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/cpu.h>
26 #include <linux/cpuset.h>
27 #include <linux/mempolicy.h>
28 #include <linux/ctype.h>
29 #include <linux/stackdepot.h>
30 #include <linux/debugobjects.h>
31 #include <linux/kallsyms.h>
32 #include <linux/kfence.h>
33 #include <linux/memory.h>
34 #include <linux/math64.h>
35 #include <linux/fault-inject.h>
36 #include <linux/stacktrace.h>
37 #include <linux/prefetch.h>
38 #include <linux/memcontrol.h>
39 #include <linux/random.h>
40 #include <kunit/test.h>
41 #include <linux/sort.h>
42 
43 #include <linux/debugfs.h>
44 #include <trace/events/kmem.h>
45 
46 #include "internal.h"
47 
48 /*
49  * Lock order:
50  *   1. slab_mutex (Global Mutex)
51  *   2. node->list_lock (Spinlock)
52  *   3. kmem_cache->cpu_slab->lock (Local lock)
53  *   4. slab_lock(slab) (Only on some arches or for debugging)
54  *   5. object_map_lock (Only for debugging)
55  *
56  *   slab_mutex
57  *
58  *   The role of the slab_mutex is to protect the list of all the slabs
59  *   and to synchronize major metadata changes to slab cache structures.
60  *   Also synchronizes memory hotplug callbacks.
61  *
62  *   slab_lock
63  *
64  *   The slab_lock is a wrapper around the page lock, thus it is a bit
65  *   spinlock.
66  *
67  *   The slab_lock is only used for debugging and on arches that do not
68  *   have the ability to do a cmpxchg_double. It only protects:
69  *	A. slab->freelist	-> List of free objects in a slab
70  *	B. slab->inuse		-> Number of objects in use
71  *	C. slab->objects	-> Number of objects in slab
72  *	D. slab->frozen		-> frozen state
73  *
74  *   Frozen slabs
75  *
76  *   If a slab is frozen then it is exempt from list management. It is not
77  *   on any list except per cpu partial list. The processor that froze the
78  *   slab is the one who can perform list operations on the slab. Other
79  *   processors may put objects onto the freelist but the processor that
80  *   froze the slab is the only one that can retrieve the objects from the
81  *   slab's freelist.
82  *
83  *   list_lock
84  *
85  *   The list_lock protects the partial and full list on each node and
86  *   the partial slab counter. If taken then no new slabs may be added or
87  *   removed from the lists nor make the number of partial slabs be modified.
88  *   (Note that the total number of slabs is an atomic value that may be
89  *   modified without taking the list lock).
90  *
91  *   The list_lock is a centralized lock and thus we avoid taking it as
92  *   much as possible. As long as SLUB does not have to handle partial
93  *   slabs, operations can continue without any centralized lock. F.e.
94  *   allocating a long series of objects that fill up slabs does not require
95  *   the list lock.
96  *
97  *   cpu_slab->lock local lock
98  *
99  *   This locks protect slowpath manipulation of all kmem_cache_cpu fields
100  *   except the stat counters. This is a percpu structure manipulated only by
101  *   the local cpu, so the lock protects against being preempted or interrupted
102  *   by an irq. Fast path operations rely on lockless operations instead.
103  *   On PREEMPT_RT, the local lock does not actually disable irqs (and thus
104  *   prevent the lockless operations), so fastpath operations also need to take
105  *   the lock and are no longer lockless.
106  *
107  *   lockless fastpaths
108  *
109  *   The fast path allocation (slab_alloc_node()) and freeing (do_slab_free())
110  *   are fully lockless when satisfied from the percpu slab (and when
111  *   cmpxchg_double is possible to use, otherwise slab_lock is taken).
112  *   They also don't disable preemption or migration or irqs. They rely on
113  *   the transaction id (tid) field to detect being preempted or moved to
114  *   another cpu.
115  *
116  *   irq, preemption, migration considerations
117  *
118  *   Interrupts are disabled as part of list_lock or local_lock operations, or
119  *   around the slab_lock operation, in order to make the slab allocator safe
120  *   to use in the context of an irq.
121  *
122  *   In addition, preemption (or migration on PREEMPT_RT) is disabled in the
123  *   allocation slowpath, bulk allocation, and put_cpu_partial(), so that the
124  *   local cpu doesn't change in the process and e.g. the kmem_cache_cpu pointer
125  *   doesn't have to be revalidated in each section protected by the local lock.
126  *
127  * SLUB assigns one slab for allocation to each processor.
128  * Allocations only occur from these slabs called cpu slabs.
129  *
130  * Slabs with free elements are kept on a partial list and during regular
131  * operations no list for full slabs is used. If an object in a full slab is
132  * freed then the slab will show up again on the partial lists.
133  * We track full slabs for debugging purposes though because otherwise we
134  * cannot scan all objects.
135  *
136  * Slabs are freed when they become empty. Teardown and setup is
137  * minimal so we rely on the page allocators per cpu caches for
138  * fast frees and allocs.
139  *
140  * slab->frozen		The slab is frozen and exempt from list processing.
141  * 			This means that the slab is dedicated to a purpose
142  * 			such as satisfying allocations for a specific
143  * 			processor. Objects may be freed in the slab while
144  * 			it is frozen but slab_free will then skip the usual
145  * 			list operations. It is up to the processor holding
146  * 			the slab to integrate the slab into the slab lists
147  * 			when the slab is no longer needed.
148  *
149  * 			One use of this flag is to mark slabs that are
150  * 			used for allocations. Then such a slab becomes a cpu
151  * 			slab. The cpu slab may be equipped with an additional
152  * 			freelist that allows lockless access to
153  * 			free objects in addition to the regular freelist
154  * 			that requires the slab lock.
155  *
156  * SLAB_DEBUG_FLAGS	Slab requires special handling due to debug
157  * 			options set. This moves	slab handling out of
158  * 			the fast path and disables lockless freelists.
159  */
160 
161 /*
162  * We could simply use migrate_disable()/enable() but as long as it's a
163  * function call even on !PREEMPT_RT, use inline preempt_disable() there.
164  */
165 #ifndef CONFIG_PREEMPT_RT
166 #define slub_get_cpu_ptr(var)	get_cpu_ptr(var)
167 #define slub_put_cpu_ptr(var)	put_cpu_ptr(var)
168 #else
169 #define slub_get_cpu_ptr(var)		\
170 ({					\
171 	migrate_disable();		\
172 	this_cpu_ptr(var);		\
173 })
174 #define slub_put_cpu_ptr(var)		\
175 do {					\
176 	(void)(var);			\
177 	migrate_enable();		\
178 } while (0)
179 #endif
180 
181 #ifdef CONFIG_SLUB_DEBUG
182 #ifdef CONFIG_SLUB_DEBUG_ON
183 DEFINE_STATIC_KEY_TRUE(slub_debug_enabled);
184 #else
185 DEFINE_STATIC_KEY_FALSE(slub_debug_enabled);
186 #endif
187 #endif		/* CONFIG_SLUB_DEBUG */
188 
kmem_cache_debug(struct kmem_cache * s)189 static inline bool kmem_cache_debug(struct kmem_cache *s)
190 {
191 	return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS);
192 }
193 
fixup_red_left(struct kmem_cache * s,void * p)194 void *fixup_red_left(struct kmem_cache *s, void *p)
195 {
196 	if (kmem_cache_debug_flags(s, SLAB_RED_ZONE))
197 		p += s->red_left_pad;
198 
199 	return p;
200 }
201 
kmem_cache_has_cpu_partial(struct kmem_cache * s)202 static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
203 {
204 #ifdef CONFIG_SLUB_CPU_PARTIAL
205 	return !kmem_cache_debug(s);
206 #else
207 	return false;
208 #endif
209 }
210 
211 /*
212  * Issues still to be resolved:
213  *
214  * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
215  *
216  * - Variable sizing of the per node arrays
217  */
218 
219 /* Enable to log cmpxchg failures */
220 #undef SLUB_DEBUG_CMPXCHG
221 
222 /*
223  * Minimum number of partial slabs. These will be left on the partial
224  * lists even if they are empty. kmem_cache_shrink may reclaim them.
225  */
226 #define MIN_PARTIAL 5
227 
228 /*
229  * Maximum number of desirable partial slabs.
230  * The existence of more partial slabs makes kmem_cache_shrink
231  * sort the partial list by the number of objects in use.
232  */
233 #define MAX_PARTIAL 10
234 
235 #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
236 				SLAB_POISON | SLAB_STORE_USER)
237 
238 /*
239  * These debug flags cannot use CMPXCHG because there might be consistency
240  * issues when checking or reading debug information
241  */
242 #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
243 				SLAB_TRACE)
244 
245 
246 /*
247  * Debugging flags that require metadata to be stored in the slab.  These get
248  * disabled when slub_debug=O is used and a cache's min order increases with
249  * metadata.
250  */
251 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
252 
253 #define OO_SHIFT	16
254 #define OO_MASK		((1 << OO_SHIFT) - 1)
255 #define MAX_OBJS_PER_PAGE	32767 /* since slab.objects is u15 */
256 
257 /* Internal SLUB flags */
258 /* Poison object */
259 #define __OBJECT_POISON		((slab_flags_t __force)0x80000000U)
260 /* Use cmpxchg_double */
261 #define __CMPXCHG_DOUBLE	((slab_flags_t __force)0x40000000U)
262 
263 /*
264  * Tracking user of a slab.
265  */
266 #define TRACK_ADDRS_COUNT 16
267 struct track {
268 	unsigned long addr;	/* Called from address */
269 #ifdef CONFIG_STACKDEPOT
270 	depot_stack_handle_t handle;
271 #endif
272 	int cpu;		/* Was running on cpu */
273 	int pid;		/* Pid context */
274 	unsigned long when;	/* When did the operation occur */
275 };
276 
277 enum track_item { TRACK_ALLOC, TRACK_FREE };
278 
279 #ifdef CONFIG_SYSFS
280 static int sysfs_slab_add(struct kmem_cache *);
281 static int sysfs_slab_alias(struct kmem_cache *, const char *);
282 #else
sysfs_slab_add(struct kmem_cache * s)283 static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
sysfs_slab_alias(struct kmem_cache * s,const char * p)284 static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
285 							{ return 0; }
286 #endif
287 
288 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
289 static void debugfs_slab_add(struct kmem_cache *);
290 #else
debugfs_slab_add(struct kmem_cache * s)291 static inline void debugfs_slab_add(struct kmem_cache *s) { }
292 #endif
293 
stat(const struct kmem_cache * s,enum stat_item si)294 static inline void stat(const struct kmem_cache *s, enum stat_item si)
295 {
296 #ifdef CONFIG_SLUB_STATS
297 	/*
298 	 * The rmw is racy on a preemptible kernel but this is acceptable, so
299 	 * avoid this_cpu_add()'s irq-disable overhead.
300 	 */
301 	raw_cpu_inc(s->cpu_slab->stat[si]);
302 #endif
303 }
304 
305 /*
306  * Tracks for which NUMA nodes we have kmem_cache_nodes allocated.
307  * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily
308  * differ during memory hotplug/hotremove operations.
309  * Protected by slab_mutex.
310  */
311 static nodemask_t slab_nodes;
312 
313 /********************************************************************
314  * 			Core slab cache functions
315  *******************************************************************/
316 
317 /*
318  * Returns freelist pointer (ptr). With hardening, this is obfuscated
319  * with an XOR of the address where the pointer is held and a per-cache
320  * random number.
321  */
freelist_ptr(const struct kmem_cache * s,void * ptr,unsigned long ptr_addr)322 static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
323 				 unsigned long ptr_addr)
324 {
325 #ifdef CONFIG_SLAB_FREELIST_HARDENED
326 	/*
327 	 * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged.
328 	 * Normally, this doesn't cause any issues, as both set_freepointer()
329 	 * and get_freepointer() are called with a pointer with the same tag.
330 	 * However, there are some issues with CONFIG_SLUB_DEBUG code. For
331 	 * example, when __free_slub() iterates over objects in a cache, it
332 	 * passes untagged pointers to check_object(). check_object() in turns
333 	 * calls get_freepointer() with an untagged pointer, which causes the
334 	 * freepointer to be restored incorrectly.
335 	 */
336 	return (void *)((unsigned long)ptr ^ s->random ^
337 			swab((unsigned long)kasan_reset_tag((void *)ptr_addr)));
338 #else
339 	return ptr;
340 #endif
341 }
342 
343 /* Returns the freelist pointer recorded at location ptr_addr. */
freelist_dereference(const struct kmem_cache * s,void * ptr_addr)344 static inline void *freelist_dereference(const struct kmem_cache *s,
345 					 void *ptr_addr)
346 {
347 	return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr),
348 			    (unsigned long)ptr_addr);
349 }
350 
get_freepointer(struct kmem_cache * s,void * object)351 static inline void *get_freepointer(struct kmem_cache *s, void *object)
352 {
353 	object = kasan_reset_tag(object);
354 	return freelist_dereference(s, object + s->offset);
355 }
356 
prefetch_freepointer(const struct kmem_cache * s,void * object)357 static void prefetch_freepointer(const struct kmem_cache *s, void *object)
358 {
359 	prefetchw(object + s->offset);
360 }
361 
get_freepointer_safe(struct kmem_cache * s,void * object)362 static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
363 {
364 	unsigned long freepointer_addr;
365 	void *p;
366 
367 	if (!debug_pagealloc_enabled_static())
368 		return get_freepointer(s, object);
369 
370 	object = kasan_reset_tag(object);
371 	freepointer_addr = (unsigned long)object + s->offset;
372 	copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p));
373 	return freelist_ptr(s, p, freepointer_addr);
374 }
375 
set_freepointer(struct kmem_cache * s,void * object,void * fp)376 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
377 {
378 	unsigned long freeptr_addr = (unsigned long)object + s->offset;
379 
380 #ifdef CONFIG_SLAB_FREELIST_HARDENED
381 	BUG_ON(object == fp); /* naive detection of double free or corruption */
382 #endif
383 
384 	freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr);
385 	*(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
386 }
387 
388 /* Loop over all objects in a slab */
389 #define for_each_object(__p, __s, __addr, __objects) \
390 	for (__p = fixup_red_left(__s, __addr); \
391 		__p < (__addr) + (__objects) * (__s)->size; \
392 		__p += (__s)->size)
393 
order_objects(unsigned int order,unsigned int size)394 static inline unsigned int order_objects(unsigned int order, unsigned int size)
395 {
396 	return ((unsigned int)PAGE_SIZE << order) / size;
397 }
398 
oo_make(unsigned int order,unsigned int size)399 static inline struct kmem_cache_order_objects oo_make(unsigned int order,
400 		unsigned int size)
401 {
402 	struct kmem_cache_order_objects x = {
403 		(order << OO_SHIFT) + order_objects(order, size)
404 	};
405 
406 	return x;
407 }
408 
oo_order(struct kmem_cache_order_objects x)409 static inline unsigned int oo_order(struct kmem_cache_order_objects x)
410 {
411 	return x.x >> OO_SHIFT;
412 }
413 
oo_objects(struct kmem_cache_order_objects x)414 static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
415 {
416 	return x.x & OO_MASK;
417 }
418 
419 #ifdef CONFIG_SLUB_CPU_PARTIAL
slub_set_cpu_partial(struct kmem_cache * s,unsigned int nr_objects)420 static void slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects)
421 {
422 	unsigned int nr_slabs;
423 
424 	s->cpu_partial = nr_objects;
425 
426 	/*
427 	 * We take the number of objects but actually limit the number of
428 	 * slabs on the per cpu partial list, in order to limit excessive
429 	 * growth of the list. For simplicity we assume that the slabs will
430 	 * be half-full.
431 	 */
432 	nr_slabs = DIV_ROUND_UP(nr_objects * 2, oo_objects(s->oo));
433 	s->cpu_partial_slabs = nr_slabs;
434 }
435 #else
436 static inline void
slub_set_cpu_partial(struct kmem_cache * s,unsigned int nr_objects)437 slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects)
438 {
439 }
440 #endif /* CONFIG_SLUB_CPU_PARTIAL */
441 
442 /*
443  * Per slab locking using the pagelock
444  */
__slab_lock(struct slab * slab)445 static __always_inline void __slab_lock(struct slab *slab)
446 {
447 	struct page *page = slab_page(slab);
448 
449 	VM_BUG_ON_PAGE(PageTail(page), page);
450 	bit_spin_lock(PG_locked, &page->flags);
451 }
452 
__slab_unlock(struct slab * slab)453 static __always_inline void __slab_unlock(struct slab *slab)
454 {
455 	struct page *page = slab_page(slab);
456 
457 	VM_BUG_ON_PAGE(PageTail(page), page);
458 	__bit_spin_unlock(PG_locked, &page->flags);
459 }
460 
slab_lock(struct slab * slab,unsigned long * flags)461 static __always_inline void slab_lock(struct slab *slab, unsigned long *flags)
462 {
463 	if (IS_ENABLED(CONFIG_PREEMPT_RT))
464 		local_irq_save(*flags);
465 	__slab_lock(slab);
466 }
467 
slab_unlock(struct slab * slab,unsigned long * flags)468 static __always_inline void slab_unlock(struct slab *slab, unsigned long *flags)
469 {
470 	__slab_unlock(slab);
471 	if (IS_ENABLED(CONFIG_PREEMPT_RT))
472 		local_irq_restore(*flags);
473 }
474 
475 /*
476  * Interrupts must be disabled (for the fallback code to work right), typically
477  * by an _irqsave() lock variant. Except on PREEMPT_RT where locks are different
478  * so we disable interrupts as part of slab_[un]lock().
479  */
__cmpxchg_double_slab(struct kmem_cache * s,struct slab * slab,void * freelist_old,unsigned long counters_old,void * freelist_new,unsigned long counters_new,const char * n)480 static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct slab *slab,
481 		void *freelist_old, unsigned long counters_old,
482 		void *freelist_new, unsigned long counters_new,
483 		const char *n)
484 {
485 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
486 		lockdep_assert_irqs_disabled();
487 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
488     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
489 	if (s->flags & __CMPXCHG_DOUBLE) {
490 		if (cmpxchg_double(&slab->freelist, &slab->counters,
491 				   freelist_old, counters_old,
492 				   freelist_new, counters_new))
493 			return true;
494 	} else
495 #endif
496 	{
497 		/* init to 0 to prevent spurious warnings */
498 		unsigned long flags = 0;
499 
500 		slab_lock(slab, &flags);
501 		if (slab->freelist == freelist_old &&
502 					slab->counters == counters_old) {
503 			slab->freelist = freelist_new;
504 			slab->counters = counters_new;
505 			slab_unlock(slab, &flags);
506 			return true;
507 		}
508 		slab_unlock(slab, &flags);
509 	}
510 
511 	cpu_relax();
512 	stat(s, CMPXCHG_DOUBLE_FAIL);
513 
514 #ifdef SLUB_DEBUG_CMPXCHG
515 	pr_info("%s %s: cmpxchg double redo ", n, s->name);
516 #endif
517 
518 	return false;
519 }
520 
cmpxchg_double_slab(struct kmem_cache * s,struct slab * slab,void * freelist_old,unsigned long counters_old,void * freelist_new,unsigned long counters_new,const char * n)521 static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct slab *slab,
522 		void *freelist_old, unsigned long counters_old,
523 		void *freelist_new, unsigned long counters_new,
524 		const char *n)
525 {
526 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
527     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
528 	if (s->flags & __CMPXCHG_DOUBLE) {
529 		if (cmpxchg_double(&slab->freelist, &slab->counters,
530 				   freelist_old, counters_old,
531 				   freelist_new, counters_new))
532 			return true;
533 	} else
534 #endif
535 	{
536 		unsigned long flags;
537 
538 		local_irq_save(flags);
539 		__slab_lock(slab);
540 		if (slab->freelist == freelist_old &&
541 					slab->counters == counters_old) {
542 			slab->freelist = freelist_new;
543 			slab->counters = counters_new;
544 			__slab_unlock(slab);
545 			local_irq_restore(flags);
546 			return true;
547 		}
548 		__slab_unlock(slab);
549 		local_irq_restore(flags);
550 	}
551 
552 	cpu_relax();
553 	stat(s, CMPXCHG_DOUBLE_FAIL);
554 
555 #ifdef SLUB_DEBUG_CMPXCHG
556 	pr_info("%s %s: cmpxchg double redo ", n, s->name);
557 #endif
558 
559 	return false;
560 }
561 
562 #ifdef CONFIG_SLUB_DEBUG
563 static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)];
564 static DEFINE_RAW_SPINLOCK(object_map_lock);
565 
__fill_map(unsigned long * obj_map,struct kmem_cache * s,struct slab * slab)566 static void __fill_map(unsigned long *obj_map, struct kmem_cache *s,
567 		       struct slab *slab)
568 {
569 	void *addr = slab_address(slab);
570 	void *p;
571 
572 	bitmap_zero(obj_map, slab->objects);
573 
574 	for (p = slab->freelist; p; p = get_freepointer(s, p))
575 		set_bit(__obj_to_index(s, addr, p), obj_map);
576 }
577 
578 #if IS_ENABLED(CONFIG_KUNIT)
slab_add_kunit_errors(void)579 static bool slab_add_kunit_errors(void)
580 {
581 	struct kunit_resource *resource;
582 
583 	if (likely(!current->kunit_test))
584 		return false;
585 
586 	resource = kunit_find_named_resource(current->kunit_test, "slab_errors");
587 	if (!resource)
588 		return false;
589 
590 	(*(int *)resource->data)++;
591 	kunit_put_resource(resource);
592 	return true;
593 }
594 #else
slab_add_kunit_errors(void)595 static inline bool slab_add_kunit_errors(void) { return false; }
596 #endif
597 
598 /*
599  * Determine a map of objects in use in a slab.
600  *
601  * Node listlock must be held to guarantee that the slab does
602  * not vanish from under us.
603  */
get_map(struct kmem_cache * s,struct slab * slab)604 static unsigned long *get_map(struct kmem_cache *s, struct slab *slab)
605 	__acquires(&object_map_lock)
606 {
607 	VM_BUG_ON(!irqs_disabled());
608 
609 	raw_spin_lock(&object_map_lock);
610 
611 	__fill_map(object_map, s, slab);
612 
613 	return object_map;
614 }
615 
put_map(unsigned long * map)616 static void put_map(unsigned long *map) __releases(&object_map_lock)
617 {
618 	VM_BUG_ON(map != object_map);
619 	raw_spin_unlock(&object_map_lock);
620 }
621 
size_from_object(struct kmem_cache * s)622 static inline unsigned int size_from_object(struct kmem_cache *s)
623 {
624 	if (s->flags & SLAB_RED_ZONE)
625 		return s->size - s->red_left_pad;
626 
627 	return s->size;
628 }
629 
restore_red_left(struct kmem_cache * s,void * p)630 static inline void *restore_red_left(struct kmem_cache *s, void *p)
631 {
632 	if (s->flags & SLAB_RED_ZONE)
633 		p -= s->red_left_pad;
634 
635 	return p;
636 }
637 
638 /*
639  * Debug settings:
640  */
641 #if defined(CONFIG_SLUB_DEBUG_ON)
642 static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
643 #else
644 static slab_flags_t slub_debug;
645 #endif
646 
647 static char *slub_debug_string;
648 static int disable_higher_order_debug;
649 
650 /*
651  * slub is about to manipulate internal object metadata.  This memory lies
652  * outside the range of the allocated object, so accessing it would normally
653  * be reported by kasan as a bounds error.  metadata_access_enable() is used
654  * to tell kasan that these accesses are OK.
655  */
metadata_access_enable(void)656 static inline void metadata_access_enable(void)
657 {
658 	kasan_disable_current();
659 }
660 
metadata_access_disable(void)661 static inline void metadata_access_disable(void)
662 {
663 	kasan_enable_current();
664 }
665 
666 /*
667  * Object debugging
668  */
669 
670 /* Verify that a pointer has an address that is valid within a slab page */
check_valid_pointer(struct kmem_cache * s,struct slab * slab,void * object)671 static inline int check_valid_pointer(struct kmem_cache *s,
672 				struct slab *slab, void *object)
673 {
674 	void *base;
675 
676 	if (!object)
677 		return 1;
678 
679 	base = slab_address(slab);
680 	object = kasan_reset_tag(object);
681 	object = restore_red_left(s, object);
682 	if (object < base || object >= base + slab->objects * s->size ||
683 		(object - base) % s->size) {
684 		return 0;
685 	}
686 
687 	return 1;
688 }
689 
print_section(char * level,char * text,u8 * addr,unsigned int length)690 static void print_section(char *level, char *text, u8 *addr,
691 			  unsigned int length)
692 {
693 	metadata_access_enable();
694 	print_hex_dump(level, text, DUMP_PREFIX_ADDRESS,
695 			16, 1, kasan_reset_tag((void *)addr), length, 1);
696 	metadata_access_disable();
697 }
698 
699 /*
700  * See comment in calculate_sizes().
701  */
freeptr_outside_object(struct kmem_cache * s)702 static inline bool freeptr_outside_object(struct kmem_cache *s)
703 {
704 	return s->offset >= s->inuse;
705 }
706 
707 /*
708  * Return offset of the end of info block which is inuse + free pointer if
709  * not overlapping with object.
710  */
get_info_end(struct kmem_cache * s)711 static inline unsigned int get_info_end(struct kmem_cache *s)
712 {
713 	if (freeptr_outside_object(s))
714 		return s->inuse + sizeof(void *);
715 	else
716 		return s->inuse;
717 }
718 
get_track(struct kmem_cache * s,void * object,enum track_item alloc)719 static struct track *get_track(struct kmem_cache *s, void *object,
720 	enum track_item alloc)
721 {
722 	struct track *p;
723 
724 	p = object + get_info_end(s);
725 
726 	return kasan_reset_tag(p + alloc);
727 }
728 
729 #ifdef CONFIG_STACKDEPOT
set_track_prepare(void)730 static noinline depot_stack_handle_t set_track_prepare(void)
731 {
732 	depot_stack_handle_t handle;
733 	unsigned long entries[TRACK_ADDRS_COUNT];
734 	unsigned int nr_entries;
735 
736 	nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 3);
737 	handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT);
738 
739 	return handle;
740 }
741 #else
set_track_prepare(void)742 static inline depot_stack_handle_t set_track_prepare(void)
743 {
744 	return 0;
745 }
746 #endif
747 
set_track_update(struct kmem_cache * s,void * object,enum track_item alloc,unsigned long addr,depot_stack_handle_t handle)748 static void set_track_update(struct kmem_cache *s, void *object,
749 			     enum track_item alloc, unsigned long addr,
750 			     depot_stack_handle_t handle)
751 {
752 	struct track *p = get_track(s, object, alloc);
753 
754 #ifdef CONFIG_STACKDEPOT
755 	p->handle = handle;
756 #endif
757 	p->addr = addr;
758 	p->cpu = smp_processor_id();
759 	p->pid = current->pid;
760 	p->when = jiffies;
761 }
762 
set_track(struct kmem_cache * s,void * object,enum track_item alloc,unsigned long addr)763 static __always_inline void set_track(struct kmem_cache *s, void *object,
764 				      enum track_item alloc, unsigned long addr)
765 {
766 	depot_stack_handle_t handle = set_track_prepare();
767 
768 	set_track_update(s, object, alloc, addr, handle);
769 }
770 
init_tracking(struct kmem_cache * s,void * object)771 static void init_tracking(struct kmem_cache *s, void *object)
772 {
773 	struct track *p;
774 
775 	if (!(s->flags & SLAB_STORE_USER))
776 		return;
777 
778 	p = get_track(s, object, TRACK_ALLOC);
779 	memset(p, 0, 2*sizeof(struct track));
780 }
781 
print_track(const char * s,struct track * t,unsigned long pr_time)782 static void print_track(const char *s, struct track *t, unsigned long pr_time)
783 {
784 	depot_stack_handle_t handle __maybe_unused;
785 
786 	if (!t->addr)
787 		return;
788 
789 	pr_err("%s in %pS age=%lu cpu=%u pid=%d\n",
790 	       s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
791 #ifdef CONFIG_STACKDEPOT
792 	handle = READ_ONCE(t->handle);
793 	if (handle)
794 		stack_depot_print(handle);
795 	else
796 		pr_err("object allocation/free stack trace missing\n");
797 #endif
798 }
799 
print_tracking(struct kmem_cache * s,void * object)800 void print_tracking(struct kmem_cache *s, void *object)
801 {
802 	unsigned long pr_time = jiffies;
803 	if (!(s->flags & SLAB_STORE_USER))
804 		return;
805 
806 	print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
807 	print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
808 }
809 
print_slab_info(const struct slab * slab)810 static void print_slab_info(const struct slab *slab)
811 {
812 	struct folio *folio = (struct folio *)slab_folio(slab);
813 
814 	pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%pGp\n",
815 	       slab, slab->objects, slab->inuse, slab->freelist,
816 	       folio_flags(folio, 0));
817 }
818 
slab_bug(struct kmem_cache * s,char * fmt,...)819 static void slab_bug(struct kmem_cache *s, char *fmt, ...)
