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/bitops.h>
19 #include <linux/slab.h>
20 #include "slab.h"
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/kasan.h>
24 #include <linux/cpu.h>
25 #include <linux/cpuset.h>
26 #include <linux/mempolicy.h>
27 #include <linux/ctype.h>
28 #include <linux/debugobjects.h>
29 #include <linux/kallsyms.h>
30 #include <linux/kfence.h>
31 #include <linux/memory.h>
32 #include <linux/math64.h>
33 #include <linux/fault-inject.h>
34 #include <linux/stacktrace.h>
35 #include <linux/prefetch.h>
36 #include <linux/memcontrol.h>
37 #include <linux/random.h>
38 
39 #include <trace/events/kmem.h>
40 
41 #include "internal.h"
42 
43 /*
44  * Lock order:
45  *   1. slab_mutex (Global Mutex)
46  *   2. node->list_lock
47  *   3. slab_lock(page) (Only on some arches and for debugging)
48  *
49  *   slab_mutex
50  *
51  *   The role of the slab_mutex is to protect the list of all the slabs
52  *   and to synchronize major metadata changes to slab cache structures.
53  *
54  *   The slab_lock is only used for debugging and on arches that do not
55  *   have the ability to do a cmpxchg_double. It only protects:
56  *	A. page->freelist	-> List of object free in a page
57  *	B. page->inuse		-> Number of objects in use
58  *	C. page->objects	-> Number of objects in page
59  *	D. page->frozen		-> frozen state
60  *
61  *   If a slab is frozen then it is exempt from list management. It is not
62  *   on any list except per cpu partial list. The processor that froze the
63  *   slab is the one who can perform list operations on the page. Other
64  *   processors may put objects onto the freelist but the processor that
65  *   froze the slab is the only one that can retrieve the objects from the
66  *   page's freelist.
67  *
68  *   The list_lock protects the partial and full list on each node and
69  *   the partial slab counter. If taken then no new slabs may be added or
70  *   removed from the lists nor make the number of partial slabs be modified.
71  *   (Note that the total number of slabs is an atomic value that may be
72  *   modified without taking the list lock).
73  *
74  *   The list_lock is a centralized lock and thus we avoid taking it as
75  *   much as possible. As long as SLUB does not have to handle partial
76  *   slabs, operations can continue without any centralized lock. F.e.
77  *   allocating a long series of objects that fill up slabs does not require
78  *   the list lock.
79  *   Interrupts are disabled during allocation and deallocation in order to
80  *   make the slab allocator safe to use in the context of an irq. In addition
81  *   interrupts are disabled to ensure that the processor does not change
82  *   while handling per_cpu slabs, due to kernel preemption.
83  *
84  * SLUB assigns one slab for allocation to each processor.
85  * Allocations only occur from these slabs called cpu slabs.
86  *
87  * Slabs with free elements are kept on a partial list and during regular
88  * operations no list for full slabs is used. If an object in a full slab is
89  * freed then the slab will show up again on the partial lists.
90  * We track full slabs for debugging purposes though because otherwise we
91  * cannot scan all objects.
92  *
93  * Slabs are freed when they become empty. Teardown and setup is
94  * minimal so we rely on the page allocators per cpu caches for
95  * fast frees and allocs.
96  *
97  * page->frozen		The slab is frozen and exempt from list processing.
98  * 			This means that the slab is dedicated to a purpose
99  * 			such as satisfying allocations for a specific
100  * 			processor. Objects may be freed in the slab while
101  * 			it is frozen but slab_free will then skip the usual
102  * 			list operations. It is up to the processor holding
103  * 			the slab to integrate the slab into the slab lists
104  * 			when the slab is no longer needed.
105  *
106  * 			One use of this flag is to mark slabs that are
107  * 			used for allocations. Then such a slab becomes a cpu
108  * 			slab. The cpu slab may be equipped with an additional
109  * 			freelist that allows lockless access to
110  * 			free objects in addition to the regular freelist
111  * 			that requires the slab lock.
112  *
113  * SLAB_DEBUG_FLAGS	Slab requires special handling due to debug
114  * 			options set. This moves	slab handling out of
115  * 			the fast path and disables lockless freelists.
116  */
117 
118 #ifdef CONFIG_SLUB_DEBUG
119 #ifdef CONFIG_SLUB_DEBUG_ON
120 DEFINE_STATIC_KEY_TRUE(slub_debug_enabled);
121 #else
122 DEFINE_STATIC_KEY_FALSE(slub_debug_enabled);
123 #endif
124 #endif
125 
kmem_cache_debug(struct kmem_cache * s)126 static inline bool kmem_cache_debug(struct kmem_cache *s)
127 {
128 	return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS);
129 }
130 
fixup_red_left(struct kmem_cache * s,void * p)131 void *fixup_red_left(struct kmem_cache *s, void *p)
132 {
133 	if (kmem_cache_debug_flags(s, SLAB_RED_ZONE))
134 		p += s->red_left_pad;
135 
136 	return p;
137 }
138 
kmem_cache_has_cpu_partial(struct kmem_cache * s)139 static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
140 {
141 #ifdef CONFIG_SLUB_CPU_PARTIAL
142 	return !kmem_cache_debug(s);
143 #else
144 	return false;
145 #endif
146 }
147 
148 /*
149  * Issues still to be resolved:
150  *
151  * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
152  *
153  * - Variable sizing of the per node arrays
154  */
155 
156 /* Enable to test recovery from slab corruption on boot */
157 #undef SLUB_RESILIENCY_TEST
158 
159 /* Enable to log cmpxchg failures */
160 #undef SLUB_DEBUG_CMPXCHG
161 
162 /*
163  * Minimum number of partial slabs. These will be left on the partial
164  * lists even if they are empty. kmem_cache_shrink may reclaim them.
165  */
166 #define MIN_PARTIAL 5
167 
168 /*
169  * Maximum number of desirable partial slabs.
170  * The existence of more partial slabs makes kmem_cache_shrink
171  * sort the partial list by the number of objects in use.
172  */
173 #define MAX_PARTIAL 10
174 
175 #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
176 				SLAB_POISON | SLAB_STORE_USER)
177 
178 /*
179  * These debug flags cannot use CMPXCHG because there might be consistency
180  * issues when checking or reading debug information
181  */
182 #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
183 				SLAB_TRACE)
184 
185 
186 /*
187  * Debugging flags that require metadata to be stored in the slab.  These get
188  * disabled when slub_debug=O is used and a cache's min order increases with
189  * metadata.
190  */
191 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
192 
193 #define OO_SHIFT	16
194 #define OO_MASK		((1 << OO_SHIFT) - 1)
195 #define MAX_OBJS_PER_PAGE	32767 /* since page.objects is u15 */
196 
197 /* Internal SLUB flags */
198 /* Poison object */
199 #define __OBJECT_POISON		((slab_flags_t __force)0x80000000U)
200 /* Use cmpxchg_double */
201 #define __CMPXCHG_DOUBLE	((slab_flags_t __force)0x40000000U)
202 
203 /*
204  * Tracking user of a slab.
205  */
206 #define TRACK_ADDRS_COUNT 16
207 struct track {
208 	unsigned long addr;	/* Called from address */
209 #ifdef CONFIG_STACKTRACE
210 	unsigned long addrs[TRACK_ADDRS_COUNT];	/* Called from address */
211 #endif
212 	int cpu;		/* Was running on cpu */
213 	int pid;		/* Pid context */
214 	unsigned long when;	/* When did the operation occur */
215 };
216 
217 enum track_item { TRACK_ALLOC, TRACK_FREE };
218 
219 #ifdef CONFIG_SYSFS
220 static int sysfs_slab_add(struct kmem_cache *);
221 static int sysfs_slab_alias(struct kmem_cache *, const char *);
222 #else
sysfs_slab_add(struct kmem_cache * s)223 static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
sysfs_slab_alias(struct kmem_cache * s,const char * p)224 static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
225 							{ return 0; }
226 #endif
227 
stat(const struct kmem_cache * s,enum stat_item si)228 static inline void stat(const struct kmem_cache *s, enum stat_item si)
229 {
230 #ifdef CONFIG_SLUB_STATS
231 	/*
232 	 * The rmw is racy on a preemptible kernel but this is acceptable, so
233 	 * avoid this_cpu_add()'s irq-disable overhead.
234 	 */
235 	raw_cpu_inc(s->cpu_slab->stat[si]);
236 #endif
237 }
238 
239 /*
240  * Tracks for which NUMA nodes we have kmem_cache_nodes allocated.
241  * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily
242  * differ during memory hotplug/hotremove operations.
243  * Protected by slab_mutex.
244  */
245 static nodemask_t slab_nodes;
246 
247 /********************************************************************
248  * 			Core slab cache functions
249  *******************************************************************/
250 
251 /*
252  * Returns freelist pointer (ptr). With hardening, this is obfuscated
253  * with an XOR of the address where the pointer is held and a per-cache
254  * random number.
255  */
freelist_ptr(const struct kmem_cache * s,void * ptr,unsigned long ptr_addr)256 static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
257 				 unsigned long ptr_addr)
258 {
259 #ifdef CONFIG_SLAB_FREELIST_HARDENED
260 	/*
261 	 * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged.
262 	 * Normally, this doesn't cause any issues, as both set_freepointer()
263 	 * and get_freepointer() are called with a pointer with the same tag.
264 	 * However, there are some issues with CONFIG_SLUB_DEBUG code. For
265 	 * example, when __free_slub() iterates over objects in a cache, it
266 	 * passes untagged pointers to check_object(). check_object() in turns
267 	 * calls get_freepointer() with an untagged pointer, which causes the
268 	 * freepointer to be restored incorrectly.
269 	 */
270 	return (void *)((unsigned long)ptr ^ s->random ^
271 			swab((unsigned long)kasan_reset_tag((void *)ptr_addr)));
272 #else
273 	return ptr;
274 #endif
275 }
276 
277 /* Returns the freelist pointer recorded at location ptr_addr. */
freelist_dereference(const struct kmem_cache * s,void * ptr_addr)278 static inline void *freelist_dereference(const struct kmem_cache *s,
279 					 void *ptr_addr)
280 {
281 	return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr),
282 			    (unsigned long)ptr_addr);
283 }
284 
get_freepointer(struct kmem_cache * s,void * object)285 static inline void *get_freepointer(struct kmem_cache *s, void *object)
286 {
287 	object = kasan_reset_tag(object);
288 	return freelist_dereference(s, object + s->offset);
289 }
290 
prefetch_freepointer(const struct kmem_cache * s,void * object)291 static void prefetch_freepointer(const struct kmem_cache *s, void *object)
292 {
293 	prefetch(object + s->offset);
294 }
295 
get_freepointer_safe(struct kmem_cache * s,void * object)296 static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
297 {
298 	unsigned long freepointer_addr;
299 	void *p;
300 
301 	if (!debug_pagealloc_enabled_static())
302 		return get_freepointer(s, object);
303 
304 	freepointer_addr = (unsigned long)object + s->offset;
305 	copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p));
306 	return freelist_ptr(s, p, freepointer_addr);
307 }
308 
set_freepointer(struct kmem_cache * s,void * object,void * fp)309 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
310 {
311 	unsigned long freeptr_addr = (unsigned long)object + s->offset;
312 
313 #ifdef CONFIG_SLAB_FREELIST_HARDENED
314 	BUG_ON(object == fp); /* naive detection of double free or corruption */
315 #endif
316 
317 	freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr);
318 	*(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
319 }
320 
321 /* Loop over all objects in a slab */
322 #define for_each_object(__p, __s, __addr, __objects) \
323 	for (__p = fixup_red_left(__s, __addr); \
324 		__p < (__addr) + (__objects) * (__s)->size; \
325 		__p += (__s)->size)
326 
order_objects(unsigned int order,unsigned int size)327 static inline unsigned int order_objects(unsigned int order, unsigned int size)
328 {
329 	return ((unsigned int)PAGE_SIZE << order) / size;
330 }
331 
oo_make(unsigned int order,unsigned int size)332 static inline struct kmem_cache_order_objects oo_make(unsigned int order,
333 		unsigned int size)
334 {
335 	struct kmem_cache_order_objects x = {
336 		(order << OO_SHIFT) + order_objects(order, size)
337 	};
338 
339 	return x;
340 }
341 
oo_order(struct kmem_cache_order_objects x)342 static inline unsigned int oo_order(struct kmem_cache_order_objects x)
343 {
344 	return x.x >> OO_SHIFT;
345 }
346 
oo_objects(struct kmem_cache_order_objects x)347 static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
348 {
349 	return x.x & OO_MASK;
350 }
351 
352 /*
353  * Per slab locking using the pagelock
354  */
slab_lock(struct page * page)355 static __always_inline void slab_lock(struct page *page)
356 {
357 	VM_BUG_ON_PAGE(PageTail(page), page);
358 	bit_spin_lock(PG_locked, &page->flags);
359 }
360 
slab_unlock(struct page * page)361 static __always_inline void slab_unlock(struct page *page)
362 {
363 	VM_BUG_ON_PAGE(PageTail(page), page);
364 	__bit_spin_unlock(PG_locked, &page->flags);
365 }
366 
367 /* Interrupts must be disabled (for the fallback code to work right) */
__cmpxchg_double_slab(struct kmem_cache * s,struct page * page,void * freelist_old,unsigned long counters_old,void * freelist_new,unsigned long counters_new,const char * n)368 static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
369 		void *freelist_old, unsigned long counters_old,
370 		void *freelist_new, unsigned long counters_new,
371 		const char *n)
372 {
373 	VM_BUG_ON(!irqs_disabled());
374 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
375     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
376 	if (s->flags & __CMPXCHG_DOUBLE) {
377 		if (cmpxchg_double(&page->freelist, &page->counters,
378 				   freelist_old, counters_old,
379 				   freelist_new, counters_new))
380 			return true;
381 	} else
382 #endif
383 	{
384 		slab_lock(page);
385 		if (page->freelist == freelist_old &&
386 					page->counters == counters_old) {
387 			page->freelist = freelist_new;
388 			page->counters = counters_new;
389 			slab_unlock(page);
390 			return true;
391 		}
392 		slab_unlock(page);
393 	}
394 
395 	cpu_relax();
396 	stat(s, CMPXCHG_DOUBLE_FAIL);
397 
398 #ifdef SLUB_DEBUG_CMPXCHG
399 	pr_info("%s %s: cmpxchg double redo ", n, s->name);
400 #endif
401 
402 	return false;
403 }
404 
cmpxchg_double_slab(struct kmem_cache * s,struct page * page,void * freelist_old,unsigned long counters_old,void * freelist_new,unsigned long counters_new,const char * n)405 static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
406 		void *freelist_old, unsigned long counters_old,
407 		void *freelist_new, unsigned long counters_new,
408 		const char *n)
409 {
410 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
411     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
412 	if (s->flags & __CMPXCHG_DOUBLE) {
413 		if (cmpxchg_double(&page->freelist, &page->counters,
414 				   freelist_old, counters_old,
415 				   freelist_new, counters_new))
416 			return true;
417 	} else
418 #endif
419 	{
420 		unsigned long flags;
421 
422 		local_irq_save(flags);
423 		slab_lock(page);
424 		if (page->freelist == freelist_old &&
425 					page->counters == counters_old) {
426 			page->freelist = freelist_new;
427 			page->counters = counters_new;
428 			slab_unlock(page);
429 			local_irq_restore(flags);
430 			return true;
431 		}
432 		slab_unlock(page);
433 		local_irq_restore(flags);
434 	}
435 
436 	cpu_relax();
437 	stat(s, CMPXCHG_DOUBLE_FAIL);
438 
439 #ifdef SLUB_DEBUG_CMPXCHG
440 	pr_info("%s %s: cmpxchg double redo ", n, s->name);
441 #endif
442 
443 	return false;
444 }
445 
446 #ifdef CONFIG_SLUB_DEBUG
447 static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)];
448 static DEFINE_SPINLOCK(object_map_lock);
449 
450 /*
451  * Determine a map of object in use on a page.
452  *
453  * Node listlock must be held to guarantee that the page does
454  * not vanish from under us.
455  */
get_map(struct kmem_cache * s,struct page * page)456 static unsigned long *get_map(struct kmem_cache *s, struct page *page)
457 	__acquires(&object_map_lock)
458 {
459 	void *p;
460 	void *addr = page_address(page);
461 
462 	VM_BUG_ON(!irqs_disabled());
463 
464 	spin_lock(&object_map_lock);
465 
466 	bitmap_zero(object_map, page->objects);
467 
468 	for (p = page->freelist; p; p = get_freepointer(s, p))
469 		set_bit(__obj_to_index(s, addr, p), object_map);
470 
471 	return object_map;
472 }
473 
put_map(unsigned long * map)474 static void put_map(unsigned long *map) __releases(&object_map_lock)
475 {
476 	VM_BUG_ON(map != object_map);
477 	spin_unlock(&object_map_lock);
478 }
479 
size_from_object(struct kmem_cache * s)480 static inline unsigned int size_from_object(struct kmem_cache *s)
481 {
482 	if (s->flags & SLAB_RED_ZONE)
483 		return s->size - s->red_left_pad;
484 
485 	return s->size;
486 }
487 
restore_red_left(struct kmem_cache * s,void * p)488 static inline void *restore_red_left(struct kmem_cache *s, void *p)
489 {
490 	if (s->flags & SLAB_RED_ZONE)
491 		p -= s->red_left_pad;
492 
493 	return p;
494 }
495 
496 /*
497  * Debug settings:
498  */
499 #if defined(CONFIG_SLUB_DEBUG_ON)
500 static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
501 #else
502 static slab_flags_t slub_debug;
503 #endif
504 
505 static char *slub_debug_string;
506 static int disable_higher_order_debug;
507 
508 /*
509  * slub is about to manipulate internal object metadata.  This memory lies
510  * outside the range of the allocated object, so accessing it would normally
511  * be reported by kasan as a bounds error.  metadata_access_enable() is used
512  * to tell kasan that these accesses are OK.
513  */
metadata_access_enable(void)514 static inline void metadata_access_enable(void)
515 {
516 	kasan_disable_current();
517 }
518 
metadata_access_disable(void)519 static inline void metadata_access_disable(void)
520 {
521 	kasan_enable_current();
522 }
523 
524 /*
525  * Object debugging
526  */
527 
528 /* Verify that a pointer has an address that is valid within a slab page */
check_valid_pointer(struct kmem_cache * s,struct page * page,void * object)529 static inline int check_valid_pointer(struct kmem_cache *s,
530 				struct page *page, void *object)
531 {
532 	void *base;
533 
534 	if (!object)
535 		return 1;
536 
537 	base = page_address(page);
538 	object = kasan_reset_tag(object);
539 	object = restore_red_left(s, object);
540 	if (object < base || object >= base + page->objects * s->size ||
541 		(object - base) % s->size) {
542 		return 0;
543 	}
544 
545 	return 1;
546 }
547 
print_section(char * level,char * text,u8 * addr,unsigned int length)548 static void print_section(char *level, char *text, u8 *addr,
549 			  unsigned int length)
550 {
551 	metadata_access_enable();
552 	print_hex_dump(level, kasan_reset_tag(text), DUMP_PREFIX_ADDRESS,
553 			16, 1, addr, length, 1);
554 	metadata_access_disable();
555 }
556 
557 /*
558  * See comment in calculate_sizes().
559  */
freeptr_outside_object(struct kmem_cache * s)560 static inline bool freeptr_outside_object(struct kmem_cache *s)
561 {
562 	return s->offset >= s->inuse;
563 }
564 
565 /*
566  * Return offset of the end of info block which is inuse + free pointer if
567  * not overlapping with object.
568  */
get_info_end(struct kmem_cache * s)569 static inline unsigned int get_info_end(struct kmem_cache *s)
570 {
571 	if (freeptr_outside_object(s))
572 		return s->inuse + sizeof(void *);
573 	else
574 		return s->inuse;
575 }
576 
get_track(struct kmem_cache * s,void * object,enum track_item alloc)577 static struct track *get_track(struct kmem_cache *s, void *object,
578 	enum track_item alloc)
579 {
580 	struct track *p;
581 
582 	p = object + get_info_end(s);
583 
584 	return kasan_reset_tag(p + alloc);
585 }
586 
set_track(struct kmem_cache * s,void * object,enum track_item alloc,unsigned long addr)587 static void set_track(struct kmem_cache *s, void *object,
588 			enum track_item alloc, unsigned long addr)
589 {
590 	struct track *p = get_track(s, object, alloc);
591 
592 	if (addr) {
593 #ifdef CONFIG_STACKTRACE
594 		unsigned int nr_entries;
595 
596 		metadata_access_enable();
597 		nr_entries = stack_trace_save(kasan_reset_tag(p->addrs),
598 					      TRACK_ADDRS_COUNT, 3);
599 		metadata_access_disable();
600 
601 		if (nr_entries < TRACK_ADDRS_COUNT)
602 			p->addrs[nr_entries] = 0;
603 #endif
604 		p->addr = addr;
605 		p->cpu = smp_processor_id();
606 		p->pid = current->pid;
607 		p->when = jiffies;
608 	} else {
609 		memset(p, 0, sizeof(struct track));
610 	}
611 }
612 
init_tracking(struct kmem_cache * s,void * object)613 static void init_tracking(struct kmem_cache *s, void *object)
614 {
615 	if (!(s->flags & SLAB_STORE_USER))
616 		return;
617 
618 	set_track(s, object, TRACK_FREE, 0UL);
619 	set_track(s, object, TRACK_ALLOC, 0UL);
620 }
621 
print_track(const char * s,struct track * t,unsigned long pr_time)622 static void print_track(const char *s, struct track *t, unsigned long pr_time)
623 {
624 	if (!t->addr)
625 		return;
626 
627 	pr_err("%s in %pS age=%lu cpu=%u pid=%d\n",
628 	       s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
629 #ifdef CONFIG_STACKTRACE
630 	{
631 		int i;
632 		for (i = 0; i < TRACK_ADDRS_COUNT; i++)
633 			if (t->addrs[i])
634 				pr_err("\t%pS\n", (void *)t->addrs[i]);
635 			else
636 				break;
637 	}
638 #endif
639 }
640 
print_tracking(struct kmem_cache * s,void * object)641 void print_tracking(struct kmem_cache *s, void *object)
642 {
643 	unsigned long pr_time = jiffies;
644 	if (!(s->flags & SLAB_STORE_USER))
645 		return;
646 
647 	print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
648 	print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
649 }
650 
print_page_info(struct page * page)651 static void print_page_info(struct page *page)
652 {
653 	pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%#lx(%pGp)\n",
654 	       page, page->objects, page->inuse, page->freelist,
655 	       page->flags, &page->flags);
656 
657 }
658 
slab_bug(struct kmem_cache * s,char * fmt,...)659 static void slab_bug(struct kmem_cache *s, char *fmt, ...)
660 {
661 	struct va_format vaf;
662 	va_list args;
663 
664 	va_start(args, fmt);
665 	vaf.fmt = fmt;
666 	vaf.va = &args;
667 	pr_err("=============================================================================\n");
668 	pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
669 	pr_err("-----------------------------------------------------------------------------\n\n");
670 
671 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
672 	va_end(args);
673 }
674 
slab_fix(struct kmem_cache * s,char * fmt,...)675 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
676 {
677 	struct va_format vaf;
678 	va_list args;
679 
680 	va_start(args, fmt);
681 	vaf.fmt = fmt;
682 	vaf.va = &args;
683 	pr_err("FIX %s: %pV\n", s->name, &vaf);
684 	va_end(args);
685 }
686 
freelist_corrupted(struct kmem_cache * s,struct page * page,void ** freelist,void * nextfree)687 static bool freelist_corrupted(struct kmem_cache *s, struct page *page,
688 			       void **freelist, void *nextfree)
689 {
690 	if ((s->flags & SLAB_CONSISTENCY_CHECKS) &&
691 	    !check_valid_pointer(s, page, nextfree) && freelist) {
692 		object_err(s, page, *freelist, "Freechain corrupt");
693 		*freelist = NULL;
694 		slab_fix(s, "Isolate corrupted freechain");
695 		return true;
696 	}
697 
698 	return false;
699 }
700 
print_trailer(struct kmem_cache * s,struct page * page,u8 * p)701 static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
702 {
703 	unsigned int off;	/* Offset of last byte */
704 	u8 *addr = page_address(page);
705 
706 	print_tracking(s, p);
707 
708 	print_page_info(page);
709 
710 	pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n",
711 	       p, p - addr, get_freepointer(s, p));
712 
713 	if (s->flags & SLAB_RED_ZONE)
714 		print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
715 			      s->red_left_pad);
716 	else if (p > addr + 16)
717 		print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
718 
719 	print_section(KERN_ERR, "Object ", p,
720 		      min_t(unsigned int, s->object_size, PAGE_SIZE));
721 	if (s->flags & SLAB_RED_ZONE)
722 		print_section(KERN_ERR, "Redzone ", p + s->object_size,
723 			s->inuse - s->object_size);
724 
725 	off = get_info_end(s);
726 
727 	if (s->flags & SLAB_STORE_USER)
728 		off += 2 * sizeof(struct track);
729 
730 	off += kasan_metadata_size(s);
731 
732 	if (off != size_from_object(s))
733 		/* Beginning of the filler is the free pointer */
734 		print_section(KERN_ERR, "Padding ", p + off,
735 			      size_from_object(s) - off);
736 
737 	dump_stack();
738 }
739 
object_err(struct kmem_cache * s,struct page * page,u8 * object,char * reason)740 void object_err(struct kmem_cache *s, struct page *page,
741 			u8 *object, char *reason)
742 {
743 	slab_bug(s, "%s", reason);
744 	print_trailer(s, page, object);
745 }
746 
slab_err(struct kmem_cache * s,struct page * page,const char * fmt,...)747 static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page,
748 			const char *fmt, ...)
