DEBUG
5960 : &total_objects_attr.attr,
5961 : &slabs_attr.attr,
5962 : &sanity_checks_attr.attr,
5963 : &trace_attr.attr,
5964 : &red_zone_attr.attr,
5965 : &poison_attr.attr,
5966 : &store_user_attr.attr,
5967 : &validate_attr.attr,
5968 : #endif
5969 : #ifdef CONFIG_ZONE_DMA
5970 : &cache_dma_attr.attr,
5971 : #endif
5972 : #ifdef CONFIG_NUMA
5973 : &remote_node_defrag_ratio_attr.attr,
5974 : #endif
5975 : #ifdef CONFIG_SLUB_STATS
5976 : &alloc_fastpath_attr.attr,
5977 : &alloc_slowpath_attr.attr,
5978 : &free_fastpath_attr.attr,
5979 : &free_slowpath_attr.attr,
5980 : &free_frozen_attr.attr,
5981 : &free_add_partial_attr.attr,
5982 : &free_remove_partial_attr.attr,
5983 : &alloc_from_partial_attr.attr,
5984 : &alloc_slab_attr.attr,
5985 : &alloc_refill_attr.attr,
5986 : &alloc_node_mismatch_attr.attr,
5987 : &free_slab_attr.attr,
5988 : &cpuslab_flush_attr.attr,
5989 : &deactivate_full_attr.attr,
5990 : &deactivate_empty_attr.attr,
5991 : &deactivate_to_head_attr.attr,
5992 : &deactivate_to_tail_attr.attr,
5993 : &deactivate_remote_frees_attr.attr,
5994 : &deactivate_bypass_attr.attr,
5995 : &order_fallback_attr.attr,
5996 : &cmpxchg_double_fail_attr.attr,
5997 : &cmpxchg_double_cpu_fail_attr.attr,
5998 : &cpu_partial_alloc_attr.attr,
5999 : &cpu_partial_free_attr.attr,
6000 : &cpu_partial_node_attr.attr,
6001 : &cpu_partial_drain_attr.attr,
6002 : #endif
6003 : #ifdef CONFIG_FAILSLAB
6004 : &failslab_attr.attr,
6005 : #endif
6006 : #ifdef CONFIG_HARDENED_USERCOPY
6007 : &usersize_attr.attr,
6008 : #endif
6009 : #ifdef CONFIG_KFENCE
6010 : &skip_kfence_attr.attr,
6011 : #endif
6012 :
6013 : NULL
6014 : };
6015 :
6016 : static const struct attribute_group slab_attr_group = {
6017 : .attrs = slab_attrs,
6018 : };
6019 :
6020 0 : static ssize_t slab_attr_show(struct kobject *kobj,
6021 : struct attribute *attr,
6022 : char *buf)
6023 : {
6024 : struct slab_attribute *attribute;
6025 : struct kmem_cache *s;
6026 :
6027 0 : attribute = to_slab_attr(attr);
6028 0 : s = to_slab(kobj);
6029 :
6030 0 : if (!attribute->show)
6031 : return -EIO;
6032 :
6033 0 : return attribute->show(s, buf);
6034 : }
6035 :
6036 0 : static ssize_t slab_attr_store(struct kobject *kobj,
6037 : struct attribute *attr,
6038 : const char *buf, size_t len)
6039 : {
6040 : struct slab_attribute *attribute;
6041 : struct kmem_cache *s;
6042 :
6043 0 : attribute = to_slab_attr(attr);
6044 0 : s = to_slab(kobj);
6045 :
6046 0 : if (!attribute->store)
6047 : return -EIO;
6048 :
6049 0 : return attribute->store(s, buf, len);
6050 : }
6051 :
6052 0 : static void kmem_cache_release(struct kobject *k)
6053 : {
6054 0 : slab_kmem_cache_release(to_slab(k));
6055 0 : }
6056 :
6057 : static const struct sysfs_ops slab_sysfs_ops = {
6058 : .show = slab_attr_show,
6059 : .store = slab_attr_store,
6060 : };
6061 :
6062 : static struct kobj_type slab_ktype = {
6063 : .sysfs_ops = &slab_sysfs_ops,
6064 : .release = kmem_cache_release,
6065 : };
6066 :
6067 : static struct kset *slab_kset;
6068 :
6069 : static inline struct kset *cache_kset(struct kmem_cache *s)
6070 : {
6071 52 : return slab_kset;
6072 : }
6073 :
6074 : #define ID_STR_LENGTH 32
6075 :
6076 : /* Create a unique string id for a slab cache:
6077 : *
6078 : * Format :[flags-]size
6079 : */
6080 41 : static char *create_unique_id(struct kmem_cache *s)
6081 : {
6082 41 : char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
6083 41 : char *p = name;
6084 :
6085 41 : if (!name)
6086 : return ERR_PTR(-ENOMEM);
6087 :
6088 41 : *p++ = ':';
6089 : /*
6090 : * First flags affecting slabcache operations. We will only
6091 : * get here for aliasable slabs so we do not need to support
6092 : * too many flags. The flags here must cover all flags that
6093 : * are matched during merging to guarantee that the id is
6094 : * unique.