820 {
821 	struct va_format vaf;
822 	va_list args;
823 
824 	va_start(args, fmt);
825 	vaf.fmt = fmt;
826 	vaf.va = &args;
827 	pr_err("=============================================================================\n");
828 	pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
829 	pr_err("-----------------------------------------------------------------------------\n\n");
830 	va_end(args);
831 }
832 
833 __printf(2, 3)
slab_fix(struct kmem_cache * s,char * fmt,...)834 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
835 {
836 	struct va_format vaf;
837 	va_list args;
838 
839 	if (slab_add_kunit_errors())
840 		return;
841 
842 	va_start(args, fmt);
843 	vaf.fmt = fmt;
844 	vaf.va = &args;
845 	pr_err("FIX %s: %pV\n", s->name, &vaf);
846 	va_end(args);
847 }
848 
print_trailer(struct kmem_cache * s,struct slab * slab,u8 * p)849 static void print_trailer(struct kmem_cache *s, struct slab *slab, u8 *p)
850 {
851 	unsigned int off;	/* Offset of last byte */
852 	u8 *addr = slab_address(slab);
853 
854 	print_tracking(s, p);
855 
856 	print_slab_info(slab);
857 
858 	pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n",
859 	       p, p - addr, get_freepointer(s, p));
860 
861 	if (s->flags & SLAB_RED_ZONE)
862 		print_section(KERN_ERR, "Redzone  ", p - s->red_left_pad,
863 			      s->red_left_pad);
864 	else if (p > addr + 16)
865 		print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
866 
867 	print_section(KERN_ERR,         "Object   ", p,
868 		      min_t(unsigned int, s->object_size, PAGE_SIZE));
869 	if (s->flags & SLAB_RED_ZONE)
870 		print_section(KERN_ERR, "Redzone  ", p + s->object_size,
871 			s->inuse - s->object_size);
872 
873 	off = get_info_end(s);
874 
875 	if (s->flags & SLAB_STORE_USER)
876 		off += 2 * sizeof(struct track);
877 
878 	off += kasan_metadata_size(s);
879 
880 	if (off != size_from_object(s))
881 		/* Beginning of the filler is the free pointer */
882 		print_section(KERN_ERR, "Padding  ", p + off,
883 			      size_from_object(s) - off);
884 
885 	dump_stack();
886 }
887 
object_err(struct kmem_cache * s,struct slab * slab,u8 * object,char * reason)888 static void object_err(struct kmem_cache *s, struct slab *slab,
889 			u8 *object, char *reason)
890 {
891 	if (slab_add_kunit_errors())
892 		return;
893 
894 	slab_bug(s, "%s", reason);
895 	print_trailer(s, slab, object);
896 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
897 }
898 
freelist_corrupted(struct kmem_cache * s,struct slab * slab,void ** freelist,void * nextfree)899 static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab,
900 			       void **freelist, void *nextfree)
901 {
902 	if ((s->flags & SLAB_CONSISTENCY_CHECKS) &&
903 	    !check_valid_pointer(s, slab, nextfree) && freelist) {
904 		object_err(s, slab, *freelist, "Freechain corrupt");
905 		*freelist = NULL;
906 		slab_fix(s, "Isolate corrupted freechain");
907 		return true;
908 	}
909 
910 	return false;
911 }
912 
slab_err(struct kmem_cache * s,struct slab * slab,const char * fmt,...)913 static __printf(3, 4) void slab_err(struct kmem_cache *s, struct slab *slab,
914 			const char *fmt, ...)
915 {
916 	va_list args;
917 	char buf[100];
918 
919 	if (slab_add_kunit_errors())
920 		return;
921 
922 	va_start(args, fmt);
923 	vsnprintf(buf, sizeof(buf), fmt, args);
924 	va_end(args);
925 	slab_bug(s, "%s", buf);
926 	print_slab_info(slab);
927 	dump_stack();
928 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
929 }
930 
init_object(struct kmem_cache * s,void * object,u8 val)931 static void init_object(struct kmem_cache *s, void *object, u8 val)
932 {
933 	u8 *p = kasan_reset_tag(object);
934 
935 	if (s->flags & SLAB_RED_ZONE)
936 		memset(p - s->red_left_pad, val, s->red_left_pad);
937 
938 	if (s->flags & __OBJECT_POISON) {
939 		memset(p, POISON_FREE, s->object_size - 1);
940 		p[s->object_size - 1] = POISON_END;
941 	}
942 
943 	if (s->flags & SLAB_RED_ZONE)
944 		memset(p + s->object_size, val, s->inuse - s->object_size);
945 }
946 
restore_bytes(struct kmem_cache * s,char * message,u8 data,void * from,void * to)947 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
948 						void *from, void *to)
949 {
950 	slab_fix(s, "Restoring %s 0x%p-0x%p=0x%x", message, from, to - 1, data);
951 	memset(from, data, to - from);
952 }
953 
check_bytes_and_report(struct kmem_cache * s,struct slab * slab,u8 * object,char * what,u8 * start,unsigned int value,unsigned int bytes)954 static int check_bytes_and_report(struct kmem_cache *s, struct slab *slab,
955 			u8 *object, char *what,
956 			u8 *start, unsigned int value, unsigned int bytes)
957 {
958 	u8 *fault;
959 	u8 *end;
960 	u8 *addr = slab_address(slab);
961 
962 	metadata_access_enable();
963 	fault = memchr_inv(kasan_reset_tag(start), value, bytes);
964 	metadata_access_disable();
965 	if (!fault)
966 		return 1;
967 
968 	end = start + bytes;
969 	while (end > fault && end[-1] == value)
970 		end--;
971 
972 	if (slab_add_kunit_errors())
973 		goto skip_bug_print;
974 
975 	slab_bug(s, "%s overwritten", what);
976 	pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n",
977 					fault, end - 1, fault - addr,
978 					fault[0], value);
979 	print_trailer(s, slab, object);
980 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
981 
982 skip_bug_print:
983 	restore_bytes(s, what, value, fault, end);
984 	return 0;
985 }
986 
987 /*
988  * Object layout:
989  *
990  * object address
991  * 	Bytes of the object to be managed.
992  * 	If the freepointer may overlay the object then the free
993  *	pointer is at the middle of the object.
994  *
995  * 	Poisoning uses 0x6b (POISON_FREE) and the last byte is
996  * 	0xa5 (POISON_END)
997  *
998  * object + s->object_size
999  * 	Padding to reach word boundary. This is also used for Redzoning.
1000  * 	Padding is extended by another word if Redzoning is enabled and
1001  * 	object_size == inuse.
1002  *
1003  * 	We fill with 0xbb (RED_INACTIVE) for inactive objects and with
1004  * 	0xcc (RED_ACTIVE) for objects in use.
1005  *
1006  * object + s->inuse
1007  * 	Meta data starts here.
1008  *
1009  * 	A. Free pointer (if we cannot overwrite object on free)
1010  * 	B. Tracking data for SLAB_STORE_USER
1011  *	C. Padding to reach required alignment boundary or at minimum
1012  * 		one word if debugging is on to be able to detect writes
1013  * 		before the word boundary.
1014  *
1015  *	Padding is done using 0x5a (POISON_INUSE)
1016  *
1017  * object + s->size
1018  * 	Nothing is used beyond s->size.
1019  *
1020  * If slabcaches are merged then the object_size and inuse boundaries are mostly
1021  * ignored. And therefore no slab options that rely on these boundaries
1022  * may be used with merged slabcaches.
1023  */
1024 
check_pad_bytes(struct kmem_cache * s,struct slab * slab,u8 * p)1025 static int check_pad_bytes(struct kmem_cache *s, struct slab *slab, u8 *p)
1026 {
1027 	unsigned long off = get_info_end(s);	/* The end of info */
1028 
1029 	if (s->flags & SLAB_STORE_USER)
1030 		/* We also have user information there */
1031 		off += 2 * sizeof(struct track);
1032 
1033 	off += kasan_metadata_size(s);
1034 
1035 	if (size_from_object(s) == off)
1036 		return 1;
1037 
1038 	return check_bytes_and_report(s, slab, p, "Object padding",
1039 			p + off, POISON_INUSE, size_from_object(s) - off);
1040 }
1041 
1042 /* Check the pad bytes at the end of a slab page */
slab_pad_check(struct kmem_cache * s,struct slab * slab)1043 static void slab_pad_check(struct kmem_cache *s, struct slab *slab)
1044 {
1045 	u8 *start;
1046 	u8 *fault;
1047 	u8 *end;
1048 	u8 *pad;
1049 	int length;
1050 	int remainder;
1051 
1052 	if (!(s->flags & SLAB_POISON))
1053 		return;
1054 
1055 	start = slab_address(slab);
1056 	length = slab_size(slab);
1057 	end = start + length;
1058 	remainder = length % s->size;
1059 	if (!remainder)
1060 		return;
1061 
1062 	pad = end - remainder;
1063 	metadata_access_enable();
1064 	fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder);
1065 	metadata_access_disable();
1066 	if (!fault)
1067 		return;
1068 	while (end > fault && end[-1] == POISON_INUSE)
1069 		end--;
1070 
1071 	slab_err(s, slab, "Padding overwritten. 0x%p-0x%p @offset=%tu",
1072 			fault, end - 1, fault - start);
1073 	print_section(KERN_ERR, "Padding ", pad, remainder);
1074 
1075 	restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
1076 }
1077 
check_object(struct kmem_cache * s,struct slab * slab,void * object,u8 val)1078 static int check_object(struct kmem_cache *s, struct slab *slab,
1079 					void *object, u8 val)
1080 {
1081 	u8 *p = object;
1082 	u8 *endobject = object + s->object_size;
1083 
1084 	if (s->flags & SLAB_RED_ZONE) {
1085 		if (!check_bytes_and_report(s, slab, object, "Left Redzone",
1086 			object - s->red_left_pad, val, s->red_left_pad))
1087 			return 0;
1088 
1089 		if (!check_bytes_and_report(s, slab, object, "Right Redzone",
1090 			endobject, val, s->inuse - s->object_size))
1091 			return 0;
1092 	} else {
1093 		if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
1094 			check_bytes_and_report(s, slab, p, "Alignment padding",
1095 				endobject, POISON_INUSE,
1096 				s->inuse - s->object_size);
1097 		}
1098 	}
1099 
1100 	if (s->flags & SLAB_POISON) {
1101 		if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
1102 			(!check_bytes_and_report(s, slab, p, "Poison", p,
1103 					POISON_FREE, s->object_size - 1) ||
1104 			 !check_bytes_and_report(s, slab, p, "End Poison",
1105 				p + s->object_size - 1, POISON_END, 1)))
1106 			return 0;
1107 		/*
1108 		 * check_pad_bytes cleans up on its own.
1109 		 */
1110 		check_pad_bytes(s, slab, p);
1111 	}
1112 
1113 	if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE)
1114 		/*
1115 		 * Object and freepointer overlap. Cannot check
1116 		 * freepointer while object is allocated.
1117 		 */
1118 		return 1;
1119 
1120 	/* Check free pointer validity */
1121 	if (!check_valid_pointer(s, slab, get_freepointer(s, p))) {
1122 		object_err(s, slab, p, "Freepointer corrupt");
1123 		/*
1124 		 * No choice but to zap it and thus lose the remainder
1125 		 * of the free objects in this slab. May cause
1126 		 * another error because the object count is now wrong.
1127 		 */
1128 		set_freepointer(s, p, NULL);
1129 		return 0;
1130 	}
1131 	return 1;
1132 }
1133 
check_slab(struct kmem_cache * s,struct slab * slab)1134 static int check_slab(struct kmem_cache *s, struct slab *slab)
1135 {
1136 	int maxobj;
1137 
1138 	if (!folio_test_slab(slab_folio(slab))) {
1139 		slab_err(s, slab, "Not a valid slab page");
1140 		return 0;
1141 	}
1142 
1143 	maxobj = order_objects(slab_order(slab), s->size);
1144 	if (slab->objects > maxobj) {
1145 		slab_err(s, slab, "objects %u > max %u",
1146 			slab->objects, maxobj);
1147 		return 0;
1148 	}
1149 	if (slab->inuse > slab->objects) {
1150 		slab_err(s, slab, "inuse %u > max %u",
1151 			slab->inuse, slab->objects);
1152 		return 0;
1153 	}
1154 	/* Slab_pad_check fixes things up after itself */
1155 	slab_pad_check(s, slab);
1156 	return 1;
1157 }
1158 
1159 /*
1160  * Determine if a certain object in a slab is on the freelist. Must hold the
1161  * slab lock to guarantee that the chains are in a consistent state.
1162  */
on_freelist(struct kmem_cache * s,struct slab * slab,void * search)1163 static int on_freelist(struct kmem_cache *s, struct slab *slab, void *search)
1164 {
1165 	int nr = 0;
1166 	void *fp;
1167 	void *object = NULL;
1168 	int max_objects;
1169 
1170 	fp = slab->freelist;
1171 	while (fp && nr <= slab->objects) {
1172 		if (fp == search)
1173 			return 1;
1174 		if (!check_valid_pointer(s, slab, fp)) {
1175 			if (object) {
1176 				object_err(s, slab, object,
1177 					"Freechain corrupt");
1178 				set_freepointer(s, object, NULL);
1179 			} else {
1180 				slab_err(s, slab, "Freepointer corrupt");
1181 				slab->freelist = NULL;
1182 				slab->inuse = slab->objects;
1183 				slab_fix(s, "Freelist cleared");
1184 				return 0;
1185 			}
1186 			break;
1187 		}
1188 		object = fp;
1189 		fp = get_freepointer(s, object);
1190 		nr++;
1191 	}
1192 
1193 	max_objects = order_objects(slab_order(slab), s->size);
1194 	if (max_objects > MAX_OBJS_PER_PAGE)
1195 		max_objects = MAX_OBJS_PER_PAGE;
1196 
1197 	if (slab->objects != max_objects) {
1198 		slab_err(s, slab, "Wrong number of objects. Found %d but should be %d",
1199 			 slab->objects, max_objects);
1200 		slab->objects = max_objects;
1201 		slab_fix(s, "Number of objects adjusted");
1202 	}
1203 	if (slab->inuse != slab->objects - nr) {
1204 		slab_err(s, slab, "Wrong object count. Counter is %d but counted were %d",
1205 			 slab->inuse, slab->objects - nr);
1206 		slab->inuse = slab->objects - nr;
1207 		slab_fix(s, "Object count adjusted");
1208 	}
1209 	return search == NULL;
1210 }
1211 
trace(struct kmem_cache * s,struct slab * slab,void * object,int alloc)1212 static void trace(struct kmem_cache *s, struct slab *slab, void *object,
1213 								int alloc)
1214 {
1215 	if (s->flags & SLAB_TRACE) {
1216 		pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
1217 			s->name,
1218 			alloc ? "alloc" : "free",
1219 			object, slab->inuse,
1220 			slab->freelist);
1221 
1222 		if (!alloc)
1223 			print_section(KERN_INFO, "Object ", (void *)object,
1224 					s->object_size);
1225 
1226 		dump_stack();
1227 	}
1228 }
1229 
1230 /*
1231  * Tracking of fully allocated slabs for debugging purposes.
1232  */
add_full(struct kmem_cache * s,struct kmem_cache_node * n,struct slab * slab)1233 static void add_full(struct kmem_cache *s,
1234 	struct kmem_cache_node *n, struct slab *slab)
1235 {
1236 	if (!(s->flags & SLAB_STORE_USER))
1237 		return;
1238 
1239 	lockdep_assert_held(&n->list_lock);
1240 	list_add(&slab->slab_list, &n->full);
1241 }
1242 
remove_full(struct kmem_cache * s,struct kmem_cache_node * n,struct slab * slab)1243 static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct slab *slab)
1244 {
1245 	if (!(s->flags & SLAB_STORE_USER))
1246 		return;
1247 
1248 	lockdep_assert_held(&n->list_lock);
1249 	list_del(&slab->slab_list);
1250 }
1251 
1252 /* Tracking of the number of slabs for debugging purposes */
slabs_node(struct kmem_cache * s,int node)1253 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1254 {
1255 	struct kmem_cache_node *n = get_node(s, node);
1256 
1257 	return atomic_long_read(&n->nr_slabs);
1258 }
1259 
node_nr_slabs(struct kmem_cache_node * n)1260 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1261 {
1262 	return atomic_long_read(&n->nr_slabs);
1263 }
1264 
inc_slabs_node(struct kmem_cache * s,int node,int objects)1265 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
1266 {
1267 	struct kmem_cache_node *n = get_node(s, node);
1268 
1269 	/*
1270 	 * May be called early in order to allocate a slab for the
1271 	 * kmem_cache_node structure. Solve the chicken-egg
1272 	 * dilemma by deferring the increment of the count during
1273 	 * bootstrap (see early_kmem_cache_node_alloc).
1274 	 */
1275 	if (likely(n)) {
1276 		atomic_long_inc(&n->nr_slabs);
1277 		atomic_long_add(objects, &n->total_objects);
1278 	}
1279 }
dec_slabs_node(struct kmem_cache * s,int node,int objects)1280 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
1281 {
1282 	struct kmem_cache_node *n = get_node(s, node);
1283 
1284 	atomic_long_dec(&n->nr_slabs);
1285 	atomic_long_sub(objects, &n->total_objects);
1286 }
1287 
1288 /* Object debug checks for alloc/free paths */
setup_object_debug(struct kmem_cache * s,void * object)1289 static void setup_object_debug(struct kmem_cache *s, void *object)
1290 {
1291 	if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))
1292 		return;
1293 
1294 	init_object(s, object, SLUB_RED_INACTIVE);
1295 	init_tracking(s, object);
1296 }
1297 
1298 static
setup_slab_debug(struct kmem_cache * s,struct slab * slab,void * addr)1299 void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr)
1300 {
1301 	if (!kmem_cache_debug_flags(s, SLAB_POISON))
1302 		return;
1303 
1304 	metadata_access_enable();
1305 	memset(kasan_reset_tag(addr), POISON_INUSE, slab_size(slab));
1306 	metadata_access_disable();
1307 }
1308 
alloc_consistency_checks(struct kmem_cache * s,struct slab * slab,void * object)1309 static inline int alloc_consistency_checks(struct kmem_cache *s,
1310 					struct slab *slab, void *object)
1311 {
1312 	if (!check_slab(s, slab))
1313 		return 0;
1314 
1315 	if (!check_valid_pointer(s, slab, object)) {
1316 		object_err(s, slab, object, "Freelist Pointer check fails");
1317 		return 0;
1318 	}
1319 
1320 	if (!check_object(s, slab, object, SLUB_RED_INACTIVE))
1321 		return 0;
1322 
1323 	return 1;
1324 }
1325 
alloc_debug_processing(struct kmem_cache * s,struct slab * slab,void * object,unsigned long addr)1326 static noinline int alloc_debug_processing(struct kmem_cache *s,
1327 					struct slab *slab,
1328 					void *object, unsigned long addr)
1329 {
1330 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1331 		if (!alloc_consistency_checks(s, slab, object))
1332 			goto bad;
1333 	}
1334 
1335 	/* Success perform special debug activities for allocs */
1336 	if (s->flags & SLAB_STORE_USER)
1337 		set_track(s, object, TRACK_ALLOC, addr);
1338 	trace(s, slab, object, 1);
1339 	init_object(s, object, SLUB_RED_ACTIVE);
1340 	return 1;
1341 
1342 bad:
1343 	if (folio_test_slab(slab_folio(slab))) {
1344 		/*
1345 		 * If this is a slab page then lets do the best we can
1346 		 * to avoid issues in the future. Marking all objects
1347 		 * as used avoids touching the remaining objects.
1348 		 */
1349 		slab_fix(s, "Marking all objects used");
1350 		slab->inuse = slab->objects;
1351 		slab->freelist = NULL;
1352 	}
1353 	return 0;
1354 }
1355 
free_consistency_checks(struct kmem_cache * s,struct slab * slab,void * object,unsigned long addr)1356 static inline int free_consistency_checks(struct kmem_cache *s,
1357 		struct slab *slab, void *object, unsigned long addr)
1358 {
1359 	if (!check_valid_pointer(s, slab, object)) {
1360 		slab_err(s, slab, "Invalid object pointer 0x%p", object);
1361 		return 0;
1362 	}
1363 
1364 	if (on_freelist(s, slab, object)) {
1365 		object_err(s, slab, object, "Object already free");
1366 		return 0;
1367 	}
1368 
1369 	if (!check_object(s, slab, object, SLUB_RED_ACTIVE))
1370 		return 0;
1371 
1372 	if (unlikely(s != slab->slab_cache)) {
1373 		if (!folio_test_slab(slab_folio(slab))) {
1374 			slab_err(s, slab, "Attempt to free object(0x%p) outside of slab",
1375 				 object);
1376 		} else if (!slab->slab_cache) {
1377 			pr_err("SLUB <none>: no slab for object 0x%p.\n",
1378 			       object);
1379 			dump_stack();
1380 		} else
1381 			object_err(s, slab, object,
1382 					"page slab pointer corrupt.");
1383 		return 0;
1384 	}
1385 	return 1;
1386 }
1387 
1388 /* Supports checking bulk free of a constructed freelist */
free_debug_processing(struct kmem_cache * s,struct slab * slab,void * head,void * tail,int bulk_cnt,unsigned long addr)1389 static noinline int free_debug_processing(
1390 	struct kmem_cache *s, struct slab *slab,
1391 	void *head, void *tail, int bulk_cnt,
1392 	unsigned long addr)
1393 {
1394 	struct kmem_cache_node *n = get_node(s, slab_nid(slab));
1395 	void *object = head;
1396 	int cnt = 0;
1397 	unsigned long flags, flags2;
1398 	int ret = 0;
1399 	depot_stack_handle_t handle = 0;
1400 
1401 	if (s->flags & SLAB_STORE_USER)
1402 		handle = set_track_prepare();
1403 
1404 	spin_lock_irqsave(&n->list_lock, flags);
1405 	slab_lock(slab, &flags2);
1406 
1407 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1408 		if (!check_slab(s, slab))
1409 			goto out;
1410 	}
1411 
1412 next_object:
1413 	cnt++;
1414 
1415 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1416 		if (!free_consistency_checks(s, slab, object, addr))
1417 			goto out;
1418 	}
1419 
1420 	if (s->flags & SLAB_STORE_USER)
1421 		set_track_update(s, object, TRACK_FREE, addr, handle);
1422 	trace(s, slab, object, 0);
1423 	/* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
1424 	init_object(s, object, SLUB_RED_INACTIVE);
1425 
1426 	/* Reached end of constructed freelist yet? */
1427 	if (object != tail) {
1428 		object = get_freepointer(s, object);
1429 		goto next_object;
1430 	}
1431 	ret = 1;
1432 
1433 out:
1434 	if (cnt != bulk_cnt)
1435 		slab_err(s, slab, "Bulk freelist count(%d) invalid(%d)\n",
1436 			 bulk_cnt, cnt);
1437 
1438 	slab_unlock(slab, &flags2);
1439 	spin_unlock_irqrestore(&n->list_lock, flags);
1440 	if (!ret)
1441 		slab_fix(s, "Object at 0x%p not freed", object);
1442 	return ret;
1443 }
1444 
1445 /*
1446  * Parse a block of slub_debug options. Blocks are delimited by ';'
1447  *
1448  * @str:    start of block
1449  * @flags:  returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified
1450  * @slabs:  return start of list of slabs, or NULL when there's no list
1451  * @init:   assume this is initial parsing and not per-kmem-create parsing
1452  *
1453  * returns the start of next block if there's any, or NULL
1454  */
1455 static char *
parse_slub_debug_flags(char * str,slab_flags_t * flags,char ** slabs,bool init)1456 parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init)
1457 {
1458 	bool higher_order_disable = false;
1459 
1460 	/* Skip any completely empty blocks */
1461 	while (*str && *str == ';')
1462 		str++;
1463 
1464 	if (*str == ',') {
1465 		/*
1466 		 * No options but restriction on slabs. This means full
1467 		 * debugging for slabs matching a pattern.
1468 		 */
1469 		*flags = DEBUG_DEFAULT_FLAGS;
1470 		goto check_slabs;
1471 	}
1472 	*flags = 0;
1473 
1474 	/* Determine which debug features should be switched on */
1475 	for (; *str && *str != ',' && *str != ';'; str++) {
1476 		switch (tolower(*str)) {
1477 		case '-':
1478 			*flags = 0;
1479 			break;
1480 		case 'f':
1481 			*flags |= SLAB_CONSISTENCY_CHECKS;
1482 			break;
1483 		case 'z':
1484 			*flags |= SLAB_RED_ZONE;
1485 			break;
1486 		case 'p':
1487 			*flags |= SLAB_POISON;
1488 			break;
1489 		case 'u':
1490 			*flags |= SLAB_STORE_USER;
1491 			break;
1492 		case 't':
1493 			*flags |= SLAB_TRACE;
1494 			break;
1495 		case 'a':
1496 			*flags |= SLAB_FAILSLAB;
1497 			break;
1498 		case 'o':
1499 			/*
1500 			 * Avoid enabling debugging on caches if its minimum
1501 			 * order would increase as a result.
1502 			 */
1503 			higher_order_disable = true;
1504 			break;
1505 		default:
1506 			if (init)
1507 				pr_err("slub_debug option '%c' unknown. skipped\n", *str);
1508 		}
1509 	}
1510 check_slabs:
1511 	if (*str == ',')
1512 		*slabs = ++str;
1513 	else
1514 		*slabs = NULL;
1515 
1516 	/* Skip over the slab list */
1517 	while (*str && *str != ';')
1518 		str++;
1519 
1520 	/* Skip any completely empty blocks */
1521 	while (*str && *str == ';')
1522 		str++;
1523 
1524 	if (init && higher_order_disable)
1525 		disable_higher_order_debug = 1;
1526 
1527 	if (*str)
1528 		return str;
1529 	else
1530 		return NULL;
1531 }
1532 
setup_slub_debug(char * str)1533 static int __init setup_slub_debug(char *str)
1534 {
1535 	slab_flags_t flags;
1536 	slab_flags_t global_flags;
1537 	char *saved_str;
1538 	char *slab_list;
1539 	bool global_slub_debug_changed = false;
1540 	bool slab_list_specified = false;
1541 
1542 	global_flags = DEBUG_DEFAULT_FLAGS;
1543 	if (*str++ != '=' || !*str)
1544 		/*
1545 		 * No options specified. Switch on full debugging.
1546 		 */
1547 		goto out;
1548 
1549 	saved_str = str;
1550 	while (str) {
1551 		str = parse_slub_debug_flags(str, &flags, &slab_list, true);
1552 
1553 		if (!slab_list) {
1554 			global_flags = flags;
1555 			global_slub_debug_changed = true;
1556 		} else {
1557 			slab_list_specified = true;
1558 			if (flags & SLAB_STORE_USER)
1559 				stack_depot_want_early_init();
1560 		}
1561 	}
1562 
1563 	/*
1564 	 * For backwards compatibility, a single list of flags with list of
1565 	 * slabs means debugging is only changed for those slabs, so the global
1566 	 * slub_debug should be unchanged (0 or DEBUG_DEFAULT_FLAGS, depending
1567 	 * on CONFIG_SLUB_DEBUG_ON). We can extended that to multiple lists as
1568 	 * long as there is no option specifying flags without a slab list.
1569 	 */
1570 	if (slab_list_specified) {
1571 		if (!global_slub_debug_changed)
1572 			global_flags = slub_debug;
1573 		slub_debug_string = saved_str;
1574 	}
1575 out:
1576 	slub_debug = global_flags;
1577 	if (slub_debug & SLAB_STORE_USER)
1578 		stack_depot_want_early_init();
1579 	if (slub_debug != 0 || slub_debug_string)
1580 		static_branch_enable(&slub_debug_enabled);
1581 	else
1582 		static_branch_disable(&slub_debug_enabled);
1583 	if ((static_branch_unlikely(&init_on_alloc) ||
1584 	     static_branch_unlikely(&init_on_free)) &&
1585 	    (slub_debug & SLAB_POISON))
1586 		pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n");
1587 	return 1;
1588 }
1589 
1590 __setup("slub_debug", setup_slub_debug);
1591 
1592 /*
1593  * kmem_cache_flags - apply debugging options to the cache
1594  * @object_size:	the size of an object without meta data
1595  * @flags:		flags to set
1596  * @name:		name of the cache
1597  *
1598  * Debug option(s) are applied to @flags. In addition to the debug
1599  * option(s), if a slab name (or multiple) is specified i.e.