749 {
750 	va_list args;
751 	char buf[100];
752 
753 	va_start(args, fmt);
754 	vsnprintf(buf, sizeof(buf), fmt, args);
755 	va_end(args);
756 	slab_bug(s, "%s", buf);
757 	print_page_info(page);
758 	dump_stack();
759 }
760 
init_object(struct kmem_cache * s,void * object,u8 val)761 static void init_object(struct kmem_cache *s, void *object, u8 val)
762 {
763 	u8 *p = kasan_reset_tag(object);
764 
765 	if (s->flags & SLAB_RED_ZONE)
766 		memset(p - s->red_left_pad, val, s->red_left_pad);
767 
768 	if (s->flags & __OBJECT_POISON) {
769 		memset(p, POISON_FREE, s->object_size - 1);
770 		p[s->object_size - 1] = POISON_END;
771 	}
772 
773 	if (s->flags & SLAB_RED_ZONE)
774 		memset(p + s->object_size, val, s->inuse - s->object_size);
775 }
776 
restore_bytes(struct kmem_cache * s,char * message,u8 data,void * from,void * to)777 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
778 						void *from, void *to)
779 {
780 	slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
781 	memset(from, data, to - from);
782 }
783 
check_bytes_and_report(struct kmem_cache * s,struct page * page,u8 * object,char * what,u8 * start,unsigned int value,unsigned int bytes)784 static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
785 			u8 *object, char *what,
786 			u8 *start, unsigned int value, unsigned int bytes)
787 {
788 	u8 *fault;
789 	u8 *end;
790 	u8 *addr = page_address(page);
791 
792 	metadata_access_enable();
793 	fault = memchr_inv(kasan_reset_tag(start), value, bytes);
794 	metadata_access_disable();
795 	if (!fault)
796 		return 1;
797 
798 	end = start + bytes;
799 	while (end > fault && end[-1] == value)
800 		end--;
801 
802 	slab_bug(s, "%s overwritten", what);
803 	pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n",
804 					fault, end - 1, fault - addr,
805 					fault[0], value);
806 	print_trailer(s, page, object);
807 
808 	restore_bytes(s, what, value, fault, end);
809 	return 0;
810 }
811 
812 /*
813  * Object layout:
814  *
815  * object address
816  * 	Bytes of the object to be managed.
817  * 	If the freepointer may overlay the object then the free
818  *	pointer is at the middle of the object.
819  *
820  * 	Poisoning uses 0x6b (POISON_FREE) and the last byte is
821  * 	0xa5 (POISON_END)
822  *
823  * object + s->object_size
824  * 	Padding to reach word boundary. This is also used for Redzoning.
825  * 	Padding is extended by another word if Redzoning is enabled and
826  * 	object_size == inuse.
827  *
828  * 	We fill with 0xbb (RED_INACTIVE) for inactive objects and with
829  * 	0xcc (RED_ACTIVE) for objects in use.
830  *
831  * object + s->inuse
832  * 	Meta data starts here.
833  *
834  * 	A. Free pointer (if we cannot overwrite object on free)
835  * 	B. Tracking data for SLAB_STORE_USER
836  *	C. Padding to reach required alignment boundary or at minimum
837  * 		one word if debugging is on to be able to detect writes
838  * 		before the word boundary.
839  *
840  *	Padding is done using 0x5a (POISON_INUSE)
841  *
842  * object + s->size
843  * 	Nothing is used beyond s->size.
844  *
845  * If slabcaches are merged then the object_size and inuse boundaries are mostly
846  * ignored. And therefore no slab options that rely on these boundaries
847  * may be used with merged slabcaches.
848  */
849 
check_pad_bytes(struct kmem_cache * s,struct page * page,u8 * p)850 static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
851 {
852 	unsigned long off = get_info_end(s);	/* The end of info */
853 
854 	if (s->flags & SLAB_STORE_USER)
855 		/* We also have user information there */
856 		off += 2 * sizeof(struct track);
857 
858 	off += kasan_metadata_size(s);
859 
860 	if (size_from_object(s) == off)
861 		return 1;
862 
863 	return check_bytes_and_report(s, page, p, "Object padding",
864 			p + off, POISON_INUSE, size_from_object(s) - off);
865 }
866 
867 /* Check the pad bytes at the end of a slab page */
slab_pad_check(struct kmem_cache * s,struct page * page)868 static int slab_pad_check(struct kmem_cache *s, struct page *page)
869 {
870 	u8 *start;
871 	u8 *fault;
872 	u8 *end;
873 	u8 *pad;
874 	int length;
875 	int remainder;
876 
877 	if (!(s->flags & SLAB_POISON))
878 		return 1;
879 
880 	start = page_address(page);
881 	length = page_size(page);
882 	end = start + length;
883 	remainder = length % s->size;
884 	if (!remainder)
885 		return 1;
886 
887 	pad = end - remainder;
888 	metadata_access_enable();
889 	fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder);
890 	metadata_access_disable();
891 	if (!fault)
892 		return 1;
893 	while (end > fault && end[-1] == POISON_INUSE)
894 		end--;
895 
896 	slab_err(s, page, "Padding overwritten. 0x%p-0x%p @offset=%tu",
897 			fault, end - 1, fault - start);
898 	print_section(KERN_ERR, "Padding ", pad, remainder);
899 
900 	restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
901 	return 0;
902 }
903 
check_object(struct kmem_cache * s,struct page * page,void * object,u8 val)904 static int check_object(struct kmem_cache *s, struct page *page,
905 					void *object, u8 val)
906 {
907 	u8 *p = object;
908 	u8 *endobject = object + s->object_size;
909 
910 	if (s->flags & SLAB_RED_ZONE) {
911 		if (!check_bytes_and_report(s, page, object, "Redzone",
912 			object - s->red_left_pad, val, s->red_left_pad))
913 			return 0;
914 
915 		if (!check_bytes_and_report(s, page, object, "Redzone",
916 			endobject, val, s->inuse - s->object_size))
917 			return 0;
918 	} else {
919 		if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
920 			check_bytes_and_report(s, page, p, "Alignment padding",
921 				endobject, POISON_INUSE,
922 				s->inuse - s->object_size);
923 		}
924 	}
925 
926 	if (s->flags & SLAB_POISON) {
927 		if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
928 			(!check_bytes_and_report(s, page, p, "Poison", p,
929 					POISON_FREE, s->object_size - 1) ||
930 			 !check_bytes_and_report(s, page, p, "Poison",
931 				p + s->object_size - 1, POISON_END, 1)))
932 			return 0;
933 		/*
934 		 * check_pad_bytes cleans up on its own.
935 		 */
936 		check_pad_bytes(s, page, p);
937 	}
938 
939 	if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE)
940 		/*
941 		 * Object and freepointer overlap. Cannot check
942 		 * freepointer while object is allocated.
943 		 */
944 		return 1;
945 
946 	/* Check free pointer validity */
947 	if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
948 		object_err(s, page, p, "Freepointer corrupt");
949 		/*
950 		 * No choice but to zap it and thus lose the remainder
951 		 * of the free objects in this slab. May cause
952 		 * another error because the object count is now wrong.
953 		 */
954 		set_freepointer(s, p, NULL);
955 		return 0;
956 	}
957 	return 1;
958 }
959 
check_slab(struct kmem_cache * s,struct page * page)960 static int check_slab(struct kmem_cache *s, struct page *page)
961 {
962 	int maxobj;
963 
964 	VM_BUG_ON(!irqs_disabled());
965 
966 	if (!PageSlab(page)) {
967 		slab_err(s, page, "Not a valid slab page");
968 		return 0;
969 	}
970 
971 	maxobj = order_objects(compound_order(page), s->size);
972 	if (page->objects > maxobj) {
973 		slab_err(s, page, "objects %u > max %u",
974 			page->objects, maxobj);
975 		return 0;
976 	}
977 	if (page->inuse > page->objects) {
978 		slab_err(s, page, "inuse %u > max %u",
979 			page->inuse, page->objects);
980 		return 0;
981 	}
982 	/* Slab_pad_check fixes things up after itself */
983 	slab_pad_check(s, page);
984 	return 1;
985 }
986 
987 /*
988  * Determine if a certain object on a page is on the freelist. Must hold the
989  * slab lock to guarantee that the chains are in a consistent state.
990  */
on_freelist(struct kmem_cache * s,struct page * page,void * search)991 static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
992 {
993 	int nr = 0;
994 	void *fp;
995 	void *object = NULL;
996 	int max_objects;
997 
998 	fp = page->freelist;
999 	while (fp && nr <= page->objects) {
1000 		if (fp == search)
1001 			return 1;
1002 		if (!check_valid_pointer(s, page, fp)) {
1003 			if (object) {
1004 				object_err(s, page, object,
1005 					"Freechain corrupt");
1006 				set_freepointer(s, object, NULL);
1007 			} else {
1008 				slab_err(s, page, "Freepointer corrupt");
1009 				page->freelist = NULL;
1010 				page->inuse = page->objects;
1011 				slab_fix(s, "Freelist cleared");
1012 				return 0;
1013 			}
1014 			break;
1015 		}
1016 		object = fp;
1017 		fp = get_freepointer(s, object);
1018 		nr++;
1019 	}
1020 
1021 	max_objects = order_objects(compound_order(page), s->size);
1022 	if (max_objects > MAX_OBJS_PER_PAGE)
1023 		max_objects = MAX_OBJS_PER_PAGE;
1024 
1025 	if (page->objects != max_objects) {
1026 		slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
1027 			 page->objects, max_objects);
1028 		page->objects = max_objects;
1029 		slab_fix(s, "Number of objects adjusted.");
1030 	}
1031 	if (page->inuse != page->objects - nr) {
1032 		slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
1033 			 page->inuse, page->objects - nr);
1034 		page->inuse = page->objects - nr;
1035 		slab_fix(s, "Object count adjusted.");
1036 	}
1037 	return search == NULL;
1038 }
1039 
trace(struct kmem_cache * s,struct page * page,void * object,int alloc)1040 static void trace(struct kmem_cache *s, struct page *page, void *object,
1041 								int alloc)
1042 {
1043 	if (s->flags & SLAB_TRACE) {
1044 		pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
1045 			s->name,
1046 			alloc ? "alloc" : "free",
1047 			object, page->inuse,
1048 			page->freelist);
1049 
1050 		if (!alloc)
1051 			print_section(KERN_INFO, "Object ", (void *)object,
1052 					s->object_size);
1053 
1054 		dump_stack();
1055 	}
1056 }
1057 
1058 /*
1059  * Tracking of fully allocated slabs for debugging purposes.
1060  */
add_full(struct kmem_cache * s,struct kmem_cache_node * n,struct page * page)1061 static void add_full(struct kmem_cache *s,
1062 	struct kmem_cache_node *n, struct page *page)
1063 {
1064 	if (!(s->flags & SLAB_STORE_USER))
1065 		return;
1066 
1067 	lockdep_assert_held(&n->list_lock);
1068 	list_add(&page->slab_list, &n->full);
1069 }
1070 
remove_full(struct kmem_cache * s,struct kmem_cache_node * n,struct page * page)1071 static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
1072 {
1073 	if (!(s->flags & SLAB_STORE_USER))
1074 		return;
1075 
1076 	lockdep_assert_held(&n->list_lock);
1077 	list_del(&page->slab_list);
1078 }
1079 
1080 /* Tracking of the number of slabs for debugging purposes */
slabs_node(struct kmem_cache * s,int node)1081 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1082 {
1083 	struct kmem_cache_node *n = get_node(s, node);
1084 
1085 	return atomic_long_read(&n->nr_slabs);
1086 }
1087 
node_nr_slabs(struct kmem_cache_node * n)1088 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1089 {
1090 	return atomic_long_read(&n->nr_slabs);
1091 }
1092 
inc_slabs_node(struct kmem_cache * s,int node,int objects)1093 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
1094 {
1095 	struct kmem_cache_node *n = get_node(s, node);
1096 
1097 	/*
1098 	 * May be called early in order to allocate a slab for the
1099 	 * kmem_cache_node structure. Solve the chicken-egg
1100 	 * dilemma by deferring the increment of the count during
1101 	 * bootstrap (see early_kmem_cache_node_alloc).
1102 	 */
1103 	if (likely(n)) {
1104 		atomic_long_inc(&n->nr_slabs);
1105 		atomic_long_add(objects, &n->total_objects);
1106 	}
1107 }
dec_slabs_node(struct kmem_cache * s,int node,int objects)1108 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
1109 {
1110 	struct kmem_cache_node *n = get_node(s, node);
1111 
1112 	atomic_long_dec(&n->nr_slabs);
1113 	atomic_long_sub(objects, &n->total_objects);
1114 }
1115 
1116 /* Object debug checks for alloc/free paths */
setup_object_debug(struct kmem_cache * s,struct page * page,void * object)1117 static void setup_object_debug(struct kmem_cache *s, struct page *page,
1118 								void *object)
1119 {
1120 	if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))
1121 		return;
1122 
1123 	init_object(s, object, SLUB_RED_INACTIVE);
1124 	init_tracking(s, object);
1125 }
1126 
1127 static
setup_page_debug(struct kmem_cache * s,struct page * page,void * addr)1128 void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr)
1129 {
1130 	if (!kmem_cache_debug_flags(s, SLAB_POISON))
1131 		return;
1132 
1133 	metadata_access_enable();
1134 	memset(kasan_reset_tag(addr), POISON_INUSE, page_size(page));
1135 	metadata_access_disable();
1136 }
1137 
alloc_consistency_checks(struct kmem_cache * s,struct page * page,void * object)1138 static inline int alloc_consistency_checks(struct kmem_cache *s,
1139 					struct page *page, void *object)
1140 {
1141 	if (!check_slab(s, page))
1142 		return 0;
1143 
1144 	if (!check_valid_pointer(s, page, object)) {
1145 		object_err(s, page, object, "Freelist Pointer check fails");
1146 		return 0;
1147 	}
1148 
1149 	if (!check_object(s, page, object, SLUB_RED_INACTIVE))
1150 		return 0;
1151 
1152 	return 1;
1153 }
1154 
alloc_debug_processing(struct kmem_cache * s,struct page * page,void * object,unsigned long addr)1155 static noinline int alloc_debug_processing(struct kmem_cache *s,
1156 					struct page *page,
1157 					void *object, unsigned long addr)
1158 {
1159 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1160 		if (!alloc_consistency_checks(s, page, object))
1161 			goto bad;
1162 	}
1163 
1164 	/* Success perform special debug activities for allocs */
1165 	if (s->flags & SLAB_STORE_USER)
1166 		set_track(s, object, TRACK_ALLOC, addr);
1167 	trace(s, page, object, 1);
1168 	init_object(s, object, SLUB_RED_ACTIVE);
1169 	return 1;
1170 
1171 bad:
1172 	if (PageSlab(page)) {
1173 		/*
1174 		 * If this is a slab page then lets do the best we can
1175 		 * to avoid issues in the future. Marking all objects
1176 		 * as used avoids touching the remaining objects.
1177 		 */
1178 		slab_fix(s, "Marking all objects used");
1179 		page->inuse = page->objects;
1180 		page->freelist = NULL;
1181 	}
1182 	return 0;
1183 }
1184 
free_consistency_checks(struct kmem_cache * s,struct page * page,void * object,unsigned long addr)1185 static inline int free_consistency_checks(struct kmem_cache *s,
1186 		struct page *page, void *object, unsigned long addr)
1187 {
1188 	if (!check_valid_pointer(s, page, object)) {
1189 		slab_err(s, page, "Invalid object pointer 0x%p", object);
1190 		return 0;
1191 	}
1192 
1193 	if (on_freelist(s, page, object)) {
1194 		object_err(s, page, object, "Object already free");
1195 		return 0;
1196 	}
1197 
1198 	if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1199 		return 0;
1200 
1201 	if (unlikely(s != page->slab_cache)) {
1202 		if (!PageSlab(page)) {
1203 			slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1204 				 object);
1205 		} else if (!page->slab_cache) {
1206 			pr_err("SLUB <none>: no slab for object 0x%p.\n",
1207 			       object);
1208 			dump_stack();
1209 		} else
1210 			object_err(s, page, object,
1211 					"page slab pointer corrupt.");
1212 		return 0;
1213 	}
1214 	return 1;
1215 }
1216 
1217 /* Supports checking bulk free of a constructed freelist */
free_debug_processing(struct kmem_cache * s,struct page * page,void * head,void * tail,int bulk_cnt,unsigned long addr)1218 static noinline int free_debug_processing(
1219 	struct kmem_cache *s, struct page *page,
1220 	void *head, void *tail, int bulk_cnt,
1221 	unsigned long addr)
1222 {
1223 	struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1224 	void *object = head;
1225 	int cnt = 0;
1226 	unsigned long flags;
1227 	int ret = 0;
1228 
1229 	spin_lock_irqsave(&n->list_lock, flags);
1230 	slab_lock(page);
1231 
1232 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1233 		if (!check_slab(s, page))
1234 			goto out;
1235 	}
1236 
1237 next_object:
1238 	cnt++;
1239 
1240 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1241 		if (!free_consistency_checks(s, page, object, addr))
1242 			goto out;
1243 	}
1244 
1245 	if (s->flags & SLAB_STORE_USER)
1246 		set_track(s, object, TRACK_FREE, addr);
1247 	trace(s, page, object, 0);
1248 	/* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
1249 	init_object(s, object, SLUB_RED_INACTIVE);
1250 
1251 	/* Reached end of constructed freelist yet? */
1252 	if (object != tail) {
1253 		object = get_freepointer(s, object);
1254 		goto next_object;
1255 	}
1256 	ret = 1;
1257 
1258 out:
1259 	if (cnt != bulk_cnt)
1260 		slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1261 			 bulk_cnt, cnt);
1262 
1263 	slab_unlock(page);
1264 	spin_unlock_irqrestore(&n->list_lock, flags);
1265 	if (!ret)
1266 		slab_fix(s, "Object at 0x%p not freed", object);
1267 	return ret;
1268 }
1269 
1270 /*
1271  * Parse a block of slub_debug options. Blocks are delimited by ';'
1272  *
1273  * @str:    start of block
1274  * @flags:  returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified
1275  * @slabs:  return start of list of slabs, or NULL when there's no list
1276  * @init:   assume this is initial parsing and not per-kmem-create parsing
1277  *
1278  * returns the start of next block if there's any, or NULL
1279  */
1280 static char *
parse_slub_debug_flags(char * str,slab_flags_t * flags,char ** slabs,bool init)1281 parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init)
1282 {
1283 	bool higher_order_disable = false;
1284 
1285 	/* Skip any completely empty blocks */
1286 	while (*str && *str == ';')
1287 		str++;
1288 
1289 	if (*str == ',') {
1290 		/*
1291 		 * No options but restriction on slabs. This means full
1292 		 * debugging for slabs matching a pattern.
1293 		 */
1294 		*flags = DEBUG_DEFAULT_FLAGS;
1295 		goto check_slabs;
1296 	}
1297 	*flags = 0;
1298 
1299 	/* Determine which debug features should be switched on */
1300 	for (; *str && *str != ',' && *str != ';'; str++) {
1301 		switch (tolower(*str)) {
1302 		case '-':
1303 			*flags = 0;
1304 			break;
1305 		case 'f':
1306 			*flags |= SLAB_CONSISTENCY_CHECKS;
1307 			break;
1308 		case 'z':
1309 			*flags |= SLAB_RED_ZONE;
1310 			break;
1311 		case 'p':
1312 			*flags |= SLAB_POISON;
1313 			break;
1314 		case 'u':
1315 			*flags |= SLAB_STORE_USER;
1316 			break;
1317 		case 't':
1318 			*flags |= SLAB_TRACE;
1319 			break;
1320 		case 'a':
1321 			*flags |= SLAB_FAILSLAB;
1322 			break;
1323 		case 'o':
1324 			/*
1325 			 * Avoid enabling debugging on caches if its minimum
1326 			 * order would increase as a result.
1327 			 */
1328 			higher_order_disable = true;
1329 			break;
1330 		default:
1331 			if (init)
1332 				pr_err("slub_debug option '%c' unknown. skipped\n", *str);
1333 		}
1334 	}
1335 check_slabs:
1336 	if (*str == ',')
1337 		*slabs = ++str;
1338 	else
1339 		*slabs = NULL;
1340 
1341 	/* Skip over the slab list */
1342 	while (*str && *str != ';')
1343 		str++;
1344 
1345 	/* Skip any completely empty blocks */
1346 	while (*str && *str == ';')
1347 		str++;
1348 
1349 	if (init && higher_order_disable)
1350 		disable_higher_order_debug = 1;
1351 
1352 	if (*str)
1353 		return str;
1354 	else
1355 		return NULL;
1356 }
1357 
setup_slub_debug(char * str)1358 static int __init setup_slub_debug(char *str)
1359 {
1360 	slab_flags_t flags;
1361 	char *saved_str;
1362 	char *slab_list;
1363 	bool global_slub_debug_changed = false;
1364 	bool slab_list_specified = false;
1365 
1366 	slub_debug = DEBUG_DEFAULT_FLAGS;
1367 	if (*str++ != '=' || !*str)
1368 		/*
1369 		 * No options specified. Switch on full debugging.
1370 		 */
1371 		goto out;
1372 
1373 	saved_str = str;
1374 	while (str) {
1375 		str = parse_slub_debug_flags(str, &flags, &slab_list, true);
1376 
1377 		if (!slab_list) {
1378 			slub_debug = flags;
1379 			global_slub_debug_changed = true;
1380 		} else {
1381 			slab_list_specified = true;
1382 		}
1383 	}
1384 
1385 	/*
1386 	 * For backwards compatibility, a single list of flags with list of
1387 	 * slabs means debugging is only enabled for those slabs, so the global
1388 	 * slub_debug should be 0. We can extended that to multiple lists as
1389 	 * long as there is no option specifying flags without a slab list.
1390 	 */
1391 	if (slab_list_specified) {
1392 		if (!global_slub_debug_changed)
1393 			slub_debug = 0;
1394 		slub_debug_string = saved_str;
1395 	}
1396 out:
1397 	if (slub_debug != 0 || slub_debug_string)
1398 		static_branch_enable(&slub_debug_enabled);
1399 	if ((static_branch_unlikely(&init_on_alloc) ||
1400 	     static_branch_unlikely(&init_on_free)) &&
1401 	    (slub_debug & SLAB_POISON))
1402 		pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n");
1403 	return 1;
1404 }
1405 
1406 __setup("slub_debug", setup_slub_debug);
1407 
1408 /*
1409  * kmem_cache_flags - apply debugging options to the cache
1410  * @object_size:	the size of an object without meta data
1411  * @flags:		flags to set
1412  * @name:		name of the cache
1413  *
1414  * Debug option(s) are applied to @flags. In addition to the debug
1415  * option(s), if a slab name (or multiple) is specified i.e.
1416  * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ...
1417  * then only the select slabs will receive the debug option(s).
1418  */
kmem_cache_flags(unsigned int object_size,slab_flags_t flags,const char * name)1419 slab_flags_t kmem_cache_flags(unsigned int object_size,
1420 	slab_flags_t flags, const char *name)
1421 {
1422 	char *iter;
1423 	size_t len;
1424 	char *next_block;
1425 	slab_flags_t block_flags;
1426 	slab_flags_t slub_debug_local = slub_debug;
1427 
1428 	/*
1429 	 * If the slab cache is for debugging (e.g. kmemleak) then
1430 	 * don't store user (stack trace) information by default,
1431 	 * but let the user enable it via the command line below.
1432 	 */
1433 	if (flags & SLAB_NOLEAKTRACE)
1434 		slub_debug_local &= ~SLAB_STORE_USER;
1435 
1436 	len = strlen(name);
1437 	next_block = slub_debug_string;
1438 	/* Go through all blocks of debug options, see if any matches our slab's name */
1439 	while (next_block) {
1440 		next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false);
1441 		if (!iter)
1442 			continue;
1443 		/* Found a block that has a slab list, search it */
1444 		while (*iter) {
1445 			char *end, *glob;
1446 			size_t cmplen;
1447 
1448 			end = strchrnul(iter, ',');
1449 			if (next_block && next_block < end)
1450 				end = next_block - 1;
1451 
1452 			glob = strnchr(iter, end - iter, '*');
1453 			if (glob)
1454 				cmplen = glob - iter;
1455 			else
1456 				cmplen = max_t(size_t, len, (end - iter));
1457 
1458 			if (!strncmp(name, iter, cmplen)) {
1459 				flags |= block_flags;
1460 				return flags;
1461 			}
1462 
1463 			if (!*end || *end == ';')
1464 				break;
1465 			iter = end + 1;
1466 		}
1467 	}
1468 
1469 	return flags | slub_debug_local;
1470 }
1471 #else /* !CONFIG_SLUB_DEBUG */
setup_object_debug(struct kmem_cache * s,struct page * page,void * object)1472 static inline void setup_object_debug(struct kmem_cache *s,
1473 			struct page *page, void *object) {}
1474 static inline
setup_page_debug(struct kmem_cache * s,struct page * page,void * addr)1475 void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {}
1476 
alloc_debug_processing(struct kmem_cache * s,struct page * page,void * object,unsigned long addr)1477 static inline int alloc_debug_processing(struct kmem_cache *s,
1478 	struct page *page, void *object, unsigned long addr) { return 0; }
1479 
free_debug_processing(struct kmem_cache * s,struct page * page,void * head,void * tail,int bulk_cnt,unsigned long addr)1480 static inline int free_debug_processing(
1481 	struct kmem_cache *s, struct page *page,
1482 	void *head, void *tail, int bulk_cnt,
1483 	unsigned long addr) { return 0; }
1484 
slab_pad_check(struct kmem_cache * s,struct page * page)1485 static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1486 			{ return 1; }
check_object(struct kmem_cache * s,struct page * page,void * object,u8 val)1487 static inline int check_object(struct kmem_cache *s, struct page *page,
1488 			void *object, u8 val) { return 1; }
add_full(struct kmem_cache * s,struct kmem_cache_node * n,struct page * page)1489 static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1490 					struct page *page) {}
remove_full(struct kmem_cache * s,struct kmem_cache_node * n,struct page * page)1491 static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1492 					struct page *page) {}
kmem_cache_flags(unsigned int object_size,slab_flags_t flags,const char * name)1493 slab_flags_t kmem_cache_flags(unsigned int object_size,
1494 	slab_flags_t flags, const char *name)
1495 {
1496 	return flags;
1497 }
1498 #define slub_debug 0
1499 
1500 #define disable_higher_order_debug 0
1501 
slabs_node(struct kmem_cache * s,int node)1502 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1503 							{ return 0; }
node_nr_slabs(struct kmem_cache_node * n)1504 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1505 							{ return 0; }
inc_slabs_node(struct kmem_cache * s,int node,int objects)1506 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1507 							int objects) {}
dec_slabs_node(struct kmem_cache * s,int node,int objects)1508 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1509 							int objects) {}
1510 
freelist_corrupted(struct kmem_cache * s,struct page * page,void ** freelist,void * nextfree)1511 static bool freelist_corrupted(struct kmem_cache *s, struct page *page,
1512 			       void **freelist, void *nextfree)
1513 {
1514 	return false;
1515 }
1516 #endif /* CONFIG_SLUB_DEBUG */
1517 
1518 /*
1519  * Hooks for other subsystems that check memory allocations. In a typical
1520  * production configuration these hooks all should produce no code at all.