6095 : */
6096 41 : if (s->flags & SLAB_CACHE_DMA)
6097 0 : *p++ = 'd';
6098 41 : if (s->flags & SLAB_CACHE_DMA32)
6099 0 : *p++ = 'D';
6100 41 : if (s->flags & SLAB_RECLAIM_ACCOUNT)
6101 14 : *p++ = 'a';
6102 41 : if (s->flags & SLAB_CONSISTENCY_CHECKS)
6103 0 : *p++ = 'F';
6104 : if (s->flags & SLAB_ACCOUNT)
6105 : *p++ = 'A';
6106 41 : if (p != name + 1)
6107 14 : *p++ = '-';
6108 41 : p += snprintf(p, ID_STR_LENGTH - (p - name), "%07u", s->size);
6109 :
6110 41 : if (WARN_ON(p > name + ID_STR_LENGTH - 1)) {
6111 0 : kfree(name);
6112 0 : return ERR_PTR(-EINVAL);
6113 : }
6114 : kmsan_unpoison_memory(name, p - name);
6115 : return name;
6116 : }
6117 :
6118 52 : static int sysfs_slab_add(struct kmem_cache *s)
6119 : {
6120 : int err;
6121 : const char *name;
6122 104 : struct kset *kset = cache_kset(s);
6123 52 : int unmergeable = slab_unmergeable(s);
6124 :
6125 52 : if (!unmergeable && disable_higher_order_debug &&
6126 0 : (slub_debug & DEBUG_METADATA_FLAGS))
6127 0 : unmergeable = 1;
6128 :
6129 52 : if (unmergeable) {
6130 : /*
6131 : * Slabcache can never be merged so we can use the name proper.
6132 : * This is typically the case for debug situations. In that
6133 : * case we can catch duplicate names easily.
6134 : */
6135 11 : sysfs_remove_link(&slab_kset->kobj, s->name);
6136 11 : name = s->name;
6137 : } else {
6138 : /*
6139 : * Create a unique name for the slab as a target
6140 : * for the symlinks.
6141 : */
6142 41 : name = create_unique_id(s);
6143 41 : if (IS_ERR(name))
6144 0 : return PTR_ERR(name);
6145 : }
6146 :
6147 52 : s->kobj.kset = kset;
6148 52 : err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
6149 52 : if (err)
6150 : goto out;
6151 :
6152 52 : err = sysfs_create_group(&s->kobj, &slab_attr_group);
6153 52 : if (err)
6154 : goto out_del_kobj;
6155 :
6156 52 : if (!unmergeable) {
6157 : /* Setup first alias */
6158 41 : sysfs_slab_alias(s, s->name);
6159 : }
6160 : out:
6161 52 : if (!unmergeable)
6162 41 : kfree(name);
6163 : return err;
6164 : out_del_kobj:
6165 0 : kobject_del(&s->kobj);
6166 0 : goto out;
6167 : }
6168 :
6169 0 : void sysfs_slab_unlink(struct kmem_cache *s)
6170 : {
6171 0 : if (slab_state >= FULL)
6172 0 : kobject_del(&s->kobj);
6173 0 : }
6174 :
6175 0 : void sysfs_slab_release(struct kmem_cache *s)
6176 : {
6177 0 : if (slab_state >= FULL)
6178 0 : kobject_put(&s->kobj);
6179 0 : }
6180 :
6181 : /*
6182 : * Need to buffer aliases during bootup until sysfs becomes
6183 : * available lest we lose that information.