1600  * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ...
1601  * then only the select slabs will receive the debug option(s).
1602  */
kmem_cache_flags(unsigned int object_size,slab_flags_t flags,const char * name)1603 slab_flags_t kmem_cache_flags(unsigned int object_size,
1604 	slab_flags_t flags, const char *name)
1605 {
1606 	char *iter;
1607 	size_t len;
1608 	char *next_block;
1609 	slab_flags_t block_flags;
1610 	slab_flags_t slub_debug_local = slub_debug;
1611 
1612 	if (flags & SLAB_NO_USER_FLAGS)
1613 		return flags;
1614 
1615 	/*
1616 	 * If the slab cache is for debugging (e.g. kmemleak) then
1617 	 * don't store user (stack trace) information by default,
1618 	 * but let the user enable it via the command line below.
1619 	 */
1620 	if (flags & SLAB_NOLEAKTRACE)
1621 		slub_debug_local &= ~SLAB_STORE_USER;
1622 
1623 	len = strlen(name);
1624 	next_block = slub_debug_string;
1625 	/* Go through all blocks of debug options, see if any matches our slab's name */
1626 	while (next_block) {
1627 		next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false);
1628 		if (!iter)
1629 			continue;
1630 		/* Found a block that has a slab list, search it */
1631 		while (*iter) {
1632 			char *end, *glob;
1633 			size_t cmplen;
1634 
1635 			end = strchrnul(iter, ',');
1636 			if (next_block && next_block < end)
1637 				end = next_block - 1;
1638 
1639 			glob = strnchr(iter, end - iter, '*');
1640 			if (glob)
1641 				cmplen = glob - iter;
1642 			else
1643 				cmplen = max_t(size_t, len, (end - iter));
1644 
1645 			if (!strncmp(name, iter, cmplen)) {
1646 				flags |= block_flags;
1647 				return flags;
1648 			}
1649 
1650 			if (!*end || *end == ';')
1651 				break;
1652 			iter = end + 1;
1653 		}
1654 	}
1655 
1656 	return flags | slub_debug_local;
1657 }
1658 #else /* !CONFIG_SLUB_DEBUG */
setup_object_debug(struct kmem_cache * s,void * object)1659 static inline void setup_object_debug(struct kmem_cache *s, void *object) {}
1660 static inline
setup_slab_debug(struct kmem_cache * s,struct slab * slab,void * addr)1661 void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr) {}
1662 
alloc_debug_processing(struct kmem_cache * s,struct slab * slab,void * object,unsigned long addr)1663 static inline int alloc_debug_processing(struct kmem_cache *s,
1664 	struct slab *slab, void *object, unsigned long addr) { return 0; }
1665 
free_debug_processing(struct kmem_cache * s,struct slab * slab,void * head,void * tail,int bulk_cnt,unsigned long addr)1666 static inline int free_debug_processing(
1667 	struct kmem_cache *s, struct slab *slab,
1668 	void *head, void *tail, int bulk_cnt,
1669 	unsigned long addr) { return 0; }
1670 
slab_pad_check(struct kmem_cache * s,struct slab * slab)1671 static inline void slab_pad_check(struct kmem_cache *s, struct slab *slab) {}
check_object(struct kmem_cache * s,struct slab * slab,void * object,u8 val)1672 static inline int check_object(struct kmem_cache *s, struct slab *slab,
1673 			void *object, u8 val) { return 1; }
add_full(struct kmem_cache * s,struct kmem_cache_node * n,struct slab * slab)1674 static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1675 					struct slab *slab) {}
remove_full(struct kmem_cache * s,struct kmem_cache_node * n,struct slab * slab)1676 static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1677 					struct slab *slab) {}
kmem_cache_flags(unsigned int object_size,slab_flags_t flags,const char * name)1678 slab_flags_t kmem_cache_flags(unsigned int object_size,
1679 	slab_flags_t flags, const char *name)
1680 {
1681 	return flags;
1682 }
1683 #define slub_debug 0
1684 
1685 #define disable_higher_order_debug 0
1686 
slabs_node(struct kmem_cache * s,int node)1687 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1688 							{ return 0; }
node_nr_slabs(struct kmem_cache_node * n)1689 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1690 							{ return 0; }
inc_slabs_node(struct kmem_cache * s,int node,int objects)1691 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1692 							int objects) {}
dec_slabs_node(struct kmem_cache * s,int node,int objects)1693 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1694 							int objects) {}
1695 
freelist_corrupted(struct kmem_cache * s,struct slab * slab,void ** freelist,void * nextfree)1696 static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab,
1697 			       void **freelist, void *nextfree)
1698 {
1699 	return false;
1700 }
1701 #endif /* CONFIG_SLUB_DEBUG */
1702 
1703 /*
1704  * Hooks for other subsystems that check memory allocations. In a typical
1705  * production configuration these hooks all should produce no code at all.
1706  */
kmalloc_large_node_hook(void * ptr,size_t size,gfp_t flags)1707 static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1708 {
1709 	ptr = kasan_kmalloc_large(ptr, size, flags);
1710 	/* As ptr might get tagged, call kmemleak hook after KASAN. */
1711 	kmemleak_alloc(ptr, size, 1, flags);
1712 	return ptr;
1713 }
1714 
kfree_hook(void * x)1715 static __always_inline void kfree_hook(void *x)
1716 {
1717 	kmemleak_free(x);
1718 	kasan_kfree_large(x);
1719 }
1720 
slab_free_hook(struct kmem_cache * s,void * x,bool init)1721 static __always_inline bool slab_free_hook(struct kmem_cache *s,
1722 						void *x, bool init)
1723 {
1724 	kmemleak_free_recursive(x, s->flags);
1725 
1726 	debug_check_no_locks_freed(x, s->object_size);
1727 
1728 	if (!(s->flags & SLAB_DEBUG_OBJECTS))
1729 		debug_check_no_obj_freed(x, s->object_size);
1730 
1731 	/* Use KCSAN to help debug racy use-after-free. */
1732 	if (!(s->flags & SLAB_TYPESAFE_BY_RCU))
1733 		__kcsan_check_access(x, s->object_size,
1734 				     KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT);
1735 
1736 	/*
1737 	 * As memory initialization might be integrated into KASAN,
1738 	 * kasan_slab_free and initialization memset's must be
1739 	 * kept together to avoid discrepancies in behavior.
1740 	 *
1741 	 * The initialization memset's clear the object and the metadata,
1742 	 * but don't touch the SLAB redzone.
1743 	 */
1744 	if (init) {
1745 		int rsize;
1746 
1747 		if (!kasan_has_integrated_init())
1748 			memset(kasan_reset_tag(x), 0, s->object_size);
1749 		rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0;
1750 		memset((char *)kasan_reset_tag(x) + s->inuse, 0,
1751 		       s->size - s->inuse - rsize);
1752 	}
1753 	/* KASAN might put x into memory quarantine, delaying its reuse. */
1754 	return kasan_slab_free(s, x, init);
1755 }
1756 
slab_free_freelist_hook(struct kmem_cache * s,void ** head,void ** tail,int * cnt)1757 static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1758 					   void **head, void **tail,
1759 					   int *cnt)
1760 {
1761 
1762 	void *object;
1763 	void *next = *head;
1764 	void *old_tail = *tail ? *tail : *head;
1765 
1766 	if (is_kfence_address(next)) {
1767 		slab_free_hook(s, next, false);
1768 		return true;
1769 	}
1770 
1771 	/* Head and tail of the reconstructed freelist */
1772 	*head = NULL;
1773 	*tail = NULL;
1774 
1775 	do {
1776 		object = next;
1777 		next = get_freepointer(s, object);
1778 
1779 		/* If object's reuse doesn't have to be delayed */
1780 		if (!slab_free_hook(s, object, slab_want_init_on_free(s))) {
1781 			/* Move object to the new freelist */
1782 			set_freepointer(s, object, *head);
1783 			*head = object;
1784 			if (!*tail)
1785 				*tail = object;
1786 		} else {
1787 			/*
1788 			 * Adjust the reconstructed freelist depth
1789 			 * accordingly if object's reuse is delayed.
1790 			 */
1791 			--(*cnt);
1792 		}
1793 	} while (object != old_tail);
1794 
1795 	if (*head == *tail)
1796 		*tail = NULL;
1797 
1798 	return *head != NULL;
1799 }
1800 
setup_object(struct kmem_cache * s,void * object)1801 static void *setup_object(struct kmem_cache *s, void *object)
1802 {
1803 	setup_object_debug(s, object);
1804 	object = kasan_init_slab_obj(s, object);
1805 	if (unlikely(s->ctor)) {
1806 		kasan_unpoison_object_data(s, object);
1807 		s->ctor(object);
1808 		kasan_poison_object_data(s, object);
1809 	}
1810 	return object;
1811 }
1812 
1813 /*
1814  * Slab allocation and freeing
1815  */
alloc_slab_page(gfp_t flags,int node,struct kmem_cache_order_objects oo)1816 static inline struct slab *alloc_slab_page(gfp_t flags, int node,
1817 		struct kmem_cache_order_objects oo)
1818 {
1819 	struct folio *folio;
1820 	struct slab *slab;
1821 	unsigned int order = oo_order(oo);
1822 
1823 	if (node == NUMA_NO_NODE)
1824 		folio = (struct folio *)alloc_pages(flags, order);
1825 	else
1826 		folio = (struct folio *)__alloc_pages_node(node, flags, order);
1827 
1828 	if (!folio)
1829 		return NULL;
1830 
1831 	slab = folio_slab(folio);
1832 	__folio_set_slab(folio);
1833 	if (page_is_pfmemalloc(folio_page(folio, 0)))
1834 		slab_set_pfmemalloc(slab);
1835 
1836 	return slab;
1837 }
1838 
1839 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1840 /* Pre-initialize the random sequence cache */
init_cache_random_seq(struct kmem_cache * s)1841 static int init_cache_random_seq(struct kmem_cache *s)
1842 {
1843 	unsigned int count = oo_objects(s->oo);
1844 	int err;
1845 
1846 	/* Bailout if already initialised */
1847 	if (s->random_seq)
1848 		return 0;
1849 
1850 	err = cache_random_seq_create(s, count, GFP_KERNEL);
1851 	if (err) {
1852 		pr_err("SLUB: Unable to initialize free list for %s\n",
1853 			s->name);
1854 		return err;
1855 	}
1856 
1857 	/* Transform to an offset on the set of pages */
1858 	if (s->random_seq) {
1859 		unsigned int i;
1860 
1861 		for (i = 0; i < count; i++)
1862 			s->random_seq[i] *= s->size;
1863 	}
1864 	return 0;
1865 }
1866 
1867 /* Initialize each random sequence freelist per cache */
init_freelist_randomization(void)1868 static void __init init_freelist_randomization(void)
1869 {
1870 	struct kmem_cache *s;
1871 
1872 	mutex_lock(&slab_mutex);
1873 
1874 	list_for_each_entry(s, &slab_caches, list)
1875 		init_cache_random_seq(s);
1876 
1877 	mutex_unlock(&slab_mutex);
1878 }
1879 
1880 /* Get the next entry on the pre-computed freelist randomized */
next_freelist_entry(struct kmem_cache * s,struct slab * slab,unsigned long * pos,void * start,unsigned long page_limit,unsigned long freelist_count)1881 static void *next_freelist_entry(struct kmem_cache *s, struct slab *slab,
1882 				unsigned long *pos, void *start,
1883 				unsigned long page_limit,
1884 				unsigned long freelist_count)
1885 {
1886 	unsigned int idx;
1887 
1888 	/*
1889 	 * If the target page allocation failed, the number of objects on the
1890 	 * page might be smaller than the usual size defined by the cache.
1891 	 */
1892 	do {
1893 		idx = s->random_seq[*pos];
1894 		*pos += 1;
1895 		if (*pos >= freelist_count)
1896 			*pos = 0;
1897 	} while (unlikely(idx >= page_limit));
1898 
1899 	return (char *)start + idx;
1900 }
1901 
1902 /* Shuffle the single linked freelist based on a random pre-computed sequence */
shuffle_freelist(struct kmem_cache * s,struct slab * slab)1903 static bool shuffle_freelist(struct kmem_cache *s, struct slab *slab)
1904 {
1905 	void *start;
1906 	void *cur;
1907 	void *next;
1908 	unsigned long idx, pos, page_limit, freelist_count;
1909 
1910 	if (slab->objects < 2 || !s->random_seq)
1911 		return false;
1912 
1913 	freelist_count = oo_objects(s->oo);
1914 	pos = get_random_int() % freelist_count;
1915 
1916 	page_limit = slab->objects * s->size;
1917 	start = fixup_red_left(s, slab_address(slab));
1918 
1919 	/* First entry is used as the base of the freelist */
1920 	cur = next_freelist_entry(s, slab, &pos, start, page_limit,
1921 				freelist_count);
1922 	cur = setup_object(s, cur);
1923 	slab->freelist = cur;
1924 
1925 	for (idx = 1; idx < slab->objects; idx++) {
1926 		next = next_freelist_entry(s, slab, &pos, start, page_limit,
1927 			freelist_count);
1928 		next = setup_object(s, next);
1929 		set_freepointer(s, cur, next);
1930 		cur = next;
1931 	}
1932 	set_freepointer(s, cur, NULL);
1933 
1934 	return true;
1935 }
1936 #else
init_cache_random_seq(struct kmem_cache * s)1937 static inline int init_cache_random_seq(struct kmem_cache *s)
1938 {
1939 	return 0;
1940 }
init_freelist_randomization(void)1941 static inline void init_freelist_randomization(void) { }
shuffle_freelist(struct kmem_cache * s,struct slab * slab)1942 static inline bool shuffle_freelist(struct kmem_cache *s, struct slab *slab)
1943 {
1944 	return false;
1945 }
1946 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1947 
allocate_slab(struct kmem_cache * s,gfp_t flags,int node)1948 static struct slab *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1949 {
1950 	struct slab *slab;
1951 	struct kmem_cache_order_objects oo = s->oo;
1952 	gfp_t alloc_gfp;
1953 	void *start, *p, *next;
1954 	int idx;
1955 	bool shuffle;
1956 
1957 	flags &= gfp_allowed_mask;
1958 
1959 	flags |= s->allocflags;
1960 
1961 	/*
1962 	 * Let the initial higher-order allocation fail under memory pressure
1963 	 * so we fall-back to the minimum order allocation.
1964 	 */
1965 	alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1966 	if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
1967 		alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_RECLAIM;
1968 
1969 	slab = alloc_slab_page(alloc_gfp, node, oo);
1970 	if (unlikely(!slab)) {
1971 		oo = s->min;
1972 		alloc_gfp = flags;
1973 		/*
1974 		 * Allocation may have failed due to fragmentation.
1975 		 * Try a lower order alloc if possible
1976 		 */
1977 		slab = alloc_slab_page(alloc_gfp, node, oo);
1978 		if (unlikely(!slab))
1979 			goto out;
1980 		stat(s, ORDER_FALLBACK);
1981 	}
1982 
1983 	slab->objects = oo_objects(oo);
1984 
1985 	account_slab(slab, oo_order(oo), s, flags);
1986 
1987 	slab->slab_cache = s;
1988 
1989 	kasan_poison_slab(slab);
1990 
1991 	start = slab_address(slab);
1992 
1993 	setup_slab_debug(s, slab, start);
1994 
1995 	shuffle = shuffle_freelist(s, slab);
1996 
1997 	if (!shuffle) {
1998 		start = fixup_red_left(s, start);
1999 		start = setup_object(s, start);
2000 		slab->freelist = start;
2001 		for (idx = 0, p = start; idx < slab->objects - 1; idx++) {
2002 			next = p + s->size;
2003 			next = setup_object(s, next);
2004 			set_freepointer(s, p, next);
2005 			p = next;
2006 		}
2007 		set_freepointer(s, p, NULL);
2008 	}
2009 
2010 	slab->inuse = slab->objects;
2011 	slab->frozen = 1;
2012 
2013 out:
2014 	if (!slab)
2015 		return NULL;
2016 
2017 	inc_slabs_node(s, slab_nid(slab), slab->objects);
2018 
2019 	return slab;
2020 }
2021 
new_slab(struct kmem_cache * s,gfp_t flags,int node)2022 static struct slab *new_slab(struct kmem_cache *s, gfp_t flags, int node)
2023 {
2024 	if (unlikely(flags & GFP_SLAB_BUG_MASK))
2025 		flags = kmalloc_fix_flags(flags);
2026 
2027 	WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2028 
2029 	return allocate_slab(s,
2030 		flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
2031 }
2032 
__free_slab(struct kmem_cache * s,struct slab * slab)2033 static void __free_slab(struct kmem_cache *s, struct slab *slab)
2034 {
2035 	struct folio *folio = slab_folio(slab);
2036 	int order = folio_order(folio);
2037 	int pages = 1 << order;
2038 
2039 	if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) {
2040 		void *p;
2041 
2042 		slab_pad_check(s, slab);
2043 		for_each_object(p, s, slab_address(slab), slab->objects)
2044 			check_object(s, slab, p, SLUB_RED_INACTIVE);
2045 	}
2046 
2047 	__slab_clear_pfmemalloc(slab);
2048 	__folio_clear_slab(folio);
2049 	folio->mapping = NULL;
2050 	if (current->reclaim_state)
2051 		current->reclaim_state->reclaimed_slab += pages;
2052 	unaccount_slab(slab, order, s);
2053 	__free_pages(folio_page(folio, 0), order);
2054 }
2055 
rcu_free_slab(struct rcu_head * h)2056 static void rcu_free_slab(struct rcu_head *h)
2057 {
2058 	struct slab *slab = container_of(h, struct slab, rcu_head);
2059 
2060 	__free_slab(slab->slab_cache, slab);
2061 }
2062 
free_slab(struct kmem_cache * s,struct slab * slab)2063 static void free_slab(struct kmem_cache *s, struct slab *slab)
2064 {
2065 	if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) {
2066 		call_rcu(&slab->rcu_head, rcu_free_slab);
2067 	} else
2068 		__free_slab(s, slab);
2069 }
2070 
discard_slab(struct kmem_cache * s,struct slab * slab)2071 static void discard_slab(struct kmem_cache *s, struct slab *slab)
2072 {
2073 	dec_slabs_node(s, slab_nid(slab), slab->objects);
2074 	free_slab(s, slab);
2075 }
2076 
2077 /*
2078  * Management of partially allocated slabs.
2079  */
2080 static inline void
__add_partial(struct kmem_cache_node * n,struct slab * slab,int tail)2081 __add_partial(struct kmem_cache_node *n, struct slab *slab, int tail)
2082 {
2083 	n->nr_partial++;
2084 	if (tail == DEACTIVATE_TO_TAIL)
2085 		list_add_tail(&slab->slab_list, &n->partial);
2086 	else
2087 		list_add(&slab->slab_list, &n->partial);
2088 }
2089 
add_partial(struct kmem_cache_node * n,struct slab * slab,int tail)2090 static inline void add_partial(struct kmem_cache_node *n,
2091 				struct slab *slab, int tail)
2092 {
2093 	lockdep_assert_held(&n->list_lock);
2094 	__add_partial(n, slab, tail);
2095 }
2096 
remove_partial(struct kmem_cache_node * n,struct slab * slab)2097 static inline void remove_partial(struct kmem_cache_node *n,
2098 					struct slab *slab)
2099 {
2100 	lockdep_assert_held(&n->list_lock);
2101 	list_del(&slab->slab_list);
2102 	n->nr_partial--;
2103 }
2104 
2105 /*
2106  * Remove slab from the partial list, freeze it and
2107  * return the pointer to the freelist.
2108  *
2109  * Returns a list of objects or NULL if it fails.
2110  */
acquire_slab(struct kmem_cache * s,struct kmem_cache_node * n,struct slab * slab,int mode)2111 static inline void *acquire_slab(struct kmem_cache *s,
2112 		struct kmem_cache_node *n, struct slab *slab,
2113 		int mode)
2114 {
2115 	void *freelist;
2116 	unsigned long counters;
2117 	struct slab new;
2118 
2119 	lockdep_assert_held(&n->list_lock);
2120 
2121 	/*
2122 	 * Zap the freelist and set the frozen bit.
2123 	 * The old freelist is the list of objects for the
2124 	 * per cpu allocation list.
2125 	 */
2126 	freelist = slab->freelist;
2127 	counters = slab->counters;
2128 	new.counters = counters;
2129 	if (mode) {
2130 		new.inuse = slab->objects;
2131 		new.freelist = NULL;
2132 	} else {
2133 		new.freelist = freelist;
2134 	}
2135 
2136 	VM_BUG_ON(new.frozen);
2137 	new.frozen = 1;
2138 
2139 	if (!__cmpxchg_double_slab(s, slab,
2140 			freelist, counters,
2141 			new.freelist, new.counters,
2142 			"acquire_slab"))
2143 		return NULL;
2144 
2145 	remove_partial(n, slab);
2146 	WARN_ON(!freelist);
2147 	return freelist;
2148 }
2149 
2150 #ifdef CONFIG_SLUB_CPU_PARTIAL
2151 static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain);
2152 #else
put_cpu_partial(struct kmem_cache * s,struct slab * slab,int drain)2153 static inline void put_cpu_partial(struct kmem_cache *s, struct slab *slab,
2154 				   int drain) { }
2155 #endif
2156 static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags);
2157 
2158 /*
2159  * Try to allocate a partial slab from a specific node.
2160  */
get_partial_node(struct kmem_cache * s,struct kmem_cache_node * n,struct slab ** ret_slab,gfp_t gfpflags)2161 static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
2162 			      struct slab **ret_slab, gfp_t gfpflags)
2163 {
2164 	struct slab *slab, *slab2;
2165 	void *object = NULL;
2166 	unsigned long flags;
2167 	unsigned int partial_slabs = 0;
2168 
2169 	/*
2170 	 * Racy check. If we mistakenly see no partial slabs then we
2171 	 * just allocate an empty slab. If we mistakenly try to get a
2172 	 * partial slab and there is none available then get_partial()
2173 	 * will return NULL.
2174 	 */
2175 	if (!n || !n->nr_partial)
2176 		return NULL;
2177 
2178 	spin_lock_irqsave(&n->list_lock, flags);
2179 	list_for_each_entry_safe(slab, slab2, &n->partial, slab_list) {
2180 		void *t;
2181 
2182 		if (!pfmemalloc_match(slab, gfpflags))
2183 			continue;
2184 
2185 		t = acquire_slab(s, n, slab, object == NULL);
2186 		if (!t)
2187 			break;
2188 
2189 		if (!object) {
2190 			*ret_slab = slab;
2191 			stat(s, ALLOC_FROM_PARTIAL);
2192 			object = t;
2193 		} else {
2194 			put_cpu_partial(s, slab, 0);
2195 			stat(s, CPU_PARTIAL_NODE);
2196 			partial_slabs++;
2197 		}
2198 #ifdef CONFIG_SLUB_CPU_PARTIAL
2199 		if (!kmem_cache_has_cpu_partial(s)
2200 			|| partial_slabs > s->cpu_partial_slabs / 2)
2201 			break;
2202 #else
2203 		break;
2204 #endif
2205 
2206 	}
2207 	spin_unlock_irqrestore(&n->list_lock, flags);
2208 	return object;
2209 }
2210 
2211 /*
2212  * Get a slab from somewhere. Search in increasing NUMA distances.
2213  */
get_any_partial(struct kmem_cache * s,gfp_t flags,struct slab ** ret_slab)2214 static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
2215 			     struct slab **ret_slab)
2216 {
2217 #ifdef CONFIG_NUMA
2218 	struct zonelist *zonelist;
2219 	struct zoneref *z;
2220 	struct zone *zone;
2221 	enum zone_type highest_zoneidx = gfp_zone(flags);
2222 	void *object;
2223 	unsigned int cpuset_mems_cookie;
2224 
2225 	/*
2226 	 * The defrag ratio allows a configuration of the tradeoffs between
2227 	 * inter node defragmentation and node local allocations. A lower
2228 	 * defrag_ratio increases the tendency to do local allocations
2229 	 * instead of attempting to obtain partial slabs from other nodes.
2230 	 *
2231 	 * If the defrag_ratio is set to 0 then kmalloc() always
2232 	 * returns node local objects. If the ratio is higher then kmalloc()
2233 	 * may return off node objects because partial slabs are obtained
2234 	 * from other nodes and filled up.
2235 	 *
2236 	 * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
2237 	 * (which makes defrag_ratio = 1000) then every (well almost)
2238 	 * allocation will first attempt to defrag slab caches on other nodes.
2239 	 * This means scanning over all nodes to look for partial slabs which
2240 	 * may be expensive if we do it every time we are trying to find a slab
2241 	 * with available objects.
2242 	 */
2243 	if (!s->remote_node_defrag_ratio ||
2244 			get_cycles() % 1024 > s->remote_node_defrag_ratio)
2245 		return NULL;
2246 
2247 	do {
2248 		cpuset_mems_cookie = read_mems_allowed_begin();
2249 		zonelist = node_zonelist(mempolicy_slab_node(), flags);
2250 		for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) {
2251 			struct kmem_cache_node *n;
2252 
2253 			n = get_node(s, zone_to_nid(zone));
2254 
2255 			if (n && cpuset_zone_allowed(zone, flags) &&
2256 					n->nr_partial > s->min_partial) {
2257 				object = get_partial_node(s, n, ret_slab, flags);
2258 				if (object) {
2259 					/*
2260 					 * Don't check read_mems_allowed_retry()
2261 					 * here - if mems_allowed was updated in
2262 					 * parallel, that was a harmless race
2263 					 * between allocation and the cpuset
2264 					 * update
2265 					 */
2266 					return object;
2267 				}
2268 			}
2269 		}
2270 	} while (read_mems_allowed_retry(cpuset_mems_cookie));
2271 #endif	/* CONFIG_NUMA */
2272 	return NULL;
2273 }
2274 
2275 /*
2276  * Get a partial slab, lock it and return it.
2277  */
get_partial(struct kmem_cache * s,gfp_t flags,int node,struct slab ** ret_slab)2278 static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
2279 			 struct slab **ret_slab)
2280 {
2281 	void *object;
2282 	int searchnode = node;
2283 
2284 	if (node == NUMA_NO_NODE)
2285 		searchnode = numa_mem_id();
2286 
2287 	object = get_partial_node(s, get_node(s, searchnode), ret_slab, flags);
2288 	if (object || node != NUMA_NO_NODE)
2289 		return object;
2290 
2291 	return get_any_partial(s, flags, ret_slab);
2292 }
2293 
2294 #ifdef CONFIG_PREEMPTION
2295 /*
2296  * Calculate the next globally unique transaction for disambiguation
2297  * during cmpxchg. The transactions start with the cpu number and are then
2298  * incremented by CONFIG_NR_CPUS.
2299  */
2300 #define TID_STEP  roundup_pow_of_two(CONFIG_NR_CPUS)
2301 #else
2302 /*
2303  * No preemption supported therefore also no need to check for
2304  * different cpus.
2305  */
2306 #define TID_STEP 1
2307 #endif
2308 
next_tid(unsigned long tid)2309 static inline unsigned long next_tid(unsigned long tid)
2310 {
2311 	return tid + TID_STEP;
2312 }
2313 
2314 #ifdef SLUB_DEBUG_CMPXCHG
tid_to_cpu(unsigned long tid)2315 static inline unsigned int tid_to_cpu(unsigned long tid)
2316 {
2317 	return tid % TID_STEP;
2318 }
2319 
tid_to_event(unsigned long tid)2320 static inline unsigned long tid_to_event(unsigned long tid)
2321 {
2322 	return tid / TID_STEP;
2323 }
2324 #endif
2325 
init_tid(int cpu)2326 static inline unsigned int init_tid(int cpu)
2327 {
2328 	return cpu;
2329 }
2330 
note_cmpxchg_failure(const char * n,const struct kmem_cache * s,unsigned long tid)2331 static inline void note_cmpxchg_failure(const char *n,
2332 		const struct kmem_cache *s, unsigned long tid)
2333 {
2334 #ifdef SLUB_DEBUG_CMPXCHG
2335 	unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
2336 
2337 	pr_info("%s %s: cmpxchg redo ", n, s->name);
2338 
2339 #ifdef CONFIG_PREEMPTION
2340 	if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
2341 		pr_warn("due to cpu change %d -> %d\n",
2342 			tid_to_cpu(tid), tid_to_cpu(actual_tid));
2343 	else
2344 #endif
2345 	if (tid_to_event(tid) != tid_to_event(actual_tid))
2346 		pr_warn("due to cpu running other code. Event %ld->%ld\n",
2347 			tid_to_event(tid), tid_to_event(actual_tid));
2348 	else
2349 		pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
2350 			actual_tid, tid, next_tid(tid));
2351 #endif
2352 	stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
2353 }
2354 
init_kmem_cache_cpus(struct kmem_cache * s)2355 static void init_kmem_cache_cpus(struct kmem_cache *s)
2356 {
2357 	int cpu;
2358 	struct kmem_cache_cpu *c;
2359 
2360 	for_each_possible_cpu(cpu) {
2361 		c = per_cpu_ptr(s->cpu_slab, cpu);
2362 		local_lock_init(&c->lock);
2363 		c->tid = init_tid(cpu);
2364 	}
2365 }
2366 
2367 /*
2368  * Finishes removing the cpu slab. Merges cpu's freelist with slab's freelist,
2369  * unfreezes the slabs and puts it on the proper list.