1521  */
kmalloc_large_node_hook(void * ptr,size_t size,gfp_t flags)1522 static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1523 {
1524 	ptr = kasan_kmalloc_large(ptr, size, flags);
1525 	/* As ptr might get tagged, call kmemleak hook after KASAN. */
1526 	kmemleak_alloc(ptr, size, 1, flags);
1527 	return ptr;
1528 }
1529 
kfree_hook(void * x)1530 static __always_inline void kfree_hook(void *x)
1531 {
1532 	kmemleak_free(x);
1533 	kasan_kfree_large(x);
1534 }
1535 
slab_free_hook(struct kmem_cache * s,void * x,bool init)1536 static __always_inline bool slab_free_hook(struct kmem_cache *s,
1537 						void *x, bool init)
1538 {
1539 	kmemleak_free_recursive(x, s->flags);
1540 
1541 	/*
1542 	 * Trouble is that we may no longer disable interrupts in the fast path
1543 	 * So in order to make the debug calls that expect irqs to be
1544 	 * disabled we need to disable interrupts temporarily.
1545 	 */
1546 #ifdef CONFIG_LOCKDEP
1547 	{
1548 		unsigned long flags;
1549 
1550 		local_irq_save(flags);
1551 		debug_check_no_locks_freed(x, s->object_size);
1552 		local_irq_restore(flags);
1553 	}
1554 #endif
1555 	if (!(s->flags & SLAB_DEBUG_OBJECTS))
1556 		debug_check_no_obj_freed(x, s->object_size);
1557 
1558 	/* Use KCSAN to help debug racy use-after-free. */
1559 	if (!(s->flags & SLAB_TYPESAFE_BY_RCU))
1560 		__kcsan_check_access(x, s->object_size,
1561 				     KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT);
1562 
1563 	/*
1564 	 * As memory initialization might be integrated into KASAN,
1565 	 * kasan_slab_free and initialization memset's must be
1566 	 * kept together to avoid discrepancies in behavior.
1567 	 *
1568 	 * The initialization memset's clear the object and the metadata,
1569 	 * but don't touch the SLAB redzone.
1570 	 */
1571 	if (init) {
1572 		int rsize;
1573 
1574 		if (!kasan_has_integrated_init())
1575 			memset(kasan_reset_tag(x), 0, s->object_size);
1576 		rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0;
1577 		memset((char *)kasan_reset_tag(x) + s->inuse, 0,
1578 		       s->size - s->inuse - rsize);
1579 	}
1580 	/* KASAN might put x into memory quarantine, delaying its reuse. */
1581 	return kasan_slab_free(s, x, init);
1582 }
1583 
slab_free_freelist_hook(struct kmem_cache * s,void ** head,void ** tail)1584 static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1585 					   void **head, void **tail)
1586 {
1587 
1588 	void *object;
1589 	void *next = *head;
1590 	void *old_tail = *tail ? *tail : *head;
1591 
1592 	if (is_kfence_address(next)) {
1593 		slab_free_hook(s, next, false);
1594 		return true;
1595 	}
1596 
1597 	/* Head and tail of the reconstructed freelist */
1598 	*head = NULL;
1599 	*tail = NULL;
1600 
1601 	do {
1602 		object = next;
1603 		next = get_freepointer(s, object);
1604 
1605 		/* If object's reuse doesn't have to be delayed */
1606 		if (!slab_free_hook(s, object, slab_want_init_on_free(s))) {
1607 			/* Move object to the new freelist */
1608 			set_freepointer(s, object, *head);
1609 			*head = object;
1610 			if (!*tail)
1611 				*tail = object;
1612 		}
1613 	} while (object != old_tail);
1614 
1615 	if (*head == *tail)
1616 		*tail = NULL;
1617 
1618 	return *head != NULL;
1619 }
1620 
setup_object(struct kmem_cache * s,struct page * page,void * object)1621 static void *setup_object(struct kmem_cache *s, struct page *page,
1622 				void *object)
1623 {
1624 	setup_object_debug(s, page, object);
1625 	object = kasan_init_slab_obj(s, object);
1626 	if (unlikely(s->ctor)) {
1627 		kasan_unpoison_object_data(s, object);
1628 		s->ctor(object);
1629 		kasan_poison_object_data(s, object);
1630 	}
1631 	return object;
1632 }
1633 
1634 /*
1635  * Slab allocation and freeing
1636  */
alloc_slab_page(struct kmem_cache * s,gfp_t flags,int node,struct kmem_cache_order_objects oo)1637 static inline struct page *alloc_slab_page(struct kmem_cache *s,
1638 		gfp_t flags, int node, struct kmem_cache_order_objects oo)
1639 {
1640 	struct page *page;
1641 	unsigned int order = oo_order(oo);
1642 
1643 	if (node == NUMA_NO_NODE)
1644 		page = alloc_pages(flags, order);
1645 	else
1646 		page = __alloc_pages_node(node, flags, order);
1647 
1648 	return page;
1649 }
1650 
1651 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1652 /* Pre-initialize the random sequence cache */
init_cache_random_seq(struct kmem_cache * s)1653 static int init_cache_random_seq(struct kmem_cache *s)
1654 {
1655 	unsigned int count = oo_objects(s->oo);
1656 	int err;
1657 
1658 	/* Bailout if already initialised */
1659 	if (s->random_seq)
1660 		return 0;
1661 
1662 	err = cache_random_seq_create(s, count, GFP_KERNEL);
1663 	if (err) {
1664 		pr_err("SLUB: Unable to initialize free list for %s\n",
1665 			s->name);
1666 		return err;
1667 	}
1668 
1669 	/* Transform to an offset on the set of pages */
1670 	if (s->random_seq) {
1671 		unsigned int i;
1672 
1673 		for (i = 0; i < count; i++)
1674 			s->random_seq[i] *= s->size;
1675 	}
1676 	return 0;
1677 }
1678 
1679 /* Initialize each random sequence freelist per cache */
init_freelist_randomization(void)1680 static void __init init_freelist_randomization(void)
1681 {
1682 	struct kmem_cache *s;
1683 
1684 	mutex_lock(&slab_mutex);
1685 
1686 	list_for_each_entry(s, &slab_caches, list)
1687 		init_cache_random_seq(s);
1688 
1689 	mutex_unlock(&slab_mutex);
1690 }
1691 
1692 /* Get the next entry on the pre-computed freelist randomized */
next_freelist_entry(struct kmem_cache * s,struct page * page,unsigned long * pos,void * start,unsigned long page_limit,unsigned long freelist_count)1693 static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1694 				unsigned long *pos, void *start,
1695 				unsigned long page_limit,
1696 				unsigned long freelist_count)
1697 {
1698 	unsigned int idx;
1699 
1700 	/*
1701 	 * If the target page allocation failed, the number of objects on the
1702 	 * page might be smaller than the usual size defined by the cache.
1703 	 */
1704 	do {
1705 		idx = s->random_seq[*pos];
1706 		*pos += 1;
1707 		if (*pos >= freelist_count)
1708 			*pos = 0;
1709 	} while (unlikely(idx >= page_limit));
1710 
1711 	return (char *)start + idx;
1712 }
1713 
1714 /* Shuffle the single linked freelist based on a random pre-computed sequence */
shuffle_freelist(struct kmem_cache * s,struct page * page)1715 static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1716 {
1717 	void *start;
1718 	void *cur;
1719 	void *next;
1720 	unsigned long idx, pos, page_limit, freelist_count;
1721 
1722 	if (page->objects < 2 || !s->random_seq)
1723 		return false;
1724 
1725 	freelist_count = oo_objects(s->oo);
1726 	pos = get_random_int() % freelist_count;
1727 
1728 	page_limit = page->objects * s->size;
1729 	start = fixup_red_left(s, page_address(page));
1730 
1731 	/* First entry is used as the base of the freelist */
1732 	cur = next_freelist_entry(s, page, &pos, start, page_limit,
1733 				freelist_count);
1734 	cur = setup_object(s, page, cur);
1735 	page->freelist = cur;
1736 
1737 	for (idx = 1; idx < page->objects; idx++) {
1738 		next = next_freelist_entry(s, page, &pos, start, page_limit,
1739 			freelist_count);
1740 		next = setup_object(s, page, next);
1741 		set_freepointer(s, cur, next);
1742 		cur = next;
1743 	}
1744 	set_freepointer(s, cur, NULL);
1745 
1746 	return true;
1747 }
1748 #else
init_cache_random_seq(struct kmem_cache * s)1749 static inline int init_cache_random_seq(struct kmem_cache *s)
1750 {
1751 	return 0;
1752 }
init_freelist_randomization(void)1753 static inline void init_freelist_randomization(void) { }
shuffle_freelist(struct kmem_cache * s,struct page * page)1754 static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1755 {
1756 	return false;
1757 }
1758 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1759 
allocate_slab(struct kmem_cache * s,gfp_t flags,int node)1760 static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1761 {
1762 	struct page *page;
1763 	struct kmem_cache_order_objects oo = s->oo;
1764 	gfp_t alloc_gfp;
1765 	void *start, *p, *next;
1766 	int idx;
1767 	bool shuffle;
1768 
1769 	flags &= gfp_allowed_mask;
1770 
1771 	if (gfpflags_allow_blocking(flags))
1772 		local_irq_enable();
1773 
1774 	flags |= s->allocflags;
1775 
1776 	/*
1777 	 * Let the initial higher-order allocation fail under memory pressure
1778 	 * so we fall-back to the minimum order allocation.
1779 	 */
1780 	alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1781 	if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
1782 		alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
1783 
1784 	page = alloc_slab_page(s, alloc_gfp, node, oo);
1785 	if (unlikely(!page)) {
1786 		oo = s->min;
1787 		alloc_gfp = flags;
1788 		/*
1789 		 * Allocation may have failed due to fragmentation.
1790 		 * Try a lower order alloc if possible
1791 		 */
1792 		page = alloc_slab_page(s, alloc_gfp, node, oo);
1793 		if (unlikely(!page))
1794 			goto out;
1795 		stat(s, ORDER_FALLBACK);
1796 	}
1797 
1798 	page->objects = oo_objects(oo);
1799 
1800 	account_slab_page(page, oo_order(oo), s, flags);
1801 
1802 	page->slab_cache = s;
1803 	__SetPageSlab(page);
1804 	if (page_is_pfmemalloc(page))
1805 		SetPageSlabPfmemalloc(page);
1806 
1807 	kasan_poison_slab(page);
1808 
1809 	start = page_address(page);
1810 
1811 	setup_page_debug(s, page, start);
1812 
1813 	shuffle = shuffle_freelist(s, page);
1814 
1815 	if (!shuffle) {
1816 		start = fixup_red_left(s, start);
1817 		start = setup_object(s, page, start);
1818 		page->freelist = start;
1819 		for (idx = 0, p = start; idx < page->objects - 1; idx++) {
1820 			next = p + s->size;
1821 			next = setup_object(s, page, next);
1822 			set_freepointer(s, p, next);
1823 			p = next;
1824 		}
1825 		set_freepointer(s, p, NULL);
1826 	}
1827 
1828 	page->inuse = page->objects;
1829 	page->frozen = 1;
1830 
1831 out:
1832 	if (gfpflags_allow_blocking(flags))
1833 		local_irq_disable();
1834 	if (!page)
1835 		return NULL;
1836 
1837 	inc_slabs_node(s, page_to_nid(page), page->objects);
1838 
1839 	return page;
1840 }
1841 
new_slab(struct kmem_cache * s,gfp_t flags,int node)1842 static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1843 {
1844 	if (unlikely(flags & GFP_SLAB_BUG_MASK))
1845 		flags = kmalloc_fix_flags(flags);
1846 
1847 	return allocate_slab(s,
1848 		flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1849 }
1850 
__free_slab(struct kmem_cache * s,struct page * page)1851 static void __free_slab(struct kmem_cache *s, struct page *page)
1852 {
1853 	int order = compound_order(page);
1854 	int pages = 1 << order;
1855 
1856 	if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) {
1857 		void *p;
1858 
1859 		slab_pad_check(s, page);
1860 		for_each_object(p, s, page_address(page),
1861 						page->objects)
1862 			check_object(s, page, p, SLUB_RED_INACTIVE);
1863 	}
1864 
1865 	__ClearPageSlabPfmemalloc(page);
1866 	__ClearPageSlab(page);
1867 	/* In union with page->mapping where page allocator expects NULL */
1868 	page->slab_cache = NULL;
1869 	if (current->reclaim_state)
1870 		current->reclaim_state->reclaimed_slab += pages;
1871 	unaccount_slab_page(page, order, s);
1872 	__free_pages(page, order);
1873 }
1874 
rcu_free_slab(struct rcu_head * h)1875 static void rcu_free_slab(struct rcu_head *h)
1876 {
1877 	struct page *page = container_of(h, struct page, rcu_head);
1878 
1879 	__free_slab(page->slab_cache, page);
1880 }
1881 
free_slab(struct kmem_cache * s,struct page * page)1882 static void free_slab(struct kmem_cache *s, struct page *page)
1883 {
1884 	if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) {
1885 		call_rcu(&page->rcu_head, rcu_free_slab);
1886 	} else
1887 		__free_slab(s, page);
1888 }
1889 
discard_slab(struct kmem_cache * s,struct page * page)1890 static void discard_slab(struct kmem_cache *s, struct page *page)
1891 {
1892 	dec_slabs_node(s, page_to_nid(page), page->objects);
1893 	free_slab(s, page);
1894 }
1895 
1896 /*
1897  * Management of partially allocated slabs.
1898  */
1899 static inline void
__add_partial(struct kmem_cache_node * n,struct page * page,int tail)1900 __add_partial(struct kmem_cache_node *n, struct page *page, int tail)
1901 {
1902 	n->nr_partial++;
1903 	if (tail == DEACTIVATE_TO_TAIL)
1904 		list_add_tail(&page->slab_list, &n->partial);
1905 	else
1906 		list_add(&page->slab_list, &n->partial);
1907 }
1908 
add_partial(struct kmem_cache_node * n,struct page * page,int tail)1909 static inline void add_partial(struct kmem_cache_node *n,
1910 				struct page *page, int tail)
1911 {
1912 	lockdep_assert_held(&n->list_lock);
1913 	__add_partial(n, page, tail);
1914 }
1915 
remove_partial(struct kmem_cache_node * n,struct page * page)1916 static inline void remove_partial(struct kmem_cache_node *n,
1917 					struct page *page)
1918 {
1919 	lockdep_assert_held(&n->list_lock);
1920 	list_del(&page->slab_list);
1921 	n->nr_partial--;
1922 }
1923 
1924 /*
1925  * Remove slab from the partial list, freeze it and
1926  * return the pointer to the freelist.
1927  *
1928  * Returns a list of objects or NULL if it fails.
1929  */
acquire_slab(struct kmem_cache * s,struct kmem_cache_node * n,struct page * page,int mode,int * objects)1930 static inline void *acquire_slab(struct kmem_cache *s,
1931 		struct kmem_cache_node *n, struct page *page,
1932 		int mode, int *objects)
1933 {
1934 	void *freelist;
1935 	unsigned long counters;
1936 	struct page new;
1937 
1938 	lockdep_assert_held(&n->list_lock);
1939 
1940 	/*
1941 	 * Zap the freelist and set the frozen bit.
1942 	 * The old freelist is the list of objects for the
1943 	 * per cpu allocation list.
1944 	 */
1945 	freelist = page->freelist;
1946 	counters = page->counters;
1947 	new.counters = counters;
1948 	*objects = new.objects - new.inuse;
1949 	if (mode) {
1950 		new.inuse = page->objects;
1951 		new.freelist = NULL;
1952 	} else {
1953 		new.freelist = freelist;
1954 	}
1955 
1956 	VM_BUG_ON(new.frozen);
1957 	new.frozen = 1;
1958 
1959 	if (!__cmpxchg_double_slab(s, page,
1960 			freelist, counters,
1961 			new.freelist, new.counters,
1962 			"acquire_slab"))
1963 		return NULL;
1964 
1965 	remove_partial(n, page);
1966 	WARN_ON(!freelist);
1967 	return freelist;
1968 }
1969 
1970 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
1971 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
1972 
1973 /*
1974  * Try to allocate a partial slab from a specific node.
1975  */
get_partial_node(struct kmem_cache * s,struct kmem_cache_node * n,struct kmem_cache_cpu * c,gfp_t flags)1976 static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1977 				struct kmem_cache_cpu *c, gfp_t flags)
1978 {
1979 	struct page *page, *page2;
1980 	void *object = NULL;
1981 	unsigned int available = 0;
1982 	int objects;
1983 
1984 	/*
1985 	 * Racy check. If we mistakenly see no partial slabs then we
1986 	 * just allocate an empty slab. If we mistakenly try to get a
1987 	 * partial slab and there is none available then get_partial()
1988 	 * will return NULL.
1989 	 */
1990 	if (!n || !n->nr_partial)
1991 		return NULL;
1992 
1993 	spin_lock(&n->list_lock);
1994 	list_for_each_entry_safe(page, page2, &n->partial, slab_list) {
1995 		void *t;
1996 
1997 		if (!pfmemalloc_match(page, flags))
1998 			continue;
1999 
2000 		t = acquire_slab(s, n, page, object == NULL, &objects);
2001 		if (!t)
2002 			break;
2003 
2004 		available += objects;
2005 		if (!object) {
2006 			c->page = page;
2007 			stat(s, ALLOC_FROM_PARTIAL);
2008 			object = t;
2009 		} else {
2010 			put_cpu_partial(s, page, 0);
2011 			stat(s, CPU_PARTIAL_NODE);
2012 		}
2013 		if (!kmem_cache_has_cpu_partial(s)
2014 			|| available > slub_cpu_partial(s) / 2)
2015 			break;
2016 
2017 	}
2018 	spin_unlock(&n->list_lock);
2019 	return object;
2020 }
2021 
2022 /*
2023  * Get a page from somewhere. Search in increasing NUMA distances.
2024  */
get_any_partial(struct kmem_cache * s,gfp_t flags,struct kmem_cache_cpu * c)2025 static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
2026 		struct kmem_cache_cpu *c)
2027 {
2028 #ifdef CONFIG_NUMA
2029 	struct zonelist *zonelist;
2030 	struct zoneref *z;
2031 	struct zone *zone;
2032 	enum zone_type highest_zoneidx = gfp_zone(flags);
2033 	void *object;
2034 	unsigned int cpuset_mems_cookie;
2035 
2036 	/*
2037 	 * The defrag ratio allows a configuration of the tradeoffs between
2038 	 * inter node defragmentation and node local allocations. A lower
2039 	 * defrag_ratio increases the tendency to do local allocations
2040 	 * instead of attempting to obtain partial slabs from other nodes.
2041 	 *
2042 	 * If the defrag_ratio is set to 0 then kmalloc() always
2043 	 * returns node local objects. If the ratio is higher then kmalloc()
2044 	 * may return off node objects because partial slabs are obtained
2045 	 * from other nodes and filled up.
2046 	 *
2047 	 * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
2048 	 * (which makes defrag_ratio = 1000) then every (well almost)
2049 	 * allocation will first attempt to defrag slab caches on other nodes.
2050 	 * This means scanning over all nodes to look for partial slabs which
2051 	 * may be expensive if we do it every time we are trying to find a slab
2052 	 * with available objects.
2053 	 */
2054 	if (!s->remote_node_defrag_ratio ||
2055 			get_cycles() % 1024 > s->remote_node_defrag_ratio)
2056 		return NULL;
2057 
2058 	do {
2059 		cpuset_mems_cookie = read_mems_allowed_begin();
2060 		zonelist = node_zonelist(mempolicy_slab_node(), flags);
2061 		for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) {
2062 			struct kmem_cache_node *n;
2063 
2064 			n = get_node(s, zone_to_nid(zone));
2065 
2066 			if (n && cpuset_zone_allowed(zone, flags) &&
2067 					n->nr_partial > s->min_partial) {
2068 				object = get_partial_node(s, n, c, flags);
2069 				if (object) {
2070 					/*
2071 					 * Don't check read_mems_allowed_retry()
2072 					 * here - if mems_allowed was updated in
2073 					 * parallel, that was a harmless race
2074 					 * between allocation and the cpuset
2075 					 * update
2076 					 */
2077 					return object;
2078 				}
2079 			}
2080 		}
2081 	} while (read_mems_allowed_retry(cpuset_mems_cookie));
2082 #endif	/* CONFIG_NUMA */
2083 	return NULL;
2084 }
2085 
2086 /*
2087  * Get a partial page, lock it and return it.
2088  */
get_partial(struct kmem_cache * s,gfp_t flags,int node,struct kmem_cache_cpu * c)2089 static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
2090 		struct kmem_cache_cpu *c)
2091 {
2092 	void *object;
2093 	int searchnode = node;
2094 
2095 	if (node == NUMA_NO_NODE)
2096 		searchnode = numa_mem_id();
2097 
2098 	object = get_partial_node(s, get_node(s, searchnode), c, flags);
2099 	if (object || node != NUMA_NO_NODE)
2100 		return object;
2101 
2102 	return get_any_partial(s, flags, c);
2103 }
2104 
2105 #ifdef CONFIG_PREEMPTION
2106 /*
2107  * Calculate the next globally unique transaction for disambiguation
2108  * during cmpxchg. The transactions start with the cpu number and are then
2109  * incremented by CONFIG_NR_CPUS.
2110  */
2111 #define TID_STEP  roundup_pow_of_two(CONFIG_NR_CPUS)
2112 #else
2113 /*
2114  * No preemption supported therefore also no need to check for
2115  * different cpus.
2116  */
2117 #define TID_STEP 1
2118 #endif
2119 
next_tid(unsigned long tid)2120 static inline unsigned long next_tid(unsigned long tid)
2121 {
2122 	return tid + TID_STEP;
2123 }
2124 
2125 #ifdef SLUB_DEBUG_CMPXCHG
tid_to_cpu(unsigned long tid)2126 static inline unsigned int tid_to_cpu(unsigned long tid)
2127 {
2128 	return tid % TID_STEP;
2129 }
2130 
tid_to_event(unsigned long tid)2131 static inline unsigned long tid_to_event(unsigned long tid)
2132 {
2133 	return tid / TID_STEP;
2134 }
2135 #endif
2136 
init_tid(int cpu)2137 static inline unsigned int init_tid(int cpu)
2138 {
2139 	return cpu;
2140 }
2141 
note_cmpxchg_failure(const char * n,const struct kmem_cache * s,unsigned long tid)2142 static inline void note_cmpxchg_failure(const char *n,
2143 		const struct kmem_cache *s, unsigned long tid)
2144 {
2145 #ifdef SLUB_DEBUG_CMPXCHG
2146 	unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
2147 
2148 	pr_info("%s %s: cmpxchg redo ", n, s->name);
2149 
2150 #ifdef CONFIG_PREEMPTION
2151 	if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
2152 		pr_warn("due to cpu change %d -> %d\n",
2153 			tid_to_cpu(tid), tid_to_cpu(actual_tid));
2154 	else
2155 #endif
2156 	if (tid_to_event(tid) != tid_to_event(actual_tid))
2157 		pr_warn("due to cpu running other code. Event %ld->%ld\n",
2158 			tid_to_event(tid), tid_to_event(actual_tid));
2159 	else
2160 		pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
2161 			actual_tid, tid, next_tid(tid));
2162 #endif
2163 	stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
2164 }
2165 
init_kmem_cache_cpus(struct kmem_cache * s)2166 static void init_kmem_cache_cpus(struct kmem_cache *s)
2167 {
2168 	int cpu;
2169 
2170 	for_each_possible_cpu(cpu)
2171 		per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
2172 }
2173 
2174 /*
2175  * Remove the cpu slab
2176  */
deactivate_slab(struct kmem_cache * s,struct page * page,void * freelist,struct kmem_cache_cpu * c)2177 static void deactivate_slab(struct kmem_cache *s, struct page *page,
2178 				void *freelist, struct kmem_cache_cpu *c)
2179 {
2180 	enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
2181 	struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2182 	int lock = 0, free_delta = 0;
2183 	enum slab_modes l = M_NONE, m = M_NONE;
2184 	void *nextfree, *freelist_iter, *freelist_tail;
2185 	int tail = DEACTIVATE_TO_HEAD;
2186 	struct page new;
2187 	struct page old;
2188 
2189 	if (page->freelist) {
2190 		stat(s, DEACTIVATE_REMOTE_FREES);
2191 		tail = DEACTIVATE_TO_TAIL;
2192 	}
2193 
2194 	/*
2195 	 * Stage one: Count the objects on cpu's freelist as free_delta and
2196 	 * remember the last object in freelist_tail for later splicing.