6184 : */
6185 : struct saved_alias {
6186 : struct kmem_cache *s;
6187 : const char *name;
6188 : struct saved_alias *next;
6189 : };
6190 :
6191 : static struct saved_alias *alias_list;
6192 :
6193 107 : static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
6194 : {
6195 : struct saved_alias *al;
6196 :
6197 107 : if (slab_state == FULL) {
6198 : /*
6199 : * If we have a leftover link then remove it.
6200 : */
6201 74 : sysfs_remove_link(&slab_kset->kobj, name);
6202 74 : return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
6203 : }
6204 :
6205 33 : al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
6206 33 : if (!al)
6207 : return -ENOMEM;
6208 :
6209 33 : al->s = s;
6210 33 : al->name = name;
6211 33 : al->next = alias_list;
6212 33 : alias_list = al;
6213 33 : kmsan_unpoison_memory(al, sizeof(*al));
6214 33 : return 0;
6215 : }
6216 :
6217 1 : static int __init slab_sysfs_init(void)
6218 : {
6219 : struct kmem_cache *s;
6220 : int err;
6221 :
6222 1 : mutex_lock(&slab_mutex);
6223 :
6224 1 : slab_kset = kset_create_and_add("slab", NULL, kernel_kobj);
6225 1 : if (!slab_kset) {
6226 0 : mutex_unlock(&slab_mutex);
6227 0 : pr_err("Cannot register slab subsystem.\n");
6228 0 : return -ENOSYS;
6229 : }
6230 :
6231 1 : slab_state = FULL;
6232 :
6233 53 : list_for_each_entry(s, &slab_caches, list) {
6234 52 : err = sysfs_slab_add(s);
6235 52 : if (err)
6236 0 : pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
6237 : s->name);
6238 : }
6239 :
6240 34 : while (alias_list) {
6241 33 : struct saved_alias *al = alias_list;
6242 :
6243 33 : alias_list = alias_list->next;
6244 33 : err = sysfs_slab_alias(al->s, al->name);
6245 33 : if (err)
6246 0 : pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
6247 : al->name);
6248 33 : kfree(al);
6249 : }
6250 :
6251 1 : mutex_unlock(&slab_mutex);
6252 1 : return 0;
6253 : }
6254 : late_initcall(slab_sysfs_init);
6255 : #endif /* SLAB_SUPPORTS_SYSFS */
6256 :
6257 : #if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS)
6258 : static int slab_debugfs_show(struct seq_file *seq, void *v)
6259 : {
6260 : struct loc_track *t = seq->private;
6261 : struct location *l;
6262 : unsigned long idx;
6263 :
6264 : idx = (unsigned long) t->idx;
6265 : if (idx < t->count) {
6266 : l = &t->loc[idx];
6267 :
6268 : seq_printf(seq, "%7ld ", l->count);
6269 :
6270 : if (l->addr)
6271 : seq_printf(seq, "%pS", (void *)l->addr);
6272 : else
6273 : seq_puts(seq, "<not-available>");
6274 :
6275 : if (l->waste)
6276 : seq_printf(seq, " waste=%lu/%lu",
6277 : l->count * l->waste, l->waste);
6278 :
6279 : if (l->sum_time != l->min_time) {
6280 : seq_printf(seq, " age=%ld/%llu/%ld",
6281 : l->min_time, div_u64(l->sum_time, l->count),
6282 : l->max_time);
6283 : } else
6284 : seq_printf(seq, " age=%ld", l->min_time);
6285 :
6286 : if (l->min_pid != l->max_pid)
6287 : seq_printf(seq, " pid=%ld-%ld", l->min_pid, l->max_pid);
6288 : else
6289 : seq_printf(seq, " pid=%ld",
6290 : l->min_pid);
6291 :
6292 : if (num_online_cpus() > 1 && !cpumask_empty(to_cpumask(l->cpus)))