2370  * Assumes the slab has been already safely taken away from kmem_cache_cpu
2371  * by the caller.
2372  */
deactivate_slab(struct kmem_cache * s,struct slab * slab,void * freelist)2373 static void deactivate_slab(struct kmem_cache *s, struct slab *slab,
2374 			    void *freelist)
2375 {
2376 	enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE, M_FULL_NOLIST };
2377 	struct kmem_cache_node *n = get_node(s, slab_nid(slab));
2378 	int free_delta = 0;
2379 	enum slab_modes mode = M_NONE;
2380 	void *nextfree, *freelist_iter, *freelist_tail;
2381 	int tail = DEACTIVATE_TO_HEAD;
2382 	unsigned long flags = 0;
2383 	struct slab new;
2384 	struct slab old;
2385 
2386 	if (slab->freelist) {
2387 		stat(s, DEACTIVATE_REMOTE_FREES);
2388 		tail = DEACTIVATE_TO_TAIL;
2389 	}
2390 
2391 	/*
2392 	 * Stage one: Count the objects on cpu's freelist as free_delta and
2393 	 * remember the last object in freelist_tail for later splicing.
2394 	 */
2395 	freelist_tail = NULL;
2396 	freelist_iter = freelist;
2397 	while (freelist_iter) {
2398 		nextfree = get_freepointer(s, freelist_iter);
2399 
2400 		/*
2401 		 * If 'nextfree' is invalid, it is possible that the object at
2402 		 * 'freelist_iter' is already corrupted.  So isolate all objects
2403 		 * starting at 'freelist_iter' by skipping them.
2404 		 */
2405 		if (freelist_corrupted(s, slab, &freelist_iter, nextfree))
2406 			break;
2407 
2408 		freelist_tail = freelist_iter;
2409 		free_delta++;
2410 
2411 		freelist_iter = nextfree;
2412 	}
2413 
2414 	/*
2415 	 * Stage two: Unfreeze the slab while splicing the per-cpu
2416 	 * freelist to the head of slab's freelist.
2417 	 *
2418 	 * Ensure that the slab is unfrozen while the list presence
2419 	 * reflects the actual number of objects during unfreeze.
2420 	 *
2421 	 * We first perform cmpxchg holding lock and insert to list
2422 	 * when it succeed. If there is mismatch then the slab is not
2423 	 * unfrozen and number of objects in the slab may have changed.
2424 	 * Then release lock and retry cmpxchg again.
2425 	 */
2426 redo:
2427 
2428 	old.freelist = READ_ONCE(slab->freelist);
2429 	old.counters = READ_ONCE(slab->counters);
2430 	VM_BUG_ON(!old.frozen);
2431 
2432 	/* Determine target state of the slab */
2433 	new.counters = old.counters;
2434 	if (freelist_tail) {
2435 		new.inuse -= free_delta;
2436 		set_freepointer(s, freelist_tail, old.freelist);
2437 		new.freelist = freelist;
2438 	} else
2439 		new.freelist = old.freelist;
2440 
2441 	new.frozen = 0;
2442 
2443 	if (!new.inuse && n->nr_partial >= s->min_partial) {
2444 		mode = M_FREE;
2445 	} else if (new.freelist) {
2446 		mode = M_PARTIAL;
2447 		/*
2448 		 * Taking the spinlock removes the possibility that
2449 		 * acquire_slab() will see a slab that is frozen
2450 		 */
2451 		spin_lock_irqsave(&n->list_lock, flags);
2452 	} else if (kmem_cache_debug_flags(s, SLAB_STORE_USER)) {
2453 		mode = M_FULL;
2454 		/*
2455 		 * This also ensures that the scanning of full
2456 		 * slabs from diagnostic functions will not see
2457 		 * any frozen slabs.
2458 		 */
2459 		spin_lock_irqsave(&n->list_lock, flags);
2460 	} else {
2461 		mode = M_FULL_NOLIST;
2462 	}
2463 
2464 
2465 	if (!cmpxchg_double_slab(s, slab,
2466 				old.freelist, old.counters,
2467 				new.freelist, new.counters,
2468 				"unfreezing slab")) {
2469 		if (mode == M_PARTIAL || mode == M_FULL)
2470 			spin_unlock_irqrestore(&n->list_lock, flags);
2471 		goto redo;
2472 	}
2473 
2474 
2475 	if (mode == M_PARTIAL) {
2476 		add_partial(n, slab, tail);
2477 		spin_unlock_irqrestore(&n->list_lock, flags);
2478 		stat(s, tail);
2479 	} else if (mode == M_FREE) {
2480 		stat(s, DEACTIVATE_EMPTY);
2481 		discard_slab(s, slab);
2482 		stat(s, FREE_SLAB);
2483 	} else if (mode == M_FULL) {
2484 		add_full(s, n, slab);
2485 		spin_unlock_irqrestore(&n->list_lock, flags);
2486 		stat(s, DEACTIVATE_FULL);
2487 	} else if (mode == M_FULL_NOLIST) {
2488 		stat(s, DEACTIVATE_FULL);
2489 	}
2490 }
2491 
2492 #ifdef CONFIG_SLUB_CPU_PARTIAL
__unfreeze_partials(struct kmem_cache * s,struct slab * partial_slab)2493 static void __unfreeze_partials(struct kmem_cache *s, struct slab *partial_slab)
2494 {
2495 	struct kmem_cache_node *n = NULL, *n2 = NULL;
2496 	struct slab *slab, *slab_to_discard = NULL;
2497 	unsigned long flags = 0;
2498 
2499 	while (partial_slab) {
2500 		struct slab new;
2501 		struct slab old;
2502 
2503 		slab = partial_slab;
2504 		partial_slab = slab->next;
2505 
2506 		n2 = get_node(s, slab_nid(slab));
2507 		if (n != n2) {
2508 			if (n)
2509 				spin_unlock_irqrestore(&n->list_lock, flags);
2510 
2511 			n = n2;
2512 			spin_lock_irqsave(&n->list_lock, flags);
2513 		}
2514 
2515 		do {
2516 
2517 			old.freelist = slab->freelist;
2518 			old.counters = slab->counters;
2519 			VM_BUG_ON(!old.frozen);
2520 
2521 			new.counters = old.counters;
2522 			new.freelist = old.freelist;
2523 
2524 			new.frozen = 0;
2525 
2526 		} while (!__cmpxchg_double_slab(s, slab,
2527 				old.freelist, old.counters,
2528 				new.freelist, new.counters,
2529 				"unfreezing slab"));
2530 
2531 		if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
2532 			slab->next = slab_to_discard;
2533 			slab_to_discard = slab;
2534 		} else {
2535 			add_partial(n, slab, DEACTIVATE_TO_TAIL);
2536 			stat(s, FREE_ADD_PARTIAL);
2537 		}
2538 	}
2539 
2540 	if (n)
2541 		spin_unlock_irqrestore(&n->list_lock, flags);
2542 
2543 	while (slab_to_discard) {
2544 		slab = slab_to_discard;
2545 		slab_to_discard = slab_to_discard->next;
2546 
2547 		stat(s, DEACTIVATE_EMPTY);
2548 		discard_slab(s, slab);
2549 		stat(s, FREE_SLAB);
2550 	}
2551 }
2552 
2553 /*
2554  * Unfreeze all the cpu partial slabs.
2555  */
unfreeze_partials(struct kmem_cache * s)2556 static void unfreeze_partials(struct kmem_cache *s)
2557 {
2558 	struct slab *partial_slab;
2559 	unsigned long flags;
2560 
2561 	local_lock_irqsave(&s->cpu_slab->lock, flags);
2562 	partial_slab = this_cpu_read(s->cpu_slab->partial);
2563 	this_cpu_write(s->cpu_slab->partial, NULL);
2564 	local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2565 
2566 	if (partial_slab)
2567 		__unfreeze_partials(s, partial_slab);
2568 }
2569 
unfreeze_partials_cpu(struct kmem_cache * s,struct kmem_cache_cpu * c)2570 static void unfreeze_partials_cpu(struct kmem_cache *s,
2571 				  struct kmem_cache_cpu *c)
2572 {
2573 	struct slab *partial_slab;
2574 
2575 	partial_slab = slub_percpu_partial(c);
2576 	c->partial = NULL;
2577 
2578 	if (partial_slab)
2579 		__unfreeze_partials(s, partial_slab);
2580 }
2581 
2582 /*
2583  * Put a slab that was just frozen (in __slab_free|get_partial_node) into a
2584  * partial slab slot if available.
2585  *
2586  * If we did not find a slot then simply move all the partials to the
2587  * per node partial list.
2588  */
put_cpu_partial(struct kmem_cache * s,struct slab * slab,int drain)2589 static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain)
2590 {
2591 	struct slab *oldslab;
2592 	struct slab *slab_to_unfreeze = NULL;
2593 	unsigned long flags;
2594 	int slabs = 0;
2595 
2596 	local_lock_irqsave(&s->cpu_slab->lock, flags);
2597 
2598 	oldslab = this_cpu_read(s->cpu_slab->partial);
2599 
2600 	if (oldslab) {
2601 		if (drain && oldslab->slabs >= s->cpu_partial_slabs) {
2602 			/*
2603 			 * Partial array is full. Move the existing set to the
2604 			 * per node partial list. Postpone the actual unfreezing
2605 			 * outside of the critical section.
2606 			 */
2607 			slab_to_unfreeze = oldslab;
2608 			oldslab = NULL;
2609 		} else {
2610 			slabs = oldslab->slabs;
2611 		}
2612 	}
2613 
2614 	slabs++;
2615 
2616 	slab->slabs = slabs;
2617 	slab->next = oldslab;
2618 
2619 	this_cpu_write(s->cpu_slab->partial, slab);
2620 
2621 	local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2622 
2623 	if (slab_to_unfreeze) {
2624 		__unfreeze_partials(s, slab_to_unfreeze);
2625 		stat(s, CPU_PARTIAL_DRAIN);
2626 	}
2627 }
2628 
2629 #else	/* CONFIG_SLUB_CPU_PARTIAL */
2630 
unfreeze_partials(struct kmem_cache * s)2631 static inline void unfreeze_partials(struct kmem_cache *s) { }
unfreeze_partials_cpu(struct kmem_cache * s,struct kmem_cache_cpu * c)2632 static inline void unfreeze_partials_cpu(struct kmem_cache *s,
2633 				  struct kmem_cache_cpu *c) { }
2634 
2635 #endif	/* CONFIG_SLUB_CPU_PARTIAL */
2636 
flush_slab(struct kmem_cache * s,struct kmem_cache_cpu * c)2637 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
2638 {
2639 	unsigned long flags;
2640 	struct slab *slab;
2641 	void *freelist;
2642 
2643 	local_lock_irqsave(&s->cpu_slab->lock, flags);
2644 
2645 	slab = c->slab;
2646 	freelist = c->freelist;
2647 
2648 	c->slab = NULL;
2649 	c->freelist = NULL;
2650 	c->tid = next_tid(c->tid);
2651 
2652 	local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2653 
2654 	if (slab) {
2655 		deactivate_slab(s, slab, freelist);
2656 		stat(s, CPUSLAB_FLUSH);
2657 	}
2658 }
2659 
__flush_cpu_slab(struct kmem_cache * s,int cpu)2660 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
2661 {
2662 	struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2663 	void *freelist = c->freelist;
2664 	struct slab *slab = c->slab;
2665 
2666 	c->slab = NULL;
2667 	c->freelist = NULL;
2668 	c->tid = next_tid(c->tid);
2669 
2670 	if (slab) {
2671 		deactivate_slab(s, slab, freelist);
2672 		stat(s, CPUSLAB_FLUSH);
2673 	}
2674 
2675 	unfreeze_partials_cpu(s, c);
2676 }
2677 
2678 struct slub_flush_work {
2679 	struct work_struct work;
2680 	struct kmem_cache *s;
2681 	bool skip;
2682 };
2683 
2684 /*
2685  * Flush cpu slab.
2686  *
2687  * Called from CPU work handler with migration disabled.
2688  */
flush_cpu_slab(struct work_struct * w)2689 static void flush_cpu_slab(struct work_struct *w)
2690 {
2691 	struct kmem_cache *s;
2692 	struct kmem_cache_cpu *c;
2693 	struct slub_flush_work *sfw;
2694 
2695 	sfw = container_of(w, struct slub_flush_work, work);
2696 
2697 	s = sfw->s;
2698 	c = this_cpu_ptr(s->cpu_slab);
2699 
2700 	if (c->slab)
2701 		flush_slab(s, c);
2702 
2703 	unfreeze_partials(s);
2704 }
2705 
has_cpu_slab(int cpu,struct kmem_cache * s)2706 static bool has_cpu_slab(int cpu, struct kmem_cache *s)
2707 {
2708 	struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2709 
2710 	return c->slab || slub_percpu_partial(c);
2711 }
2712 
2713 static DEFINE_MUTEX(flush_lock);
2714 static DEFINE_PER_CPU(struct slub_flush_work, slub_flush);
2715 
flush_all_cpus_locked(struct kmem_cache * s)2716 static void flush_all_cpus_locked(struct kmem_cache *s)
2717 {
2718 	struct slub_flush_work *sfw;
2719 	unsigned int cpu;
2720 
2721 	lockdep_assert_cpus_held();
2722 	mutex_lock(&flush_lock);
2723 
2724 	for_each_online_cpu(cpu) {
2725 		sfw = &per_cpu(slub_flush, cpu);
2726 		if (!has_cpu_slab(cpu, s)) {
2727 			sfw->skip = true;
2728 			continue;
2729 		}
2730 		INIT_WORK(&sfw->work, flush_cpu_slab);
2731 		sfw->skip = false;
2732 		sfw->s = s;
2733 		schedule_work_on(cpu, &sfw->work);
2734 	}
2735 
2736 	for_each_online_cpu(cpu) {
2737 		sfw = &per_cpu(slub_flush, cpu);
2738 		if (sfw->skip)
2739 			continue;
2740 		flush_work(&sfw->work);
2741 	}
2742 
2743 	mutex_unlock(&flush_lock);
2744 }
2745 
flush_all(struct kmem_cache * s)2746 static void flush_all(struct kmem_cache *s)
2747 {
2748 	cpus_read_lock();
2749 	flush_all_cpus_locked(s);
2750 	cpus_read_unlock();
2751 }
2752 
2753 /*
2754  * Use the cpu notifier to insure that the cpu slabs are flushed when
2755  * necessary.
2756  */
slub_cpu_dead(unsigned int cpu)2757 static int slub_cpu_dead(unsigned int cpu)
2758 {
2759 	struct kmem_cache *s;
2760 
2761 	mutex_lock(&slab_mutex);
2762 	list_for_each_entry(s, &slab_caches, list)
2763 		__flush_cpu_slab(s, cpu);
2764 	mutex_unlock(&slab_mutex);
2765 	return 0;
2766 }
2767 
2768 /*
2769  * Check if the objects in a per cpu structure fit numa
2770  * locality expectations.
2771  */
node_match(struct slab * slab,int node)2772 static inline int node_match(struct slab *slab, int node)
2773 {
2774 #ifdef CONFIG_NUMA
2775 	if (node != NUMA_NO_NODE && slab_nid(slab) != node)
2776 		return 0;
2777 #endif
2778 	return 1;
2779 }
2780 
2781 #ifdef CONFIG_SLUB_DEBUG
count_free(struct slab * slab)2782 static int count_free(struct slab *slab)
2783 {
2784 	return slab->objects - slab->inuse;
2785 }
2786 
node_nr_objs(struct kmem_cache_node * n)2787 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2788 {
2789 	return atomic_long_read(&n->total_objects);
2790 }
2791 #endif /* CONFIG_SLUB_DEBUG */
2792 
2793 #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
count_partial(struct kmem_cache_node * n,int (* get_count)(struct slab *))2794 static unsigned long count_partial(struct kmem_cache_node *n,
2795 					int (*get_count)(struct slab *))
2796 {
2797 	unsigned long flags;
2798 	unsigned long x = 0;
2799 	struct slab *slab;
2800 
2801 	spin_lock_irqsave(&n->list_lock, flags);
2802 	list_for_each_entry(slab, &n->partial, slab_list)
2803 		x += get_count(slab);
2804 	spin_unlock_irqrestore(&n->list_lock, flags);
2805 	return x;
2806 }
2807 #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
2808 
2809 static noinline void
slab_out_of_memory(struct kmem_cache * s,gfp_t gfpflags,int nid)2810 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2811 {
2812 #ifdef CONFIG_SLUB_DEBUG
2813 	static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2814 				      DEFAULT_RATELIMIT_BURST);
2815 	int node;
2816 	struct kmem_cache_node *n;
2817 
2818 	if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2819 		return;
2820 
2821 	pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2822 		nid, gfpflags, &gfpflags);
2823 	pr_warn("  cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
2824 		s->name, s->object_size, s->size, oo_order(s->oo),
2825 		oo_order(s->min));
2826 
2827 	if (oo_order(s->min) > get_order(s->object_size))
2828 		pr_warn("  %s debugging increased min order, use slub_debug=O to disable.\n",
2829 			s->name);
2830 
2831 	for_each_kmem_cache_node(s, node, n) {
2832 		unsigned long nr_slabs;
2833 		unsigned long nr_objs;
2834 		unsigned long nr_free;
2835 
2836 		nr_free  = count_partial(n, count_free);
2837 		nr_slabs = node_nr_slabs(n);
2838 		nr_objs  = node_nr_objs(n);
2839 
2840 		pr_warn("  node %d: slabs: %ld, objs: %ld, free: %ld\n",
2841 			node, nr_slabs, nr_objs, nr_free);
2842 	}
2843 #endif
2844 }
2845 
pfmemalloc_match(struct slab * slab,gfp_t gfpflags)2846 static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags)
2847 {
2848 	if (unlikely(slab_test_pfmemalloc(slab)))
2849 		return gfp_pfmemalloc_allowed(gfpflags);
2850 
2851 	return true;
2852 }
2853 
2854 /*
2855  * Check the slab->freelist and either transfer the freelist to the
2856  * per cpu freelist or deactivate the slab.
2857  *
2858  * The slab is still frozen if the return value is not NULL.
2859  *
2860  * If this function returns NULL then the slab has been unfrozen.
2861  */
get_freelist(struct kmem_cache * s,struct slab * slab)2862 static inline void *get_freelist(struct kmem_cache *s, struct slab *slab)
2863 {
2864 	struct slab new;
2865 	unsigned long counters;
2866 	void *freelist;
2867 
2868 	lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock));
2869 
2870 	do {
2871 		freelist = slab->freelist;
2872 		counters = slab->counters;
2873 
2874 		new.counters = counters;
2875 		VM_BUG_ON(!new.frozen);
2876 
2877 		new.inuse = slab->objects;
2878 		new.frozen = freelist != NULL;
2879 
2880 	} while (!__cmpxchg_double_slab(s, slab,
2881 		freelist, counters,
2882 		NULL, new.counters,
2883 		"get_freelist"));
2884 
2885 	return freelist;
2886 }
2887 
2888 /*
2889  * Slow path. The lockless freelist is empty or we need to perform
2890  * debugging duties.
2891  *
2892  * Processing is still very fast if new objects have been freed to the
2893  * regular freelist. In that case we simply take over the regular freelist
2894  * as the lockless freelist and zap the regular freelist.
2895  *
2896  * If that is not working then we fall back to the partial lists. We take the
2897  * first element of the freelist as the object to allocate now and move the
2898  * rest of the freelist to the lockless freelist.
2899  *
2900  * And if we were unable to get a new slab from the partial slab lists then
2901  * we need to allocate a new slab. This is the slowest path since it involves
2902  * a call to the page allocator and the setup of a new slab.
2903  *
2904  * Version of __slab_alloc to use when we know that preemption is
2905  * already disabled (which is the case for bulk allocation).
2906  */
___slab_alloc(struct kmem_cache * s,gfp_t gfpflags,int node,unsigned long addr,struct kmem_cache_cpu * c)2907 static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2908 			  unsigned long addr, struct kmem_cache_cpu *c)
2909 {
2910 	void *freelist;
2911 	struct slab *slab;
2912 	unsigned long flags;
2913 
2914 	stat(s, ALLOC_SLOWPATH);
2915 
2916 reread_slab:
2917 
2918 	slab = READ_ONCE(c->slab);
2919 	if (!slab) {
2920 		/*
2921 		 * if the node is not online or has no normal memory, just
2922 		 * ignore the node constraint
2923 		 */
2924 		if (unlikely(node != NUMA_NO_NODE &&
2925 			     !node_isset(node, slab_nodes)))
2926 			node = NUMA_NO_NODE;
2927 		goto new_slab;
2928 	}
2929 redo:
2930 
2931 	if (unlikely(!node_match(slab, node))) {
2932 		/*
2933 		 * same as above but node_match() being false already
2934 		 * implies node != NUMA_NO_NODE
2935 		 */
2936 		if (!node_isset(node, slab_nodes)) {
2937 			node = NUMA_NO_NODE;
2938 		} else {
2939 			stat(s, ALLOC_NODE_MISMATCH);
2940 			goto deactivate_slab;
2941 		}
2942 	}
2943 
2944 	/*
2945 	 * By rights, we should be searching for a slab page that was
2946 	 * PFMEMALLOC but right now, we are losing the pfmemalloc
2947 	 * information when the page leaves the per-cpu allocator
2948 	 */
2949 	if (unlikely(!pfmemalloc_match(slab, gfpflags)))
2950 		goto deactivate_slab;
2951 
2952 	/* must check again c->slab in case we got preempted and it changed */
2953 	local_lock_irqsave(&s->cpu_slab->lock, flags);
2954 	if (unlikely(slab != c->slab)) {
2955 		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2956 		goto reread_slab;
2957 	}
2958 	freelist = c->freelist;
2959 	if (freelist)
2960 		goto load_freelist;
2961 
2962 	freelist = get_freelist(s, slab);
2963 
2964 	if (!freelist) {
2965 		c->slab = NULL;
2966 		c->tid = next_tid(c->tid);
2967 		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2968 		stat(s, DEACTIVATE_BYPASS);
2969 		goto new_slab;
2970 	}
2971 
2972 	stat(s, ALLOC_REFILL);
2973 
2974 load_freelist:
2975 
2976 	lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock));
2977 
2978 	/*
2979 	 * freelist is pointing to the list of objects to be used.
2980 	 * slab is pointing to the slab from which the objects are obtained.
2981 	 * That slab must be frozen for per cpu allocations to work.
2982 	 */
2983 	VM_BUG_ON(!c->slab->frozen);
2984 	c->freelist = get_freepointer(s, freelist);
2985 	c->tid = next_tid(c->tid);
2986 	local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2987 	return freelist;
2988 
2989 deactivate_slab:
2990 
2991 	local_lock_irqsave(&s->cpu_slab->lock, flags);
2992 	if (slab != c->slab) {
2993 		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2994 		goto reread_slab;
2995 	}
2996 	freelist = c->freelist;
2997 	c->slab = NULL;
2998 	c->freelist = NULL;
2999 	c->tid = next_tid(c->tid);
3000 	local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3001 	deactivate_slab(s, slab, freelist);
3002 
3003 new_slab:
3004 
3005 	if (slub_percpu_partial(c)) {
3006 		local_lock_irqsave(&s->cpu_slab->lock, flags);
3007 		if (unlikely(c->slab)) {
3008 			local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3009 			goto reread_slab;
3010 		}
3011 		if (unlikely(!slub_percpu_partial(c))) {
3012 			local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3013 			/* we were preempted and partial list got empty */
3014 			goto new_objects;
3015 		}
3016 
3017 		slab = c->slab = slub_percpu_partial(c);
3018 		slub_set_percpu_partial(c, slab);
3019 		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3020 		stat(s, CPU_PARTIAL_ALLOC);
3021 		goto redo;
3022 	}
3023 
3024 new_objects:
3025 
3026 	freelist = get_partial(s, gfpflags, node, &slab);
3027 	if (freelist)
3028 		goto check_new_slab;
3029 
3030 	slub_put_cpu_ptr(s->cpu_slab);
3031 	slab = new_slab(s, gfpflags, node);
3032 	c = slub_get_cpu_ptr(s->cpu_slab);
3033 
3034 	if (unlikely(!slab)) {
3035 		slab_out_of_memory(s, gfpflags, node);
3036 		return NULL;
3037 	}
3038 
3039 	/*
3040 	 * No other reference to the slab yet so we can
3041 	 * muck around with it freely without cmpxchg
3042 	 */
3043 	freelist = slab->freelist;
3044 	slab->freelist = NULL;
3045 
3046 	stat(s, ALLOC_SLAB);
3047 
3048 check_new_slab:
3049 
3050 	if (kmem_cache_debug(s)) {
3051 		if (!alloc_debug_processing(s, slab, freelist, addr)) {
3052 			/* Slab failed checks. Next slab needed */
3053 			goto new_slab;
3054 		} else {
3055 			/*
3056 			 * For debug case, we don't load freelist so that all
3057 			 * allocations go through alloc_debug_processing()
3058 			 */
3059 			goto return_single;
3060 		}
3061 	}
3062 
3063 	if (unlikely(!pfmemalloc_match(slab, gfpflags)))
3064 		/*
3065 		 * For !pfmemalloc_match() case we don't load freelist so that
3066 		 * we don't make further mismatched allocations easier.
3067 		 */
3068 		goto return_single;
3069 
3070 retry_load_slab:
3071 
3072 	local_lock_irqsave(&s->cpu_slab->lock, flags);
3073 	if (unlikely(c->slab)) {
3074 		void *flush_freelist = c->freelist;
3075 		struct slab *flush_slab = c->slab;
3076 
3077 		c->slab = NULL;
3078 		c->freelist = NULL;
3079 		c->tid = next_tid(c->tid);
3080 
3081 		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3082 
3083 		deactivate_slab(s, flush_slab, flush_freelist);
3084 
3085 		stat(s, CPUSLAB_FLUSH);
3086 
3087 		goto retry_load_slab;
3088 	}
3089 	c->slab = slab;
3090 
3091 	goto load_freelist;
3092 
3093 return_single:
3094 
3095 	deactivate_slab(s, slab, get_freepointer(s, freelist));
3096 	return freelist;
3097 }
3098 
3099 /*
3100  * A wrapper for ___slab_alloc() for contexts where preemption is not yet
3101  * disabled. Compensates for possible cpu changes by refetching the per cpu area
3102  * pointer.
3103  */
__slab_alloc(struct kmem_cache * s,gfp_t gfpflags,int node,unsigned long addr,struct kmem_cache_cpu * c)3104 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
3105 			  unsigned long addr, struct kmem_cache_cpu *c)
3106 {
3107 	void *p;
3108 
3109 #ifdef CONFIG_PREEMPT_COUNT
3110 	/*
3111 	 * We may have been preempted and rescheduled on a different
3112 	 * cpu before disabling preemption. Need to reload cpu area
3113 	 * pointer.
3114 	 */
3115 	c = slub_get_cpu_ptr(s->cpu_slab);
3116 #endif
3117 
3118 	p = ___slab_alloc(s, gfpflags, node, addr, c);
3119 #ifdef CONFIG_PREEMPT_COUNT
3120 	slub_put_cpu_ptr(s->cpu_slab);
3121 #endif
3122 	return p;
3123 }
3124 
3125 /*
3126  * If the object has been wiped upon free, make sure it's fully initialized by
3127  * zeroing out freelist pointer.
3128  */
maybe_wipe_obj_freeptr(struct kmem_cache * s,void * obj)3129 static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s,
3130 						   void *obj)
3131 {
3132 	if (unlikely(slab_want_init_on_free(s)) && obj)
3133 		memset((void *)((char *)kasan_reset_tag(obj) + s->offset),
3134 			0, sizeof(void *));
3135 }
3136 
3137 /*
3138  * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
3139  * have the fastpath folded into their functions. So no function call
3140  * overhead for requests that can be satisfied on the fastpath.
3141  *
3142  * The fastpath works by first checking if the lockless freelist can be used.
3143  * If not then __slab_alloc is called for slow processing.
3144  *
3145  * Otherwise we can simply pick the next object from the lockless free list.