2197 	 */
2198 	freelist_tail = NULL;
2199 	freelist_iter = freelist;
2200 	while (freelist_iter) {
2201 		nextfree = get_freepointer(s, freelist_iter);
2202 
2203 		/*
2204 		 * If 'nextfree' is invalid, it is possible that the object at
2205 		 * 'freelist_iter' is already corrupted.  So isolate all objects
2206 		 * starting at 'freelist_iter' by skipping them.
2207 		 */
2208 		if (freelist_corrupted(s, page, &freelist_iter, nextfree))
2209 			break;
2210 
2211 		freelist_tail = freelist_iter;
2212 		free_delta++;
2213 
2214 		freelist_iter = nextfree;
2215 	}
2216 
2217 	/*
2218 	 * Stage two: Unfreeze the page while splicing the per-cpu
2219 	 * freelist to the head of page's freelist.
2220 	 *
2221 	 * Ensure that the page is unfrozen while the list presence
2222 	 * reflects the actual number of objects during unfreeze.
2223 	 *
2224 	 * We setup the list membership and then perform a cmpxchg
2225 	 * with the count. If there is a mismatch then the page
2226 	 * is not unfrozen but the page is on the wrong list.
2227 	 *
2228 	 * Then we restart the process which may have to remove
2229 	 * the page from the list that we just put it on again
2230 	 * because the number of objects in the slab may have
2231 	 * changed.
2232 	 */
2233 redo:
2234 
2235 	old.freelist = READ_ONCE(page->freelist);
2236 	old.counters = READ_ONCE(page->counters);
2237 	VM_BUG_ON(!old.frozen);
2238 
2239 	/* Determine target state of the slab */
2240 	new.counters = old.counters;
2241 	if (freelist_tail) {
2242 		new.inuse -= free_delta;
2243 		set_freepointer(s, freelist_tail, old.freelist);
2244 		new.freelist = freelist;
2245 	} else
2246 		new.freelist = old.freelist;
2247 
2248 	new.frozen = 0;
2249 
2250 	if (!new.inuse && n->nr_partial >= s->min_partial)
2251 		m = M_FREE;
2252 	else if (new.freelist) {
2253 		m = M_PARTIAL;
2254 		if (!lock) {
2255 			lock = 1;
2256 			/*
2257 			 * Taking the spinlock removes the possibility
2258 			 * that acquire_slab() will see a slab page that
2259 			 * is frozen
2260 			 */
2261 			spin_lock(&n->list_lock);
2262 		}
2263 	} else {
2264 		m = M_FULL;
2265 		if (kmem_cache_debug_flags(s, SLAB_STORE_USER) && !lock) {
2266 			lock = 1;
2267 			/*
2268 			 * This also ensures that the scanning of full
2269 			 * slabs from diagnostic functions will not see
2270 			 * any frozen slabs.
2271 			 */
2272 			spin_lock(&n->list_lock);
2273 		}
2274 	}
2275 
2276 	if (l != m) {
2277 		if (l == M_PARTIAL)
2278 			remove_partial(n, page);
2279 		else if (l == M_FULL)
2280 			remove_full(s, n, page);
2281 
2282 		if (m == M_PARTIAL)
2283 			add_partial(n, page, tail);
2284 		else if (m == M_FULL)
2285 			add_full(s, n, page);
2286 	}
2287 
2288 	l = m;
2289 	if (!__cmpxchg_double_slab(s, page,
2290 				old.freelist, old.counters,
2291 				new.freelist, new.counters,
2292 				"unfreezing slab"))
2293 		goto redo;
2294 
2295 	if (lock)
2296 		spin_unlock(&n->list_lock);
2297 
2298 	if (m == M_PARTIAL)
2299 		stat(s, tail);
2300 	else if (m == M_FULL)
2301 		stat(s, DEACTIVATE_FULL);
2302 	else if (m == M_FREE) {
2303 		stat(s, DEACTIVATE_EMPTY);
2304 		discard_slab(s, page);
2305 		stat(s, FREE_SLAB);
2306 	}
2307 
2308 	c->page = NULL;
2309 	c->freelist = NULL;
2310 }
2311 
2312 /*
2313  * Unfreeze all the cpu partial slabs.
2314  *
2315  * This function must be called with interrupts disabled
2316  * for the cpu using c (or some other guarantee must be there
2317  * to guarantee no concurrent accesses).
2318  */
unfreeze_partials(struct kmem_cache * s,struct kmem_cache_cpu * c)2319 static void unfreeze_partials(struct kmem_cache *s,
2320 		struct kmem_cache_cpu *c)
2321 {
2322 #ifdef CONFIG_SLUB_CPU_PARTIAL
2323 	struct kmem_cache_node *n = NULL, *n2 = NULL;
2324 	struct page *page, *discard_page = NULL;
2325 
2326 	while ((page = slub_percpu_partial(c))) {
2327 		struct page new;
2328 		struct page old;
2329 
2330 		slub_set_percpu_partial(c, page);
2331 
2332 		n2 = get_node(s, page_to_nid(page));
2333 		if (n != n2) {
2334 			if (n)
2335 				spin_unlock(&n->list_lock);
2336 
2337 			n = n2;
2338 			spin_lock(&n->list_lock);
2339 		}
2340 
2341 		do {
2342 
2343 			old.freelist = page->freelist;
2344 			old.counters = page->counters;
2345 			VM_BUG_ON(!old.frozen);
2346 
2347 			new.counters = old.counters;
2348 			new.freelist = old.freelist;
2349 
2350 			new.frozen = 0;
2351 
2352 		} while (!__cmpxchg_double_slab(s, page,
2353 				old.freelist, old.counters,
2354 				new.freelist, new.counters,
2355 				"unfreezing slab"));
2356 
2357 		if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
2358 			page->next = discard_page;
2359 			discard_page = page;
2360 		} else {
2361 			add_partial(n, page, DEACTIVATE_TO_TAIL);
2362 			stat(s, FREE_ADD_PARTIAL);
2363 		}
2364 	}
2365 
2366 	if (n)
2367 		spin_unlock(&n->list_lock);
2368 
2369 	while (discard_page) {
2370 		page = discard_page;
2371 		discard_page = discard_page->next;
2372 
2373 		stat(s, DEACTIVATE_EMPTY);
2374 		discard_slab(s, page);
2375 		stat(s, FREE_SLAB);
2376 	}
2377 #endif	/* CONFIG_SLUB_CPU_PARTIAL */
2378 }
2379 
2380 /*
2381  * Put a page that was just frozen (in __slab_free|get_partial_node) into a
2382  * partial page slot if available.
2383  *
2384  * If we did not find a slot then simply move all the partials to the
2385  * per node partial list.
2386  */
put_cpu_partial(struct kmem_cache * s,struct page * page,int drain)2387 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
2388 {
2389 #ifdef CONFIG_SLUB_CPU_PARTIAL
2390 	struct page *oldpage;
2391 	int pages;
2392 	int pobjects;
2393 
2394 	preempt_disable();
2395 	do {
2396 		pages = 0;
2397 		pobjects = 0;
2398 		oldpage = this_cpu_read(s->cpu_slab->partial);
2399 
2400 		if (oldpage) {
2401 			pobjects = oldpage->pobjects;
2402 			pages = oldpage->pages;
2403 			if (drain && pobjects > slub_cpu_partial(s)) {
2404 				unsigned long flags;
2405 				/*
2406 				 * partial array is full. Move the existing
2407 				 * set to the per node partial list.
2408 				 */
2409 				local_irq_save(flags);
2410 				unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2411 				local_irq_restore(flags);
2412 				oldpage = NULL;
2413 				pobjects = 0;
2414 				pages = 0;
2415 				stat(s, CPU_PARTIAL_DRAIN);
2416 			}
2417 		}
2418 
2419 		pages++;
2420 		pobjects += page->objects - page->inuse;
2421 
2422 		page->pages = pages;
2423 		page->pobjects = pobjects;
2424 		page->next = oldpage;
2425 
2426 	} while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2427 								!= oldpage);
2428 	if (unlikely(!slub_cpu_partial(s))) {
2429 		unsigned long flags;
2430 
2431 		local_irq_save(flags);
2432 		unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2433 		local_irq_restore(flags);
2434 	}
2435 	preempt_enable();
2436 #endif	/* CONFIG_SLUB_CPU_PARTIAL */
2437 }
2438 
flush_slab(struct kmem_cache * s,struct kmem_cache_cpu * c)2439 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
2440 {
2441 	stat(s, CPUSLAB_FLUSH);
2442 	deactivate_slab(s, c->page, c->freelist, c);
2443 
2444 	c->tid = next_tid(c->tid);
2445 }
2446 
2447 /*
2448  * Flush cpu slab.
2449  *
2450  * Called from IPI handler with interrupts disabled.
2451  */
__flush_cpu_slab(struct kmem_cache * s,int cpu)2452 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
2453 {
2454 	struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2455 
2456 	if (c->page)
2457 		flush_slab(s, c);
2458 
2459 	unfreeze_partials(s, c);
2460 }
2461 
flush_cpu_slab(void * d)2462 static void flush_cpu_slab(void *d)
2463 {
2464 	struct kmem_cache *s = d;
2465 
2466 	__flush_cpu_slab(s, smp_processor_id());
2467 }
2468 
has_cpu_slab(int cpu,void * info)2469 static bool has_cpu_slab(int cpu, void *info)
2470 {
2471 	struct kmem_cache *s = info;
2472 	struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2473 
2474 	return c->page || slub_percpu_partial(c);
2475 }
2476 
flush_all(struct kmem_cache * s)2477 static void flush_all(struct kmem_cache *s)
2478 {
2479 	on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1);
2480 }
2481 
2482 /*
2483  * Use the cpu notifier to insure that the cpu slabs are flushed when
2484  * necessary.
2485  */
slub_cpu_dead(unsigned int cpu)2486 static int slub_cpu_dead(unsigned int cpu)
2487 {
2488 	struct kmem_cache *s;
2489 	unsigned long flags;
2490 
2491 	mutex_lock(&slab_mutex);
2492 	list_for_each_entry(s, &slab_caches, list) {
2493 		local_irq_save(flags);
2494 		__flush_cpu_slab(s, cpu);
2495 		local_irq_restore(flags);
2496 	}
2497 	mutex_unlock(&slab_mutex);
2498 	return 0;
2499 }
2500 
2501 /*
2502  * Check if the objects in a per cpu structure fit numa
2503  * locality expectations.
2504  */
node_match(struct page * page,int node)2505 static inline int node_match(struct page *page, int node)
2506 {
2507 #ifdef CONFIG_NUMA
2508 	if (node != NUMA_NO_NODE && page_to_nid(page) != node)
2509 		return 0;
2510 #endif
2511 	return 1;
2512 }
2513 
2514 #ifdef CONFIG_SLUB_DEBUG
count_free(struct page * page)2515 static int count_free(struct page *page)
2516 {
2517 	return page->objects - page->inuse;
2518 }
2519 
node_nr_objs(struct kmem_cache_node * n)2520 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2521 {
2522 	return atomic_long_read(&n->total_objects);
2523 }
2524 #endif /* CONFIG_SLUB_DEBUG */
2525 
2526 #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
count_partial(struct kmem_cache_node * n,int (* get_count)(struct page *))2527 static unsigned long count_partial(struct kmem_cache_node *n,
2528 					int (*get_count)(struct page *))
2529 {
2530 	unsigned long flags;
2531 	unsigned long x = 0;
2532 	struct page *page;
2533 
2534 	spin_lock_irqsave(&n->list_lock, flags);
2535 	list_for_each_entry(page, &n->partial, slab_list)
2536 		x += get_count(page);
2537 	spin_unlock_irqrestore(&n->list_lock, flags);
2538 	return x;
2539 }
2540 #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
2541 
2542 static noinline void
slab_out_of_memory(struct kmem_cache * s,gfp_t gfpflags,int nid)2543 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2544 {
2545 #ifdef CONFIG_SLUB_DEBUG
2546 	static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2547 				      DEFAULT_RATELIMIT_BURST);
2548 	int node;
2549 	struct kmem_cache_node *n;
2550 
2551 	if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2552 		return;
2553 
2554 	pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2555 		nid, gfpflags, &gfpflags);
2556 	pr_warn("  cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
2557 		s->name, s->object_size, s->size, oo_order(s->oo),
2558 		oo_order(s->min));
2559 
2560 	if (oo_order(s->min) > get_order(s->object_size))
2561 		pr_warn("  %s debugging increased min order, use slub_debug=O to disable.\n",
2562 			s->name);
2563 
2564 	for_each_kmem_cache_node(s, node, n) {
2565 		unsigned long nr_slabs;
2566 		unsigned long nr_objs;
2567 		unsigned long nr_free;
2568 
2569 		nr_free  = count_partial(n, count_free);
2570 		nr_slabs = node_nr_slabs(n);
2571 		nr_objs  = node_nr_objs(n);
2572 
2573 		pr_warn("  node %d: slabs: %ld, objs: %ld, free: %ld\n",
2574 			node, nr_slabs, nr_objs, nr_free);
2575 	}
2576 #endif
2577 }
2578 
new_slab_objects(struct kmem_cache * s,gfp_t flags,int node,struct kmem_cache_cpu ** pc)2579 static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2580 			int node, struct kmem_cache_cpu **pc)
2581 {
2582 	void *freelist;
2583 	struct kmem_cache_cpu *c = *pc;
2584 	struct page *page;
2585 
2586 	WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2587 
2588 	freelist = get_partial(s, flags, node, c);
2589 
2590 	if (freelist)
2591 		return freelist;
2592 
2593 	page = new_slab(s, flags, node);
2594 	if (page) {
2595 		c = raw_cpu_ptr(s->cpu_slab);
2596 		if (c->page)
2597 			flush_slab(s, c);
2598 
2599 		/*
2600 		 * No other reference to the page yet so we can
2601 		 * muck around with it freely without cmpxchg
2602 		 */
2603 		freelist = page->freelist;
2604 		page->freelist = NULL;
2605 
2606 		stat(s, ALLOC_SLAB);
2607 		c->page = page;
2608 		*pc = c;
2609 	}
2610 
2611 	return freelist;
2612 }
2613 
pfmemalloc_match(struct page * page,gfp_t gfpflags)2614 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2615 {
2616 	if (unlikely(PageSlabPfmemalloc(page)))
2617 		return gfp_pfmemalloc_allowed(gfpflags);
2618 
2619 	return true;
2620 }
2621 
2622 /*
2623  * Check the page->freelist of a page and either transfer the freelist to the
2624  * per cpu freelist or deactivate the page.
2625  *
2626  * The page is still frozen if the return value is not NULL.
2627  *
2628  * If this function returns NULL then the page has been unfrozen.
2629  *
2630  * This function must be called with interrupt disabled.
2631  */
get_freelist(struct kmem_cache * s,struct page * page)2632 static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2633 {
2634 	struct page new;
2635 	unsigned long counters;
2636 	void *freelist;
2637 
2638 	do {
2639 		freelist = page->freelist;
2640 		counters = page->counters;
2641 
2642 		new.counters = counters;
2643 		VM_BUG_ON(!new.frozen);
2644 
2645 		new.inuse = page->objects;
2646 		new.frozen = freelist != NULL;
2647 
2648 	} while (!__cmpxchg_double_slab(s, page,
2649 		freelist, counters,
2650 		NULL, new.counters,
2651 		"get_freelist"));
2652 
2653 	return freelist;
2654 }
2655 
2656 /*
2657  * Slow path. The lockless freelist is empty or we need to perform
2658  * debugging duties.
2659  *
2660  * Processing is still very fast if new objects have been freed to the
2661  * regular freelist. In that case we simply take over the regular freelist
2662  * as the lockless freelist and zap the regular freelist.
2663  *
2664  * If that is not working then we fall back to the partial lists. We take the
2665  * first element of the freelist as the object to allocate now and move the
2666  * rest of the freelist to the lockless freelist.
2667  *
2668  * And if we were unable to get a new slab from the partial slab lists then
2669  * we need to allocate a new slab. This is the slowest path since it involves
2670  * a call to the page allocator and the setup of a new slab.
2671  *
2672  * Version of __slab_alloc to use when we know that interrupts are
2673  * already disabled (which is the case for bulk allocation).
2674  */
___slab_alloc(struct kmem_cache * s,gfp_t gfpflags,int node,unsigned long addr,struct kmem_cache_cpu * c)2675 static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2676 			  unsigned long addr, struct kmem_cache_cpu *c)
2677 {
2678 	void *freelist;
2679 	struct page *page;
2680 
2681 	stat(s, ALLOC_SLOWPATH);
2682 
2683 	page = c->page;
2684 	if (!page) {
2685 		/*
2686 		 * if the node is not online or has no normal memory, just
2687 		 * ignore the node constraint
2688 		 */
2689 		if (unlikely(node != NUMA_NO_NODE &&
2690 			     !node_isset(node, slab_nodes)))
2691 			node = NUMA_NO_NODE;
2692 		goto new_slab;
2693 	}
2694 redo:
2695 
2696 	if (unlikely(!node_match(page, node))) {
2697 		/*
2698 		 * same as above but node_match() being false already
2699 		 * implies node != NUMA_NO_NODE
2700 		 */
2701 		if (!node_isset(node, slab_nodes)) {
2702 			node = NUMA_NO_NODE;
2703 			goto redo;
2704 		} else {
2705 			stat(s, ALLOC_NODE_MISMATCH);
2706 			deactivate_slab(s, page, c->freelist, c);
2707 			goto new_slab;
2708 		}
2709 	}
2710 
2711 	/*
2712 	 * By rights, we should be searching for a slab page that was
2713 	 * PFMEMALLOC but right now, we are losing the pfmemalloc
2714 	 * information when the page leaves the per-cpu allocator
2715 	 */
2716 	if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2717 		deactivate_slab(s, page, c->freelist, c);
2718 		goto new_slab;
2719 	}
2720 
2721 	/* must check again c->freelist in case of cpu migration or IRQ */
2722 	freelist = c->freelist;
2723 	if (freelist)
2724 		goto load_freelist;
2725 
2726 	freelist = get_freelist(s, page);
2727 
2728 	if (!freelist) {
2729 		c->page = NULL;
2730 		stat(s, DEACTIVATE_BYPASS);
2731 		goto new_slab;
2732 	}
2733 
2734 	stat(s, ALLOC_REFILL);
2735 
2736 load_freelist:
2737 	/*
2738 	 * freelist is pointing to the list of objects to be used.
2739 	 * page is pointing to the page from which the objects are obtained.
2740 	 * That page must be frozen for per cpu allocations to work.
2741 	 */
2742 	VM_BUG_ON(!c->page->frozen);
2743 	c->freelist = get_freepointer(s, freelist);
2744 	c->tid = next_tid(c->tid);
2745 	return freelist;
2746 
2747 new_slab:
2748 
2749 	if (slub_percpu_partial(c)) {
2750 		page = c->page = slub_percpu_partial(c);
2751 		slub_set_percpu_partial(c, page);
2752 		stat(s, CPU_PARTIAL_ALLOC);
2753 		goto redo;
2754 	}
2755 
2756 	freelist = new_slab_objects(s, gfpflags, node, &c);
2757 
2758 	if (unlikely(!freelist)) {
2759 		slab_out_of_memory(s, gfpflags, node);
2760 		return NULL;
2761 	}
2762 
2763 	page = c->page;
2764 	if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
2765 		goto load_freelist;
2766 
2767 	/* Only entered in the debug case */
2768 	if (kmem_cache_debug(s) &&
2769 			!alloc_debug_processing(s, page, freelist, addr))
2770 		goto new_slab;	/* Slab failed checks. Next slab needed */
2771 
2772 	deactivate_slab(s, page, get_freepointer(s, freelist), c);
2773 	return freelist;
2774 }
2775 
2776 /*
2777  * Another one that disabled interrupt and compensates for possible
2778  * cpu changes by refetching the per cpu area pointer.
2779  */
__slab_alloc(struct kmem_cache * s,gfp_t gfpflags,int node,unsigned long addr,struct kmem_cache_cpu * c)2780 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2781 			  unsigned long addr, struct kmem_cache_cpu *c)
2782 {
2783 	void *p;
2784 	unsigned long flags;
2785 
2786 	local_irq_save(flags);
2787 #ifdef CONFIG_PREEMPTION
2788 	/*
2789 	 * We may have been preempted and rescheduled on a different
2790 	 * cpu before disabling interrupts. Need to reload cpu area
2791 	 * pointer.
2792 	 */
2793 	c = this_cpu_ptr(s->cpu_slab);
2794 #endif
2795 
2796 	p = ___slab_alloc(s, gfpflags, node, addr, c);
2797 	local_irq_restore(flags);
2798 	return p;
2799 }
2800 
2801 /*
2802  * If the object has been wiped upon free, make sure it's fully initialized by
2803  * zeroing out freelist pointer.
2804  */
maybe_wipe_obj_freeptr(struct kmem_cache * s,void * obj)2805 static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s,
2806 						   void *obj)
2807 {
2808 	if (unlikely(slab_want_init_on_free(s)) && obj)
2809 		memset((void *)((char *)kasan_reset_tag(obj) + s->offset),
2810 			0, sizeof(void *));
2811 }
2812 
2813 /*
2814  * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2815  * have the fastpath folded into their functions. So no function call
2816  * overhead for requests that can be satisfied on the fastpath.
2817  *
2818  * The fastpath works by first checking if the lockless freelist can be used.
2819  * If not then __slab_alloc is called for slow processing.
2820  *
2821  * Otherwise we can simply pick the next object from the lockless free list.
2822  */
slab_alloc_node(struct kmem_cache * s,gfp_t gfpflags,int node,unsigned long addr,size_t orig_size)2823 static __always_inline void *slab_alloc_node(struct kmem_cache *s,
2824 		gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
2825 {
2826 	void *object;
2827 	struct kmem_cache_cpu *c;
2828 	struct page *page;
2829 	unsigned long tid;
2830 	struct obj_cgroup *objcg = NULL;
2831 	bool init = false;
2832 
2833 	s = slab_pre_alloc_hook(s, &objcg, 1, gfpflags);
2834 	if (!s)
2835 		return NULL;
2836 
2837 	object = kfence_alloc(s, orig_size, gfpflags);
2838 	if (unlikely(object))
2839 		goto out;
2840 
2841 redo:
2842 	/*
2843 	 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2844 	 * enabled. We may switch back and forth between cpus while
2845 	 * reading from one cpu area. That does not matter as long
2846 	 * as we end up on the original cpu again when doing the cmpxchg.
2847 	 *
2848 	 * We should guarantee that tid and kmem_cache are retrieved on
2849 	 * the same cpu. It could be different if CONFIG_PREEMPTION so we need
2850 	 * to check if it is matched or not.
2851 	 */
2852 	do {
2853 		tid = this_cpu_read(s->cpu_slab->tid);
2854 		c = raw_cpu_ptr(s->cpu_slab);
2855 	} while (IS_ENABLED(CONFIG_PREEMPTION) &&
2856 		 unlikely(tid != READ_ONCE(c->tid)));
2857 
2858 	/*
2859 	 * Irqless object alloc/free algorithm used here depends on sequence
2860 	 * of fetching cpu_slab's data. tid should be fetched before anything
2861 	 * on c to guarantee that object and page associated with previous tid
2862 	 * won't be used with current tid. If we fetch tid first, object and
2863 	 * page could be one associated with next tid and our alloc/free
2864 	 * request will be failed. In this case, we will retry. So, no problem.
2865 	 */
2866 	barrier();
2867 
2868 	/*
2869 	 * The transaction ids are globally unique per cpu and per operation on
2870 	 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2871 	 * occurs on the right processor and that there was no operation on the
2872 	 * linked list in between.
2873 	 */
2874 
2875 	object = c->freelist;
2876 	page = c->page;
2877 	if (unlikely(!object || !page || !node_match(page, node))) {
2878 		object = __slab_alloc(s, gfpflags, node, addr, c);
2879 	} else {
2880 		void *next_object = get_freepointer_safe(s, object);
2881 
2882 		/*
2883 		 * The cmpxchg will only match if there was no additional
2884 		 * operation and if we are on the right processor.
2885 		 *
2886 		 * The cmpxchg does the following atomically (without lock
2887 		 * semantics!)
2888 		 * 1. Relocate first pointer to the current per cpu area.
2889 		 * 2. Verify that tid and freelist have not been changed
2890 		 * 3. If they were not changed replace tid and freelist
2891 		 *
2892 		 * Since this is without lock semantics the protection is only
2893 		 * against code executing on this cpu *not* from access by
2894 		 * other cpus.