6293 : seq_printf(seq, " cpus=%*pbl",
6294 : cpumask_pr_args(to_cpumask(l->cpus)));
6295 :
6296 : if (nr_online_nodes > 1 && !nodes_empty(l->nodes))
6297 : seq_printf(seq, " nodes=%*pbl",
6298 : nodemask_pr_args(&l->nodes));
6299 :
6300 : #ifdef CONFIG_STACKDEPOT
6301 : {
6302 : depot_stack_handle_t handle;
6303 : unsigned long *entries;
6304 : unsigned int nr_entries, j;
6305 :
6306 : handle = READ_ONCE(l->handle);
6307 : if (handle) {
6308 : nr_entries = stack_depot_fetch(handle, &entries);
6309 : seq_puts(seq, "\n");
6310 : for (j = 0; j < nr_entries; j++)
6311 : seq_printf(seq, " %pS\n", (void *)entries[j]);
6312 : }
6313 : }
6314 : #endif
6315 : seq_puts(seq, "\n");
6316 : }
6317 :
6318 : if (!idx && !t->count)
6319 : seq_puts(seq, "No data\n");
6320 :
6321 : return 0;
6322 : }
6323 :
6324 : static void slab_debugfs_stop(struct seq_file *seq, void *v)
6325 : {
6326 : }
6327 :
6328 : static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos)
6329 : {
6330 : struct loc_track *t = seq->private;
6331 :
6332 : t->idx = ++(*ppos);
6333 : if (*ppos <= t->count)
6334 : return ppos;
6335 :
6336 : return NULL;
6337 : }
6338 :
6339 : static int cmp_loc_by_count(const void *a, const void *b, const void *data)
6340 : {
6341 : struct location *loc1 = (struct location *)a;
6342 : struct location *loc2 = (struct location *)b;
6343 :
6344 : if (loc1->count > loc2->count)
6345 : return -1;
6346 : else
6347 : return 1;
6348 : }
6349 :
6350 : static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos)
6351 : {
6352 : struct loc_track *t = seq->private;
6353 :
6354 : t->idx = *ppos;
6355 : return ppos;
6356 : }
6357 :
6358 : static const struct seq_operations slab_debugfs_sops = {
6359 : .start = slab_debugfs_start,
6360 : .next = slab_debugfs_next,
6361 : .stop = slab_debugfs_stop,
6362 : .show = slab_debugfs_show,
6363 : };
6364 :
6365 : static int slab_debug_trace_open(struct inode *inode, struct file *filep)
6366 : {
6367 :
6368 : struct kmem_cache_node *n;
6369 : enum track_item alloc;
6370 : int node;
6371 : struct loc_track *t = __seq_open_private(filep, &slab_debugfs_sops,
6372 : sizeof(struct loc_track));
6373 : struct kmem_cache *s = file_inode(filep)->i_private;
6374 : unsigned long *obj_map;
6375 :
6376 : if (!t)
6377 : return -ENOMEM;
6378 :
6379 : obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL);
6380 : if (!obj_map) {
6381 : seq_release_private(inode, filep);
6382 : return -ENOMEM;
6383 : }
6384 :
6385 : if (strcmp(filep->f_path.dentry->d_name.name, "alloc_traces") == 0)
6386 : alloc = TRACK_ALLOC;
6387 : else
6388 : alloc = TRACK_FREE;
6389 :
6390 : if (!alloc_loc_track(t, PAGE_SIZE / sizeof(struct location), GFP_KERNEL)) {
6391 : bitmap_free(obj_map);
6392 : seq_release_private(inode, filep);
6393 : return -ENOMEM;
6394 : }
6395 :
6396 : for_each_kmem_cache_node(s, node, n) {
6397 : unsigned long flags;
6398 : struct slab *slab;
6399 :
6400 : if (!atomic_long_read(&n->nr_slabs))
6401 : continue;
6402 :
6403 : spin_lock_irqsave(&n->list_lock, flags);