3146  */
slab_alloc_node(struct kmem_cache * s,struct list_lru * lru,gfp_t gfpflags,int node,unsigned long addr,size_t orig_size)3147 static __always_inline void *slab_alloc_node(struct kmem_cache *s, struct list_lru *lru,
3148 		gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
3149 {
3150 	void *object;
3151 	struct kmem_cache_cpu *c;
3152 	struct slab *slab;
3153 	unsigned long tid;
3154 	struct obj_cgroup *objcg = NULL;
3155 	bool init = false;
3156 
3157 	s = slab_pre_alloc_hook(s, lru, &objcg, 1, gfpflags);
3158 	if (!s)
3159 		return NULL;
3160 
3161 	object = kfence_alloc(s, orig_size, gfpflags);
3162 	if (unlikely(object))
3163 		goto out;
3164 
3165 redo:
3166 	/*
3167 	 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
3168 	 * enabled. We may switch back and forth between cpus while
3169 	 * reading from one cpu area. That does not matter as long
3170 	 * as we end up on the original cpu again when doing the cmpxchg.
3171 	 *
3172 	 * We must guarantee that tid and kmem_cache_cpu are retrieved on the
3173 	 * same cpu. We read first the kmem_cache_cpu pointer and use it to read
3174 	 * the tid. If we are preempted and switched to another cpu between the
3175 	 * two reads, it's OK as the two are still associated with the same cpu
3176 	 * and cmpxchg later will validate the cpu.
3177 	 */
3178 	c = raw_cpu_ptr(s->cpu_slab);
3179 	tid = READ_ONCE(c->tid);
3180 
3181 	/*
3182 	 * Irqless object alloc/free algorithm used here depends on sequence
3183 	 * of fetching cpu_slab's data. tid should be fetched before anything
3184 	 * on c to guarantee that object and slab associated with previous tid
3185 	 * won't be used with current tid. If we fetch tid first, object and
3186 	 * slab could be one associated with next tid and our alloc/free
3187 	 * request will be failed. In this case, we will retry. So, no problem.
3188 	 */
3189 	barrier();
3190 
3191 	/*
3192 	 * The transaction ids are globally unique per cpu and per operation on
3193 	 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
3194 	 * occurs on the right processor and that there was no operation on the
3195 	 * linked list in between.
3196 	 */
3197 
3198 	object = c->freelist;
3199 	slab = c->slab;
3200 	/*
3201 	 * We cannot use the lockless fastpath on PREEMPT_RT because if a
3202 	 * slowpath has taken the local_lock_irqsave(), it is not protected
3203 	 * against a fast path operation in an irq handler. So we need to take
3204 	 * the slow path which uses local_lock. It is still relatively fast if
3205 	 * there is a suitable cpu freelist.
3206 	 */
3207 	if (IS_ENABLED(CONFIG_PREEMPT_RT) ||
3208 	    unlikely(!object || !slab || !node_match(slab, node))) {
3209 		object = __slab_alloc(s, gfpflags, node, addr, c);
3210 	} else {
3211 		void *next_object = get_freepointer_safe(s, object);
3212 
3213 		/*
3214 		 * The cmpxchg will only match if there was no additional
3215 		 * operation and if we are on the right processor.
3216 		 *
3217 		 * The cmpxchg does the following atomically (without lock
3218 		 * semantics!)
3219 		 * 1. Relocate first pointer to the current per cpu area.
3220 		 * 2. Verify that tid and freelist have not been changed
3221 		 * 3. If they were not changed replace tid and freelist
3222 		 *
3223 		 * Since this is without lock semantics the protection is only
3224 		 * against code executing on this cpu *not* from access by
3225 		 * other cpus.
3226 		 */
3227 		if (unlikely(!this_cpu_cmpxchg_double(
3228 				s->cpu_slab->freelist, s->cpu_slab->tid,
3229 				object, tid,
3230 				next_object, next_tid(tid)))) {
3231 
3232 			note_cmpxchg_failure("slab_alloc", s, tid);
3233 			goto redo;
3234 		}
3235 		prefetch_freepointer(s, next_object);
3236 		stat(s, ALLOC_FASTPATH);
3237 	}
3238 
3239 	maybe_wipe_obj_freeptr(s, object);
3240 	init = slab_want_init_on_alloc(gfpflags, s);
3241 
3242 out:
3243 	slab_post_alloc_hook(s, objcg, gfpflags, 1, &object, init);
3244 
3245 	return object;
3246 }
3247 
slab_alloc(struct kmem_cache * s,struct list_lru * lru,gfp_t gfpflags,unsigned long addr,size_t orig_size)3248 static __always_inline void *slab_alloc(struct kmem_cache *s, struct list_lru *lru,
3249 		gfp_t gfpflags, unsigned long addr, size_t orig_size)
3250 {
3251 	return slab_alloc_node(s, lru, gfpflags, NUMA_NO_NODE, addr, orig_size);
3252 }
3253 
3254 static __always_inline
__kmem_cache_alloc_lru(struct kmem_cache * s,struct list_lru * lru,gfp_t gfpflags)3255 void *__kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
3256 			     gfp_t gfpflags)
3257 {
3258 	void *ret = slab_alloc(s, lru, gfpflags, _RET_IP_, s->object_size);
3259 
3260 	trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
3261 				s->size, gfpflags);
3262 
3263 	return ret;
3264 }
3265 
kmem_cache_alloc(struct kmem_cache * s,gfp_t gfpflags)3266 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
3267 {
3268 	return __kmem_cache_alloc_lru(s, NULL, gfpflags);
3269 }
3270 EXPORT_SYMBOL(kmem_cache_alloc);
3271 
kmem_cache_alloc_lru(struct kmem_cache * s,struct list_lru * lru,gfp_t gfpflags)3272 void *kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
3273 			   gfp_t gfpflags)
3274 {
3275 	return __kmem_cache_alloc_lru(s, lru, gfpflags);
3276 }
3277 EXPORT_SYMBOL(kmem_cache_alloc_lru);
3278 
3279 #ifdef CONFIG_TRACING
kmem_cache_alloc_trace(struct kmem_cache * s,gfp_t gfpflags,size_t size)3280 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
3281 {
3282 	void *ret = slab_alloc(s, NULL, gfpflags, _RET_IP_, size);
3283 	trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
3284 	ret = kasan_kmalloc(s, ret, size, gfpflags);
3285 	return ret;
3286 }
3287 EXPORT_SYMBOL(kmem_cache_alloc_trace);
3288 #endif
3289 
3290 #ifdef CONFIG_NUMA
kmem_cache_alloc_node(struct kmem_cache * s,gfp_t gfpflags,int node)3291 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
3292 {
3293 	void *ret = slab_alloc_node(s, NULL, gfpflags, node, _RET_IP_, s->object_size);
3294 
3295 	trace_kmem_cache_alloc_node(_RET_IP_, ret,
3296 				    s->object_size, s->size, gfpflags, node);
3297 
3298 	return ret;
3299 }
3300 EXPORT_SYMBOL(kmem_cache_alloc_node);
3301 
3302 #ifdef CONFIG_TRACING
kmem_cache_alloc_node_trace(struct kmem_cache * s,gfp_t gfpflags,int node,size_t size)3303 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
3304 				    gfp_t gfpflags,
3305 				    int node, size_t size)
3306 {
3307 	void *ret = slab_alloc_node(s, NULL, gfpflags, node, _RET_IP_, size);
3308 
3309 	trace_kmalloc_node(_RET_IP_, ret,
3310 			   size, s->size, gfpflags, node);
3311 
3312 	ret = kasan_kmalloc(s, ret, size, gfpflags);
3313 	return ret;
3314 }
3315 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
3316 #endif
3317 #endif	/* CONFIG_NUMA */
3318 
3319 /*
3320  * Slow path handling. This may still be called frequently since objects
3321  * have a longer lifetime than the cpu slabs in most processing loads.
3322  *
3323  * So we still attempt to reduce cache line usage. Just take the slab
3324  * lock and free the item. If there is no additional partial slab
3325  * handling required then we can return immediately.
3326  */
__slab_free(struct kmem_cache * s,struct slab * slab,void * head,void * tail,int cnt,unsigned long addr)3327 static void __slab_free(struct kmem_cache *s, struct slab *slab,
3328 			void *head, void *tail, int cnt,
3329 			unsigned long addr)
3330 
3331 {
3332 	void *prior;
3333 	int was_frozen;
3334 	struct slab new;
3335 	unsigned long counters;
3336 	struct kmem_cache_node *n = NULL;
3337 	unsigned long flags;
3338 
3339 	stat(s, FREE_SLOWPATH);
3340 
3341 	if (kfence_free(head))
3342 		return;
3343 
3344 	if (kmem_cache_debug(s) &&
3345 	    !free_debug_processing(s, slab, head, tail, cnt, addr))
3346 		return;
3347 
3348 	do {
3349 		if (unlikely(n)) {
3350 			spin_unlock_irqrestore(&n->list_lock, flags);
3351 			n = NULL;
3352 		}
3353 		prior = slab->freelist;
3354 		counters = slab->counters;
3355 		set_freepointer(s, tail, prior);
3356 		new.counters = counters;
3357 		was_frozen = new.frozen;
3358 		new.inuse -= cnt;
3359 		if ((!new.inuse || !prior) && !was_frozen) {
3360 
3361 			if (kmem_cache_has_cpu_partial(s) && !prior) {
3362 
3363 				/*
3364 				 * Slab was on no list before and will be
3365 				 * partially empty
3366 				 * We can defer the list move and instead
3367 				 * freeze it.
3368 				 */
3369 				new.frozen = 1;
3370 
3371 			} else { /* Needs to be taken off a list */
3372 
3373 				n = get_node(s, slab_nid(slab));
3374 				/*
3375 				 * Speculatively acquire the list_lock.
3376 				 * If the cmpxchg does not succeed then we may
3377 				 * drop the list_lock without any processing.
3378 				 *
3379 				 * Otherwise the list_lock will synchronize with
3380 				 * other processors updating the list of slabs.
3381 				 */
3382 				spin_lock_irqsave(&n->list_lock, flags);
3383 
3384 			}
3385 		}
3386 
3387 	} while (!cmpxchg_double_slab(s, slab,
3388 		prior, counters,
3389 		head, new.counters,
3390 		"__slab_free"));
3391 
3392 	if (likely(!n)) {
3393 
3394 		if (likely(was_frozen)) {
3395 			/*
3396 			 * The list lock was not taken therefore no list
3397 			 * activity can be necessary.
3398 			 */
3399 			stat(s, FREE_FROZEN);
3400 		} else if (new.frozen) {
3401 			/*
3402 			 * If we just froze the slab then put it onto the
3403 			 * per cpu partial list.
3404 			 */
3405 			put_cpu_partial(s, slab, 1);
3406 			stat(s, CPU_PARTIAL_FREE);
3407 		}
3408 
3409 		return;
3410 	}
3411 
3412 	if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
3413 		goto slab_empty;
3414 
3415 	/*
3416 	 * Objects left in the slab. If it was not on the partial list before
3417 	 * then add it.
3418 	 */
3419 	if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
3420 		remove_full(s, n, slab);
3421 		add_partial(n, slab, DEACTIVATE_TO_TAIL);
3422 		stat(s, FREE_ADD_PARTIAL);
3423 	}
3424 	spin_unlock_irqrestore(&n->list_lock, flags);
3425 	return;
3426 
3427 slab_empty:
3428 	if (prior) {
3429 		/*
3430 		 * Slab on the partial list.
3431 		 */
3432 		remove_partial(n, slab);
3433 		stat(s, FREE_REMOVE_PARTIAL);
3434 	} else {
3435 		/* Slab must be on the full list */
3436 		remove_full(s, n, slab);
3437 	}
3438 
3439 	spin_unlock_irqrestore(&n->list_lock, flags);
3440 	stat(s, FREE_SLAB);
3441 	discard_slab(s, slab);
3442 }
3443 
3444 /*
3445  * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
3446  * can perform fastpath freeing without additional function calls.
3447  *
3448  * The fastpath is only possible if we are freeing to the current cpu slab
3449  * of this processor. This typically the case if we have just allocated
3450  * the item before.
3451  *
3452  * If fastpath is not possible then fall back to __slab_free where we deal
3453  * with all sorts of special processing.
3454  *
3455  * Bulk free of a freelist with several objects (all pointing to the
3456  * same slab) possible by specifying head and tail ptr, plus objects
3457  * count (cnt). Bulk free indicated by tail pointer being set.
3458  */
do_slab_free(struct kmem_cache * s,struct slab * slab,void * head,void * tail,int cnt,unsigned long addr)3459 static __always_inline void do_slab_free(struct kmem_cache *s,
3460 				struct slab *slab, void *head, void *tail,
3461 				int cnt, unsigned long addr)
3462 {
3463 	void *tail_obj = tail ? : head;
3464 	struct kmem_cache_cpu *c;
3465 	unsigned long tid;
3466 
3467 	/* memcg_slab_free_hook() is already called for bulk free. */
3468 	if (!tail)
3469 		memcg_slab_free_hook(s, &head, 1);
3470 redo:
3471 	/*
3472 	 * Determine the currently cpus per cpu slab.
3473 	 * The cpu may change afterward. However that does not matter since
3474 	 * data is retrieved via this pointer. If we are on the same cpu
3475 	 * during the cmpxchg then the free will succeed.
3476 	 */
3477 	c = raw_cpu_ptr(s->cpu_slab);
3478 	tid = READ_ONCE(c->tid);
3479 
3480 	/* Same with comment on barrier() in slab_alloc_node() */
3481 	barrier();
3482 
3483 	if (likely(slab == c->slab)) {
3484 #ifndef CONFIG_PREEMPT_RT
3485 		void **freelist = READ_ONCE(c->freelist);
3486 
3487 		set_freepointer(s, tail_obj, freelist);
3488 
3489 		if (unlikely(!this_cpu_cmpxchg_double(
3490 				s->cpu_slab->freelist, s->cpu_slab->tid,
3491 				freelist, tid,
3492 				head, next_tid(tid)))) {
3493 
3494 			note_cmpxchg_failure("slab_free", s, tid);
3495 			goto redo;
3496 		}
3497 #else /* CONFIG_PREEMPT_RT */
3498 		/*
3499 		 * We cannot use the lockless fastpath on PREEMPT_RT because if
3500 		 * a slowpath has taken the local_lock_irqsave(), it is not
3501 		 * protected against a fast path operation in an irq handler. So
3502 		 * we need to take the local_lock. We shouldn't simply defer to
3503 		 * __slab_free() as that wouldn't use the cpu freelist at all.
3504 		 */
3505 		void **freelist;
3506 
3507 		local_lock(&s->cpu_slab->lock);
3508 		c = this_cpu_ptr(s->cpu_slab);
3509 		if (unlikely(slab != c->slab)) {
3510 			local_unlock(&s->cpu_slab->lock);
3511 			goto redo;
3512 		}
3513 		tid = c->tid;
3514 		freelist = c->freelist;
3515 
3516 		set_freepointer(s, tail_obj, freelist);
3517 		c->freelist = head;
3518 		c->tid = next_tid(tid);
3519 
3520 		local_unlock(&s->cpu_slab->lock);
3521 #endif
3522 		stat(s, FREE_FASTPATH);
3523 	} else
3524 		__slab_free(s, slab, head, tail_obj, cnt, addr);
3525 
3526 }
3527 
slab_free(struct kmem_cache * s,struct slab * slab,void * head,void * tail,int cnt,unsigned long addr)3528 static __always_inline void slab_free(struct kmem_cache *s, struct slab *slab,
3529 				      void *head, void *tail, int cnt,
3530 				      unsigned long addr)
3531 {
3532 	/*
3533 	 * With KASAN enabled slab_free_freelist_hook modifies the freelist
3534 	 * to remove objects, whose reuse must be delayed.
3535 	 */
3536 	if (slab_free_freelist_hook(s, &head, &tail, &cnt))
3537 		do_slab_free(s, slab, head, tail, cnt, addr);
3538 }
3539 
3540 #ifdef CONFIG_KASAN_GENERIC
___cache_free(struct kmem_cache * cache,void * x,unsigned long addr)3541 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
3542 {
3543 	do_slab_free(cache, virt_to_slab(x), x, NULL, 1, addr);
3544 }
3545 #endif
3546 
kmem_cache_free(struct kmem_cache * s,void * x)3547 void kmem_cache_free(struct kmem_cache *s, void *x)
3548 {
3549 	s = cache_from_obj(s, x);
3550 	if (!s)
3551 		return;
3552 	trace_kmem_cache_free(_RET_IP_, x, s->name);
3553 	slab_free(s, virt_to_slab(x), x, NULL, 1, _RET_IP_);
3554 }
3555 EXPORT_SYMBOL(kmem_cache_free);
3556 
3557 struct detached_freelist {
3558 	struct slab *slab;
3559 	void *tail;
3560 	void *freelist;
3561 	int cnt;
3562 	struct kmem_cache *s;
3563 };
3564 
free_large_kmalloc(struct folio * folio,void * object)3565 static inline void free_large_kmalloc(struct folio *folio, void *object)
3566 {
3567 	unsigned int order = folio_order(folio);
3568 
3569 	if (WARN_ON_ONCE(order == 0))
3570 		pr_warn_once("object pointer: 0x%p\n", object);
3571 
3572 	kfree_hook(object);
3573 	mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B,
3574 			      -(PAGE_SIZE << order));
3575 	__free_pages(folio_page(folio, 0), order);
3576 }
3577 
3578 /*
3579  * This function progressively scans the array with free objects (with
3580  * a limited look ahead) and extract objects belonging to the same
3581  * slab.  It builds a detached freelist directly within the given
3582  * slab/objects.  This can happen without any need for
3583  * synchronization, because the objects are owned by running process.
3584  * The freelist is build up as a single linked list in the objects.
3585  * The idea is, that this detached freelist can then be bulk
3586  * transferred to the real freelist(s), but only requiring a single
3587  * synchronization primitive.  Look ahead in the array is limited due
3588  * to performance reasons.
3589  */
3590 static inline
build_detached_freelist(struct kmem_cache * s,size_t size,void ** p,struct detached_freelist * df)3591 int build_detached_freelist(struct kmem_cache *s, size_t size,
3592 			    void **p, struct detached_freelist *df)
3593 {
3594 	size_t first_skipped_index = 0;
3595 	int lookahead = 3;
3596 	void *object;
3597 	struct folio *folio;
3598 	struct slab *slab;
3599 
3600 	/* Always re-init detached_freelist */
3601 	df->slab = NULL;
3602 
3603 	do {
3604 		object = p[--size];
3605 		/* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
3606 	} while (!object && size);
3607 
3608 	if (!object)
3609 		return 0;
3610 
3611 	folio = virt_to_folio(object);
3612 	if (!s) {
3613 		/* Handle kalloc'ed objects */
3614 		if (unlikely(!folio_test_slab(folio))) {
3615 			free_large_kmalloc(folio, object);
3616 			p[size] = NULL; /* mark object processed */
3617 			return size;
3618 		}
3619 		/* Derive kmem_cache from object */
3620 		slab = folio_slab(folio);
3621 		df->s = slab->slab_cache;
3622 	} else {
3623 		slab = folio_slab(folio);
3624 		df->s = cache_from_obj(s, object); /* Support for memcg */
3625 	}
3626 
3627 	if (is_kfence_address(object)) {
3628 		slab_free_hook(df->s, object, false);
3629 		__kfence_free(object);
3630 		p[size] = NULL; /* mark object processed */
3631 		return size;
3632 	}
3633 
3634 	/* Start new detached freelist */
3635 	df->slab = slab;
3636 	set_freepointer(df->s, object, NULL);
3637 	df->tail = object;
3638 	df->freelist = object;
3639 	p[size] = NULL; /* mark object processed */
3640 	df->cnt = 1;
3641 
3642 	while (size) {
3643 		object = p[--size];
3644 		if (!object)
3645 			continue; /* Skip processed objects */
3646 
3647 		/* df->slab is always set at this point */
3648 		if (df->slab == virt_to_slab(object)) {
3649 			/* Opportunity build freelist */
3650 			set_freepointer(df->s, object, df->freelist);
3651 			df->freelist = object;
3652 			df->cnt++;
3653 			p[size] = NULL; /* mark object processed */
3654 
3655 			continue;
3656 		}
3657 
3658 		/* Limit look ahead search */
3659 		if (!--lookahead)
3660 			break;
3661 
3662 		if (!first_skipped_index)
3663 			first_skipped_index = size + 1;
3664 	}
3665 
3666 	return first_skipped_index;
3667 }
3668 
3669 /* Note that interrupts must be enabled when calling this function. */
kmem_cache_free_bulk(struct kmem_cache * s,size_t size,void ** p)3670 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
3671 {
3672 	if (WARN_ON(!size))
3673 		return;
3674 
3675 	memcg_slab_free_hook(s, p, size);
3676 	do {
3677 		struct detached_freelist df;
3678 
3679 		size = build_detached_freelist(s, size, p, &df);
3680 		if (!df.slab)
3681 			continue;
3682 
3683 		slab_free(df.s, df.slab, df.freelist, df.tail, df.cnt, _RET_IP_);
3684 	} while (likely(size));
3685 }
3686 EXPORT_SYMBOL(kmem_cache_free_bulk);
3687 
3688 /* Note that interrupts must be enabled when calling this function. */
kmem_cache_alloc_bulk(struct kmem_cache * s,gfp_t flags,size_t size,void ** p)3689 int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3690 			  void **p)
3691 {
3692 	struct kmem_cache_cpu *c;
3693 	int i;
3694 	struct obj_cgroup *objcg = NULL;
3695 
3696 	/* memcg and kmem_cache debug support */
3697 	s = slab_pre_alloc_hook(s, NULL, &objcg, size, flags);
3698 	if (unlikely(!s))
3699 		return false;
3700 	/*
3701 	 * Drain objects in the per cpu slab, while disabling local
3702 	 * IRQs, which protects against PREEMPT and interrupts
3703 	 * handlers invoking normal fastpath.
3704 	 */
3705 	c = slub_get_cpu_ptr(s->cpu_slab);
3706 	local_lock_irq(&s->cpu_slab->lock);
3707 
3708 	for (i = 0; i < size; i++) {
3709 		void *object = kfence_alloc(s, s->object_size, flags);
3710 
3711 		if (unlikely(object)) {
3712 			p[i] = object;
3713 			continue;
3714 		}
3715 
3716 		object = c->freelist;
3717 		if (unlikely(!object)) {
3718 			/*
3719 			 * We may have removed an object from c->freelist using
3720 			 * the fastpath in the previous iteration; in that case,
3721 			 * c->tid has not been bumped yet.
3722 			 * Since ___slab_alloc() may reenable interrupts while
3723 			 * allocating memory, we should bump c->tid now.
3724 			 */
3725 			c->tid = next_tid(c->tid);
3726 
3727 			local_unlock_irq(&s->cpu_slab->lock);
3728 
3729 			/*
3730 			 * Invoking slow path likely have side-effect
3731 			 * of re-populating per CPU c->freelist
3732 			 */
3733 			p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
3734 					    _RET_IP_, c);
3735 			if (unlikely(!p[i]))
3736 				goto error;
3737 
3738 			c = this_cpu_ptr(s->cpu_slab);
3739 			maybe_wipe_obj_freeptr(s, p[i]);
3740 
3741 			local_lock_irq(&s->cpu_slab->lock);
3742 
3743 			continue; /* goto for-loop */
3744 		}
3745 		c->freelist = get_freepointer(s, object);
3746 		p[i] = object;
3747 		maybe_wipe_obj_freeptr(s, p[i]);
3748 	}
3749 	c->tid = next_tid(c->tid);
3750 	local_unlock_irq(&s->cpu_slab->lock);
3751 	slub_put_cpu_ptr(s->cpu_slab);
3752 
3753 	/*
3754 	 * memcg and kmem_cache debug support and memory initialization.
3755 	 * Done outside of the IRQ disabled fastpath loop.
3756 	 */
3757 	slab_post_alloc_hook(s, objcg, flags, size, p,
3758 				slab_want_init_on_alloc(flags, s));
3759 	return i;
3760 error:
3761 	slub_put_cpu_ptr(s->cpu_slab);
3762 	slab_post_alloc_hook(s, objcg, flags, i, p, false);
3763 	__kmem_cache_free_bulk(s, i, p);
3764 	return 0;
3765 }
3766 EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3767 
3768 
3769 /*
3770  * Object placement in a slab is made very easy because we always start at
3771  * offset 0. If we tune the size of the object to the alignment then we can
3772  * get the required alignment by putting one properly sized object after
3773  * another.
3774  *
3775  * Notice that the allocation order determines the sizes of the per cpu
3776  * caches. Each processor has always one slab available for allocations.
3777  * Increasing the allocation order reduces the number of times that slabs
3778  * must be moved on and off the partial lists and is therefore a factor in
3779  * locking overhead.
3780  */
3781 
3782 /*
3783  * Minimum / Maximum order of slab pages. This influences locking overhead
3784  * and slab fragmentation. A higher order reduces the number of partial slabs
3785  * and increases the number of allocations possible without having to
3786  * take the list_lock.
3787  */
3788 static unsigned int slub_min_order;
3789 static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
3790 static unsigned int slub_min_objects;
3791 
3792 /*
3793  * Calculate the order of allocation given an slab object size.
3794  *
3795  * The order of allocation has significant impact on performance and other
3796  * system components. Generally order 0 allocations should be preferred since
3797  * order 0 does not cause fragmentation in the page allocator. Larger objects
3798  * be problematic to put into order 0 slabs because there may be too much
3799  * unused space left. We go to a higher order if more than 1/16th of the slab
3800  * would be wasted.
3801  *
3802  * In order to reach satisfactory performance we must ensure that a minimum
3803  * number of objects is in one slab. Otherwise we may generate too much
3804  * activity on the partial lists which requires taking the list_lock. This is
3805  * less a concern for large slabs though which are rarely used.
3806  *
3807  * slub_max_order specifies the order where we begin to stop considering the
3808  * number of objects in a slab as critical. If we reach slub_max_order then
3809  * we try to keep the page order as low as possible. So we accept more waste
3810  * of space in favor of a small page order.
3811  *
3812  * Higher order allocations also allow the placement of more objects in a
3813  * slab and thereby reduce object handling overhead. If the user has
3814  * requested a higher minimum order then we start with that one instead of
3815  * the smallest order which will fit the object.
3816  */
calc_slab_order(unsigned int size,unsigned int min_objects,unsigned int max_order,unsigned int fract_leftover)3817 static inline unsigned int calc_slab_order(unsigned int size,
3818 		unsigned int min_objects, unsigned int max_order,
3819 		unsigned int fract_leftover)
3820 {
3821 	unsigned int min_order = slub_min_order;
3822 	unsigned int order;
3823 
3824 	if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
3825 		return get_order(size * MAX_OBJS_PER_PAGE) - 1;
3826 
3827 	for (order = max(min_order, (unsigned int)get_order(min_objects * size));
3828 			order <= max_order; order++) {
3829 
3830 		unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
3831 		unsigned int rem;
3832 
3833 		rem = slab_size % size;
3834 
3835 		if (rem <= slab_size / fract_leftover)
3836 			break;
3837 	}
3838 
3839 	return order;
3840 }
3841 
calculate_order(unsigned int size)3842 static inline int calculate_order(unsigned int size)
3843 {
3844 	unsigned int order;
3845 	unsigned int min_objects;
3846 	unsigned int max_objects;
3847 	unsigned int nr_cpus;
3848 
3849 	/*
3850 	 * Attempt to find best configuration for a slab. This
3851 	 * works by first attempting to generate a layout with
3852 	 * the best configuration and backing off gradually.
3853 	 *
3854 	 * First we increase the acceptable waste in a slab. Then
3855 	 * we reduce the minimum objects required in a slab.
3856 	 */
3857 	min_objects = slub_min_objects;
3858 	if (!min_objects) {
3859 		/*
3860 		 * Some architectures will only update present cpus when
3861 		 * onlining them, so don't trust the number if it's just 1. But
3862 		 * we also don't want to use nr_cpu_ids always, as on some other
3863 		 * architectures, there can be many possible cpus, but never
3864 		 * onlined. Here we compromise between trying to avoid too high
3865 		 * order on systems that appear larger than they are, and too
3866 		 * low order on systems that appear smaller than they are.
3867 		 */
3868 		nr_cpus = num_present_cpus();
3869 		if (nr_cpus <= 1)
3870 			nr_cpus = nr_cpu_ids;
3871 		min_objects = 4 * (fls(nr_cpus) + 1);
3872 	}
3873 	max_objects = order_objects(slub_max_order, size);
3874 	min_objects = min(min_objects, max_objects);
3875 
3876 	while (min_objects > 1) {
3877 		unsigned int fraction;
3878 
3879 		fraction = 16;
3880 		while (fraction >= 4) {
3881 			order = calc_slab_order(size, min_objects,
3882 					slub_max_order, fraction);
3883 			if (order <= slub_max_order)
3884 				return order;
3885 			fraction /= 2;
3886 		}
3887 		min_objects--;
3888 	}
3889 
3890 	/*
3891 	 * We were unable to place multiple objects in a slab. Now
3892 	 * lets see if we can place a single object there.