2895 		 */
2896 		if (unlikely(!this_cpu_cmpxchg_double(
2897 				s->cpu_slab->freelist, s->cpu_slab->tid,
2898 				object, tid,
2899 				next_object, next_tid(tid)))) {
2900 
2901 			note_cmpxchg_failure("slab_alloc", s, tid);
2902 			goto redo;
2903 		}
2904 		prefetch_freepointer(s, next_object);
2905 		stat(s, ALLOC_FASTPATH);
2906 	}
2907 
2908 	maybe_wipe_obj_freeptr(s, object);
2909 	init = slab_want_init_on_alloc(gfpflags, s);
2910 
2911 out:
2912 	slab_post_alloc_hook(s, objcg, gfpflags, 1, &object, init);
2913 
2914 	return object;
2915 }
2916 
slab_alloc(struct kmem_cache * s,gfp_t gfpflags,unsigned long addr,size_t orig_size)2917 static __always_inline void *slab_alloc(struct kmem_cache *s,
2918 		gfp_t gfpflags, unsigned long addr, size_t orig_size)
2919 {
2920 	return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr, orig_size);
2921 }
2922 
kmem_cache_alloc(struct kmem_cache * s,gfp_t gfpflags)2923 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2924 {
2925 	void *ret = slab_alloc(s, gfpflags, _RET_IP_, s->object_size);
2926 
2927 	trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2928 				s->size, gfpflags);
2929 
2930 	return ret;
2931 }
2932 EXPORT_SYMBOL(kmem_cache_alloc);
2933 
2934 #ifdef CONFIG_TRACING
kmem_cache_alloc_trace(struct kmem_cache * s,gfp_t gfpflags,size_t size)2935 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
2936 {
2937 	void *ret = slab_alloc(s, gfpflags, _RET_IP_, size);
2938 	trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2939 	ret = kasan_kmalloc(s, ret, size, gfpflags);
2940 	return ret;
2941 }
2942 EXPORT_SYMBOL(kmem_cache_alloc_trace);
2943 #endif
2944 
2945 #ifdef CONFIG_NUMA
kmem_cache_alloc_node(struct kmem_cache * s,gfp_t gfpflags,int node)2946 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2947 {
2948 	void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, s->object_size);
2949 
2950 	trace_kmem_cache_alloc_node(_RET_IP_, ret,
2951 				    s->object_size, s->size, gfpflags, node);
2952 
2953 	return ret;
2954 }
2955 EXPORT_SYMBOL(kmem_cache_alloc_node);
2956 
2957 #ifdef CONFIG_TRACING
kmem_cache_alloc_node_trace(struct kmem_cache * s,gfp_t gfpflags,int node,size_t size)2958 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
2959 				    gfp_t gfpflags,
2960 				    int node, size_t size)
2961 {
2962 	void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, size);
2963 
2964 	trace_kmalloc_node(_RET_IP_, ret,
2965 			   size, s->size, gfpflags, node);
2966 
2967 	ret = kasan_kmalloc(s, ret, size, gfpflags);
2968 	return ret;
2969 }
2970 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
2971 #endif
2972 #endif	/* CONFIG_NUMA */
2973 
2974 /*
2975  * Slow path handling. This may still be called frequently since objects
2976  * have a longer lifetime than the cpu slabs in most processing loads.
2977  *
2978  * So we still attempt to reduce cache line usage. Just take the slab
2979  * lock and free the item. If there is no additional partial page
2980  * handling required then we can return immediately.
2981  */
__slab_free(struct kmem_cache * s,struct page * page,void * head,void * tail,int cnt,unsigned long addr)2982 static void __slab_free(struct kmem_cache *s, struct page *page,
2983 			void *head, void *tail, int cnt,
2984 			unsigned long addr)
2985 
2986 {
2987 	void *prior;
2988 	int was_frozen;
2989 	struct page new;
2990 	unsigned long counters;
2991 	struct kmem_cache_node *n = NULL;
2992 	unsigned long flags;
2993 
2994 	stat(s, FREE_SLOWPATH);
2995 
2996 	if (kfence_free(head))
2997 		return;
2998 
2999 	if (kmem_cache_debug(s) &&
3000 	    !free_debug_processing(s, page, head, tail, cnt, addr))
3001 		return;
3002 
3003 	do {
3004 		if (unlikely(n)) {
3005 			spin_unlock_irqrestore(&n->list_lock, flags);
3006 			n = NULL;
3007 		}
3008 		prior = page->freelist;
3009 		counters = page->counters;
3010 		set_freepointer(s, tail, prior);
3011 		new.counters = counters;
3012 		was_frozen = new.frozen;
3013 		new.inuse -= cnt;
3014 		if ((!new.inuse || !prior) && !was_frozen) {
3015 
3016 			if (kmem_cache_has_cpu_partial(s) && !prior) {
3017 
3018 				/*
3019 				 * Slab was on no list before and will be
3020 				 * partially empty
3021 				 * We can defer the list move and instead
3022 				 * freeze it.
3023 				 */
3024 				new.frozen = 1;
3025 
3026 			} else { /* Needs to be taken off a list */
3027 
3028 				n = get_node(s, page_to_nid(page));
3029 				/*
3030 				 * Speculatively acquire the list_lock.
3031 				 * If the cmpxchg does not succeed then we may
3032 				 * drop the list_lock without any processing.
3033 				 *
3034 				 * Otherwise the list_lock will synchronize with
3035 				 * other processors updating the list of slabs.
3036 				 */
3037 				spin_lock_irqsave(&n->list_lock, flags);
3038 
3039 			}
3040 		}
3041 
3042 	} while (!cmpxchg_double_slab(s, page,
3043 		prior, counters,
3044 		head, new.counters,
3045 		"__slab_free"));
3046 
3047 	if (likely(!n)) {
3048 
3049 		if (likely(was_frozen)) {
3050 			/*
3051 			 * The list lock was not taken therefore no list
3052 			 * activity can be necessary.
3053 			 */
3054 			stat(s, FREE_FROZEN);
3055 		} else if (new.frozen) {
3056 			/*
3057 			 * If we just froze the page then put it onto the
3058 			 * per cpu partial list.
3059 			 */
3060 			put_cpu_partial(s, page, 1);
3061 			stat(s, CPU_PARTIAL_FREE);
3062 		}
3063 
3064 		return;
3065 	}
3066 
3067 	if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
3068 		goto slab_empty;
3069 
3070 	/*
3071 	 * Objects left in the slab. If it was not on the partial list before
3072 	 * then add it.
3073 	 */
3074 	if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
3075 		remove_full(s, n, page);
3076 		add_partial(n, page, DEACTIVATE_TO_TAIL);
3077 		stat(s, FREE_ADD_PARTIAL);
3078 	}
3079 	spin_unlock_irqrestore(&n->list_lock, flags);
3080 	return;
3081 
3082 slab_empty:
3083 	if (prior) {
3084 		/*
3085 		 * Slab on the partial list.
3086 		 */
3087 		remove_partial(n, page);
3088 		stat(s, FREE_REMOVE_PARTIAL);
3089 	} else {
3090 		/* Slab must be on the full list */
3091 		remove_full(s, n, page);
3092 	}
3093 
3094 	spin_unlock_irqrestore(&n->list_lock, flags);
3095 	stat(s, FREE_SLAB);
3096 	discard_slab(s, page);
3097 }
3098 
3099 /*
3100  * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
3101  * can perform fastpath freeing without additional function calls.
3102  *
3103  * The fastpath is only possible if we are freeing to the current cpu slab
3104  * of this processor. This typically the case if we have just allocated
3105  * the item before.
3106  *
3107  * If fastpath is not possible then fall back to __slab_free where we deal
3108  * with all sorts of special processing.
3109  *
3110  * Bulk free of a freelist with several objects (all pointing to the
3111  * same page) possible by specifying head and tail ptr, plus objects
3112  * count (cnt). Bulk free indicated by tail pointer being set.
3113  */
do_slab_free(struct kmem_cache * s,struct page * page,void * head,void * tail,int cnt,unsigned long addr)3114 static __always_inline void do_slab_free(struct kmem_cache *s,
3115 				struct page *page, void *head, void *tail,
3116 				int cnt, unsigned long addr)
3117 {
3118 	void *tail_obj = tail ? : head;
3119 	struct kmem_cache_cpu *c;
3120 	unsigned long tid;
3121 
3122 	memcg_slab_free_hook(s, &head, 1);
3123 redo:
3124 	/*
3125 	 * Determine the currently cpus per cpu slab.
3126 	 * The cpu may change afterward. However that does not matter since
3127 	 * data is retrieved via this pointer. If we are on the same cpu
3128 	 * during the cmpxchg then the free will succeed.
3129 	 */
3130 	do {
3131 		tid = this_cpu_read(s->cpu_slab->tid);
3132 		c = raw_cpu_ptr(s->cpu_slab);
3133 	} while (IS_ENABLED(CONFIG_PREEMPTION) &&
3134 		 unlikely(tid != READ_ONCE(c->tid)));
3135 
3136 	/* Same with comment on barrier() in slab_alloc_node() */
3137 	barrier();
3138 
3139 	if (likely(page == c->page)) {
3140 		void **freelist = READ_ONCE(c->freelist);
3141 
3142 		set_freepointer(s, tail_obj, freelist);
3143 
3144 		if (unlikely(!this_cpu_cmpxchg_double(
3145 				s->cpu_slab->freelist, s->cpu_slab->tid,
3146 				freelist, tid,
3147 				head, next_tid(tid)))) {
3148 
3149 			note_cmpxchg_failure("slab_free", s, tid);
3150 			goto redo;
3151 		}
3152 		stat(s, FREE_FASTPATH);
3153 	} else
3154 		__slab_free(s, page, head, tail_obj, cnt, addr);
3155 
3156 }
3157 
slab_free(struct kmem_cache * s,struct page * page,void * head,void * tail,int cnt,unsigned long addr)3158 static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
3159 				      void *head, void *tail, int cnt,
3160 				      unsigned long addr)
3161 {
3162 	/*
3163 	 * With KASAN enabled slab_free_freelist_hook modifies the freelist
3164 	 * to remove objects, whose reuse must be delayed.
3165 	 */
3166 	if (slab_free_freelist_hook(s, &head, &tail))
3167 		do_slab_free(s, page, head, tail, cnt, addr);
3168 }
3169 
3170 #ifdef CONFIG_KASAN_GENERIC
___cache_free(struct kmem_cache * cache,void * x,unsigned long addr)3171 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
3172 {
3173 	do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
3174 }
3175 #endif
3176 
kmem_cache_free(struct kmem_cache * s,void * x)3177 void kmem_cache_free(struct kmem_cache *s, void *x)
3178 {
3179 	s = cache_from_obj(s, x);
3180 	if (!s)
3181 		return;
3182 	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
3183 	trace_kmem_cache_free(_RET_IP_, x, s->name);
3184 }
3185 EXPORT_SYMBOL(kmem_cache_free);
3186 
3187 struct detached_freelist {
3188 	struct page *page;
3189 	void *tail;
3190 	void *freelist;
3191 	int cnt;
3192 	struct kmem_cache *s;
3193 };
3194 
3195 /*
3196  * This function progressively scans the array with free objects (with
3197  * a limited look ahead) and extract objects belonging to the same
3198  * page.  It builds a detached freelist directly within the given
3199  * page/objects.  This can happen without any need for
3200  * synchronization, because the objects are owned by running process.
3201  * The freelist is build up as a single linked list in the objects.
3202  * The idea is, that this detached freelist can then be bulk
3203  * transferred to the real freelist(s), but only requiring a single
3204  * synchronization primitive.  Look ahead in the array is limited due
3205  * to performance reasons.
3206  */
3207 static inline
build_detached_freelist(struct kmem_cache * s,size_t size,void ** p,struct detached_freelist * df)3208 int build_detached_freelist(struct kmem_cache *s, size_t size,
3209 			    void **p, struct detached_freelist *df)
3210 {
3211 	size_t first_skipped_index = 0;
3212 	int lookahead = 3;
3213 	void *object;
3214 	struct page *page;
3215 
3216 	/* Always re-init detached_freelist */
3217 	df->page = NULL;
3218 
3219 	do {
3220 		object = p[--size];
3221 		/* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
3222 	} while (!object && size);
3223 
3224 	if (!object)
3225 		return 0;
3226 
3227 	page = virt_to_head_page(object);
3228 	if (!s) {
3229 		/* Handle kalloc'ed objects */
3230 		if (unlikely(!PageSlab(page))) {
3231 			BUG_ON(!PageCompound(page));
3232 			kfree_hook(object);
3233 			__free_pages(page, compound_order(page));
3234 			p[size] = NULL; /* mark object processed */
3235 			return size;
3236 		}
3237 		/* Derive kmem_cache from object */
3238 		df->s = page->slab_cache;
3239 	} else {
3240 		df->s = cache_from_obj(s, object); /* Support for memcg */
3241 	}
3242 
3243 	if (is_kfence_address(object)) {
3244 		slab_free_hook(df->s, object, false);
3245 		__kfence_free(object);
3246 		p[size] = NULL; /* mark object processed */
3247 		return size;
3248 	}
3249 
3250 	/* Start new detached freelist */
3251 	df->page = page;
3252 	set_freepointer(df->s, object, NULL);
3253 	df->tail = object;
3254 	df->freelist = object;
3255 	p[size] = NULL; /* mark object processed */
3256 	df->cnt = 1;
3257 
3258 	while (size) {
3259 		object = p[--size];
3260 		if (!object)
3261 			continue; /* Skip processed objects */
3262 
3263 		/* df->page is always set at this point */
3264 		if (df->page == virt_to_head_page(object)) {
3265 			/* Opportunity build freelist */
3266 			set_freepointer(df->s, object, df->freelist);
3267 			df->freelist = object;
3268 			df->cnt++;
3269 			p[size] = NULL; /* mark object processed */
3270 
3271 			continue;
3272 		}
3273 
3274 		/* Limit look ahead search */
3275 		if (!--lookahead)
3276 			break;
3277 
3278 		if (!first_skipped_index)
3279 			first_skipped_index = size + 1;
3280 	}
3281 
3282 	return first_skipped_index;
3283 }
3284 
3285 /* Note that interrupts must be enabled when calling this function. */
kmem_cache_free_bulk(struct kmem_cache * s,size_t size,void ** p)3286 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
3287 {
3288 	if (WARN_ON(!size))
3289 		return;
3290 
3291 	memcg_slab_free_hook(s, p, size);
3292 	do {
3293 		struct detached_freelist df;
3294 
3295 		size = build_detached_freelist(s, size, p, &df);
3296 		if (!df.page)
3297 			continue;
3298 
3299 		slab_free(df.s, df.page, df.freelist, df.tail, df.cnt, _RET_IP_);
3300 	} while (likely(size));
3301 }
3302 EXPORT_SYMBOL(kmem_cache_free_bulk);
3303 
3304 /* 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)3305 int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3306 			  void **p)
3307 {
3308 	struct kmem_cache_cpu *c;
3309 	int i;
3310 	struct obj_cgroup *objcg = NULL;
3311 
3312 	/* memcg and kmem_cache debug support */
3313 	s = slab_pre_alloc_hook(s, &objcg, size, flags);
3314 	if (unlikely(!s))
3315 		return false;
3316 	/*
3317 	 * Drain objects in the per cpu slab, while disabling local
3318 	 * IRQs, which protects against PREEMPT and interrupts
3319 	 * handlers invoking normal fastpath.
3320 	 */
3321 	local_irq_disable();
3322 	c = this_cpu_ptr(s->cpu_slab);
3323 
3324 	for (i = 0; i < size; i++) {
3325 		void *object = kfence_alloc(s, s->object_size, flags);
3326 
3327 		if (unlikely(object)) {
3328 			p[i] = object;
3329 			continue;
3330 		}
3331 
3332 		object = c->freelist;
3333 		if (unlikely(!object)) {
3334 			/*
3335 			 * We may have removed an object from c->freelist using
3336 			 * the fastpath in the previous iteration; in that case,
3337 			 * c->tid has not been bumped yet.
3338 			 * Since ___slab_alloc() may reenable interrupts while
3339 			 * allocating memory, we should bump c->tid now.
3340 			 */
3341 			c->tid = next_tid(c->tid);
3342 
3343 			/*
3344 			 * Invoking slow path likely have side-effect
3345 			 * of re-populating per CPU c->freelist
3346 			 */
3347 			p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
3348 					    _RET_IP_, c);
3349 			if (unlikely(!p[i]))
3350 				goto error;
3351 
3352 			c = this_cpu_ptr(s->cpu_slab);
3353 			maybe_wipe_obj_freeptr(s, p[i]);
3354 
3355 			continue; /* goto for-loop */
3356 		}
3357 		c->freelist = get_freepointer(s, object);
3358 		p[i] = object;
3359 		maybe_wipe_obj_freeptr(s, p[i]);
3360 	}
3361 	c->tid = next_tid(c->tid);
3362 	local_irq_enable();
3363 
3364 	/*
3365 	 * memcg and kmem_cache debug support and memory initialization.
3366 	 * Done outside of the IRQ disabled fastpath loop.
3367 	 */
3368 	slab_post_alloc_hook(s, objcg, flags, size, p,
3369 				slab_want_init_on_alloc(flags, s));
3370 	return i;
3371 error:
3372 	local_irq_enable();
3373 	slab_post_alloc_hook(s, objcg, flags, i, p, false);
3374 	__kmem_cache_free_bulk(s, i, p);
3375 	return 0;
3376 }
3377 EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3378 
3379 
3380 /*
3381  * Object placement in a slab is made very easy because we always start at
3382  * offset 0. If we tune the size of the object to the alignment then we can
3383  * get the required alignment by putting one properly sized object after
3384  * another.
3385  *
3386  * Notice that the allocation order determines the sizes of the per cpu
3387  * caches. Each processor has always one slab available for allocations.
3388  * Increasing the allocation order reduces the number of times that slabs
3389  * must be moved on and off the partial lists and is therefore a factor in
3390  * locking overhead.
3391  */
3392 
3393 /*
3394  * Minimum / Maximum order of slab pages. This influences locking overhead
3395  * and slab fragmentation. A higher order reduces the number of partial slabs
3396  * and increases the number of allocations possible without having to
3397  * take the list_lock.
3398  */
3399 static unsigned int slub_min_order;
3400 static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
3401 static unsigned int slub_min_objects;
3402 
3403 /*
3404  * Calculate the order of allocation given an slab object size.
3405  *
3406  * The order of allocation has significant impact on performance and other
3407  * system components. Generally order 0 allocations should be preferred since
3408  * order 0 does not cause fragmentation in the page allocator. Larger objects
3409  * be problematic to put into order 0 slabs because there may be too much
3410  * unused space left. We go to a higher order if more than 1/16th of the slab
3411  * would be wasted.
3412  *
3413  * In order to reach satisfactory performance we must ensure that a minimum
3414  * number of objects is in one slab. Otherwise we may generate too much
3415  * activity on the partial lists which requires taking the list_lock. This is
3416  * less a concern for large slabs though which are rarely used.
3417  *
3418  * slub_max_order specifies the order where we begin to stop considering the
3419  * number of objects in a slab as critical. If we reach slub_max_order then
3420  * we try to keep the page order as low as possible. So we accept more waste
3421  * of space in favor of a small page order.
3422  *
3423  * Higher order allocations also allow the placement of more objects in a
3424  * slab and thereby reduce object handling overhead. If the user has
3425  * requested a higher minimum order then we start with that one instead of
3426  * the smallest order which will fit the object.
3427  */
slab_order(unsigned int size,unsigned int min_objects,unsigned int max_order,unsigned int fract_leftover)3428 static inline unsigned int slab_order(unsigned int size,
3429 		unsigned int min_objects, unsigned int max_order,
3430 		unsigned int fract_leftover)
3431 {
3432 	unsigned int min_order = slub_min_order;
3433 	unsigned int order;
3434 
3435 	if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
3436 		return get_order(size * MAX_OBJS_PER_PAGE) - 1;
3437 
3438 	for (order = max(min_order, (unsigned int)get_order(min_objects * size));
3439 			order <= max_order; order++) {
3440 
3441 		unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
3442 		unsigned int rem;
3443 
3444 		rem = slab_size % size;
3445 
3446 		if (rem <= slab_size / fract_leftover)
3447 			break;
3448 	}
3449 
3450 	return order;
3451 }
3452 
calculate_order(unsigned int size)3453 static inline int calculate_order(unsigned int size)
3454 {
3455 	unsigned int order;
3456 	unsigned int min_objects;
3457 	unsigned int max_objects;
3458 	unsigned int nr_cpus;
3459 
3460 	/*
3461 	 * Attempt to find best configuration for a slab. This
3462 	 * works by first attempting to generate a layout with
3463 	 * the best configuration and backing off gradually.
3464 	 *
3465 	 * First we increase the acceptable waste in a slab. Then
3466 	 * we reduce the minimum objects required in a slab.
3467 	 */
3468 	min_objects = slub_min_objects;
3469 	if (!min_objects) {
3470 		/*
3471 		 * Some architectures will only update present cpus when
3472 		 * onlining them, so don't trust the number if it's just 1. But
3473 		 * we also don't want to use nr_cpu_ids always, as on some other
3474 		 * architectures, there can be many possible cpus, but never
3475 		 * onlined. Here we compromise between trying to avoid too high
3476 		 * order on systems that appear larger than they are, and too
3477 		 * low order on systems that appear smaller than they are.
3478 		 */
3479 		nr_cpus = num_present_cpus();
3480 		if (nr_cpus <= 1)
3481 			nr_cpus = nr_cpu_ids;
3482 		min_objects = 4 * (fls(nr_cpus) + 1);
3483 	}
3484 	max_objects = order_objects(slub_max_order, size);
3485 	min_objects = min(min_objects, max_objects);
3486 
3487 	while (min_objects > 1) {
3488 		unsigned int fraction;
3489 
3490 		fraction = 16;
3491 		while (fraction >= 4) {
3492 			order = slab_order(size, min_objects,
3493 					slub_max_order, fraction);
3494 			if (order <= slub_max_order)
3495 				return order;
3496 			fraction /= 2;
3497 		}
3498 		min_objects--;
3499 	}
3500 
3501 	/*
3502 	 * We were unable to place multiple objects in a slab. Now
3503 	 * lets see if we can place a single object there.
3504 	 */
3505 	order = slab_order(size, 1, slub_max_order, 1);
3506 	if (order <= slub_max_order)
3507 		return order;
3508 
3509 	/*
3510 	 * Doh this slab cannot be placed using slub_max_order.
3511 	 */
3512 	order = slab_order(size, 1, MAX_ORDER, 1);
3513 	if (order < MAX_ORDER)
3514 		return order;
3515 	return -ENOSYS;
3516 }
3517 
3518 static void
init_kmem_cache_node(struct kmem_cache_node * n)3519 init_kmem_cache_node(struct kmem_cache_node *n)
3520 {
3521 	n->nr_partial = 0;
3522 	spin_lock_init(&n->list_lock);
3523 	INIT_LIST_HEAD(&n->partial);
3524 #ifdef CONFIG_SLUB_DEBUG
3525 	atomic_long_set(&n->nr_slabs, 0);
3526 	atomic_long_set(&n->total_objects, 0);
3527 	INIT_LIST_HEAD(&n->full);
3528 #endif
3529 }
3530 
alloc_kmem_cache_cpus(struct kmem_cache * s)3531 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
3532 {
3533 	BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
3534 			KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
3535 
3536 	/*
3537 	 * Must align to double word boundary for the double cmpxchg
3538 	 * instructions to work; see __pcpu_double_call_return_bool().
3539 	 */
3540 	s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3541 				     2 * sizeof(void *));
3542 
3543 	if (!s->cpu_slab)
3544 		return 0;
3545 
3546 	init_kmem_cache_cpus(s);
3547 
3548 	return 1;
3549 }
3550 
3551 static struct kmem_cache *kmem_cache_node;
3552 
3553 /*
3554  * No kmalloc_node yet so do it by hand. We know that this is the first
3555  * slab on the node for this slabcache. There are no concurrent accesses
3556  * possible.
3557  *
3558  * Note that this function only works on the kmem_cache_node
3559  * when allocating for the kmem_cache_node. This is used for bootstrapping
3560  * memory on a fresh node that has no slab structures yet.
3561  */
early_kmem_cache_node_alloc(int node)3562 static void early_kmem_cache_node_alloc(int node)
3563 {
3564 	struct page *page;
3565 	struct kmem_cache_node *n;
3566 
3567 	BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
3568 
3569 	page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
3570 
3571 	BUG_ON(!page);
3572 	if (page_to_nid(page) != node) {
3573 		pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3574 		pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
3575 	}
3576 
3577 	n = page->freelist;
3578 	BUG_ON(!n);
3579 #ifdef CONFIG_SLUB_DEBUG
3580 	init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
3581 	init_tracking(kmem_cache_node, n);
3582 #endif
3583 	n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false);
3584 	page->freelist = get_freepointer(kmem_cache_node, n);
3585 	page->inuse = 1;
3586 	page->frozen = 0;
3587 	kmem_cache_node->node[node] = n;
3588 	init_kmem_cache_node(n);
3589 	inc_slabs_node(kmem_cache_node, node, page->objects);
3590 
3591 	/*
3592 	 * No locks need to be taken here as it has just been
3593 	 * initialized and there is no concurrent access.
3594 	 */
3595 	__add_partial(n, page, DEACTIVATE_TO_HEAD);
3596 }
3597 
free_kmem_cache_nodes(struct kmem_cache * s)3598 static void free_kmem_cache_nodes(struct kmem_cache *s)
3599 {
3600 	int node;
3601 	struct kmem_cache_node *n;
3602 
3603 	for_each_kmem_cache_node(s, node, n) {
3604 		s->node[node] = NULL;
3605 		kmem_cache_free(kmem_cache_node, n);
3606 	}
3607 }
3608 
__kmem_cache_release(struct kmem_cache * s)3609 void __kmem_cache_release(struct kmem_cache *s)
3610 {
3611 	cache_random_seq_destroy(s);
3612 	free_percpu(s->cpu_slab);
3613 	free_kmem_cache_nodes(s);
3614 }
3615 
init_kmem_cache_nodes(struct kmem_cache * s)3616 static int init_kmem_cache_nodes(struct kmem_cache *s)
3617 {
3618 	int node;
3619 
3620 	for_each_node_mask(node, slab_nodes) {
3621 		struct kmem_cache_node *n;
3622 
3623 		if (slab_state == DOWN) {
3624 			early_kmem_cache_node_alloc(node);
3625 			continue;
3626 		}
3627 		n = kmem_cache_alloc_node(kmem_cache_node,
3628 						GFP_KERNEL, node);
3629 
3630 		if (!n) {
3631 			free_kmem_cache_nodes(s);
3632 			return 0;
3633 		}
3634 
3635 		init_kmem_cache_node(n);
3636 		s->node[node] = n;
3637 	}
3638 	return 1;
3639 }
3640 
set_min_partial(struct kmem_cache * s,unsigned long min)3641 static void set_min_partial(struct kmem_cache *s, unsigned long min)
3642 {
3643 	if (min < MIN_PARTIAL)
3644 		min = MIN_PARTIAL;
3645 	else if (min > MAX_PARTIAL)
3646 		min = MAX_PARTIAL;
3647 	s->min_partial = min;
3648 }
3649 
set_cpu_partial(struct kmem_cache * s)3650 static void set_cpu_partial(struct kmem_cache *s)
3651 {
3652 #ifdef CONFIG_SLUB_CPU_PARTIAL
3653 	/*
3654 	 * cpu_partial determined the maximum number of objects kept in the
3655 	 * per cpu partial lists of a processor.