6404 : list_for_each_entry(slab, &n->partial, slab_list)
6405 : process_slab(t, s, slab, alloc, obj_map);
6406 : list_for_each_entry(slab, &n->full, slab_list)
6407 : process_slab(t, s, slab, alloc, obj_map);
6408 : spin_unlock_irqrestore(&n->list_lock, flags);
6409 : }
6410 :
6411 : /* Sort locations by count */
6412 : sort_r(t->loc, t->count, sizeof(struct location),
6413 : cmp_loc_by_count, NULL, NULL);
6414 :
6415 : bitmap_free(obj_map);
6416 : return 0;
6417 : }
6418 :
6419 : static int slab_debug_trace_release(struct inode *inode, struct file *file)
6420 : {
6421 : struct seq_file *seq = file->private_data;
6422 : struct loc_track *t = seq->private;
6423 :
6424 : free_loc_track(t);
6425 : return seq_release_private(inode, file);
6426 : }
6427 :
6428 : static const struct file_operations slab_debugfs_fops = {
6429 : .open = slab_debug_trace_open,
6430 : .read = seq_read,
6431 : .llseek = seq_lseek,
6432 : .release = slab_debug_trace_release,
6433 : };
6434 :
6435 : static void debugfs_slab_add(struct kmem_cache *s)
6436 : {
6437 : struct dentry *slab_cache_dir;
6438 :
6439 : if (unlikely(!slab_debugfs_root))
6440 : return;
6441 :
6442 : slab_cache_dir = debugfs_create_dir(s->name, slab_debugfs_root);
6443 :
6444 : debugfs_create_file("alloc_traces", 0400,
6445 : slab_cache_dir, s, &slab_debugfs_fops);
6446 :
6447 : debugfs_create_file("free_traces", 0400,
6448 : slab_cache_dir, s, &slab_debugfs_fops);
6449 : }
6450 :
6451 : void debugfs_slab_release(struct kmem_cache *s)
6452 : {
6453 : debugfs_lookup_and_remove(s->name, slab_debugfs_root);
6454 : }
6455 :
6456 : static int __init slab_debugfs_init(void)
6457 : {
6458 : struct kmem_cache *s;
6459 :
6460 : slab_debugfs_root = debugfs_create_dir("slab", NULL);
6461 :
6462 : list_for_each_entry(s, &slab_caches, list)
6463 : if (s->flags & SLAB_STORE_USER)
6464 : debugfs_slab_add(s);
6465 :
6466 : return 0;
6467 :
6468 : }
6469 : __initcall(slab_debugfs_init);
6470 : #endif
6471 : /*
6472 : * The /proc/slabinfo ABI
6473 : */
6474 : #ifdef CONFIG_SLUB_DEBUG
6475 0 : void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
6476 : {
6477 0 : unsigned long nr_slabs = 0;
6478 0 : unsigned long nr_objs = 0;
6479 0 : unsigned long nr_free = 0;
6480 : int node;
6481 : struct kmem_cache_node *n;
6482 :
6483 0 : for_each_kmem_cache_node(s, node, n) {
6484 0 : nr_slabs += node_nr_slabs(n);
6485 0 : nr_objs += node_nr_objs(n);
6486 0 : nr_free += count_partial(n, count_free);
6487 : }
6488 :
6489 0 : sinfo->active_objs = nr_objs - nr_free;
6490 0 : sinfo->num_objs = nr_objs;
6491 0 : sinfo->active_slabs = nr_slabs;
6492 0 : sinfo->num_slabs = nr_slabs;
6493 0 : sinfo->objects_per_slab = oo_objects(s->oo);
6494 0 : sinfo->cache_order = oo_order(s->oo);
6495 0 : }
6496 :
6497 0 : void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
6498 : {
6499 0 : }
6500 :
6501 0 : ssize_t slabinfo_write(struct file *file, const char __user *buffer,
6502 : size_t count, loff_t *ppos)
6503 : {
6504 0 : return -EIO;
6505 : }
6506 : #endif /* CONFIG_SLUB_DEBUG */
|