3893 	 */
3894 	order = calc_slab_order(size, 1, slub_max_order, 1);
3895 	if (order <= slub_max_order)
3896 		return order;
3897 
3898 	/*
3899 	 * Doh this slab cannot be placed using slub_max_order.
3900 	 */
3901 	order = calc_slab_order(size, 1, MAX_ORDER, 1);
3902 	if (order < MAX_ORDER)
3903 		return order;
3904 	return -ENOSYS;
3905 }
3906 
3907 static void
init_kmem_cache_node(struct kmem_cache_node * n)3908 init_kmem_cache_node(struct kmem_cache_node *n)
3909 {
3910 	n->nr_partial = 0;
3911 	spin_lock_init(&n->list_lock);
3912 	INIT_LIST_HEAD(&n->partial);
3913 #ifdef CONFIG_SLUB_DEBUG
3914 	atomic_long_set(&n->nr_slabs, 0);
3915 	atomic_long_set(&n->total_objects, 0);
3916 	INIT_LIST_HEAD(&n->full);
3917 #endif
3918 }
3919 
alloc_kmem_cache_cpus(struct kmem_cache * s)3920 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
3921 {
3922 	BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
3923 			KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
3924 
3925 	/*
3926 	 * Must align to double word boundary for the double cmpxchg
3927 	 * instructions to work; see __pcpu_double_call_return_bool().
3928 	 */
3929 	s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3930 				     2 * sizeof(void *));
3931 
3932 	if (!s->cpu_slab)
3933 		return 0;
3934 
3935 	init_kmem_cache_cpus(s);
3936 
3937 	return 1;
3938 }
3939 
3940 static struct kmem_cache *kmem_cache_node;
3941 
3942 /*
3943  * No kmalloc_node yet so do it by hand. We know that this is the first
3944  * slab on the node for this slabcache. There are no concurrent accesses
3945  * possible.
3946  *
3947  * Note that this function only works on the kmem_cache_node
3948  * when allocating for the kmem_cache_node. This is used for bootstrapping
3949  * memory on a fresh node that has no slab structures yet.
3950  */
early_kmem_cache_node_alloc(int node)3951 static void early_kmem_cache_node_alloc(int node)
3952 {
3953 	struct slab *slab;
3954 	struct kmem_cache_node *n;
3955 
3956 	BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
3957 
3958 	slab = new_slab(kmem_cache_node, GFP_NOWAIT, node);
3959 
3960 	BUG_ON(!slab);
3961 	if (slab_nid(slab) != node) {
3962 		pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3963 		pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
3964 	}
3965 
3966 	n = slab->freelist;
3967 	BUG_ON(!n);
3968 #ifdef CONFIG_SLUB_DEBUG
3969 	init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
3970 	init_tracking(kmem_cache_node, n);
3971 #endif
3972 	n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false);
3973 	slab->freelist = get_freepointer(kmem_cache_node, n);
3974 	slab->inuse = 1;
3975 	slab->frozen = 0;
3976 	kmem_cache_node->node[node] = n;
3977 	init_kmem_cache_node(n);
3978 	inc_slabs_node(kmem_cache_node, node, slab->objects);
3979 
3980 	/*
3981 	 * No locks need to be taken here as it has just been
3982 	 * initialized and there is no concurrent access.
3983 	 */
3984 	__add_partial(n, slab, DEACTIVATE_TO_HEAD);
3985 }
3986 
free_kmem_cache_nodes(struct kmem_cache * s)3987 static void free_kmem_cache_nodes(struct kmem_cache *s)
3988 {
3989 	int node;
3990 	struct kmem_cache_node *n;
3991 
3992 	for_each_kmem_cache_node(s, node, n) {
3993 		s->node[node] = NULL;
3994 		kmem_cache_free(kmem_cache_node, n);
3995 	}
3996 }
3997 
__kmem_cache_release(struct kmem_cache * s)3998 void __kmem_cache_release(struct kmem_cache *s)
3999 {
4000 	cache_random_seq_destroy(s);
4001 	free_percpu(s->cpu_slab);
4002 	free_kmem_cache_nodes(s);
4003 }
4004 
init_kmem_cache_nodes(struct kmem_cache * s)4005 static int init_kmem_cache_nodes(struct kmem_cache *s)
4006 {
4007 	int node;
4008 
4009 	for_each_node_mask(node, slab_nodes) {
4010 		struct kmem_cache_node *n;
4011 
4012 		if (slab_state == DOWN) {
4013 			early_kmem_cache_node_alloc(node);
4014 			continue;
4015 		}
4016 		n = kmem_cache_alloc_node(kmem_cache_node,
4017 						GFP_KERNEL, node);
4018 
4019 		if (!n) {
4020 			free_kmem_cache_nodes(s);
4021 			return 0;
4022 		}
4023 
4024 		init_kmem_cache_node(n);
4025 		s->node[node] = n;
4026 	}
4027 	return 1;
4028 }
4029 
set_cpu_partial(struct kmem_cache * s)4030 static void set_cpu_partial(struct kmem_cache *s)
4031 {
4032 #ifdef CONFIG_SLUB_CPU_PARTIAL
4033 	unsigned int nr_objects;
4034 
4035 	/*
4036 	 * cpu_partial determined the maximum number of objects kept in the
4037 	 * per cpu partial lists of a processor.
4038 	 *
4039 	 * Per cpu partial lists mainly contain slabs that just have one
4040 	 * object freed. If they are used for allocation then they can be
4041 	 * filled up again with minimal effort. The slab will never hit the
4042 	 * per node partial lists and therefore no locking will be required.
4043 	 *
4044 	 * For backwards compatibility reasons, this is determined as number
4045 	 * of objects, even though we now limit maximum number of pages, see
4046 	 * slub_set_cpu_partial()
4047 	 */
4048 	if (!kmem_cache_has_cpu_partial(s))
4049 		nr_objects = 0;
4050 	else if (s->size >= PAGE_SIZE)
4051 		nr_objects = 6;
4052 	else if (s->size >= 1024)
4053 		nr_objects = 24;
4054 	else if (s->size >= 256)
4055 		nr_objects = 52;
4056 	else
4057 		nr_objects = 120;
4058 
4059 	slub_set_cpu_partial(s, nr_objects);
4060 #endif
4061 }
4062 
4063 /*
4064  * calculate_sizes() determines the order and the distribution of data within
4065  * a slab object.
4066  */
calculate_sizes(struct kmem_cache * s)4067 static int calculate_sizes(struct kmem_cache *s)
4068 {
4069 	slab_flags_t flags = s->flags;
4070 	unsigned int size = s->object_size;
4071 	unsigned int order;
4072 
4073 	/*
4074 	 * Round up object size to the next word boundary. We can only
4075 	 * place the free pointer at word boundaries and this determines
4076 	 * the possible location of the free pointer.
4077 	 */
4078 	size = ALIGN(size, sizeof(void *));
4079 
4080 #ifdef CONFIG_SLUB_DEBUG
4081 	/*
4082 	 * Determine if we can poison the object itself. If the user of
4083 	 * the slab may touch the object after free or before allocation
4084 	 * then we should never poison the object itself.
4085 	 */
4086 	if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
4087 			!s->ctor)
4088 		s->flags |= __OBJECT_POISON;
4089 	else
4090 		s->flags &= ~__OBJECT_POISON;
4091 
4092 
4093 	/*
4094 	 * If we are Redzoning then check if there is some space between the
4095 	 * end of the object and the free pointer. If not then add an
4096 	 * additional word to have some bytes to store Redzone information.
4097 	 */
4098 	if ((flags & SLAB_RED_ZONE) && size == s->object_size)
4099 		size += sizeof(void *);
4100 #endif
4101 
4102 	/*
4103 	 * With that we have determined the number of bytes in actual use
4104 	 * by the object and redzoning.
4105 	 */
4106 	s->inuse = size;
4107 
4108 	if ((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
4109 	    ((flags & SLAB_RED_ZONE) && s->object_size < sizeof(void *)) ||
4110 	    s->ctor) {
4111 		/*
4112 		 * Relocate free pointer after the object if it is not
4113 		 * permitted to overwrite the first word of the object on
4114 		 * kmem_cache_free.
4115 		 *
4116 		 * This is the case if we do RCU, have a constructor or
4117 		 * destructor, are poisoning the objects, or are
4118 		 * redzoning an object smaller than sizeof(void *).
4119 		 *
4120 		 * The assumption that s->offset >= s->inuse means free
4121 		 * pointer is outside of the object is used in the
4122 		 * freeptr_outside_object() function. If that is no
4123 		 * longer true, the function needs to be modified.
4124 		 */
4125 		s->offset = size;
4126 		size += sizeof(void *);
4127 	} else {
4128 		/*
4129 		 * Store freelist pointer near middle of object to keep
4130 		 * it away from the edges of the object to avoid small
4131 		 * sized over/underflows from neighboring allocations.
4132 		 */
4133 		s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *));
4134 	}
4135 
4136 #ifdef CONFIG_SLUB_DEBUG
4137 	if (flags & SLAB_STORE_USER)
4138 		/*
4139 		 * Need to store information about allocs and frees after
4140 		 * the object.
4141 		 */
4142 		size += 2 * sizeof(struct track);
4143 #endif
4144 
4145 	kasan_cache_create(s, &size, &s->flags);
4146 #ifdef CONFIG_SLUB_DEBUG
4147 	if (flags & SLAB_RED_ZONE) {
4148 		/*
4149 		 * Add some empty padding so that we can catch
4150 		 * overwrites from earlier objects rather than let
4151 		 * tracking information or the free pointer be
4152 		 * corrupted if a user writes before the start
4153 		 * of the object.
4154 		 */
4155 		size += sizeof(void *);
4156 
4157 		s->red_left_pad = sizeof(void *);
4158 		s->red_left_pad = ALIGN(s->red_left_pad, s->align);
4159 		size += s->red_left_pad;
4160 	}
4161 #endif
4162 
4163 	/*
4164 	 * SLUB stores one object immediately after another beginning from
4165 	 * offset 0. In order to align the objects we have to simply size
4166 	 * each object to conform to the alignment.
4167 	 */
4168 	size = ALIGN(size, s->align);
4169 	s->size = size;
4170 	s->reciprocal_size = reciprocal_value(size);
4171 	order = calculate_order(size);
4172 
4173 	if ((int)order < 0)
4174 		return 0;
4175 
4176 	s->allocflags = 0;
4177 	if (order)
4178 		s->allocflags |= __GFP_COMP;
4179 
4180 	if (s->flags & SLAB_CACHE_DMA)
4181 		s->allocflags |= GFP_DMA;
4182 
4183 	if (s->flags & SLAB_CACHE_DMA32)
4184 		s->allocflags |= GFP_DMA32;
4185 
4186 	if (s->flags & SLAB_RECLAIM_ACCOUNT)
4187 		s->allocflags |= __GFP_RECLAIMABLE;
4188 
4189 	/*
4190 	 * Determine the number of objects per slab
4191 	 */
4192 	s->oo = oo_make(order, size);
4193 	s->min = oo_make(get_order(size), size);
4194 
4195 	return !!oo_objects(s->oo);
4196 }
4197 
kmem_cache_open(struct kmem_cache * s,slab_flags_t flags)4198 static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
4199 {
4200 	s->flags = kmem_cache_flags(s->size, flags, s->name);
4201 #ifdef CONFIG_SLAB_FREELIST_HARDENED
4202 	s->random = get_random_long();
4203 #endif
4204 
4205 	if (!calculate_sizes(s))
4206 		goto error;
4207 	if (disable_higher_order_debug) {
4208 		/*
4209 		 * Disable debugging flags that store metadata if the min slab
4210 		 * order increased.
4211 		 */
4212 		if (get_order(s->size) > get_order(s->object_size)) {
4213 			s->flags &= ~DEBUG_METADATA_FLAGS;
4214 			s->offset = 0;
4215 			if (!calculate_sizes(s))
4216 				goto error;
4217 		}
4218 	}
4219 
4220 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
4221     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
4222 	if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
4223 		/* Enable fast mode */
4224 		s->flags |= __CMPXCHG_DOUBLE;
4225 #endif
4226 
4227 	/*
4228 	 * The larger the object size is, the more slabs we want on the partial
4229 	 * list to avoid pounding the page allocator excessively.
4230 	 */
4231 	s->min_partial = min_t(unsigned long, MAX_PARTIAL, ilog2(s->size) / 2);
4232 	s->min_partial = max_t(unsigned long, MIN_PARTIAL, s->min_partial);
4233 
4234 	set_cpu_partial(s);
4235 
4236 #ifdef CONFIG_NUMA
4237 	s->remote_node_defrag_ratio = 1000;
4238 #endif
4239 
4240 	/* Initialize the pre-computed randomized freelist if slab is up */
4241 	if (slab_state >= UP) {
4242 		if (init_cache_random_seq(s))
4243 			goto error;
4244 	}
4245 
4246 	if (!init_kmem_cache_nodes(s))
4247 		goto error;
4248 
4249 	if (alloc_kmem_cache_cpus(s))
4250 		return 0;
4251 
4252 error:
4253 	__kmem_cache_release(s);
4254 	return -EINVAL;
4255 }
4256 
list_slab_objects(struct kmem_cache * s,struct slab * slab,const char * text)4257 static void list_slab_objects(struct kmem_cache *s, struct slab *slab,
4258 			      const char *text)
4259 {
4260 #ifdef CONFIG_SLUB_DEBUG
4261 	void *addr = slab_address(slab);
4262 	unsigned long flags;
4263 	unsigned long *map;
4264 	void *p;
4265 
4266 	slab_err(s, slab, text, s->name);
4267 	slab_lock(slab, &flags);
4268 
4269 	map = get_map(s, slab);
4270 	for_each_object(p, s, addr, slab->objects) {
4271 
4272 		if (!test_bit(__obj_to_index(s, addr, p), map)) {
4273 			pr_err("Object 0x%p @offset=%tu\n", p, p - addr);
4274 			print_tracking(s, p);
4275 		}
4276 	}
4277 	put_map(map);
4278 	slab_unlock(slab, &flags);
4279 #endif
4280 }
4281 
4282 /*
4283  * Attempt to free all partial slabs on a node.
4284  * This is called from __kmem_cache_shutdown(). We must take list_lock
4285  * because sysfs file might still access partial list after the shutdowning.
4286  */
free_partial(struct kmem_cache * s,struct kmem_cache_node * n)4287 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
4288 {
4289 	LIST_HEAD(discard);
4290 	struct slab *slab, *h;
4291 
4292 	BUG_ON(irqs_disabled());
4293 	spin_lock_irq(&n->list_lock);
4294 	list_for_each_entry_safe(slab, h, &n->partial, slab_list) {
4295 		if (!slab->inuse) {
4296 			remove_partial(n, slab);
4297 			list_add(&slab->slab_list, &discard);
4298 		} else {
4299 			list_slab_objects(s, slab,
4300 			  "Objects remaining in %s on __kmem_cache_shutdown()");
4301 		}
4302 	}
4303 	spin_unlock_irq(&n->list_lock);
4304 
4305 	list_for_each_entry_safe(slab, h, &discard, slab_list)
4306 		discard_slab(s, slab);
4307 }
4308 
__kmem_cache_empty(struct kmem_cache * s)4309 bool __kmem_cache_empty(struct kmem_cache *s)
4310 {
4311 	int node;
4312 	struct kmem_cache_node *n;
4313 
4314 	for_each_kmem_cache_node(s, node, n)
4315 		if (n->nr_partial || slabs_node(s, node))
4316 			return false;
4317 	return true;
4318 }
4319 
4320 /*
4321  * Release all resources used by a slab cache.
4322  */
__kmem_cache_shutdown(struct kmem_cache * s)4323 int __kmem_cache_shutdown(struct kmem_cache *s)
4324 {
4325 	int node;
4326 	struct kmem_cache_node *n;
4327 
4328 	flush_all_cpus_locked(s);
4329 	/* Attempt to free all objects */
4330 	for_each_kmem_cache_node(s, node, n) {
4331 		free_partial(s, n);
4332 		if (n->nr_partial || slabs_node(s, node))
4333 			return 1;
4334 	}
4335 	return 0;
4336 }
4337 
4338 #ifdef CONFIG_PRINTK
__kmem_obj_info(struct kmem_obj_info * kpp,void * object,struct slab * slab)4339 void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
4340 {
4341 	void *base;
4342 	int __maybe_unused i;
4343 	unsigned int objnr;
4344 	void *objp;
4345 	void *objp0;
4346 	struct kmem_cache *s = slab->slab_cache;
4347 	struct track __maybe_unused *trackp;
4348 
4349 	kpp->kp_ptr = object;
4350 	kpp->kp_slab = slab;
4351 	kpp->kp_slab_cache = s;
4352 	base = slab_address(slab);
4353 	objp0 = kasan_reset_tag(object);
4354 #ifdef CONFIG_SLUB_DEBUG
4355 	objp = restore_red_left(s, objp0);
4356 #else
4357 	objp = objp0;
4358 #endif
4359 	objnr = obj_to_index(s, slab, objp);
4360 	kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp);
4361 	objp = base + s->size * objnr;
4362 	kpp->kp_objp = objp;
4363 	if (WARN_ON_ONCE(objp < base || objp >= base + slab->objects * s->size
4364 			 || (objp - base) % s->size) ||
4365 	    !(s->flags & SLAB_STORE_USER))
4366 		return;
4367 #ifdef CONFIG_SLUB_DEBUG
4368 	objp = fixup_red_left(s, objp);
4369 	trackp = get_track(s, objp, TRACK_ALLOC);
4370 	kpp->kp_ret = (void *)trackp->addr;
4371 #ifdef CONFIG_STACKDEPOT
4372 	{
4373 		depot_stack_handle_t handle;
4374 		unsigned long *entries;
4375 		unsigned int nr_entries;
4376 
4377 		handle = READ_ONCE(trackp->handle);
4378 		if (handle) {
4379 			nr_entries = stack_depot_fetch(handle, &entries);
4380 			for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
4381 				kpp->kp_stack[i] = (void *)entries[i];
4382 		}
4383 
4384 		trackp = get_track(s, objp, TRACK_FREE);
4385 		handle = READ_ONCE(trackp->handle);
4386 		if (handle) {
4387 			nr_entries = stack_depot_fetch(handle, &entries);
4388 			for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
4389 				kpp->kp_free_stack[i] = (void *)entries[i];
4390 		}
4391 	}
4392 #endif
4393 #endif
4394 }
4395 #endif
4396 
4397 /********************************************************************
4398  *		Kmalloc subsystem
4399  *******************************************************************/
4400 
setup_slub_min_order(char * str)4401 static int __init setup_slub_min_order(char *str)
4402 {
4403 	get_option(&str, (int *)&slub_min_order);
4404 
4405 	return 1;
4406 }
4407 
4408 __setup("slub_min_order=", setup_slub_min_order);
4409 
setup_slub_max_order(char * str)4410 static int __init setup_slub_max_order(char *str)
4411 {
4412 	get_option(&str, (int *)&slub_max_order);
4413 	slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1);
4414 
4415 	return 1;
4416 }
4417 
4418 __setup("slub_max_order=", setup_slub_max_order);
4419 
setup_slub_min_objects(char * str)4420 static int __init setup_slub_min_objects(char *str)
4421 {
4422 	get_option(&str, (int *)&slub_min_objects);
4423 
4424 	return 1;
4425 }
4426 
4427 __setup("slub_min_objects=", setup_slub_min_objects);
4428 
__kmalloc(size_t size,gfp_t flags)4429 void *__kmalloc(size_t size, gfp_t flags)
4430 {
4431 	struct kmem_cache *s;
4432 	void *ret;
4433 
4434 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
4435 		return kmalloc_large(size, flags);
4436 
4437 	s = kmalloc_slab(size, flags);
4438 
4439 	if (unlikely(ZERO_OR_NULL_PTR(s)))
4440 		return s;
4441 
4442 	ret = slab_alloc(s, NULL, flags, _RET_IP_, size);
4443 
4444 	trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
4445 
4446 	ret = kasan_kmalloc(s, ret, size, flags);
4447 
4448 	return ret;
4449 }
4450 EXPORT_SYMBOL(__kmalloc);
4451 
4452 #ifdef CONFIG_NUMA
kmalloc_large_node(size_t size,gfp_t flags,int node)4453 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
4454 {
4455 	struct page *page;
4456 	void *ptr = NULL;
4457 	unsigned int order = get_order(size);
4458 
4459 	flags |= __GFP_COMP;
4460 	page = alloc_pages_node(node, flags, order);
4461 	if (page) {
4462 		ptr = page_address(page);
4463 		mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
4464 				      PAGE_SIZE << order);
4465 	}
4466 
4467 	return kmalloc_large_node_hook(ptr, size, flags);
4468 }
4469 
__kmalloc_node(size_t size,gfp_t flags,int node)4470 void *__kmalloc_node(size_t size, gfp_t flags, int node)
4471 {
4472 	struct kmem_cache *s;
4473 	void *ret;
4474 
4475 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
4476 		ret = kmalloc_large_node(size, flags, node);
4477 
4478 		trace_kmalloc_node(_RET_IP_, ret,
4479 				   size, PAGE_SIZE << get_order(size),
4480 				   flags, node);
4481 
4482 		return ret;
4483 	}
4484 
4485 	s = kmalloc_slab(size, flags);
4486 
4487 	if (unlikely(ZERO_OR_NULL_PTR(s)))
4488 		return s;
4489 
4490 	ret = slab_alloc_node(s, NULL, flags, node, _RET_IP_, size);
4491 
4492 	trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
4493 
4494 	ret = kasan_kmalloc(s, ret, size, flags);
4495 
4496 	return ret;
4497 }
4498 EXPORT_SYMBOL(__kmalloc_node);
4499 #endif	/* CONFIG_NUMA */
4500 
4501 #ifdef CONFIG_HARDENED_USERCOPY
4502 /*
4503  * Rejects incorrectly sized objects and objects that are to be copied
4504  * to/from userspace but do not fall entirely within the containing slab
4505  * cache's usercopy region.
4506  *
4507  * Returns NULL if check passes, otherwise const char * to name of cache
4508  * to indicate an error.
4509  */
__check_heap_object(const void * ptr,unsigned long n,const struct slab * slab,bool to_user)4510 void __check_heap_object(const void *ptr, unsigned long n,
4511 			 const struct slab *slab, bool to_user)
4512 {
4513 	struct kmem_cache *s;
4514 	unsigned int offset;
4515 	bool is_kfence = is_kfence_address(ptr);
4516 
4517 	ptr = kasan_reset_tag(ptr);
4518 
4519 	/* Find object and usable object size. */
4520 	s = slab->slab_cache;
4521 
4522 	/* Reject impossible pointers. */
4523 	if (ptr < slab_address(slab))
4524 		usercopy_abort("SLUB object not in SLUB page?!", NULL,
4525 			       to_user, 0, n);
4526 
4527 	/* Find offset within object. */
4528 	if (is_kfence)
4529 		offset = ptr - kfence_object_start(ptr);
4530 	else
4531 		offset = (ptr - slab_address(slab)) % s->size;
4532 
4533 	/* Adjust for redzone and reject if within the redzone. */
4534 	if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) {
4535 		if (offset < s->red_left_pad)
4536 			usercopy_abort("SLUB object in left red zone",
4537 				       s->name, to_user, offset, n);
4538 		offset -= s->red_left_pad;
4539 	}
4540 
4541 	/* Allow address range falling entirely within usercopy region. */
4542 	if (offset >= s->useroffset &&
4543 	    offset - s->useroffset <= s->usersize &&
4544 	    n <= s->useroffset - offset + s->usersize)
4545 		return;
4546 
4547 	usercopy_abort("SLUB object", s->name, to_user, offset, n);
4548 }
4549 #endif /* CONFIG_HARDENED_USERCOPY */
4550 
__ksize(const void * object)4551 size_t __ksize(const void *object)
4552 {
4553 	struct folio *folio;
4554 
4555 	if (unlikely(object == ZERO_SIZE_PTR))
4556 		return 0;
4557 
4558 	folio = virt_to_folio(object);
4559 
4560 	if (unlikely(!folio_test_slab(folio)))
4561 		return folio_size(folio);
4562 
4563 	return slab_ksize(folio_slab(folio)->slab_cache);
4564 }
4565 EXPORT_SYMBOL(__ksize);
4566 
kfree(const void * x)4567 void kfree(const void *x)
4568 {
4569 	struct folio *folio;
4570 	struct slab *slab;
4571 	void *object = (void *)x;
4572 
4573 	trace_kfree(_RET_IP_, x);
4574 
4575 	if (unlikely(ZERO_OR_NULL_PTR(x)))
4576 		return;
4577 
4578 	folio = virt_to_folio(x);
4579 	if (unlikely(!folio_test_slab(folio))) {
4580 		free_large_kmalloc(folio, object);
4581 		return;
4582 	}
4583 	slab = folio_slab(folio);
4584 	slab_free(slab->slab_cache, slab, object, NULL, 1, _RET_IP_);
4585 }
4586 EXPORT_SYMBOL(kfree);
4587 
4588 #define SHRINK_PROMOTE_MAX 32
4589 
4590 /*
4591  * kmem_cache_shrink discards empty slabs and promotes the slabs filled
4592  * up most to the head of the partial lists. New allocations will then
4593  * fill those up and thus they can be removed from the partial lists.
4594  *
4595  * The slabs with the least items are placed last. This results in them
4596  * being allocated from last increasing the chance that the last objects
4597  * are freed in them.
4598  */
__kmem_cache_do_shrink(struct kmem_cache * s)4599 static int __kmem_cache_do_shrink(struct kmem_cache *s)
4600 {
4601 	int node;
4602 	int i;
4603 	struct kmem_cache_node *n;
4604 	struct slab *slab;
4605 	struct slab *t;
4606 	struct list_head discard;
4607 	struct list_head promote[SHRINK_PROMOTE_MAX];
4608 	unsigned long flags;
4609 	int ret = 0;
4610 
4611 	for_each_kmem_cache_node(s, node, n) {
4612 		INIT_LIST_HEAD(&discard);
4613 		for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
4614 			INIT_LIST_HEAD(promote + i);
4615 
4616 		spin_lock_irqsave(&n->list_lock, flags);
4617 
4618 		/*
4619 		 * Build lists of slabs to discard or promote.
4620 		 *
4621 		 * Note that concurrent frees may occur while we hold the
4622 		 * list_lock. slab->inuse here is the upper limit.
4623 		 */
4624 		list_for_each_entry_safe(slab, t, &n->partial, slab_list) {
4625 			int free = slab->objects - slab->inuse;
4626 
4627 			/* Do not reread slab->inuse */
4628 			barrier();
4629 
4630 			/* We do not keep full slabs on the list */
4631 			BUG_ON(free <= 0);
4632 
4633 			if (free == slab->objects) {
4634 				list_move(&slab->slab_list, &discard);
4635 				n->nr_partial--;
4636 			} else if (free <= SHRINK_PROMOTE_MAX)
4637 				list_move(&slab->slab_list, promote + free - 1);
4638 		}
4639 
4640 		/*
4641 		 * Promote the slabs filled up most to the head of the
4642 		 * partial list.
4643 		 */
4644 		for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
4645 			list_splice(promote + i, &n->partial);
4646 
4647 		spin_unlock_irqrestore(&n->list_lock, flags);
4648 
4649 		/* Release empty slabs */
4650 		list_for_each_entry_safe(slab, t, &discard, slab_list)
4651 			discard_slab(s, slab);
4652 
4653 		if (slabs_node(s, node))
4654 			ret = 1;
4655 	}
4656 
4657 	return ret;
4658 }
4659 
__kmem_cache_shrink(struct kmem_cache * s)4660 int __kmem_cache_shrink(struct kmem_cache *s)
4661 {
4662 	flush_all(s);
4663 	return __kmem_cache_do_shrink(s);
4664 }
4665 
slab_mem_going_offline_callback(void * arg)4666 static int slab_mem_going_offline_callback(void *arg)
4667 {
4668 	struct kmem_cache *s;
4669 
4670 	mutex_lock(&slab_mutex);
4671 	list_for_each_entry(s, &slab_caches, list) {
4672 		flush_all_cpus_locked(s);
4673 		__kmem_cache_do_shrink(s);
4674 	}
4675 	mutex_unlock(&slab_mutex);
4676 
4677 	return 0;
4678 }
4679 
slab_mem_offline_callback(void * arg)4680 static void slab_mem_offline_callback(void *arg)
4681 {
4682 	struct memory_notify *marg = arg;
4683 	int offline_node;
4684 
4685 	offline_node = marg->status_change_nid_normal;
4686 
4687 	/*
4688 	 * If the node still has available memory. we need kmem_cache_node
4689 	 * for it yet.