3656 	 *
3657 	 * Per cpu partial lists mainly contain slabs that just have one
3658 	 * object freed. If they are used for allocation then they can be
3659 	 * filled up again with minimal effort. The slab will never hit the
3660 	 * per node partial lists and therefore no locking will be required.
3661 	 *
3662 	 * This setting also determines
3663 	 *
3664 	 * A) The number of objects from per cpu partial slabs dumped to the
3665 	 *    per node list when we reach the limit.
3666 	 * B) The number of objects in cpu partial slabs to extract from the
3667 	 *    per node list when we run out of per cpu objects. We only fetch
3668 	 *    50% to keep some capacity around for frees.
3669 	 */
3670 	if (!kmem_cache_has_cpu_partial(s))
3671 		slub_set_cpu_partial(s, 0);
3672 	else if (s->size >= PAGE_SIZE)
3673 		slub_set_cpu_partial(s, 2);
3674 	else if (s->size >= 1024)
3675 		slub_set_cpu_partial(s, 6);
3676 	else if (s->size >= 256)
3677 		slub_set_cpu_partial(s, 13);
3678 	else
3679 		slub_set_cpu_partial(s, 30);
3680 #endif
3681 }
3682 
3683 /*
3684  * calculate_sizes() determines the order and the distribution of data within
3685  * a slab object.
3686  */
calculate_sizes(struct kmem_cache * s,int forced_order)3687 static int calculate_sizes(struct kmem_cache *s, int forced_order)
3688 {
3689 	slab_flags_t flags = s->flags;
3690 	unsigned int size = s->object_size;
3691 	unsigned int freepointer_area;
3692 	unsigned int order;
3693 
3694 	/*
3695 	 * Round up object size to the next word boundary. We can only
3696 	 * place the free pointer at word boundaries and this determines
3697 	 * the possible location of the free pointer.
3698 	 */
3699 	size = ALIGN(size, sizeof(void *));
3700 	/*
3701 	 * This is the area of the object where a freepointer can be
3702 	 * safely written. If redzoning adds more to the inuse size, we
3703 	 * can't use that portion for writing the freepointer, so
3704 	 * s->offset must be limited within this for the general case.
3705 	 */
3706 	freepointer_area = size;
3707 
3708 #ifdef CONFIG_SLUB_DEBUG
3709 	/*
3710 	 * Determine if we can poison the object itself. If the user of
3711 	 * the slab may touch the object after free or before allocation
3712 	 * then we should never poison the object itself.
3713 	 */
3714 	if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
3715 			!s->ctor)
3716 		s->flags |= __OBJECT_POISON;
3717 	else
3718 		s->flags &= ~__OBJECT_POISON;
3719 
3720 
3721 	/*
3722 	 * If we are Redzoning then check if there is some space between the
3723 	 * end of the object and the free pointer. If not then add an
3724 	 * additional word to have some bytes to store Redzone information.
3725 	 */
3726 	if ((flags & SLAB_RED_ZONE) && size == s->object_size)
3727 		size += sizeof(void *);
3728 #endif
3729 
3730 	/*
3731 	 * With that we have determined the number of bytes in actual use
3732 	 * by the object. This is the potential offset to the free pointer.
3733 	 */
3734 	s->inuse = size;
3735 
3736 	if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
3737 		s->ctor)) {
3738 		/*
3739 		 * Relocate free pointer after the object if it is not
3740 		 * permitted to overwrite the first word of the object on
3741 		 * kmem_cache_free.
3742 		 *
3743 		 * This is the case if we do RCU, have a constructor or
3744 		 * destructor or are poisoning the objects.
3745 		 *
3746 		 * The assumption that s->offset >= s->inuse means free
3747 		 * pointer is outside of the object is used in the
3748 		 * freeptr_outside_object() function. If that is no
3749 		 * longer true, the function needs to be modified.
3750 		 */
3751 		s->offset = size;
3752 		size += sizeof(void *);
3753 	} else if (freepointer_area > sizeof(void *)) {
3754 		/*
3755 		 * Store freelist pointer near middle of object to keep
3756 		 * it away from the edges of the object to avoid small
3757 		 * sized over/underflows from neighboring allocations.
3758 		 */
3759 		s->offset = ALIGN(freepointer_area / 2, sizeof(void *));
3760 	}
3761 
3762 #ifdef CONFIG_SLUB_DEBUG
3763 	if (flags & SLAB_STORE_USER)
3764 		/*
3765 		 * Need to store information about allocs and frees after
3766 		 * the object.
3767 		 */
3768 		size += 2 * sizeof(struct track);
3769 #endif
3770 
3771 	kasan_cache_create(s, &size, &s->flags);
3772 #ifdef CONFIG_SLUB_DEBUG
3773 	if (flags & SLAB_RED_ZONE) {
3774 		/*
3775 		 * Add some empty padding so that we can catch
3776 		 * overwrites from earlier objects rather than let
3777 		 * tracking information or the free pointer be
3778 		 * corrupted if a user writes before the start
3779 		 * of the object.
3780 		 */
3781 		size += sizeof(void *);
3782 
3783 		s->red_left_pad = sizeof(void *);
3784 		s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3785 		size += s->red_left_pad;
3786 	}
3787 #endif
3788 
3789 	/*
3790 	 * SLUB stores one object immediately after another beginning from
3791 	 * offset 0. In order to align the objects we have to simply size
3792 	 * each object to conform to the alignment.
3793 	 */
3794 	size = ALIGN(size, s->align);
3795 	s->size = size;
3796 	s->reciprocal_size = reciprocal_value(size);
3797 	if (forced_order >= 0)
3798 		order = forced_order;
3799 	else
3800 		order = calculate_order(size);
3801 
3802 	if ((int)order < 0)
3803 		return 0;
3804 
3805 	s->allocflags = 0;
3806 	if (order)
3807 		s->allocflags |= __GFP_COMP;
3808 
3809 	if (s->flags & SLAB_CACHE_DMA)
3810 		s->allocflags |= GFP_DMA;
3811 
3812 	if (s->flags & SLAB_CACHE_DMA32)
3813 		s->allocflags |= GFP_DMA32;
3814 
3815 	if (s->flags & SLAB_RECLAIM_ACCOUNT)
3816 		s->allocflags |= __GFP_RECLAIMABLE;
3817 
3818 	/*
3819 	 * Determine the number of objects per slab
3820 	 */
3821 	s->oo = oo_make(order, size);
3822 	s->min = oo_make(get_order(size), size);
3823 	if (oo_objects(s->oo) > oo_objects(s->max))
3824 		s->max = s->oo;
3825 
3826 	return !!oo_objects(s->oo);
3827 }
3828 
kmem_cache_open(struct kmem_cache * s,slab_flags_t flags)3829 static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
3830 {
3831 	s->flags = kmem_cache_flags(s->size, flags, s->name);
3832 #ifdef CONFIG_SLAB_FREELIST_HARDENED
3833 	s->random = get_random_long();
3834 #endif
3835 
3836 	if (!calculate_sizes(s, -1))
3837 		goto error;
3838 	if (disable_higher_order_debug) {
3839 		/*
3840 		 * Disable debugging flags that store metadata if the min slab
3841 		 * order increased.
3842 		 */
3843 		if (get_order(s->size) > get_order(s->object_size)) {
3844 			s->flags &= ~DEBUG_METADATA_FLAGS;
3845 			s->offset = 0;
3846 			if (!calculate_sizes(s, -1))
3847 				goto error;
3848 		}
3849 	}
3850 
3851 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3852     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3853 	if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
3854 		/* Enable fast mode */
3855 		s->flags |= __CMPXCHG_DOUBLE;
3856 #endif
3857 
3858 	/*
3859 	 * The larger the object size is, the more pages we want on the partial
3860 	 * list to avoid pounding the page allocator excessively.
3861 	 */
3862 	set_min_partial(s, ilog2(s->size) / 2);
3863 
3864 	set_cpu_partial(s);
3865 
3866 #ifdef CONFIG_NUMA
3867 	s->remote_node_defrag_ratio = 1000;
3868 #endif
3869 
3870 	/* Initialize the pre-computed randomized freelist if slab is up */
3871 	if (slab_state >= UP) {
3872 		if (init_cache_random_seq(s))
3873 			goto error;
3874 	}
3875 
3876 	if (!init_kmem_cache_nodes(s))
3877 		goto error;
3878 
3879 	if (alloc_kmem_cache_cpus(s))
3880 		return 0;
3881 
3882 	free_kmem_cache_nodes(s);
3883 error:
3884 	return -EINVAL;
3885 }
3886 
list_slab_objects(struct kmem_cache * s,struct page * page,const char * text)3887 static void list_slab_objects(struct kmem_cache *s, struct page *page,
3888 			      const char *text)
3889 {
3890 #ifdef CONFIG_SLUB_DEBUG
3891 	void *addr = page_address(page);
3892 	unsigned long *map;
3893 	void *p;
3894 
3895 	slab_err(s, page, text, s->name);
3896 	slab_lock(page);
3897 
3898 	map = get_map(s, page);
3899 	for_each_object(p, s, addr, page->objects) {
3900 
3901 		if (!test_bit(__obj_to_index(s, addr, p), map)) {
3902 			pr_err("Object 0x%p @offset=%tu\n", p, p - addr);
3903 			print_tracking(s, p);
3904 		}
3905 	}
3906 	put_map(map);
3907 	slab_unlock(page);
3908 #endif
3909 }
3910 
3911 /*
3912  * Attempt to free all partial slabs on a node.
3913  * This is called from __kmem_cache_shutdown(). We must take list_lock
3914  * because sysfs file might still access partial list after the shutdowning.
3915  */
free_partial(struct kmem_cache * s,struct kmem_cache_node * n)3916 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
3917 {
3918 	LIST_HEAD(discard);
3919 	struct page *page, *h;
3920 
3921 	BUG_ON(irqs_disabled());
3922 	spin_lock_irq(&n->list_lock);
3923 	list_for_each_entry_safe(page, h, &n->partial, slab_list) {
3924 		if (!page->inuse) {
3925 			remove_partial(n, page);
3926 			list_add(&page->slab_list, &discard);
3927 		} else {
3928 			list_slab_objects(s, page,
3929 			  "Objects remaining in %s on __kmem_cache_shutdown()");
3930 		}
3931 	}
3932 	spin_unlock_irq(&n->list_lock);
3933 
3934 	list_for_each_entry_safe(page, h, &discard, slab_list)
3935 		discard_slab(s, page);
3936 }
3937 
__kmem_cache_empty(struct kmem_cache * s)3938 bool __kmem_cache_empty(struct kmem_cache *s)
3939 {
3940 	int node;
3941 	struct kmem_cache_node *n;
3942 
3943 	for_each_kmem_cache_node(s, node, n)
3944 		if (n->nr_partial || slabs_node(s, node))
3945 			return false;
3946 	return true;
3947 }
3948 
3949 /*
3950  * Release all resources used by a slab cache.
3951  */
__kmem_cache_shutdown(struct kmem_cache * s)3952 int __kmem_cache_shutdown(struct kmem_cache *s)
3953 {
3954 	int node;
3955 	struct kmem_cache_node *n;
3956 
3957 	flush_all(s);
3958 	/* Attempt to free all objects */
3959 	for_each_kmem_cache_node(s, node, n) {
3960 		free_partial(s, n);
3961 		if (n->nr_partial || slabs_node(s, node))
3962 			return 1;
3963 	}
3964 	return 0;
3965 }
3966 
3967 #ifdef CONFIG_PRINTK
kmem_obj_info(struct kmem_obj_info * kpp,void * object,struct page * page)3968 void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct page *page)
3969 {
3970 	void *base;
3971 	int __maybe_unused i;
3972 	unsigned int objnr;
3973 	void *objp;
3974 	void *objp0;
3975 	struct kmem_cache *s = page->slab_cache;
3976 	struct track __maybe_unused *trackp;
3977 
3978 	kpp->kp_ptr = object;
3979 	kpp->kp_page = page;
3980 	kpp->kp_slab_cache = s;
3981 	base = page_address(page);
3982 	objp0 = kasan_reset_tag(object);
3983 #ifdef CONFIG_SLUB_DEBUG
3984 	objp = restore_red_left(s, objp0);
3985 #else
3986 	objp = objp0;
3987 #endif
3988 	objnr = obj_to_index(s, page, objp);
3989 	kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp);
3990 	objp = base + s->size * objnr;
3991 	kpp->kp_objp = objp;
3992 	if (WARN_ON_ONCE(objp < base || objp >= base + page->objects * s->size || (objp - base) % s->size) ||
3993 	    !(s->flags & SLAB_STORE_USER))
3994 		return;
3995 #ifdef CONFIG_SLUB_DEBUG
3996 	trackp = get_track(s, objp, TRACK_ALLOC);
3997 	kpp->kp_ret = (void *)trackp->addr;
3998 #ifdef CONFIG_STACKTRACE
3999 	for (i = 0; i < KS_ADDRS_COUNT && i < TRACK_ADDRS_COUNT; i++) {
4000 		kpp->kp_stack[i] = (void *)trackp->addrs[i];
4001 		if (!kpp->kp_stack[i])
4002 			break;
4003 	}
4004 #endif
4005 #endif
4006 }
4007 #endif
4008 
4009 /********************************************************************
4010  *		Kmalloc subsystem
4011  *******************************************************************/
4012 
setup_slub_min_order(char * str)4013 static int __init setup_slub_min_order(char *str)
4014 {
4015 	get_option(&str, (int *)&slub_min_order);
4016 
4017 	return 1;
4018 }
4019 
4020 __setup("slub_min_order=", setup_slub_min_order);
4021 
setup_slub_max_order(char * str)4022 static int __init setup_slub_max_order(char *str)
4023 {
4024 	get_option(&str, (int *)&slub_max_order);
4025 	slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1);
4026 
4027 	return 1;
4028 }
4029 
4030 __setup("slub_max_order=", setup_slub_max_order);
4031 
setup_slub_min_objects(char * str)4032 static int __init setup_slub_min_objects(char *str)
4033 {
4034 	get_option(&str, (int *)&slub_min_objects);
4035 
4036 	return 1;
4037 }
4038 
4039 __setup("slub_min_objects=", setup_slub_min_objects);
4040 
__kmalloc(size_t size,gfp_t flags)4041 void *__kmalloc(size_t size, gfp_t flags)
4042 {
4043 	struct kmem_cache *s;
4044 	void *ret;
4045 
4046 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
4047 		return kmalloc_large(size, flags);
4048 
4049 	s = kmalloc_slab(size, flags);
4050 
4051 	if (unlikely(ZERO_OR_NULL_PTR(s)))
4052 		return s;
4053 
4054 	ret = slab_alloc(s, flags, _RET_IP_, size);
4055 
4056 	trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
4057 
4058 	ret = kasan_kmalloc(s, ret, size, flags);
4059 
4060 	return ret;
4061 }
4062 EXPORT_SYMBOL(__kmalloc);
4063 
4064 #ifdef CONFIG_NUMA
kmalloc_large_node(size_t size,gfp_t flags,int node)4065 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
4066 {
4067 	struct page *page;
4068 	void *ptr = NULL;
4069 	unsigned int order = get_order(size);
4070 
4071 	flags |= __GFP_COMP;
4072 	page = alloc_pages_node(node, flags, order);
4073 	if (page) {
4074 		ptr = page_address(page);
4075 		mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
4076 				      PAGE_SIZE << order);
4077 	}
4078 
4079 	return kmalloc_large_node_hook(ptr, size, flags);
4080 }
4081 
__kmalloc_node(size_t size,gfp_t flags,int node)4082 void *__kmalloc_node(size_t size, gfp_t flags, int node)
4083 {
4084 	struct kmem_cache *s;
4085 	void *ret;
4086 
4087 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
4088 		ret = kmalloc_large_node(size, flags, node);
4089 
4090 		trace_kmalloc_node(_RET_IP_, ret,
4091 				   size, PAGE_SIZE << get_order(size),
4092 				   flags, node);
4093 
4094 		return ret;
4095 	}
4096 
4097 	s = kmalloc_slab(size, flags);
4098 
4099 	if (unlikely(ZERO_OR_NULL_PTR(s)))
4100 		return s;
4101 
4102 	ret = slab_alloc_node(s, flags, node, _RET_IP_, size);
4103 
4104 	trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
4105 
4106 	ret = kasan_kmalloc(s, ret, size, flags);
4107 
4108 	return ret;
4109 }
4110 EXPORT_SYMBOL(__kmalloc_node);
4111 #endif	/* CONFIG_NUMA */
4112 
4113 #ifdef CONFIG_HARDENED_USERCOPY
4114 /*
4115  * Rejects incorrectly sized objects and objects that are to be copied
4116  * to/from userspace but do not fall entirely within the containing slab
4117  * cache's usercopy region.
4118  *
4119  * Returns NULL if check passes, otherwise const char * to name of cache
4120  * to indicate an error.
4121  */
__check_heap_object(const void * ptr,unsigned long n,struct page * page,bool to_user)4122 void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
4123 			 bool to_user)
4124 {
4125 	struct kmem_cache *s;
4126 	unsigned int offset;
4127 	size_t object_size;
4128 	bool is_kfence = is_kfence_address(ptr);
4129 
4130 	ptr = kasan_reset_tag(ptr);
4131 
4132 	/* Find object and usable object size. */
4133 	s = page->slab_cache;
4134 
4135 	/* Reject impossible pointers. */
4136 	if (ptr < page_address(page))
4137 		usercopy_abort("SLUB object not in SLUB page?!", NULL,
4138 			       to_user, 0, n);
4139 
4140 	/* Find offset within object. */
4141 	if (is_kfence)
4142 		offset = ptr - kfence_object_start(ptr);
4143 	else
4144 		offset = (ptr - page_address(page)) % s->size;
4145 
4146 	/* Adjust for redzone and reject if within the redzone. */
4147 	if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) {
4148 		if (offset < s->red_left_pad)
4149 			usercopy_abort("SLUB object in left red zone",
4150 				       s->name, to_user, offset, n);
4151 		offset -= s->red_left_pad;
4152 	}
4153 
4154 	/* Allow address range falling entirely within usercopy region. */
4155 	if (offset >= s->useroffset &&
4156 	    offset - s->useroffset <= s->usersize &&
4157 	    n <= s->useroffset - offset + s->usersize)
4158 		return;
4159 
4160 	/*
4161 	 * If the copy is still within the allocated object, produce
4162 	 * a warning instead of rejecting the copy. This is intended
4163 	 * to be a temporary method to find any missing usercopy
4164 	 * whitelists.
4165 	 */
4166 	object_size = slab_ksize(s);
4167 	if (usercopy_fallback &&
4168 	    offset <= object_size && n <= object_size - offset) {
4169 		usercopy_warn("SLUB object", s->name, to_user, offset, n);
4170 		return;
4171 	}
4172 
4173 	usercopy_abort("SLUB object", s->name, to_user, offset, n);
4174 }
4175 #endif /* CONFIG_HARDENED_USERCOPY */
4176 
__ksize(const void * object)4177 size_t __ksize(const void *object)
4178 {
4179 	struct page *page;
4180 
4181 	if (unlikely(object == ZERO_SIZE_PTR))
4182 		return 0;
4183 
4184 	page = virt_to_head_page(object);
4185 
4186 	if (unlikely(!PageSlab(page))) {
4187 		WARN_ON(!PageCompound(page));
4188 		return page_size(page);
4189 	}
4190 
4191 	return slab_ksize(page->slab_cache);
4192 }
4193 EXPORT_SYMBOL(__ksize);
4194 
kfree(const void * x)4195 void kfree(const void *x)
4196 {
4197 	struct page *page;
4198 	void *object = (void *)x;
4199 
4200 	trace_kfree(_RET_IP_, x);
4201 
4202 	if (unlikely(ZERO_OR_NULL_PTR(x)))
4203 		return;
4204 
4205 	page = virt_to_head_page(x);
4206 	if (unlikely(!PageSlab(page))) {
4207 		unsigned int order = compound_order(page);
4208 
4209 		BUG_ON(!PageCompound(page));
4210 		kfree_hook(object);
4211 		mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
4212 				      -(PAGE_SIZE << order));
4213 		__free_pages(page, order);
4214 		return;
4215 	}
4216 	slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
4217 }
4218 EXPORT_SYMBOL(kfree);
4219 
4220 #define SHRINK_PROMOTE_MAX 32
4221 
4222 /*
4223  * kmem_cache_shrink discards empty slabs and promotes the slabs filled
4224  * up most to the head of the partial lists. New allocations will then
4225  * fill those up and thus they can be removed from the partial lists.
4226  *
4227  * The slabs with the least items are placed last. This results in them
4228  * being allocated from last increasing the chance that the last objects
4229  * are freed in them.
4230  */
__kmem_cache_shrink(struct kmem_cache * s)4231 int __kmem_cache_shrink(struct kmem_cache *s)
4232 {
4233 	int node;
4234 	int i;
4235 	struct kmem_cache_node *n;
4236 	struct page *page;
4237 	struct page *t;
4238 	struct list_head discard;
4239 	struct list_head promote[SHRINK_PROMOTE_MAX];
4240 	unsigned long flags;
4241 	int ret = 0;
4242 
4243 	flush_all(s);
4244 	for_each_kmem_cache_node(s, node, n) {
4245 		INIT_LIST_HEAD(&discard);
4246 		for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
4247 			INIT_LIST_HEAD(promote + i);
4248 
4249 		spin_lock_irqsave(&n->list_lock, flags);
4250 
4251 		/*
4252 		 * Build lists of slabs to discard or promote.
4253 		 *
4254 		 * Note that concurrent frees may occur while we hold the
4255 		 * list_lock. page->inuse here is the upper limit.
4256 		 */
4257 		list_for_each_entry_safe(page, t, &n->partial, slab_list) {
4258 			int free = page->objects - page->inuse;
4259 
4260 			/* Do not reread page->inuse */
4261 			barrier();
4262 
4263 			/* We do not keep full slabs on the list */
4264 			BUG_ON(free <= 0);
4265 
4266 			if (free == page->objects) {
4267 				list_move(&page->slab_list, &discard);
4268 				n->nr_partial--;
4269 			} else if (free <= SHRINK_PROMOTE_MAX)
4270 				list_move(&page->slab_list, promote + free - 1);
4271 		}
4272 
4273 		/*
4274 		 * Promote the slabs filled up most to the head of the
4275 		 * partial list.
4276 		 */
4277 		for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
4278 			list_splice(promote + i, &n->partial);
4279 
4280 		spin_unlock_irqrestore(&n->list_lock, flags);
4281 
4282 		/* Release empty slabs */
4283 		list_for_each_entry_safe(page, t, &discard, slab_list)
4284 			discard_slab(s, page);
4285 
4286 		if (slabs_node(s, node))
4287 			ret = 1;
4288 	}
4289 
4290 	return ret;
4291 }
4292 
slab_mem_going_offline_callback(void * arg)4293 static int slab_mem_going_offline_callback(void *arg)
4294 {
4295 	struct kmem_cache *s;
4296 
4297 	mutex_lock(&slab_mutex);
4298 	list_for_each_entry(s, &slab_caches, list)
4299 		__kmem_cache_shrink(s);
4300 	mutex_unlock(&slab_mutex);
4301 
4302 	return 0;
4303 }
4304 
slab_mem_offline_callback(void * arg)4305 static void slab_mem_offline_callback(void *arg)
4306 {
4307 	struct memory_notify *marg = arg;
4308 	int offline_node;
4309 
4310 	offline_node = marg->status_change_nid_normal;
4311 
4312 	/*
4313 	 * If the node still has available memory. we need kmem_cache_node
4314 	 * for it yet.
4315 	 */
4316 	if (offline_node < 0)
4317 		return;
4318 
4319 	mutex_lock(&slab_mutex);
4320 	node_clear(offline_node, slab_nodes);
4321 	/*
4322 	 * We no longer free kmem_cache_node structures here, as it would be
4323 	 * racy with all get_node() users, and infeasible to protect them with
4324 	 * slab_mutex.
4325 	 */
4326 	mutex_unlock(&slab_mutex);
4327 }
4328 
slab_mem_going_online_callback(void * arg)4329 static int slab_mem_going_online_callback(void *arg)
4330 {
4331 	struct kmem_cache_node *n;
4332 	struct kmem_cache *s;
4333 	struct memory_notify *marg = arg;
4334 	int nid = marg->status_change_nid_normal;
4335 	int ret = 0;
4336 
4337 	/*
4338 	 * If the node's memory is already available, then kmem_cache_node is
4339 	 * already created. Nothing to do.
4340 	 */
4341 	if (nid < 0)
4342 		return 0;
4343 
4344 	/*
4345 	 * We are bringing a node online. No memory is available yet. We must
4346 	 * allocate a kmem_cache_node structure in order to bring the node
4347 	 * online.
4348 	 */
4349 	mutex_lock(&slab_mutex);
4350 	list_for_each_entry(s, &slab_caches, list) {
4351 		/*
4352 		 * The structure may already exist if the node was previously
4353 		 * onlined and offlined.