4690 	 */
4691 	if (offline_node < 0)
4692 		return;
4693 
4694 	mutex_lock(&slab_mutex);
4695 	node_clear(offline_node, slab_nodes);
4696 	/*
4697 	 * We no longer free kmem_cache_node structures here, as it would be
4698 	 * racy with all get_node() users, and infeasible to protect them with
4699 	 * slab_mutex.
4700 	 */
4701 	mutex_unlock(&slab_mutex);
4702 }
4703 
slab_mem_going_online_callback(void * arg)4704 static int slab_mem_going_online_callback(void *arg)
4705 {
4706 	struct kmem_cache_node *n;
4707 	struct kmem_cache *s;
4708 	struct memory_notify *marg = arg;
4709 	int nid = marg->status_change_nid_normal;
4710 	int ret = 0;
4711 
4712 	/*
4713 	 * If the node's memory is already available, then kmem_cache_node is
4714 	 * already created. Nothing to do.
4715 	 */
4716 	if (nid < 0)
4717 		return 0;
4718 
4719 	/*
4720 	 * We are bringing a node online. No memory is available yet. We must
4721 	 * allocate a kmem_cache_node structure in order to bring the node
4722 	 * online.
4723 	 */
4724 	mutex_lock(&slab_mutex);
4725 	list_for_each_entry(s, &slab_caches, list) {
4726 		/*
4727 		 * The structure may already exist if the node was previously
4728 		 * onlined and offlined.
4729 		 */
4730 		if (get_node(s, nid))
4731 			continue;
4732 		/*
4733 		 * XXX: kmem_cache_alloc_node will fallback to other nodes
4734 		 *      since memory is not yet available from the node that
4735 		 *      is brought up.
4736 		 */
4737 		n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
4738 		if (!n) {
4739 			ret = -ENOMEM;
4740 			goto out;
4741 		}
4742 		init_kmem_cache_node(n);
4743 		s->node[nid] = n;
4744 	}
4745 	/*
4746 	 * Any cache created after this point will also have kmem_cache_node
4747 	 * initialized for the new node.
4748 	 */
4749 	node_set(nid, slab_nodes);
4750 out:
4751 	mutex_unlock(&slab_mutex);
4752 	return ret;
4753 }
4754 
slab_memory_callback(struct notifier_block * self,unsigned long action,void * arg)4755 static int slab_memory_callback(struct notifier_block *self,
4756 				unsigned long action, void *arg)
4757 {
4758 	int ret = 0;
4759 
4760 	switch (action) {
4761 	case MEM_GOING_ONLINE:
4762 		ret = slab_mem_going_online_callback(arg);
4763 		break;
4764 	case MEM_GOING_OFFLINE:
4765 		ret = slab_mem_going_offline_callback(arg);
4766 		break;
4767 	case MEM_OFFLINE:
4768 	case MEM_CANCEL_ONLINE:
4769 		slab_mem_offline_callback(arg);
4770 		break;
4771 	case MEM_ONLINE:
4772 	case MEM_CANCEL_OFFLINE:
4773 		break;
4774 	}
4775 	if (ret)
4776 		ret = notifier_from_errno(ret);
4777 	else
4778 		ret = NOTIFY_OK;
4779 	return ret;
4780 }
4781 
4782 static struct notifier_block slab_memory_callback_nb = {
4783 	.notifier_call = slab_memory_callback,
4784 	.priority = SLAB_CALLBACK_PRI,
4785 };
4786 
4787 /********************************************************************
4788  *			Basic setup of slabs
4789  *******************************************************************/
4790 
4791 /*
4792  * Used for early kmem_cache structures that were allocated using
4793  * the page allocator. Allocate them properly then fix up the pointers
4794  * that may be pointing to the wrong kmem_cache structure.
4795  */
4796 
bootstrap(struct kmem_cache * static_cache)4797 static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
4798 {
4799 	int node;
4800 	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
4801 	struct kmem_cache_node *n;
4802 
4803 	memcpy(s, static_cache, kmem_cache->object_size);
4804 
4805 	/*
4806 	 * This runs very early, and only the boot processor is supposed to be
4807 	 * up.  Even if it weren't true, IRQs are not up so we couldn't fire
4808 	 * IPIs around.
4809 	 */
4810 	__flush_cpu_slab(s, smp_processor_id());
4811 	for_each_kmem_cache_node(s, node, n) {
4812 		struct slab *p;
4813 
4814 		list_for_each_entry(p, &n->partial, slab_list)
4815 			p->slab_cache = s;
4816 
4817 #ifdef CONFIG_SLUB_DEBUG
4818 		list_for_each_entry(p, &n->full, slab_list)
4819 			p->slab_cache = s;
4820 #endif
4821 	}
4822 	list_add(&s->list, &slab_caches);
4823 	return s;
4824 }
4825 
kmem_cache_init(void)4826 void __init kmem_cache_init(void)
4827 {
4828 	static __initdata struct kmem_cache boot_kmem_cache,
4829 		boot_kmem_cache_node;
4830 	int node;
4831 
4832 	if (debug_guardpage_minorder())
4833 		slub_max_order = 0;
4834 
4835 	/* Print slub debugging pointers without hashing */
4836 	if (__slub_debug_enabled())
4837 		no_hash_pointers_enable(NULL);
4838 
4839 	kmem_cache_node = &boot_kmem_cache_node;
4840 	kmem_cache = &boot_kmem_cache;
4841 
4842 	/*
4843 	 * Initialize the nodemask for which we will allocate per node
4844 	 * structures. Here we don't need taking slab_mutex yet.
4845 	 */
4846 	for_each_node_state(node, N_NORMAL_MEMORY)
4847 		node_set(node, slab_nodes);
4848 
4849 	create_boot_cache(kmem_cache_node, "kmem_cache_node",
4850 		sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
4851 
4852 	register_hotmemory_notifier(&slab_memory_callback_nb);
4853 
4854 	/* Able to allocate the per node structures */
4855 	slab_state = PARTIAL;
4856 
4857 	create_boot_cache(kmem_cache, "kmem_cache",
4858 			offsetof(struct kmem_cache, node) +
4859 				nr_node_ids * sizeof(struct kmem_cache_node *),
4860 		       SLAB_HWCACHE_ALIGN, 0, 0);
4861 
4862 	kmem_cache = bootstrap(&boot_kmem_cache);
4863 	kmem_cache_node = bootstrap(&boot_kmem_cache_node);
4864 
4865 	/* Now we can use the kmem_cache to allocate kmalloc slabs */
4866 	setup_kmalloc_cache_index_table();
4867 	create_kmalloc_caches(0);
4868 
4869 	/* Setup random freelists for each cache */
4870 	init_freelist_randomization();
4871 
4872 	cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4873 				  slub_cpu_dead);
4874 
4875 	pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n",
4876 		cache_line_size(),
4877 		slub_min_order, slub_max_order, slub_min_objects,
4878 		nr_cpu_ids, nr_node_ids);
4879 }
4880 
kmem_cache_init_late(void)4881 void __init kmem_cache_init_late(void)
4882 {
4883 }
4884 
4885 struct kmem_cache *
__kmem_cache_alias(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,void (* ctor)(void *))4886 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
4887 		   slab_flags_t flags, void (*ctor)(void *))
4888 {
4889 	struct kmem_cache *s;
4890 
4891 	s = find_mergeable(size, align, flags, name, ctor);
4892 	if (s) {
4893 		s->refcount++;
4894 
4895 		/*
4896 		 * Adjust the object sizes so that we clear
4897 		 * the complete object on kzalloc.
4898 		 */
4899 		s->object_size = max(s->object_size, size);
4900 		s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
4901 
4902 		if (sysfs_slab_alias(s, name)) {
4903 			s->refcount--;
4904 			s = NULL;
4905 		}
4906 	}
4907 
4908 	return s;
4909 }
4910 
__kmem_cache_create(struct kmem_cache * s,slab_flags_t flags)4911 int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
4912 {
4913 	int err;
4914 
4915 	err = kmem_cache_open(s, flags);
4916 	if (err)
4917 		return err;
4918 
4919 	/* Mutex is not taken during early boot */
4920 	if (slab_state <= UP)
4921 		return 0;
4922 
4923 	err = sysfs_slab_add(s);
4924 	if (err) {
4925 		__kmem_cache_release(s);
4926 		return err;
4927 	}
4928 
4929 	if (s->flags & SLAB_STORE_USER)
4930 		debugfs_slab_add(s);
4931 
4932 	return 0;
4933 }
4934 
__kmalloc_track_caller(size_t size,gfp_t gfpflags,unsigned long caller)4935 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
4936 {
4937 	struct kmem_cache *s;
4938 	void *ret;
4939 
4940 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
4941 		return kmalloc_large(size, gfpflags);
4942 
4943 	s = kmalloc_slab(size, gfpflags);
4944 
4945 	if (unlikely(ZERO_OR_NULL_PTR(s)))
4946 		return s;
4947 
4948 	ret = slab_alloc(s, NULL, gfpflags, caller, size);
4949 
4950 	/* Honor the call site pointer we received. */
4951 	trace_kmalloc(caller, ret, size, s->size, gfpflags);
4952 
4953 	return ret;
4954 }
4955 EXPORT_SYMBOL(__kmalloc_track_caller);
4956 
4957 #ifdef CONFIG_NUMA
__kmalloc_node_track_caller(size_t size,gfp_t gfpflags,int node,unsigned long caller)4958 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
4959 					int node, unsigned long caller)
4960 {
4961 	struct kmem_cache *s;
4962 	void *ret;
4963 
4964 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
4965 		ret = kmalloc_large_node(size, gfpflags, node);
4966 
4967 		trace_kmalloc_node(caller, ret,
4968 				   size, PAGE_SIZE << get_order(size),
4969 				   gfpflags, node);
4970 
4971 		return ret;
4972 	}
4973 
4974 	s = kmalloc_slab(size, gfpflags);
4975 
4976 	if (unlikely(ZERO_OR_NULL_PTR(s)))
4977 		return s;
4978 
4979 	ret = slab_alloc_node(s, NULL, gfpflags, node, caller, size);
4980 
4981 	/* Honor the call site pointer we received. */
4982 	trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
4983 
4984 	return ret;
4985 }
4986 EXPORT_SYMBOL(__kmalloc_node_track_caller);
4987 #endif
4988 
4989 #ifdef CONFIG_SYSFS
count_inuse(struct slab * slab)4990 static int count_inuse(struct slab *slab)
4991 {
4992 	return slab->inuse;
4993 }
4994 
count_total(struct slab * slab)4995 static int count_total(struct slab *slab)
4996 {
4997 	return slab->objects;
4998 }
4999 #endif
5000 
5001 #ifdef CONFIG_SLUB_DEBUG
validate_slab(struct kmem_cache * s,struct slab * slab,unsigned long * obj_map)5002 static void validate_slab(struct kmem_cache *s, struct slab *slab,
5003 			  unsigned long *obj_map)
5004 {
5005 	void *p;
5006 	void *addr = slab_address(slab);
5007 	unsigned long flags;
5008 
5009 	slab_lock(slab, &flags);
5010 
5011 	if (!check_slab(s, slab) || !on_freelist(s, slab, NULL))
5012 		goto unlock;
5013 
5014 	/* Now we know that a valid freelist exists */
5015 	__fill_map(obj_map, s, slab);
5016 	for_each_object(p, s, addr, slab->objects) {
5017 		u8 val = test_bit(__obj_to_index(s, addr, p), obj_map) ?
5018 			 SLUB_RED_INACTIVE : SLUB_RED_ACTIVE;
5019 
5020 		if (!check_object(s, slab, p, val))
5021 			break;
5022 	}
5023 unlock:
5024 	slab_unlock(slab, &flags);
5025 }
5026 
validate_slab_node(struct kmem_cache * s,struct kmem_cache_node * n,unsigned long * obj_map)5027 static int validate_slab_node(struct kmem_cache *s,
5028 		struct kmem_cache_node *n, unsigned long *obj_map)
5029 {
5030 	unsigned long count = 0;
5031 	struct slab *slab;
5032 	unsigned long flags;
5033 
5034 	spin_lock_irqsave(&n->list_lock, flags);
5035 
5036 	list_for_each_entry(slab, &n->partial, slab_list) {
5037 		validate_slab(s, slab, obj_map);
5038 		count++;
5039 	}
5040 	if (count != n->nr_partial) {
5041 		pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
5042 		       s->name, count, n->nr_partial);
5043 		slab_add_kunit_errors();
5044 	}
5045 
5046 	if (!(s->flags & SLAB_STORE_USER))
5047 		goto out;
5048 
5049 	list_for_each_entry(slab, &n->full, slab_list) {
5050 		validate_slab(s, slab, obj_map);
5051 		count++;
5052 	}
5053 	if (count != atomic_long_read(&n->nr_slabs)) {
5054 		pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
5055 		       s->name, count, atomic_long_read(&n->nr_slabs));
5056 		slab_add_kunit_errors();
5057 	}
5058 
5059 out:
5060 	spin_unlock_irqrestore(&n->list_lock, flags);
5061 	return count;
5062 }
5063 
validate_slab_cache(struct kmem_cache * s)5064 long validate_slab_cache(struct kmem_cache *s)
5065 {
5066 	int node;
5067 	unsigned long count = 0;
5068 	struct kmem_cache_node *n;
5069 	unsigned long *obj_map;
5070 
5071 	obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL);
5072 	if (!obj_map)
5073 		return -ENOMEM;
5074 
5075 	flush_all(s);
5076 	for_each_kmem_cache_node(s, node, n)
5077 		count += validate_slab_node(s, n, obj_map);
5078 
5079 	bitmap_free(obj_map);
5080 
5081 	return count;
5082 }
5083 EXPORT_SYMBOL(validate_slab_cache);
5084 
5085 #ifdef CONFIG_DEBUG_FS
5086 /*
5087  * Generate lists of code addresses where slabcache objects are allocated
5088  * and freed.
5089  */
5090 
5091 struct location {
5092 	depot_stack_handle_t handle;
5093 	unsigned long count;
5094 	unsigned long addr;
5095 	long long sum_time;
5096 	long min_time;
5097 	long max_time;
5098 	long min_pid;
5099 	long max_pid;
5100 	DECLARE_BITMAP(cpus, NR_CPUS);
5101 	nodemask_t nodes;
5102 };
5103 
5104 struct loc_track {
5105 	unsigned long max;
5106 	unsigned long count;
5107 	struct location *loc;
5108 	loff_t idx;
5109 };
5110 
5111 static struct dentry *slab_debugfs_root;
5112 
free_loc_track(struct loc_track * t)5113 static void free_loc_track(struct loc_track *t)
5114 {
5115 	if (t->max)
5116 		free_pages((unsigned long)t->loc,
5117 			get_order(sizeof(struct location) * t->max));
5118 }
5119 
alloc_loc_track(struct loc_track * t,unsigned long max,gfp_t flags)5120 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
5121 {
5122 	struct location *l;
5123 	int order;
5124 
5125 	order = get_order(sizeof(struct location) * max);
5126 
5127 	l = (void *)__get_free_pages(flags, order);
5128 	if (!l)
5129 		return 0;
5130 
5131 	if (t->count) {
5132 		memcpy(l, t->loc, sizeof(struct location) * t->count);
5133 		free_loc_track(t);
5134 	}
5135 	t->max = max;
5136 	t->loc = l;
5137 	return 1;
5138 }
5139 
add_location(struct loc_track * t,struct kmem_cache * s,const struct track * track)5140 static int add_location(struct loc_track *t, struct kmem_cache *s,
5141 				const struct track *track)
5142 {
5143 	long start, end, pos;
5144 	struct location *l;
5145 	unsigned long caddr, chandle;
5146 	unsigned long age = jiffies - track->when;
5147 	depot_stack_handle_t handle = 0;
5148 
5149 #ifdef CONFIG_STACKDEPOT
5150 	handle = READ_ONCE(track->handle);
5151 #endif
5152 	start = -1;
5153 	end = t->count;
5154 
5155 	for ( ; ; ) {
5156 		pos = start + (end - start + 1) / 2;
5157 
5158 		/*
5159 		 * There is nothing at "end". If we end up there
5160 		 * we need to add something to before end.
5161 		 */
5162 		if (pos == end)
5163 			break;
5164 
5165 		caddr = t->loc[pos].addr;
5166 		chandle = t->loc[pos].handle;
5167 		if ((track->addr == caddr) && (handle == chandle)) {
5168 
5169 			l = &t->loc[pos];
5170 			l->count++;
5171 			if (track->when) {
5172 				l->sum_time += age;
5173 				if (age < l->min_time)
5174 					l->min_time = age;
5175 				if (age > l->max_time)
5176 					l->max_time = age;
5177 
5178 				if (track->pid < l->min_pid)
5179 					l->min_pid = track->pid;
5180 				if (track->pid > l->max_pid)
5181 					l->max_pid = track->pid;
5182 
5183 				cpumask_set_cpu(track->cpu,
5184 						to_cpumask(l->cpus));
5185 			}
5186 			node_set(page_to_nid(virt_to_page(track)), l->nodes);
5187 			return 1;
5188 		}
5189 
5190 		if (track->addr < caddr)
5191 			end = pos;
5192 		else if (track->addr == caddr && handle < chandle)
5193 			end = pos;
5194 		else
5195 			start = pos;
5196 	}
5197 
5198 	/*
5199 	 * Not found. Insert new tracking element.
5200 	 */
5201 	if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
5202 		return 0;
5203 
5204 	l = t->loc + pos;
5205 	if (pos < t->count)
5206 		memmove(l + 1, l,
5207 			(t->count - pos) * sizeof(struct location));
5208 	t->count++;
5209 	l->count = 1;
5210 	l->addr = track->addr;
5211 	l->sum_time = age;
5212 	l->min_time = age;
5213 	l->max_time = age;
5214 	l->min_pid = track->pid;
5215 	l->max_pid = track->pid;
5216 	l->handle = handle;
5217 	cpumask_clear(to_cpumask(l->cpus));
5218 	cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
5219 	nodes_clear(l->nodes);
5220 	node_set(page_to_nid(virt_to_page(track)), l->nodes);
5221 	return 1;
5222 }
5223 
process_slab(struct loc_track * t,struct kmem_cache * s,struct slab * slab,enum track_item alloc,unsigned long * obj_map)5224 static void process_slab(struct loc_track *t, struct kmem_cache *s,
5225 		struct slab *slab, enum track_item alloc,
5226 		unsigned long *obj_map)
5227 {
5228 	void *addr = slab_address(slab);
5229 	void *p;
5230 
5231 	__fill_map(obj_map, s, slab);
5232 
5233 	for_each_object(p, s, addr, slab->objects)
5234 		if (!test_bit(__obj_to_index(s, addr, p), obj_map))
5235 			add_location(t, s, get_track(s, p, alloc));
5236 }
5237 #endif  /* CONFIG_DEBUG_FS   */
5238 #endif	/* CONFIG_SLUB_DEBUG */
5239 
5240 #ifdef CONFIG_SYSFS
5241 enum slab_stat_type {
5242 	SL_ALL,			/* All slabs */
5243 	SL_PARTIAL,		/* Only partially allocated slabs */
5244 	SL_CPU,			/* Only slabs used for cpu caches */
5245 	SL_OBJECTS,		/* Determine allocated objects not slabs */
5246 	SL_TOTAL		/* Determine object capacity not slabs */
5247 };
5248 
5249 #define SO_ALL		(1 << SL_ALL)
5250 #define SO_PARTIAL	(1 << SL_PARTIAL)
5251 #define SO_CPU		(1 << SL_CPU)
5252 #define SO_OBJECTS	(1 << SL_OBJECTS)
5253 #define SO_TOTAL	(1 << SL_TOTAL)
5254 
show_slab_objects(struct kmem_cache * s,char * buf,unsigned long flags)5255 static ssize_t show_slab_objects(struct kmem_cache *s,
5256 				 char *buf, unsigned long flags)
5257 {
5258 	unsigned long total = 0;
5259 	int node;
5260 	int x;
5261 	unsigned long *nodes;
5262 	int len = 0;
5263 
5264 	nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
5265 	if (!nodes)
5266 		return -ENOMEM;
5267 
5268 	if (flags & SO_CPU) {
5269 		int cpu;
5270 
5271 		for_each_possible_cpu(cpu) {
5272 			struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
5273 							       cpu);
5274 			int node;
5275 			struct slab *slab;
5276 
5277 			slab = READ_ONCE(c->slab);
5278 			if (!slab)
5279 				continue;
5280 
5281 			node = slab_nid(slab);
5282 			if (flags & SO_TOTAL)
5283 				x = slab->objects;
5284 			else if (flags & SO_OBJECTS)
5285 				x = slab->inuse;
5286 			else
5287 				x = 1;
5288 
5289 			total += x;
5290 			nodes[node] += x;
5291 
5292 #ifdef CONFIG_SLUB_CPU_PARTIAL
5293 			slab = slub_percpu_partial_read_once(c);
5294 			if (slab) {
5295 				node = slab_nid(slab);
5296 				if (flags & SO_TOTAL)
5297 					WARN_ON_ONCE(1);
5298 				else if (flags & SO_OBJECTS)
5299 					WARN_ON_ONCE(1);
5300 				else
5301 					x = slab->slabs;
5302 				total += x;
5303 				nodes[node] += x;
5304 			}
5305 #endif
5306 		}
5307 	}
5308 
5309 	/*
5310 	 * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex"
5311 	 * already held which will conflict with an existing lock order:
5312 	 *
5313 	 * mem_hotplug_lock->slab_mutex->kernfs_mutex
5314 	 *
5315 	 * We don't really need mem_hotplug_lock (to hold off
5316 	 * slab_mem_going_offline_callback) here because slab's memory hot
5317 	 * unplug code doesn't destroy the kmem_cache->node[] data.