4354 		 */
4355 		if (get_node(s, nid))
4356 			continue;
4357 		/*
4358 		 * XXX: kmem_cache_alloc_node will fallback to other nodes
4359 		 *      since memory is not yet available from the node that
4360 		 *      is brought up.
4361 		 */
4362 		n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
4363 		if (!n) {
4364 			ret = -ENOMEM;
4365 			goto out;
4366 		}
4367 		init_kmem_cache_node(n);
4368 		s->node[nid] = n;
4369 	}
4370 	/*
4371 	 * Any cache created after this point will also have kmem_cache_node
4372 	 * initialized for the new node.
4373 	 */
4374 	node_set(nid, slab_nodes);
4375 out:
4376 	mutex_unlock(&slab_mutex);
4377 	return ret;
4378 }
4379 
slab_memory_callback(struct notifier_block * self,unsigned long action,void * arg)4380 static int slab_memory_callback(struct notifier_block *self,
4381 				unsigned long action, void *arg)
4382 {
4383 	int ret = 0;
4384 
4385 	switch (action) {
4386 	case MEM_GOING_ONLINE:
4387 		ret = slab_mem_going_online_callback(arg);
4388 		break;
4389 	case MEM_GOING_OFFLINE:
4390 		ret = slab_mem_going_offline_callback(arg);
4391 		break;
4392 	case MEM_OFFLINE:
4393 	case MEM_CANCEL_ONLINE:
4394 		slab_mem_offline_callback(arg);
4395 		break;
4396 	case MEM_ONLINE:
4397 	case MEM_CANCEL_OFFLINE:
4398 		break;
4399 	}
4400 	if (ret)
4401 		ret = notifier_from_errno(ret);
4402 	else
4403 		ret = NOTIFY_OK;
4404 	return ret;
4405 }
4406 
4407 static struct notifier_block slab_memory_callback_nb = {
4408 	.notifier_call = slab_memory_callback,
4409 	.priority = SLAB_CALLBACK_PRI,
4410 };
4411 
4412 /********************************************************************
4413  *			Basic setup of slabs
4414  *******************************************************************/
4415 
4416 /*
4417  * Used for early kmem_cache structures that were allocated using
4418  * the page allocator. Allocate them properly then fix up the pointers
4419  * that may be pointing to the wrong kmem_cache structure.
4420  */
4421 
bootstrap(struct kmem_cache * static_cache)4422 static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
4423 {
4424 	int node;
4425 	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
4426 	struct kmem_cache_node *n;
4427 
4428 	memcpy(s, static_cache, kmem_cache->object_size);
4429 
4430 	/*
4431 	 * This runs very early, and only the boot processor is supposed to be
4432 	 * up.  Even if it weren't true, IRQs are not up so we couldn't fire
4433 	 * IPIs around.
4434 	 */
4435 	__flush_cpu_slab(s, smp_processor_id());
4436 	for_each_kmem_cache_node(s, node, n) {
4437 		struct page *p;
4438 
4439 		list_for_each_entry(p, &n->partial, slab_list)
4440 			p->slab_cache = s;
4441 
4442 #ifdef CONFIG_SLUB_DEBUG
4443 		list_for_each_entry(p, &n->full, slab_list)
4444 			p->slab_cache = s;
4445 #endif
4446 	}
4447 	list_add(&s->list, &slab_caches);
4448 	return s;
4449 }
4450 
kmem_cache_init(void)4451 void __init kmem_cache_init(void)
4452 {
4453 	static __initdata struct kmem_cache boot_kmem_cache,
4454 		boot_kmem_cache_node;
4455 	int node;
4456 
4457 	if (debug_guardpage_minorder())
4458 		slub_max_order = 0;
4459 
4460 	kmem_cache_node = &boot_kmem_cache_node;
4461 	kmem_cache = &boot_kmem_cache;
4462 
4463 	/*
4464 	 * Initialize the nodemask for which we will allocate per node
4465 	 * structures. Here we don't need taking slab_mutex yet.
4466 	 */
4467 	for_each_node_state(node, N_NORMAL_MEMORY)
4468 		node_set(node, slab_nodes);
4469 
4470 	create_boot_cache(kmem_cache_node, "kmem_cache_node",
4471 		sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
4472 
4473 	register_hotmemory_notifier(&slab_memory_callback_nb);
4474 
4475 	/* Able to allocate the per node structures */
4476 	slab_state = PARTIAL;
4477 
4478 	create_boot_cache(kmem_cache, "kmem_cache",
4479 			offsetof(struct kmem_cache, node) +
4480 				nr_node_ids * sizeof(struct kmem_cache_node *),
4481 		       SLAB_HWCACHE_ALIGN, 0, 0);
4482 
4483 	kmem_cache = bootstrap(&boot_kmem_cache);
4484 	kmem_cache_node = bootstrap(&boot_kmem_cache_node);
4485 
4486 	/* Now we can use the kmem_cache to allocate kmalloc slabs */
4487 	setup_kmalloc_cache_index_table();
4488 	create_kmalloc_caches(0);
4489 
4490 	/* Setup random freelists for each cache */
4491 	init_freelist_randomization();
4492 
4493 	cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4494 				  slub_cpu_dead);
4495 
4496 	pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n",
4497 		cache_line_size(),
4498 		slub_min_order, slub_max_order, slub_min_objects,
4499 		nr_cpu_ids, nr_node_ids);
4500 }
4501 
kmem_cache_init_late(void)4502 void __init kmem_cache_init_late(void)
4503 {
4504 }
4505 
4506 struct kmem_cache *
__kmem_cache_alias(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,void (* ctor)(void *))4507 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
4508 		   slab_flags_t flags, void (*ctor)(void *))
4509 {
4510 	struct kmem_cache *s;
4511 
4512 	s = find_mergeable(size, align, flags, name, ctor);
4513 	if (s) {
4514 		s->refcount++;
4515 
4516 		/*
4517 		 * Adjust the object sizes so that we clear
4518 		 * the complete object on kzalloc.
4519 		 */
4520 		s->object_size = max(s->object_size, size);
4521 		s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
4522 
4523 		if (sysfs_slab_alias(s, name)) {
4524 			s->refcount--;
4525 			s = NULL;
4526 		}
4527 	}
4528 
4529 	return s;
4530 }
4531 
__kmem_cache_create(struct kmem_cache * s,slab_flags_t flags)4532 int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
4533 {
4534 	int err;
4535 
4536 	err = kmem_cache_open(s, flags);
4537 	if (err)
4538 		return err;
4539 
4540 	/* Mutex is not taken during early boot */
4541 	if (slab_state <= UP)
4542 		return 0;
4543 
4544 	err = sysfs_slab_add(s);
4545 	if (err)
4546 		__kmem_cache_release(s);
4547 
4548 	return err;
4549 }
4550 
__kmalloc_track_caller(size_t size,gfp_t gfpflags,unsigned long caller)4551 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
4552 {
4553 	struct kmem_cache *s;
4554 	void *ret;
4555 
4556 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
4557 		return kmalloc_large(size, gfpflags);
4558 
4559 	s = kmalloc_slab(size, gfpflags);
4560 
4561 	if (unlikely(ZERO_OR_NULL_PTR(s)))
4562 		return s;
4563 
4564 	ret = slab_alloc(s, gfpflags, caller, size);
4565 
4566 	/* Honor the call site pointer we received. */
4567 	trace_kmalloc(caller, ret, size, s->size, gfpflags);
4568 
4569 	return ret;
4570 }
4571 EXPORT_SYMBOL(__kmalloc_track_caller);
4572 
4573 #ifdef CONFIG_NUMA
__kmalloc_node_track_caller(size_t size,gfp_t gfpflags,int node,unsigned long caller)4574 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
4575 					int node, unsigned long caller)
4576 {
4577 	struct kmem_cache *s;
4578 	void *ret;
4579 
4580 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
4581 		ret = kmalloc_large_node(size, gfpflags, node);
4582 
4583 		trace_kmalloc_node(caller, ret,
4584 				   size, PAGE_SIZE << get_order(size),
4585 				   gfpflags, node);
4586 
4587 		return ret;
4588 	}
4589 
4590 	s = kmalloc_slab(size, gfpflags);
4591 
4592 	if (unlikely(ZERO_OR_NULL_PTR(s)))
4593 		return s;
4594 
4595 	ret = slab_alloc_node(s, gfpflags, node, caller, size);
4596 
4597 	/* Honor the call site pointer we received. */
4598 	trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
4599 
4600 	return ret;
4601 }
4602 EXPORT_SYMBOL(__kmalloc_node_track_caller);
4603 #endif
4604 
4605 #ifdef CONFIG_SYSFS
count_inuse(struct page * page)4606 static int count_inuse(struct page *page)
4607 {
4608 	return page->inuse;
4609 }
4610 
count_total(struct page * page)4611 static int count_total(struct page *page)
4612 {
4613 	return page->objects;
4614 }
4615 #endif
4616 
4617 #ifdef CONFIG_SLUB_DEBUG
validate_slab(struct kmem_cache * s,struct page * page)4618 static void validate_slab(struct kmem_cache *s, struct page *page)
4619 {
4620 	void *p;
4621 	void *addr = page_address(page);
4622 	unsigned long *map;
4623 
4624 	slab_lock(page);
4625 
4626 	if (!check_slab(s, page) || !on_freelist(s, page, NULL))
4627 		goto unlock;
4628 
4629 	/* Now we know that a valid freelist exists */
4630 	map = get_map(s, page);
4631 	for_each_object(p, s, addr, page->objects) {
4632 		u8 val = test_bit(__obj_to_index(s, addr, p), map) ?
4633 			 SLUB_RED_INACTIVE : SLUB_RED_ACTIVE;
4634 
4635 		if (!check_object(s, page, p, val))
4636 			break;
4637 	}
4638 	put_map(map);
4639 unlock:
4640 	slab_unlock(page);
4641 }
4642 
validate_slab_node(struct kmem_cache * s,struct kmem_cache_node * n)4643 static int validate_slab_node(struct kmem_cache *s,
4644 		struct kmem_cache_node *n)
4645 {
4646 	unsigned long count = 0;
4647 	struct page *page;
4648 	unsigned long flags;
4649 
4650 	spin_lock_irqsave(&n->list_lock, flags);
4651 
4652 	list_for_each_entry(page, &n->partial, slab_list) {
4653 		validate_slab(s, page);
4654 		count++;
4655 	}
4656 	if (count != n->nr_partial)
4657 		pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4658 		       s->name, count, n->nr_partial);
4659 
4660 	if (!(s->flags & SLAB_STORE_USER))
4661 		goto out;
4662 
4663 	list_for_each_entry(page, &n->full, slab_list) {
4664 		validate_slab(s, page);
4665 		count++;
4666 	}
4667 	if (count != atomic_long_read(&n->nr_slabs))
4668 		pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4669 		       s->name, count, atomic_long_read(&n->nr_slabs));
4670 
4671 out:
4672 	spin_unlock_irqrestore(&n->list_lock, flags);
4673 	return count;
4674 }
4675 
validate_slab_cache(struct kmem_cache * s)4676 static long validate_slab_cache(struct kmem_cache *s)
4677 {
4678 	int node;
4679 	unsigned long count = 0;
4680 	struct kmem_cache_node *n;
4681 
4682 	flush_all(s);
4683 	for_each_kmem_cache_node(s, node, n)
4684 		count += validate_slab_node(s, n);
4685 
4686 	return count;
4687 }
4688 /*
4689  * Generate lists of code addresses where slabcache objects are allocated
4690  * and freed.
4691  */
4692 
4693 struct location {
4694 	unsigned long count;
4695 	unsigned long addr;
4696 	long long sum_time;
4697 	long min_time;
4698 	long max_time;
4699 	long min_pid;
4700 	long max_pid;
4701 	DECLARE_BITMAP(cpus, NR_CPUS);
4702 	nodemask_t nodes;
4703 };
4704 
4705 struct loc_track {
4706 	unsigned long max;
4707 	unsigned long count;
4708 	struct location *loc;
4709 };
4710 
free_loc_track(struct loc_track * t)4711 static void free_loc_track(struct loc_track *t)
4712 {
4713 	if (t->max)
4714 		free_pages((unsigned long)t->loc,
4715 			get_order(sizeof(struct location) * t->max));
4716 }
4717 
alloc_loc_track(struct loc_track * t,unsigned long max,gfp_t flags)4718 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
4719 {
4720 	struct location *l;
4721 	int order;
4722 
4723 	order = get_order(sizeof(struct location) * max);
4724 
4725 	l = (void *)__get_free_pages(flags, order);
4726 	if (!l)
4727 		return 0;
4728 
4729 	if (t->count) {
4730 		memcpy(l, t->loc, sizeof(struct location) * t->count);
4731 		free_loc_track(t);
4732 	}
4733 	t->max = max;
4734 	t->loc = l;
4735 	return 1;
4736 }
4737 
add_location(struct loc_track * t,struct kmem_cache * s,const struct track * track)4738 static int add_location(struct loc_track *t, struct kmem_cache *s,
4739 				const struct track *track)
4740 {
4741 	long start, end, pos;
4742 	struct location *l;
4743 	unsigned long caddr;
4744 	unsigned long age = jiffies - track->when;
4745 
4746 	start = -1;
4747 	end = t->count;
4748 
4749 	for ( ; ; ) {
4750 		pos = start + (end - start + 1) / 2;
4751 
4752 		/*
4753 		 * There is nothing at "end". If we end up there
4754 		 * we need to add something to before end.
4755 		 */
4756 		if (pos == end)
4757 			break;
4758 
4759 		caddr = t->loc[pos].addr;
4760 		if (track->addr == caddr) {
4761 
4762 			l = &t->loc[pos];
4763 			l->count++;
4764 			if (track->when) {
4765 				l->sum_time += age;
4766 				if (age < l->min_time)
4767 					l->min_time = age;
4768 				if (age > l->max_time)
4769 					l->max_time = age;
4770 
4771 				if (track->pid < l->min_pid)
4772 					l->min_pid = track->pid;
4773 				if (track->pid > l->max_pid)
4774 					l->max_pid = track->pid;
4775 
4776 				cpumask_set_cpu(track->cpu,
4777 						to_cpumask(l->cpus));
4778 			}
4779 			node_set(page_to_nid(virt_to_page(track)), l->nodes);
4780 			return 1;
4781 		}
4782 
4783 		if (track->addr < caddr)
4784 			end = pos;
4785 		else
4786 			start = pos;
4787 	}
4788 
4789 	/*
4790 	 * Not found. Insert new tracking element.
4791 	 */
4792 	if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
4793 		return 0;
4794 
4795 	l = t->loc + pos;
4796 	if (pos < t->count)
4797 		memmove(l + 1, l,
4798 			(t->count - pos) * sizeof(struct location));
4799 	t->count++;
4800 	l->count = 1;
4801 	l->addr = track->addr;
4802 	l->sum_time = age;
4803 	l->min_time = age;
4804 	l->max_time = age;
4805 	l->min_pid = track->pid;
4806 	l->max_pid = track->pid;
4807 	cpumask_clear(to_cpumask(l->cpus));
4808 	cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
4809 	nodes_clear(l->nodes);
4810 	node_set(page_to_nid(virt_to_page(track)), l->nodes);
4811 	return 1;
4812 }
4813 
process_slab(struct loc_track * t,struct kmem_cache * s,struct page * page,enum track_item alloc)4814 static void process_slab(struct loc_track *t, struct kmem_cache *s,
4815 		struct page *page, enum track_item alloc)
4816 {
4817 	void *addr = page_address(page);
4818 	void *p;
4819 	unsigned long *map;
4820 
4821 	map = get_map(s, page);
4822 	for_each_object(p, s, addr, page->objects)
4823 		if (!test_bit(__obj_to_index(s, addr, p), map))
4824 			add_location(t, s, get_track(s, p, alloc));
4825 	put_map(map);
4826 }
4827 
list_locations(struct kmem_cache * s,char * buf,enum track_item alloc)4828 static int list_locations(struct kmem_cache *s, char *buf,
4829 			  enum track_item alloc)
4830 {
4831 	int len = 0;
4832 	unsigned long i;
4833 	struct loc_track t = { 0, 0, NULL };
4834 	int node;
4835 	struct kmem_cache_node *n;
4836 
4837 	if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4838 			     GFP_KERNEL)) {
4839 		return sysfs_emit(buf, "Out of memory\n");
4840 	}
4841 	/* Push back cpu slabs */
4842 	flush_all(s);
4843 
4844 	for_each_kmem_cache_node(s, node, n) {
4845 		unsigned long flags;
4846 		struct page *page;
4847 
4848 		if (!atomic_long_read(&n->nr_slabs))
4849 			continue;
4850 
4851 		spin_lock_irqsave(&n->list_lock, flags);
4852 		list_for_each_entry(page, &n->partial, slab_list)
4853 			process_slab(&t, s, page, alloc);
4854 		list_for_each_entry(page, &n->full, slab_list)
4855 			process_slab(&t, s, page, alloc);
4856 		spin_unlock_irqrestore(&n->list_lock, flags);
4857 	}
4858 
4859 	for (i = 0; i < t.count; i++) {
4860 		struct location *l = &t.loc[i];
4861 
4862 		len += sysfs_emit_at(buf, len, "%7ld ", l->count);
4863 
4864 		if (l->addr)
4865 			len += sysfs_emit_at(buf, len, "%pS", (void *)l->addr);
4866 		else
4867 			len += sysfs_emit_at(buf, len, "<not-available>");
4868 
4869 		if (l->sum_time != l->min_time)
4870 			len += sysfs_emit_at(buf, len, " age=%ld/%ld/%ld",
4871 					     l->min_time,
4872 					     (long)div_u64(l->sum_time,
4873 							   l->count),
4874 					     l->max_time);
4875 		else
4876 			len += sysfs_emit_at(buf, len, " age=%ld", l->min_time);
4877 
4878 		if (l->min_pid != l->max_pid)
4879 			len += sysfs_emit_at(buf, len, " pid=%ld-%ld",
4880 					     l->min_pid, l->max_pid);
4881 		else
4882 			len += sysfs_emit_at(buf, len, " pid=%ld",
4883 					     l->min_pid);
4884 
4885 		if (num_online_cpus() > 1 &&
4886 		    !cpumask_empty(to_cpumask(l->cpus)))
4887 			len += sysfs_emit_at(buf, len, " cpus=%*pbl",
4888 					     cpumask_pr_args(to_cpumask(l->cpus)));
4889 
4890 		if (nr_online_nodes > 1 && !nodes_empty(l->nodes))
4891 			len += sysfs_emit_at(buf, len, " nodes=%*pbl",
4892 					     nodemask_pr_args(&l->nodes));
4893 
4894 		len += sysfs_emit_at(buf, len, "\n");
4895 	}
4896 
4897 	free_loc_track(&t);
4898 	if (!t.count)
4899 		len += sysfs_emit_at(buf, len, "No data\n");
4900 
4901 	return len;
4902 }
4903 #endif	/* CONFIG_SLUB_DEBUG */
4904 
4905 #ifdef SLUB_RESILIENCY_TEST
resiliency_test(void)4906 static void __init resiliency_test(void)
4907 {
4908 	u8 *p;
4909 	int type = KMALLOC_NORMAL;
4910 
4911 	BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
4912 
4913 	pr_err("SLUB resiliency testing\n");
4914 	pr_err("-----------------------\n");
4915 	pr_err("A. Corruption after allocation\n");
4916 
4917 	p = kzalloc(16, GFP_KERNEL);
4918 	p[16] = 0x12;
4919 	pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4920 	       p + 16);
4921 
4922 	validate_slab_cache(kmalloc_caches[type][4]);
4923 
4924 	/* Hmmm... The next two are dangerous */
4925 	p = kzalloc(32, GFP_KERNEL);
4926 	p[32 + sizeof(void *)] = 0x34;
4927 	pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4928 	       p);
4929 	pr_err("If allocated object is overwritten then not detectable\n\n");
4930 
4931 	validate_slab_cache(kmalloc_caches[type][5]);
4932 	p = kzalloc(64, GFP_KERNEL);
4933 	p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4934 	*p = 0x56;
4935 	pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4936 	       p);
4937 	pr_err("If allocated object is overwritten then not detectable\n\n");
4938 	validate_slab_cache(kmalloc_caches[type][6]);
4939 
4940 	pr_err("\nB. Corruption after free\n");
4941 	p = kzalloc(128, GFP_KERNEL);
4942 	kfree(p);
4943 	*p = 0x78;
4944 	pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
4945 	validate_slab_cache(kmalloc_caches[type][7]);
4946 
4947 	p = kzalloc(256, GFP_KERNEL);
4948 	kfree(p);
4949 	p[50] = 0x9a;
4950 	pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
4951 	validate_slab_cache(kmalloc_caches[type][8]);
4952 
4953 	p = kzalloc(512, GFP_KERNEL);
4954 	kfree(p);
4955 	p[512] = 0xab;
4956 	pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
4957 	validate_slab_cache(kmalloc_caches[type][9]);
4958 }
4959 #else
4960 #ifdef CONFIG_SYSFS
resiliency_test(void)4961 static void resiliency_test(void) {};
4962 #endif
4963 #endif	/* SLUB_RESILIENCY_TEST */
4964 
4965 #ifdef CONFIG_SYSFS
4966 enum slab_stat_type {
4967 	SL_ALL,			/* All slabs */
4968 	SL_PARTIAL,		/* Only partially allocated slabs */
4969 	SL_CPU,			/* Only slabs used for cpu caches */
4970 	SL_OBJECTS,		/* Determine allocated objects not slabs */
4971 	SL_TOTAL		/* Determine object capacity not slabs */
4972 };
4973 
4974 #define SO_ALL		(1 << SL_ALL)
4975 #define SO_PARTIAL	(1 << SL_PARTIAL)
4976 #define SO_CPU		(1 << SL_CPU)
4977 #define SO_OBJECTS	(1 << SL_OBJECTS)
4978 #define SO_TOTAL	(1 << SL_TOTAL)
4979 
show_slab_objects(struct kmem_cache * s,char * buf,unsigned long flags)4980 static ssize_t show_slab_objects(struct kmem_cache *s,
4981 				 char *buf, unsigned long flags)
4982 {
4983 	unsigned long total = 0;
4984 	int node;
4985 	int x;
4986 	unsigned long *nodes;
4987 	int len = 0;
4988 
4989 	nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
4990 	if (!nodes)
4991 		return -ENOMEM;
4992 
4993 	if (flags & SO_CPU) {
4994 		int cpu;
4995 
4996 		for_each_possible_cpu(cpu) {
4997 			struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4998 							       cpu);
4999 			int node;
5000 			struct page *page;
5001 
5002 			page = READ_ONCE(c->page);
5003 			if (!page)
5004 				continue;
5005 
5006 			node = page_to_nid(page);
5007 			if (flags & SO_TOTAL)
5008 				x = page->objects;
5009 			else if (flags & SO_OBJECTS)
5010 				x = page->inuse;
5011 			else
5012 				x = 1;
5013 
5014 			total += x;
5015 			nodes[node] += x;
5016 
5017 			page = slub_percpu_partial_read_once(c);
5018 			if (page) {
5019 				node = page_to_nid(page);
5020 				if (flags & SO_TOTAL)
5021 					WARN_ON_ONCE(1);
5022 				else if (flags & SO_OBJECTS)
5023 					WARN_ON_ONCE(1);
5024 				else
5025 					x = page->pages;
5026 				total += x;
5027 				nodes[node] += x;
5028 			}
5029 		}
5030 	}
5031 
5032 	/*
5033 	 * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex"
5034 	 * already held which will conflict with an existing lock order:
5035 	 *
5036 	 * mem_hotplug_lock->slab_mutex->kernfs_mutex
5037 	 *
5038 	 * We don't really need mem_hotplug_lock (to hold off
5039 	 * slab_mem_going_offline_callback) here because slab's memory hot
5040 	 * unplug code doesn't destroy the kmem_cache->node[] data.
5041 	 */
5042 
5043 #ifdef CONFIG_SLUB_DEBUG
5044 	if (flags & SO_ALL) {
5045 		struct kmem_cache_node *n;
5046 
5047 		for_each_kmem_cache_node(s, node, n) {
5048 
5049 			if (flags & SO_TOTAL)
5050 				x = atomic_long_read(&n->total_objects);
5051 			else if (flags & SO_OBJECTS)
5052 				x = atomic_long_read(&n->total_objects) -
5053 					count_partial(n, count_free);
5054 			else
5055 				x = atomic_long_read(&n->nr_slabs);
5056 			total += x;
5057 			nodes[node] += x;
5058 		}
5059 
5060 	} else
5061 #endif
5062 	if (flags & SO_PARTIAL) {
5063 		struct kmem_cache_node *n;
5064 
5065 		for_each_kmem_cache_node(s, node, n) {
5066 			if (flags & SO_TOTAL)
5067 				x = count_partial(n, count_total);
5068 			else if (flags & SO_OBJECTS)
5069 				x = count_partial(n, count_inuse);
5070 			else
5071 				x = n->nr_partial;
5072 			total += x;
5073 			nodes[node] += x;
5074 		}
5075 	}
5076 
5077 	len += sysfs_emit_at(buf, len, "%lu", total);
5078 #ifdef CONFIG_NUMA
5079 	for (node = 0; node < nr_node_ids; node++) {
5080 		if (nodes[node])
5081 			len += sysfs_emit_at(buf, len, " N%d=%lu",
5082 					     node, nodes[node]);
5083 	}
5084 #endif
5085 	len += sysfs_emit_at(buf, len, "\n");
5086 	kfree(nodes);
5087 
5088 	return len;
5089 }
5090 
5091 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
5092 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
5093 
5094 struct slab_attribute {
5095 	struct attribute attr;
5096 	ssize_t (*show)(struct kmem_cache *s, char *buf);
5097 	ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
5098 };
5099 
5100 #define SLAB_ATTR_RO(_name) \
5101 	static struct slab_attribute _name##_attr = \
5102 	__ATTR(_name, 0400, _name##_show, NULL)
5103 
5104 #define SLAB_ATTR(_name) \
5105 	static struct slab_attribute _name##_attr =  \
5106 	__ATTR(_name, 0600, _name##_show, _name##_store)
5107 
slab_size_show(struct kmem_cache * s,char * buf)5108 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
5109 {
5110 	return sysfs_emit(buf, "%u\n", s->size);
5111 }
5112 SLAB_ATTR_RO(slab_size);
5113 
align_show(struct kmem_cache * s,char * buf)5114 static ssize_t align_show(struct kmem_cache *s, char *buf)
5115 {
5116 	return sysfs_emit(buf, "%u\n", s->align);
5117 }
5118 SLAB_ATTR_RO(align);
5119 
object_size_show(struct kmem_cache * s,char * buf)5120 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
5121 {
5122 	return sysfs_emit(buf, "%u\n", s->object_size);
5123 }
5124 SLAB_ATTR_RO(object_size);
5125 
objs_per_slab_show(struct kmem_cache * s,char * buf)5126 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
5127 {
5128 	return sysfs_emit(buf, "%u\n", oo_objects(s->oo));
5129 }
5130 SLAB_ATTR_RO(objs_per_slab);
5131 
order_show(struct kmem_cache * s,char * buf)5132 static ssize_t order_show(struct kmem_cache *s, char *buf)
5133 {
5134 	return sysfs_emit(buf, "%u\n", oo_order(s->oo));
5135 }
5136 SLAB_ATTR_RO(order);
5137 
min_partial_show(struct kmem_cache * s,char * buf)5138 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
5139 {
5140 	return sysfs_emit(buf, "%lu\n", s->min_partial);
5141 }
5142 
min_partial_store(struct kmem_cache * s,const char * buf,size_t length)5143 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
5144 				 size_t length)
5145 {
5146 	unsigned long min;
5147 	int err;
5148 
5149 	err = kstrtoul(buf, 10, &min);
5150 	if (err)
5151 		return err;
5152 
5153 	set_min_partial(s, min);
5154 	return length;
5155 }
5156 SLAB_ATTR(min_partial);
5157 
cpu_partial_show(struct kmem_cache * s,char * buf)5158 static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
5159 {
5160 	return sysfs_emit(buf, "%u\n", slub_cpu_partial(s));
5161 }
5162 
cpu_partial_store(struct kmem_cache * s,const char * buf,size_t length)5163 static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
5164 				 size_t length)
5165 {
5166 	unsigned int objects;
5167 	int err;
5168 
5169 	err = kstrtouint(buf, 10, &objects);
5170 	if (err)
5171 		return err;
5172 	if (objects && !kmem_cache_has_cpu_partial(s))
5173 		return -EINVAL;
5174 
5175 	slub_set_cpu_partial(s, objects);
5176 	flush_all(s);
5177 	return length;
5178 }
5179 SLAB_ATTR(cpu_partial);
5180 
ctor_show(struct kmem_cache * s,char * buf)5181 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
5182 {
5183 	if (!s->ctor)
5184 		return 0;
5185 	return sysfs_emit(buf, "%pS\n", s->ctor);
5186 }
5187 SLAB_ATTR_RO(ctor);
5188 
aliases_show(struct kmem_cache * s,char * buf)5189 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5190 {
5191 	return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
5192 }
5193 SLAB_ATTR_RO(aliases);
5194 
partial_show(struct kmem_cache * s,char * buf)5195 static ssize_t partial_show(struct kmem_cache *s, char *buf)
5196 {
5197 	return show_slab_objects(s, buf, SO_PARTIAL);
5198 }
5199 SLAB_ATTR_RO(partial);
5200 
cpu_slabs_show(struct kmem_cache * s,char * buf)5201 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5202 {
5203 	return show_slab_objects(s, buf, SO_CPU);
5204 }
5205 SLAB_ATTR_RO(cpu_slabs);
5206 
objects_show(struct kmem_cache * s,char * buf)5207 static ssize_t objects_show(struct kmem_cache *s, char *buf)
5208 {
5209 	return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
5210 }
5211 SLAB_ATTR_RO(objects);
5212 
objects_partial_show(struct kmem_cache * s,char * buf)5213 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5214 {
5215 	return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5216 }
5217 SLAB_ATTR_RO(objects_partial);
5218 
slabs_cpu_partial_show(struct kmem_cache * s,char * buf)5219 static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5220 {
5221 	int objects = 0;
5222 	int pages = 0;
5223 	int cpu;
5224 	int len = 0;
5225 
5226 	for_each_online_cpu(cpu) {
5227 		struct page *page;
5228 
5229 		page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5230 
5231 		if (page) {
5232 			pages += page->pages;
5233 			objects += page->pobjects;
5234 		}
5235 	}
5236 
5237 	len += sysfs_emit_at(buf, len, "%d(%d)", objects, pages);
5238 
5239 #ifdef CONFIG_SMP
5240 	for_each_online_cpu(cpu) {
5241 		struct page *page;
5242 
5243 		page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5244 		if (page)
5245 			len += sysfs_emit_at(buf, len, " C%d=%d(%d)",
5246 					     cpu, page->pobjects, page->pages);
5247 	}
5248 #endif
5249 	len += sysfs_emit_at(buf, len, "\n");
5250 
5251 	return len;
5252 }
5253 SLAB_ATTR_RO(slabs_cpu_partial);
5254 
reclaim_account_show(struct kmem_cache * s,char * buf)5255 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5256 {
5257 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5258 }
5259 SLAB_ATTR_RO(reclaim_account);
5260 
hwcache_align_show(struct kmem_cache * s,char * buf)5261 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5262 {
5263 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
5264 }
5265 SLAB_ATTR_RO(hwcache_align);
5266 
5267 #ifdef CONFIG_ZONE_DMA
cache_dma_show(struct kmem_cache * s,char * buf)5268 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5269 {
5270 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5271 }
5272 SLAB_ATTR_RO(cache_dma);
5273 #endif
5274 
usersize_show(struct kmem_cache * s,char * buf)5275 static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5276 {
5277 	return sysfs_emit(buf, "%u\n", s->usersize);
5278 }
5279 SLAB_ATTR_RO(usersize);
5280 
destroy_by_rcu_show(struct kmem_cache * s,char * buf)5281 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5282 {
5283 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
5284 }
5285 SLAB_ATTR_RO(destroy_by_rcu);
5286 
5287 #ifdef CONFIG_SLUB_DEBUG
slabs_show(struct kmem_cache * s,char * buf)5288 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5289 {
5290 	return show_slab_objects(s, buf, SO_ALL);
5291 }
5292 SLAB_ATTR_RO(slabs);
5293 
total_objects_show(struct kmem_cache * s,char * buf)5294 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5295 {
5296 	return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5297 }
5298 SLAB_ATTR_RO(total_objects);
5299 
sanity_checks_show(struct kmem_cache * s,char * buf)5300 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5301 {
5302 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
5303 }
5304 SLAB_ATTR_RO(sanity_checks);
5305 
trace_show(struct kmem_cache * s,char * buf)5306 static ssize_t trace_show(struct kmem_cache *s, char *buf)
5307 {
5308 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5309 }
5310 SLAB_ATTR_RO(trace);
5311 
red_zone_show(struct kmem_cache * s,char * buf)5312 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5313 {
5314 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5315 }
5316 
5317 SLAB_ATTR_RO(red_zone);
5318 
poison_show(struct kmem_cache * s,char * buf)5319 static ssize_t poison_show(struct kmem_cache *s, char *buf)
5320 {
5321 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON));
5322 }
5323 
5324 SLAB_ATTR_RO(poison);
5325 
store_user_show(struct kmem_cache * s,char * buf)5326 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5327 {
5328 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5329 }
5330 
5331 SLAB_ATTR_RO(store_user);
5332 
validate_show(struct kmem_cache * s,char * buf)5333 static ssize_t validate_show(struct kmem_cache *s, char *buf)
5334 {
5335 	return 0;
5336 }
5337 
validate_store(struct kmem_cache * s,const char * buf,size_t length)5338 static ssize_t validate_store(struct kmem_cache *s,
5339 			const char *buf, size_t length)
5340 {
5341 	int ret = -EINVAL;
5342 
5343 	if (buf[0] == '1') {
5344 		ret = validate_slab_cache(s);
5345 		if (ret >= 0)
5346 			ret = length;
5347 	}
5348 	return ret;
5349 }
5350 SLAB_ATTR(validate);
5351 
alloc_calls_show(struct kmem_cache * s,char * buf)5352 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
5353 {
5354 	if (!(s->flags & SLAB_STORE_USER))
5355 		return -ENOSYS;
5356 	return list_locations(s, buf, TRACK_ALLOC);
5357 }
5358 SLAB_ATTR_RO(alloc_calls);
5359 
free_calls_show(struct kmem_cache * s,char * buf)5360 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
5361 {
5362 	if (!(s->flags & SLAB_STORE_USER))
5363 		return -ENOSYS;
5364 	return list_locations(s, buf, TRACK_FREE);
5365 }
5366 SLAB_ATTR_RO(free_calls);
5367 #endif /* CONFIG_SLUB_DEBUG */
5368 
5369 #ifdef CONFIG_FAILSLAB
failslab_show(struct kmem_cache * s,char * buf)5370 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5371 {
5372 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5373 }
5374 SLAB_ATTR_RO(failslab);
5375 #endif
5376 
shrink_show(struct kmem_cache * s,char * buf)5377 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5378 {
5379 	return 0;
5380 }
5381 
shrink_store(struct kmem_cache * s,const char * buf,size_t length)5382 static ssize_t shrink_store(struct kmem_cache *s,
5383 			const char *buf, size_t length)
5384 {
5385 	if (buf[0] == '1')
5386 		kmem_cache_shrink(s);
5387 	else
5388 		return -EINVAL;
5389 	return length;
5390 }
5391 SLAB_ATTR(shrink);
5392 
5393 #ifdef CONFIG_NUMA
remote_node_defrag_ratio_show(struct kmem_cache * s,char * buf)5394 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
5395 {
5396 	return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10);
5397 }
5398 
remote_node_defrag_ratio_store(struct kmem_cache * s,const char * buf,size_t length)5399 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
5400 				const char *buf, size_t length)
5401 {
5402 	unsigned int ratio;
5403 	int err;
5404 
5405 	err = kstrtouint(buf, 10, &ratio);
5406 	if (err)
5407 		return err;
5408 	if (ratio > 100)
5409 		return -ERANGE;
5410 
5411 	s->remote_node_defrag_ratio = ratio * 10;
5412 
5413 	return length;
5414 }
5415 SLAB_ATTR(remote_node_defrag_ratio);
5416 #endif
5417 
5418 #ifdef CONFIG_SLUB_STATS
show_stat(struct kmem_cache * s,char * buf,enum stat_item si)5419 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5420 {
5421 	unsigned long sum  = 0;
5422 	int cpu;
5423 	int len = 0;
5424 	int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
5425 
5426 	if (!data)
5427 		return -ENOMEM;
5428 
5429 	for_each_online_cpu(cpu) {
5430 		unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
5431 
5432 		data[cpu] = x;
5433 		sum += x;
5434 	}
5435 
5436 	len += sysfs_emit_at(buf, len, "%lu", sum);
5437 
5438 #ifdef CONFIG_SMP
5439 	for_each_online_cpu(cpu) {
5440 		if (data[cpu])
5441 			len += sysfs_emit_at(buf, len, " C%d=%u",
5442 					     cpu, data[cpu]);
5443 	}
5444 #endif
5445 	kfree(data);
5446 	len += sysfs_emit_at(buf, len, "\n");
5447 
5448 	return len;
5449 }
5450 
clear_stat(struct kmem_cache * s,enum stat_item si)5451 static void clear_stat(struct kmem_cache *s, enum stat_item si)
5452 {
5453 	int cpu;
5454 
5455 	for_each_online_cpu(cpu)
5456 		per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
5457 }
5458 
5459 #define STAT_ATTR(si, text) 					\
5460 static ssize_t text##_show(struct kmem_cache *s, char *buf)	\
5461 {								\
5462 	return show_stat(s, buf, si);				\
5463 }								\
5464 static ssize_t text##_store(struct kmem_cache *s,		\
5465 				const char *buf, size_t length)	\
5466 {								\
5467 	if (buf[0] != '0')					\
5468 		return -EINVAL;					\
5469 	clear_stat(s, si);					\
5470 	return length;						\
5471 }								\
5472 SLAB_ATTR(text);						\
5473 
5474 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5475 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5476 STAT_ATTR(FREE_FASTPATH, free_fastpath);
5477 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5478 STAT_ATTR(FREE_FROZEN, free_frozen);
5479 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5480 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5481 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5482 STAT_ATTR(ALLOC_SLAB, alloc_slab);
5483 STAT_ATTR(ALLOC_REFILL, alloc_refill);
5484 STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
5485 STAT_ATTR(FREE_SLAB, free_slab);
5486 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5487 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5488 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5489 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5490 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5491 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
5492 STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
5493 STAT_ATTR(ORDER_FALLBACK, order_fallback);
5494 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5495 STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
5496 STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5497 STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
5498 STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5499 STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
5500 #endif	/* CONFIG_SLUB_STATS */
5501 
5502 static struct attribute *slab_attrs[] = {
5503 	&slab_size_attr.attr,
5504 	&object_size_attr.attr,
5505 	&objs_per_slab_attr.attr,
5506 	&order_attr.attr,
5507 	&min_partial_attr.attr,
5508 	&cpu_partial_attr.attr,
5509 	&objects_attr.attr,
5510 	&objects_partial_attr.attr,
5511 	&partial_attr.attr,
5512 	&cpu_slabs_attr.attr,
5513 	&ctor_attr.attr,
5514 	&aliases_attr.attr,
5515 	&align_attr.attr,
5516 	&hwcache_align_attr.attr,
5517 	&reclaim_account_attr.attr,
5518 	&destroy_by_rcu_attr.attr,
5519 	&shrink_attr.attr,
5520 	&slabs_cpu_partial_attr.attr,
5521 #ifdef CONFIG_SLUB_DEBUG
5522 	&total_objects_attr.attr,
5523 	&slabs_attr.attr,
5524 	&sanity_checks_attr.attr,
5525 	&trace_attr.attr,
5526 	&red_zone_attr.attr,
5527 	&poison_attr.attr,
5528 	&store_user_attr.attr,
5529 	&validate_attr.attr,
5530 	&alloc_calls_attr.attr,
5531 	&free_calls_attr.attr,
5532 #endif
5533 #ifdef CONFIG_ZONE_DMA
5534 	&cache_dma_attr.attr,
5535 #endif
5536 #ifdef CONFIG_NUMA
5537 	&remote_node_defrag_ratio_attr.attr,
5538 #endif
5539 #ifdef CONFIG_SLUB_STATS
5540 	&alloc_fastpath_attr.attr,
5541 	&alloc_slowpath_attr.attr,
5542 	&free_fastpath_attr.attr,
5543 	&free_slowpath_attr.attr,
5544 	&free_frozen_attr.attr,
5545 	&free_add_partial_attr.attr,
5546 	&free_remove_partial_attr.attr,
5547 	&alloc_from_partial_attr.attr,
5548 	&alloc_slab_attr.attr,
5549 	&alloc_refill_attr.attr,
5550 	&alloc_node_mismatch_attr.attr,
5551 	&free_slab_attr.attr,
5552 	&cpuslab_flush_attr.attr,
5553 	&deactivate_full_attr.attr,
5554 	&deactivate_empty_attr.attr,
5555 	&deactivate_to_head_attr.attr,
5556 	&deactivate_to_tail_attr.attr,
5557 	&deactivate_remote_frees_attr.attr,
5558 	&deactivate_bypass_attr.attr,
5559 	&order_fallback_attr.attr,
5560 	&cmpxchg_double_fail_attr.attr,
5561 	&cmpxchg_double_cpu_fail_attr.attr,
5562 	&cpu_partial_alloc_attr.attr,
5563 	&cpu_partial_free_attr.attr,
5564 	&cpu_partial_node_attr.attr,
5565 	&cpu_partial_drain_attr.attr,
5566 #endif
5567 #ifdef CONFIG_FAILSLAB
5568 	&failslab_attr.attr,
5569 #endif
5570 	&usersize_attr.attr,
5571 
5572 	NULL
5573 };
5574 
5575 static const struct attribute_group slab_attr_group = {
5576 	.attrs = slab_attrs,
5577 };
5578 
slab_attr_show(struct kobject * kobj,struct attribute * attr,char * buf)5579 static ssize_t slab_attr_show(struct kobject *kobj,
5580 				struct attribute *attr,
5581 				char *buf)
5582 {
5583 	struct slab_attribute *attribute;
5584 	struct kmem_cache *s;
5585 	int err;
5586 
5587 	attribute = to_slab_attr(attr);
5588 	s = to_slab(kobj);
5589 
5590 	if (!attribute->show)
5591 		return -EIO;
5592 
5593 	err = attribute->show(s, buf);
5594 
5595 	return err;
5596 }
5597 
slab_attr_store(struct kobject * kobj,struct attribute * attr,const char * buf,size_t len)5598 static ssize_t slab_attr_store(struct kobject *kobj,
5599 				struct attribute *attr,
5600 				const char *buf, size_t len)
5601 {
5602 	struct slab_attribute *attribute;
5603 	struct kmem_cache *s;
5604 	int err;
5605 
5606 	attribute = to_slab_attr(attr);
5607 	s = to_slab(kobj);
5608 
5609 	if (!attribute->store)
5610 		return -EIO;
5611 
5612 	err = attribute->store(s, buf, len);
5613 	return err;
5614 }
5615 
kmem_cache_release(struct kobject * k)5616 static void kmem_cache_release(struct kobject *k)
5617 {
5618 	slab_kmem_cache_release(to_slab(k));
5619 }
5620 
5621 static const struct sysfs_ops slab_sysfs_ops = {
5622 	.show = slab_attr_show,
5623 	.store = slab_attr_store,
5624 };
5625 
5626 static struct kobj_type slab_ktype = {
5627 	.sysfs_ops = &slab_sysfs_ops,
5628 	.release = kmem_cache_release,
5629 };
5630 
5631 static struct kset *slab_kset;
5632 
cache_kset(struct kmem_cache * s)5633 static inline struct kset *cache_kset(struct kmem_cache *s)
5634 {
5635 	return slab_kset;
5636 }
5637 
5638 #define ID_STR_LENGTH 64
5639 
5640 /* Create a unique string id for a slab cache:
5641  *
5642  * Format	:[flags-]size
5643  */
create_unique_id(struct kmem_cache * s)5644 static char *create_unique_id(struct kmem_cache *s)
5645 {
5646 	char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5647 	char *p = name;
5648 
5649 	BUG_ON(!name);
5650 
5651 	*p++ = ':';
5652 	/*
5653 	 * First flags affecting slabcache operations. We will only
5654 	 * get here for aliasable slabs so we do not need to support
5655 	 * too many flags. The flags here must cover all flags that
5656 	 * are matched during merging to guarantee that the id is
5657 	 * unique.
5658 	 */
5659 	if (s->flags & SLAB_CACHE_DMA)
5660 		*p++ = 'd';
5661 	if (s->flags & SLAB_CACHE_DMA32)
5662 		*p++ = 'D';
5663 	if (s->flags & SLAB_RECLAIM_ACCOUNT)
5664 		*p++ = 'a';
5665 	if (s->flags & SLAB_CONSISTENCY_CHECKS)
5666 		*p++ = 'F';
5667 	if (s->flags & SLAB_ACCOUNT)
5668 		*p++ = 'A';
5669 	if (p != name + 1)
5670 		*p++ = '-';
5671 	p += sprintf(p, "%07u", s->size);
5672 
5673 	BUG_ON(p > name + ID_STR_LENGTH - 1);
5674 	return name;
5675 }
5676 
sysfs_slab_add(struct kmem_cache * s)5677 static int sysfs_slab_add(struct kmem_cache *s)
5678 {
5679 	int err;
5680 	const char *name;
5681 	struct kset *kset = cache_kset(s);
5682 	int unmergeable = slab_unmergeable(s);
5683 
5684 	if (!kset) {
5685 		kobject_init(&s->kobj, &slab_ktype);
5686 		return 0;
5687 	}
5688 
5689 	if (!unmergeable && disable_higher_order_debug &&
5690 			(slub_debug & DEBUG_METADATA_FLAGS))
5691 		unmergeable = 1;
5692 
5693 	if (unmergeable) {
5694 		/*
5695 		 * Slabcache can never be merged so we can use the name proper.
5696 		 * This is typically the case for debug situations. In that
5697 		 * case we can catch duplicate names easily.
5698 		 */
5699 		sysfs_remove_link(&slab_kset->kobj, s->name);
5700 		name = s->name;
5701 	} else {
5702 		/*
5703 		 * Create a unique name for the slab as a target
5704 		 * for the symlinks.
5705 		 */
5706 		name = create_unique_id(s);
5707 	}
5708 
5709 	s->kobj.kset = kset;
5710 	err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
5711 	if (err)
5712 		goto out;
5713 
5714 	err = sysfs_create_group(&s->kobj, &slab_attr_group);
5715 	if (err)
5716 		goto out_del_kobj;
5717 
5718 	if (!unmergeable) {
5719 		/* Setup first alias */
5720 		sysfs_slab_alias(s, s->name);
5721 	}
5722 out:
5723 	if (!unmergeable)
5724 		kfree(name);
5725 	return err;
5726 out_del_kobj:
5727 	kobject_del(&s->kobj);
5728 	goto out;
5729 }
5730 
sysfs_slab_unlink(struct kmem_cache * s)5731 void sysfs_slab_unlink(struct kmem_cache *s)
5732 {
5733 	if (slab_state >= FULL)
5734 		kobject_del(&s->kobj);
5735 }
5736 
sysfs_slab_release(struct kmem_cache * s)5737 void sysfs_slab_release(struct kmem_cache *s)
5738 {
5739 	if (slab_state >= FULL)
5740 		kobject_put(&s->kobj);
5741 }
5742 
5743 /*
5744  * Need to buffer aliases during bootup until sysfs becomes
5745  * available lest we lose that information.
5746  */
5747 struct saved_alias {
5748 	struct kmem_cache *s;
5749 	const char *name;
5750 	struct saved_alias *next;
5751 };
5752 
5753 static struct saved_alias *alias_list;
5754 
sysfs_slab_alias(struct kmem_cache * s,const char * name)5755 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5756 {
5757 	struct saved_alias *al;
5758 
5759 	if (slab_state == FULL) {
5760 		/*
5761 		 * If we have a leftover link then remove it.
5762 		 */
5763 		sysfs_remove_link(&slab_kset->kobj, name);
5764 		return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
5765 	}
5766 
5767 	al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5768 	if (!al)
5769 		return -ENOMEM;
5770 
5771 	al->s = s;
5772 	al->name = name;
5773 	al->next = alias_list;
5774 	alias_list = al;
5775 	return 0;
5776 }
5777 
slab_sysfs_init(void)5778 static int __init slab_sysfs_init(void)
5779 {
5780 	struct kmem_cache *s;
5781 	int err;
5782 
5783 	mutex_lock(&slab_mutex);
5784 
5785 	slab_kset = kset_create_and_add("slab", NULL, kernel_kobj);
5786 	if (!slab_kset) {
5787 		mutex_unlock(&slab_mutex);
5788 		pr_err("Cannot register slab subsystem.\n");
5789 		return -ENOSYS;
5790 	}
5791 
5792 	slab_state = FULL;
5793 
5794 	list_for_each_entry(s, &slab_caches, list) {
5795 		err = sysfs_slab_add(s);
5796 		if (err)
5797 			pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5798 			       s->name);
5799 	}
5800 
5801 	while (alias_list) {
5802 		struct saved_alias *al = alias_list;
5803 
5804 		alias_list = alias_list->next;
5805 		err = sysfs_slab_alias(al->s, al->name);
5806 		if (err)
5807 			pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5808 			       al->name);
5809 		kfree(al);
5810 	}
5811 
5812 	mutex_unlock(&slab_mutex);
5813 	resiliency_test();
5814 	return 0;
5815 }
5816 
5817 __initcall(slab_sysfs_init);
5818 #endif /* CONFIG_SYSFS */
5819 
5820 /*
5821  * The /proc/slabinfo ABI
5822  */
5823 #ifdef CONFIG_SLUB_DEBUG
get_slabinfo(struct kmem_cache * s,struct slabinfo * sinfo)5824 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
5825 {
5826 	unsigned long nr_slabs = 0;
5827 	unsigned long nr_objs = 0;
5828 	unsigned long nr_free = 0;
5829 	int node;
5830 	struct kmem_cache_node *n;
5831 
5832 	for_each_kmem_cache_node(s, node, n) {
5833 		nr_slabs += node_nr_slabs(n);
5834 		nr_objs += node_nr_objs(n);
5835 		nr_free += count_partial(n, count_free);
5836 	}
5837 
5838 	sinfo->active_objs = nr_objs - nr_free;
5839 	sinfo->num_objs = nr_objs;
5840 	sinfo->active_slabs = nr_slabs;
5841 	sinfo->num_slabs = nr_slabs;
5842 	sinfo->objects_per_slab = oo_objects(s->oo);
5843 	sinfo->cache_order = oo_order(s->oo);
5844 }
5845 
slabinfo_show_stats(struct seq_file * m,struct kmem_cache * s)5846 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
5847 {
5848 }
5849 
slabinfo_write(struct file * file,const char __user * buffer,size_t count,loff_t * ppos)5850 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5851 		       size_t count, loff_t *ppos)
5852 {
5853 	return -EIO;
5854 }
5855 #endif /* CONFIG_SLUB_DEBUG */
5856