5318 	 */
5319 
5320 #ifdef CONFIG_SLUB_DEBUG
5321 	if (flags & SO_ALL) {
5322 		struct kmem_cache_node *n;
5323 
5324 		for_each_kmem_cache_node(s, node, n) {
5325 
5326 			if (flags & SO_TOTAL)
5327 				x = atomic_long_read(&n->total_objects);
5328 			else if (flags & SO_OBJECTS)
5329 				x = atomic_long_read(&n->total_objects) -
5330 					count_partial(n, count_free);
5331 			else
5332 				x = atomic_long_read(&n->nr_slabs);
5333 			total += x;
5334 			nodes[node] += x;
5335 		}
5336 
5337 	} else
5338 #endif
5339 	if (flags & SO_PARTIAL) {
5340 		struct kmem_cache_node *n;
5341 
5342 		for_each_kmem_cache_node(s, node, n) {
5343 			if (flags & SO_TOTAL)
5344 				x = count_partial(n, count_total);
5345 			else if (flags & SO_OBJECTS)
5346 				x = count_partial(n, count_inuse);
5347 			else
5348 				x = n->nr_partial;
5349 			total += x;
5350 			nodes[node] += x;
5351 		}
5352 	}
5353 
5354 	len += sysfs_emit_at(buf, len, "%lu", total);
5355 #ifdef CONFIG_NUMA
5356 	for (node = 0; node < nr_node_ids; node++) {
5357 		if (nodes[node])
5358 			len += sysfs_emit_at(buf, len, " N%d=%lu",
5359 					     node, nodes[node]);
5360 	}
5361 #endif
5362 	len += sysfs_emit_at(buf, len, "\n");
5363 	kfree(nodes);
5364 
5365 	return len;
5366 }
5367 
5368 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
5369 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
5370 
5371 struct slab_attribute {
5372 	struct attribute attr;
5373 	ssize_t (*show)(struct kmem_cache *s, char *buf);
5374 	ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
5375 };
5376 
5377 #define SLAB_ATTR_RO(_name) \
5378 	static struct slab_attribute _name##_attr = __ATTR_RO_MODE(_name, 0400)
5379 
5380 #define SLAB_ATTR(_name) \
5381 	static struct slab_attribute _name##_attr = __ATTR_RW_MODE(_name, 0600)
5382 
slab_size_show(struct kmem_cache * s,char * buf)5383 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
5384 {
5385 	return sysfs_emit(buf, "%u\n", s->size);
5386 }
5387 SLAB_ATTR_RO(slab_size);
5388 
align_show(struct kmem_cache * s,char * buf)5389 static ssize_t align_show(struct kmem_cache *s, char *buf)
5390 {
5391 	return sysfs_emit(buf, "%u\n", s->align);
5392 }
5393 SLAB_ATTR_RO(align);
5394 
object_size_show(struct kmem_cache * s,char * buf)5395 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
5396 {
5397 	return sysfs_emit(buf, "%u\n", s->object_size);
5398 }
5399 SLAB_ATTR_RO(object_size);
5400 
objs_per_slab_show(struct kmem_cache * s,char * buf)5401 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
5402 {
5403 	return sysfs_emit(buf, "%u\n", oo_objects(s->oo));
5404 }
5405 SLAB_ATTR_RO(objs_per_slab);
5406 
order_show(struct kmem_cache * s,char * buf)5407 static ssize_t order_show(struct kmem_cache *s, char *buf)
5408 {
5409 	return sysfs_emit(buf, "%u\n", oo_order(s->oo));
5410 }
5411 SLAB_ATTR_RO(order);
5412 
min_partial_show(struct kmem_cache * s,char * buf)5413 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
5414 {
5415 	return sysfs_emit(buf, "%lu\n", s->min_partial);
5416 }
5417 
min_partial_store(struct kmem_cache * s,const char * buf,size_t length)5418 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
5419 				 size_t length)
5420 {
5421 	unsigned long min;
5422 	int err;
5423 
5424 	err = kstrtoul(buf, 10, &min);
5425 	if (err)
5426 		return err;
5427 
5428 	s->min_partial = min;
5429 	return length;
5430 }
5431 SLAB_ATTR(min_partial);
5432 
cpu_partial_show(struct kmem_cache * s,char * buf)5433 static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
5434 {
5435 	unsigned int nr_partial = 0;
5436 #ifdef CONFIG_SLUB_CPU_PARTIAL
5437 	nr_partial = s->cpu_partial;
5438 #endif
5439 
5440 	return sysfs_emit(buf, "%u\n", nr_partial);
5441 }
5442 
cpu_partial_store(struct kmem_cache * s,const char * buf,size_t length)5443 static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
5444 				 size_t length)
5445 {
5446 	unsigned int objects;
5447 	int err;
5448 
5449 	err = kstrtouint(buf, 10, &objects);
5450 	if (err)
5451 		return err;
5452 	if (objects && !kmem_cache_has_cpu_partial(s))
5453 		return -EINVAL;
5454 
5455 	slub_set_cpu_partial(s, objects);
5456 	flush_all(s);
5457 	return length;
5458 }
5459 SLAB_ATTR(cpu_partial);
5460 
ctor_show(struct kmem_cache * s,char * buf)5461 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
5462 {
5463 	if (!s->ctor)
5464 		return 0;
5465 	return sysfs_emit(buf, "%pS\n", s->ctor);
5466 }
5467 SLAB_ATTR_RO(ctor);
5468 
aliases_show(struct kmem_cache * s,char * buf)5469 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5470 {
5471 	return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
5472 }
5473 SLAB_ATTR_RO(aliases);
5474 
partial_show(struct kmem_cache * s,char * buf)5475 static ssize_t partial_show(struct kmem_cache *s, char *buf)
5476 {
5477 	return show_slab_objects(s, buf, SO_PARTIAL);
5478 }
5479 SLAB_ATTR_RO(partial);
5480 
cpu_slabs_show(struct kmem_cache * s,char * buf)5481 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5482 {
5483 	return show_slab_objects(s, buf, SO_CPU);
5484 }
5485 SLAB_ATTR_RO(cpu_slabs);
5486 
objects_show(struct kmem_cache * s,char * buf)5487 static ssize_t objects_show(struct kmem_cache *s, char *buf)
5488 {
5489 	return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
5490 }
5491 SLAB_ATTR_RO(objects);
5492 
objects_partial_show(struct kmem_cache * s,char * buf)5493 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5494 {
5495 	return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5496 }
5497 SLAB_ATTR_RO(objects_partial);
5498 
slabs_cpu_partial_show(struct kmem_cache * s,char * buf)5499 static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5500 {
5501 	int objects = 0;
5502 	int slabs = 0;
5503 	int cpu __maybe_unused;
5504 	int len = 0;
5505 
5506 #ifdef CONFIG_SLUB_CPU_PARTIAL
5507 	for_each_online_cpu(cpu) {
5508 		struct slab *slab;
5509 
5510 		slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5511 
5512 		if (slab)
5513 			slabs += slab->slabs;
5514 	}
5515 #endif
5516 
5517 	/* Approximate half-full slabs, see slub_set_cpu_partial() */
5518 	objects = (slabs * oo_objects(s->oo)) / 2;
5519 	len += sysfs_emit_at(buf, len, "%d(%d)", objects, slabs);
5520 
5521 #if defined(CONFIG_SLUB_CPU_PARTIAL) && defined(CONFIG_SMP)
5522 	for_each_online_cpu(cpu) {
5523 		struct slab *slab;
5524 
5525 		slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5526 		if (slab) {
5527 			slabs = READ_ONCE(slab->slabs);
5528 			objects = (slabs * oo_objects(s->oo)) / 2;
5529 			len += sysfs_emit_at(buf, len, " C%d=%d(%d)",
5530 					     cpu, objects, slabs);
5531 		}
5532 	}
5533 #endif
5534 	len += sysfs_emit_at(buf, len, "\n");
5535 
5536 	return len;
5537 }
5538 SLAB_ATTR_RO(slabs_cpu_partial);
5539 
reclaim_account_show(struct kmem_cache * s,char * buf)5540 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5541 {
5542 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5543 }
5544 SLAB_ATTR_RO(reclaim_account);
5545 
hwcache_align_show(struct kmem_cache * s,char * buf)5546 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5547 {
5548 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
5549 }
5550 SLAB_ATTR_RO(hwcache_align);
5551 
5552 #ifdef CONFIG_ZONE_DMA
cache_dma_show(struct kmem_cache * s,char * buf)5553 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5554 {
5555 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5556 }
5557 SLAB_ATTR_RO(cache_dma);
5558 #endif
5559 
usersize_show(struct kmem_cache * s,char * buf)5560 static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5561 {
5562 	return sysfs_emit(buf, "%u\n", s->usersize);
5563 }
5564 SLAB_ATTR_RO(usersize);
5565 
destroy_by_rcu_show(struct kmem_cache * s,char * buf)5566 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5567 {
5568 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
5569 }
5570 SLAB_ATTR_RO(destroy_by_rcu);
5571 
5572 #ifdef CONFIG_SLUB_DEBUG
slabs_show(struct kmem_cache * s,char * buf)5573 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5574 {
5575 	return show_slab_objects(s, buf, SO_ALL);
5576 }
5577 SLAB_ATTR_RO(slabs);
5578 
total_objects_show(struct kmem_cache * s,char * buf)5579 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5580 {
5581 	return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5582 }
5583 SLAB_ATTR_RO(total_objects);
5584 
sanity_checks_show(struct kmem_cache * s,char * buf)5585 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5586 {
5587 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
5588 }
5589 SLAB_ATTR_RO(sanity_checks);
5590 
trace_show(struct kmem_cache * s,char * buf)5591 static ssize_t trace_show(struct kmem_cache *s, char *buf)
5592 {
5593 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5594 }
5595 SLAB_ATTR_RO(trace);
5596 
red_zone_show(struct kmem_cache * s,char * buf)5597 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5598 {
5599 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5600 }
5601 
5602 SLAB_ATTR_RO(red_zone);
5603 
poison_show(struct kmem_cache * s,char * buf)5604 static ssize_t poison_show(struct kmem_cache *s, char *buf)
5605 {
5606 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON));
5607 }
5608 
5609 SLAB_ATTR_RO(poison);
5610 
store_user_show(struct kmem_cache * s,char * buf)5611 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5612 {
5613 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5614 }
5615 
5616 SLAB_ATTR_RO(store_user);
5617 
validate_show(struct kmem_cache * s,char * buf)5618 static ssize_t validate_show(struct kmem_cache *s, char *buf)
5619 {
5620 	return 0;
5621 }
5622 
validate_store(struct kmem_cache * s,const char * buf,size_t length)5623 static ssize_t validate_store(struct kmem_cache *s,
5624 			const char *buf, size_t length)
5625 {
5626 	int ret = -EINVAL;
5627 
5628 	if (buf[0] == '1') {
5629 		ret = validate_slab_cache(s);
5630 		if (ret >= 0)
5631 			ret = length;
5632 	}
5633 	return ret;
5634 }
5635 SLAB_ATTR(validate);
5636 
5637 #endif /* CONFIG_SLUB_DEBUG */
5638 
5639 #ifdef CONFIG_FAILSLAB
failslab_show(struct kmem_cache * s,char * buf)5640 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5641 {
5642 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5643 }
5644 SLAB_ATTR_RO(failslab);
5645 #endif
5646 
shrink_show(struct kmem_cache * s,char * buf)5647 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5648 {
5649 	return 0;
5650 }
5651 
shrink_store(struct kmem_cache * s,const char * buf,size_t length)5652 static ssize_t shrink_store(struct kmem_cache *s,
5653 			const char *buf, size_t length)
5654 {
5655 	if (buf[0] == '1')
5656 		kmem_cache_shrink(s);
5657 	else
5658 		return -EINVAL;
5659 	return length;
5660 }
5661 SLAB_ATTR(shrink);
5662 
5663 #ifdef CONFIG_NUMA
remote_node_defrag_ratio_show(struct kmem_cache * s,char * buf)5664 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
5665 {
5666 	return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10);
5667 }
5668 
remote_node_defrag_ratio_store(struct kmem_cache * s,const char * buf,size_t length)5669 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
5670 				const char *buf, size_t length)
5671 {
5672 	unsigned int ratio;
5673 	int err;
5674 
5675 	err = kstrtouint(buf, 10, &ratio);
5676 	if (err)
5677 		return err;
5678 	if (ratio > 100)
5679 		return -ERANGE;
5680 
5681 	s->remote_node_defrag_ratio = ratio * 10;
5682 
5683 	return length;
5684 }
5685 SLAB_ATTR(remote_node_defrag_ratio);
5686 #endif
5687 
5688 #ifdef CONFIG_SLUB_STATS
show_stat(struct kmem_cache * s,char * buf,enum stat_item si)5689 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5690 {
5691 	unsigned long sum  = 0;
5692 	int cpu;
5693 	int len = 0;
5694 	int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
5695 
5696 	if (!data)
5697 		return -ENOMEM;
5698 
5699 	for_each_online_cpu(cpu) {
5700 		unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
5701 
5702 		data[cpu] = x;
5703 		sum += x;
5704 	}
5705 
5706 	len += sysfs_emit_at(buf, len, "%lu", sum);
5707 
5708 #ifdef CONFIG_SMP
5709 	for_each_online_cpu(cpu) {
5710 		if (data[cpu])
5711 			len += sysfs_emit_at(buf, len, " C%d=%u",
5712 					     cpu, data[cpu]);
5713 	}
5714 #endif
5715 	kfree(data);
5716 	len += sysfs_emit_at(buf, len, "\n");
5717 
5718 	return len;
5719 }
5720 
clear_stat(struct kmem_cache * s,enum stat_item si)5721 static void clear_stat(struct kmem_cache *s, enum stat_item si)
5722 {
5723 	int cpu;
5724 
5725 	for_each_online_cpu(cpu)
5726 		per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
5727 }
5728 
5729 #define STAT_ATTR(si, text) 					\
5730 static ssize_t text##_show(struct kmem_cache *s, char *buf)	\
5731 {								\
5732 	return show_stat(s, buf, si);				\
5733 }								\
5734 static ssize_t text##_store(struct kmem_cache *s,		\
5735 				const char *buf, size_t length)	\
5736 {								\
5737 	if (buf[0] != '0')					\
5738 		return -EINVAL;					\
5739 	clear_stat(s, si);					\
5740 	return length;						\
5741 }								\
5742 SLAB_ATTR(text);						\
5743 
5744 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5745 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5746 STAT_ATTR(FREE_FASTPATH, free_fastpath);
5747 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5748 STAT_ATTR(FREE_FROZEN, free_frozen);
5749 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5750 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5751 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5752 STAT_ATTR(ALLOC_SLAB, alloc_slab);
5753 STAT_ATTR(ALLOC_REFILL, alloc_refill);
5754 STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
5755 STAT_ATTR(FREE_SLAB, free_slab);
5756 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5757 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5758 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5759 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5760 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5761 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
5762 STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
5763 STAT_ATTR(ORDER_FALLBACK, order_fallback);
5764 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5765 STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
5766 STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5767 STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
5768 STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5769 STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
5770 #endif	/* CONFIG_SLUB_STATS */
5771 
5772 static struct attribute *slab_attrs[] = {
5773 	&slab_size_attr.attr,
5774 	&object_size_attr.attr,
5775 	&objs_per_slab_attr.attr,
5776 	&order_attr.attr,
5777 	&min_partial_attr.attr,
5778 	&cpu_partial_attr.attr,
5779 	&objects_attr.attr,
5780 	&objects_partial_attr.attr,
5781 	&partial_attr.attr,
5782 	&cpu_slabs_attr.attr,
5783 	&ctor_attr.attr,
5784 	&aliases_attr.attr,
5785 	&align_attr.attr,
5786 	&hwcache_align_attr.attr,
5787 	&reclaim_account_attr.attr,
5788 	&destroy_by_rcu_attr.attr,
5789 	&shrink_attr.attr,
5790 	&slabs_cpu_partial_attr.attr,
5791 #ifdef CONFIG_SLUB_DEBUG
5792 	&total_objects_attr.attr,
5793 	&slabs_attr.attr,
5794 	&sanity_checks_attr.attr,
5795 	&trace_attr.attr,
5796 	&red_zone_attr.attr,
5797 	&poison_attr.attr,
5798 	&store_user_attr.attr,
5799 	&validate_attr.attr,
5800 #endif
5801 #ifdef CONFIG_ZONE_DMA
5802 	&cache_dma_attr.attr,
5803 #endif
5804 #ifdef CONFIG_NUMA
5805 	&remote_node_defrag_ratio_attr.attr,
5806 #endif
5807 #ifdef CONFIG_SLUB_STATS
5808 	&alloc_fastpath_attr.attr,
5809 	&alloc_slowpath_attr.attr,
5810 	&free_fastpath_attr.attr,
5811 	&free_slowpath_attr.attr,
5812 	&free_frozen_attr.attr,
5813 	&free_add_partial_attr.attr,
5814 	&free_remove_partial_attr.attr,
5815 	&alloc_from_partial_attr.attr,
5816 	&alloc_slab_attr.attr,
5817 	&alloc_refill_attr.attr,
5818 	&alloc_node_mismatch_attr.attr,
5819 	&free_slab_attr.attr,
5820 	&cpuslab_flush_attr.attr,
5821 	&deactivate_full_attr.attr,
5822 	&deactivate_empty_attr.attr,
5823 	&deactivate_to_head_attr.attr,
5824 	&deactivate_to_tail_attr.attr,
5825 	&deactivate_remote_frees_attr.attr,
5826 	&deactivate_bypass_attr.attr,
5827 	&order_fallback_attr.attr,
5828 	&cmpxchg_double_fail_attr.attr,
5829 	&cmpxchg_double_cpu_fail_attr.attr,
5830 	&cpu_partial_alloc_attr.attr,
5831 	&cpu_partial_free_attr.attr,
5832 	&cpu_partial_node_attr.attr,
5833 	&cpu_partial_drain_attr.attr,
5834 #endif
5835 #ifdef CONFIG_FAILSLAB
5836 	&failslab_attr.attr,
5837 #endif
5838 	&usersize_attr.attr,
5839 
5840 	NULL
5841 };
5842 
5843 static const struct attribute_group slab_attr_group = {
5844 	.attrs = slab_attrs,
5845 };
5846 
slab_attr_show(struct kobject * kobj,struct attribute * attr,char * buf)5847 static ssize_t slab_attr_show(struct kobject *kobj,
5848 				struct attribute *attr,
5849 				char *buf)
5850 {
5851 	struct slab_attribute *attribute;
5852 	struct kmem_cache *s;
5853 	int err;
5854 
5855 	attribute = to_slab_attr(attr);
5856 	s = to_slab(kobj);
5857 
5858 	if (!attribute->show)
5859 		return -EIO;
5860 
5861 	err = attribute->show(s, buf);
5862 
5863 	return err;
5864 }
5865 
slab_attr_store(struct kobject * kobj,struct attribute * attr,const char * buf,size_t len)5866 static ssize_t slab_attr_store(struct kobject *kobj,
5867 				struct attribute *attr,
5868 				const char *buf, size_t len)
5869 {
5870 	struct slab_attribute *attribute;
5871 	struct kmem_cache *s;
5872 	int err;
5873 
5874 	attribute = to_slab_attr(attr);
5875 	s = to_slab(kobj);
5876 
5877 	if (!attribute->store)
5878 		return -EIO;
5879 
5880 	err = attribute->store(s, buf, len);
5881 	return err;
5882 }
5883 
kmem_cache_release(struct kobject * k)5884 static void kmem_cache_release(struct kobject *k)
5885 {
5886 	slab_kmem_cache_release(to_slab(k));
5887 }
5888 
5889 static const struct sysfs_ops slab_sysfs_ops = {
5890 	.show = slab_attr_show,
5891 	.store = slab_attr_store,
5892 };
5893 
5894 static struct kobj_type slab_ktype = {
5895 	.sysfs_ops = &slab_sysfs_ops,
5896 	.release = kmem_cache_release,
5897 };
5898 
5899 static struct kset *slab_kset;
5900 
cache_kset(struct kmem_cache * s)5901 static inline struct kset *cache_kset(struct kmem_cache *s)
5902 {
5903 	return slab_kset;
5904 }
5905 
5906 #define ID_STR_LENGTH 64
5907 
5908 /* Create a unique string id for a slab cache:
5909  *
5910  * Format	:[flags-]size
5911  */
create_unique_id(struct kmem_cache * s)5912 static char *create_unique_id(struct kmem_cache *s)
5913 {
5914 	char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5915 	char *p = name;
5916 
5917 	BUG_ON(!name);
5918 
5919 	*p++ = ':';
5920 	/*
5921 	 * First flags affecting slabcache operations. We will only
5922 	 * get here for aliasable slabs so we do not need to support
5923 	 * too many flags. The flags here must cover all flags that
5924 	 * are matched during merging to guarantee that the id is
5925 	 * unique.
5926 	 */
5927 	if (s->flags & SLAB_CACHE_DMA)
5928 		*p++ = 'd';
5929 	if (s->flags & SLAB_CACHE_DMA32)
5930 		*p++ = 'D';
5931 	if (s->flags & SLAB_RECLAIM_ACCOUNT)
5932 		*p++ = 'a';
5933 	if (s->flags & SLAB_CONSISTENCY_CHECKS)
5934 		*p++ = 'F';
5935 	if (s->flags & SLAB_ACCOUNT)
5936 		*p++ = 'A';
5937 	if (p != name + 1)
5938 		*p++ = '-';
5939 	p += sprintf(p, "%07u", s->size);
5940 
5941 	BUG_ON(p > name + ID_STR_LENGTH - 1);
5942 	return name;
5943 }
5944 
sysfs_slab_add(struct kmem_cache * s)5945 static int sysfs_slab_add(struct kmem_cache *s)
5946 {
5947 	int err;
5948 	const char *name;
5949 	struct kset *kset = cache_kset(s);
5950 	int unmergeable = slab_unmergeable(s);
5951 
5952 	if (!kset) {
5953 		kobject_init(&s->kobj, &slab_ktype);
5954 		return 0;
5955 	}
5956 
5957 	if (!unmergeable && disable_higher_order_debug &&
5958 			(slub_debug & DEBUG_METADATA_FLAGS))
5959 		unmergeable = 1;
5960 
5961 	if (unmergeable) {
5962 		/*
5963 		 * Slabcache can never be merged so we can use the name proper.
5964 		 * This is typically the case for debug situations. In that
5965 		 * case we can catch duplicate names easily.
5966 		 */
5967 		sysfs_remove_link(&slab_kset->kobj, s->name);
5968 		name = s->name;
5969 	} else {
5970 		/*
5971 		 * Create a unique name for the slab as a target
5972 		 * for the symlinks.
5973 		 */
5974 		name = create_unique_id(s);
5975 	}
5976 
5977 	s->kobj.kset = kset;
5978 	err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
5979 	if (err)
5980 		goto out;
5981 
5982 	err = sysfs_create_group(&s->kobj, &slab_attr_group);
5983 	if (err)
5984 		goto out_del_kobj;
5985 
5986 	if (!unmergeable) {
5987 		/* Setup first alias */
5988 		sysfs_slab_alias(s, s->name);
5989 	}
5990 out:
5991 	if (!unmergeable)
5992 		kfree(name);
5993 	return err;
5994 out_del_kobj:
5995 	kobject_del(&s->kobj);
5996 	goto out;
5997 }
5998 
sysfs_slab_unlink(struct kmem_cache * s)5999 void sysfs_slab_unlink(struct kmem_cache *s)
6000 {
6001 	if (slab_state >= FULL)
6002 		kobject_del(&s->kobj);
6003 }
6004 
sysfs_slab_release(struct kmem_cache * s)6005 void sysfs_slab_release(struct kmem_cache *s)
6006 {
6007 	if (slab_state >= FULL)
6008 		kobject_put(&s->kobj);
6009 }
6010 
6011 /*
6012  * Need to buffer aliases during bootup until sysfs becomes
6013  * available lest we lose that information.
6014  */
6015 struct saved_alias {
6016 	struct kmem_cache *s;
6017 	const char *name;
6018 	struct saved_alias *next;
6019 };
6020 
6021 static struct saved_alias *alias_list;
6022 
sysfs_slab_alias(struct kmem_cache * s,const char * name)6023 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
6024 {
6025 	struct saved_alias *al;
6026 
6027 	if (slab_state == FULL) {
6028 		/*
6029 		 * If we have a leftover link then remove it.
6030 		 */
6031 		sysfs_remove_link(&slab_kset->kobj, name);
6032 		return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
6033 	}
6034 
6035 	al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
6036 	if (!al)
6037 		return -ENOMEM;
6038 
6039 	al->s = s;
6040 	al->name = name;
6041 	al->next = alias_list;
6042 	alias_list = al;
6043 	return 0;
6044 }
6045 
slab_sysfs_init(void)6046 static int __init slab_sysfs_init(void)
6047 {
6048 	struct kmem_cache *s;
6049 	int err;
6050 
6051 	mutex_lock(&slab_mutex);
6052 
6053 	slab_kset = kset_create_and_add("slab", NULL, kernel_kobj);
6054 	if (!slab_kset) {
6055 		mutex_unlock(&slab_mutex);
6056 		pr_err("Cannot register slab subsystem.\n");
6057 		return -ENOSYS;
6058 	}
6059 
6060 	slab_state = FULL;
6061 
6062 	list_for_each_entry(s, &slab_caches, list) {
6063 		err = sysfs_slab_add(s);
6064 		if (err)
6065 			pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
6066 			       s->name);
6067 	}
6068 
6069 	while (alias_list) {
6070 		struct saved_alias *al = alias_list;
6071 
6072 		alias_list = alias_list->next;
6073 		err = sysfs_slab_alias(al->s, al->name);
6074 		if (err)
6075 			pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
6076 			       al->name);
6077 		kfree(al);
6078 	}
6079 
6080 	mutex_unlock(&slab_mutex);
6081 	return 0;
6082 }
6083 
6084 __initcall(slab_sysfs_init);
6085 #endif /* CONFIG_SYSFS */
6086 
6087 #if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS)
slab_debugfs_show(struct seq_file * seq,void * v)6088 static int slab_debugfs_show(struct seq_file *seq, void *v)
6089 {
6090 	struct loc_track *t = seq->private;
6091 	struct location *l;
6092 	unsigned long idx;
6093 
6094 	idx = (unsigned long) t->idx;
6095 	if (idx < t->count) {
6096 		l = &t->loc[idx];
6097 
6098 		seq_printf(seq, "%7ld ", l->count);
6099 
6100 		if (l->addr)
6101 			seq_printf(seq, "%pS", (void *)l->addr);
6102 		else
6103 			seq_puts(seq, "<not-available>");
6104 
6105 		if (l->sum_time != l->min_time) {
6106 			seq_printf(seq, " age=%ld/%llu/%ld",
6107 				l->min_time, div_u64(l->sum_time, l->count),
6108 				l->max_time);
6109 		} else
6110 			seq_printf(seq, " age=%ld", l->min_time);
6111 
6112 		if (l->min_pid != l->max_pid)
6113 			seq_printf(seq, " pid=%ld-%ld", l->min_pid, l->max_pid);
6114 		else
6115 			seq_printf(seq, " pid=%ld",
6116 				l->min_pid);
6117 
6118 		if (num_online_cpus() > 1 && !cpumask_empty(to_cpumask(l->cpus)))
6119 			seq_printf(seq, " cpus=%*pbl",
6120 				 cpumask_pr_args(to_cpumask(l->cpus)));
6121 
6122 		if (nr_online_nodes > 1 && !nodes_empty(l->nodes))
6123 			seq_printf(seq, " nodes=%*pbl",
6124 				 nodemask_pr_args(&l->nodes));
6125 
6126 #ifdef CONFIG_STACKDEPOT
6127 		{
6128 			depot_stack_handle_t handle;
6129 			unsigned long *entries;
6130 			unsigned int nr_entries, j;
6131 
6132 			handle = READ_ONCE(l->handle);
6133 			if (handle) {
6134 				nr_entries = stack_depot_fetch(handle, &entries);
6135 				seq_puts(seq, "\n");
6136 				for (j = 0; j < nr_entries; j++)
6137 					seq_printf(seq, "        %pS\n", (void *)entries[j]);
6138 			}
6139 		}
6140 #endif
6141 		seq_puts(seq, "\n");
6142 	}
6143 
6144 	if (!idx && !t->count)
6145 		seq_puts(seq, "No data\n");
6146 
6147 	return 0;
6148 }
6149 
slab_debugfs_stop(struct seq_file * seq,void * v)6150 static void slab_debugfs_stop(struct seq_file *seq, void *v)
6151 {
6152 }
6153 
slab_debugfs_next(struct seq_file * seq,void * v,loff_t * ppos)6154 static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos)
6155 {
6156 	struct loc_track *t = seq->private;
6157 
6158 	t->idx = ++(*ppos);
6159 	if (*ppos <= t->count)
6160 		return ppos;
6161 
6162 	return NULL;
6163 }
6164 
cmp_loc_by_count(const void * a,const void * b,const void * data)6165 static int cmp_loc_by_count(const void *a, const void *b, const void *data)
6166 {
6167 	struct location *loc1 = (struct location *)a;
6168 	struct location *loc2 = (struct location *)b;
6169 
6170 	if (loc1->count > loc2->count)
6171 		return -1;
6172 	else
6173 		return 1;
6174 }
6175 
slab_debugfs_start(struct seq_file * seq,loff_t * ppos)6176 static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos)
6177 {
6178 	struct loc_track *t = seq->private;
6179 
6180 	t->idx = *ppos;
6181 	return ppos;
6182 }
6183 
6184 static const struct seq_operations slab_debugfs_sops = {
6185 	.start  = slab_debugfs_start,
6186 	.next   = slab_debugfs_next,
6187 	.stop   = slab_debugfs_stop,
6188 	.show   = slab_debugfs_show,
6189 };
6190 
slab_debug_trace_open(struct inode * inode,struct file * filep)6191 static int slab_debug_trace_open(struct inode *inode, struct file *filep)
6192 {
6193 
6194 	struct kmem_cache_node *n;
6195 	enum track_item alloc;
6196 	int node;
6197 	struct loc_track *t = __seq_open_private(filep, &slab_debugfs_sops,
6198 						sizeof(struct loc_track));
6199 	struct kmem_cache *s = file_inode(filep)->i_private;
6200 	unsigned long *obj_map;
6201 
6202 	if (!t)
6203 		return -ENOMEM;
6204 
6205 	obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL);
6206 	if (!obj_map) {
6207 		seq_release_private(inode, filep);
6208 		return -ENOMEM;
6209 	}
6210 
6211 	if (strcmp(filep->f_path.dentry->d_name.name, "alloc_traces") == 0)
6212 		alloc = TRACK_ALLOC;
6213 	else
6214 		alloc = TRACK_FREE;
6215 
6216 	if (!alloc_loc_track(t, PAGE_SIZE / sizeof(struct location), GFP_KERNEL)) {
6217 		bitmap_free(obj_map);
6218 		seq_release_private(inode, filep);
6219 		return -ENOMEM;
6220 	}
6221 
6222 	for_each_kmem_cache_node(s, node, n) {
6223 		unsigned long flags;
6224 		struct slab *slab;
6225 
6226 		if (!atomic_long_read(&n->nr_slabs))
6227 			continue;
6228 
6229 		spin_lock_irqsave(&n->list_lock, flags);
6230 		list_for_each_entry(slab, &n->partial, slab_list)
6231 			process_slab(t, s, slab, alloc, obj_map);
6232 		list_for_each_entry(slab, &n->full, slab_list)
6233 			process_slab(t, s, slab, alloc, obj_map);
6234 		spin_unlock_irqrestore(&n->list_lock, flags);
6235 	}
6236 
6237 	/* Sort locations by count */
6238 	sort_r(t->loc, t->count, sizeof(struct location),
6239 		cmp_loc_by_count, NULL, NULL);
6240 
6241 	bitmap_free(obj_map);
6242 	return 0;
6243 }
6244 
slab_debug_trace_release(struct inode * inode,struct file * file)6245 static int slab_debug_trace_release(struct inode *inode, struct file *file)
6246 {
6247 	struct seq_file *seq = file->private_data;
6248 	struct loc_track *t = seq->private;
6249 
6250 	free_loc_track(t);
6251 	return seq_release_private(inode, file);
6252 }
6253 
6254 static const struct file_operations slab_debugfs_fops = {
6255 	.open    = slab_debug_trace_open,
6256 	.read    = seq_read,
6257 	.llseek  = seq_lseek,
6258 	.release = slab_debug_trace_release,
6259 };
6260 
debugfs_slab_add(struct kmem_cache * s)6261 static void debugfs_slab_add(struct kmem_cache *s)
6262 {
6263 	struct dentry *slab_cache_dir;
6264 
6265 	if (unlikely(!slab_debugfs_root))
6266 		return;
6267 
6268 	slab_cache_dir = debugfs_create_dir(s->name, slab_debugfs_root);
6269 
6270 	debugfs_create_file("alloc_traces", 0400,
6271 		slab_cache_dir, s, &slab_debugfs_fops);
6272 
6273 	debugfs_create_file("free_traces", 0400,
6274 		slab_cache_dir, s, &slab_debugfs_fops);
6275 }
6276 
debugfs_slab_release(struct kmem_cache * s)6277 void debugfs_slab_release(struct kmem_cache *s)
6278 {
6279 	debugfs_remove_recursive(debugfs_lookup(s->name, slab_debugfs_root));
6280 }
6281 
slab_debugfs_init(void)6282 static int __init slab_debugfs_init(void)
6283 {
6284 	struct kmem_cache *s;
6285 
6286 	slab_debugfs_root = debugfs_create_dir("slab", NULL);
6287 
6288 	list_for_each_entry(s, &slab_caches, list)
6289 		if (s->flags & SLAB_STORE_USER)
6290 			debugfs_slab_add(s);
6291 
6292 	return 0;
6293 
6294 }
6295 __initcall(slab_debugfs_init);
6296 #endif
6297 /*
6298  * The /proc/slabinfo ABI
6299  */
6300 #ifdef CONFIG_SLUB_DEBUG
get_slabinfo(struct kmem_cache * s,struct slabinfo * sinfo)6301 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
6302 {
6303 	unsigned long nr_slabs = 0;
6304 	unsigned long nr_objs = 0;
6305 	unsigned long nr_free = 0;
6306 	int node;
6307 	struct kmem_cache_node *n;
6308 
6309 	for_each_kmem_cache_node(s, node, n) {
6310 		nr_slabs += node_nr_slabs(n);
6311 		nr_objs += node_nr_objs(n);
6312 		nr_free += count_partial(n, count_free);
6313 	}
6314 
6315 	sinfo->active_objs = nr_objs - nr_free;
6316 	sinfo->num_objs = nr_objs;
6317 	sinfo->active_slabs = nr_slabs;
6318 	sinfo->num_slabs = nr_slabs;
6319 	sinfo->objects_per_slab = oo_objects(s->oo);
6320 	sinfo->cache_order = oo_order(s->oo);
6321 }
6322 
slabinfo_show_stats(struct seq_file * m,struct kmem_cache * s)6323 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
6324 {
6325 }
6326 
slabinfo_write(struct file * file,const char __user * buffer,size_t count,loff_t * ppos)6327 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
6328 		       size_t count, loff_t *ppos)
6329 {
6330 	return -EIO;
6331 }
6332 #endif /* CONFIG_SLUB_DEBUG */
6333