xref: /linux/mm/slub.c (revision 2c1ed907520c50326b8f604907a8478b27881a2e)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * SLUB: A slab allocator that limits cache line use instead of queuing
4  * objects in per cpu and per node lists.
5  *
6  * The allocator synchronizes using per slab locks or atomic operations
7  * and only uses a centralized lock to manage a pool of partial slabs.
8  *
9  * (C) 2007 SGI, Christoph Lameter
10  * (C) 2011 Linux Foundation, Christoph Lameter
11  */
12 
13 #include <linux/mm.h>
14 #include <linux/swap.h> /* mm_account_reclaimed_pages() */
15 #include <linux/module.h>
16 #include <linux/bit_spinlock.h>
17 #include <linux/interrupt.h>
18 #include <linux/swab.h>
19 #include <linux/bitops.h>
20 #include <linux/slab.h>
21 #include "slab.h"
22 #include <linux/proc_fs.h>
23 #include <linux/seq_file.h>
24 #include <linux/kasan.h>
25 #include <linux/kmsan.h>
26 #include <linux/cpu.h>
27 #include <linux/cpuset.h>
28 #include <linux/mempolicy.h>
29 #include <linux/ctype.h>
30 #include <linux/stackdepot.h>
31 #include <linux/debugobjects.h>
32 #include <linux/kallsyms.h>
33 #include <linux/kfence.h>
34 #include <linux/memory.h>
35 #include <linux/math64.h>
36 #include <linux/fault-inject.h>
37 #include <linux/kmemleak.h>
38 #include <linux/stacktrace.h>
39 #include <linux/prefetch.h>
40 #include <linux/memcontrol.h>
41 #include <linux/random.h>
42 #include <kunit/test.h>
43 #include <kunit/test-bug.h>
44 #include <linux/sort.h>
45 
46 #include <linux/debugfs.h>
47 #include <trace/events/kmem.h>
48 
49 #include "internal.h"
50 
51 /*
52  * Lock order:
53  *   1. slab_mutex (Global Mutex)
54  *   2. node->list_lock (Spinlock)
55  *   3. kmem_cache->cpu_slab->lock (Local lock)
56  *   4. slab_lock(slab) (Only on some arches)
57  *   5. object_map_lock (Only for debugging)
58  *
59  *   slab_mutex
60  *
61  *   The role of the slab_mutex is to protect the list of all the slabs
62  *   and to synchronize major metadata changes to slab cache structures.
63  *   Also synchronizes memory hotplug callbacks.
64  *
65  *   slab_lock
66  *
67  *   The slab_lock is a wrapper around the page lock, thus it is a bit
68  *   spinlock.
69  *
70  *   The slab_lock is only used on arches that do not have the ability
71  *   to do a cmpxchg_double. It only protects:
72  *
73  *	A. slab->freelist	-> List of free objects in a slab
74  *	B. slab->inuse		-> Number of objects in use
75  *	C. slab->objects	-> Number of objects in slab
76  *	D. slab->frozen		-> frozen state
77  *
78  *   Frozen slabs
79  *
80  *   If a slab is frozen then it is exempt from list management. It is
81  *   the cpu slab which is actively allocated from by the processor that
82  *   froze it and it is not on any list. The processor that froze the
83  *   slab is the one who can perform list operations on the slab. Other
84  *   processors may put objects onto the freelist but the processor that
85  *   froze the slab is the only one that can retrieve the objects from the
86  *   slab's freelist.
87  *
88  *   CPU partial slabs
89  *
90  *   The partially empty slabs cached on the CPU partial list are used
91  *   for performance reasons, which speeds up the allocation process.
92  *   These slabs are not frozen, but are also exempt from list management,
93  *   by clearing the PG_workingset flag when moving out of the node
94  *   partial list. Please see __slab_free() for more details.
95  *
96  *   To sum up, the current scheme is:
97  *   - node partial slab: PG_Workingset && !frozen
98  *   - cpu partial slab: !PG_Workingset && !frozen
99  *   - cpu slab: !PG_Workingset && frozen
100  *   - full slab: !PG_Workingset && !frozen
101  *
102  *   list_lock
103  *
104  *   The list_lock protects the partial and full list on each node and
105  *   the partial slab counter. If taken then no new slabs may be added or
106  *   removed from the lists nor make the number of partial slabs be modified.
107  *   (Note that the total number of slabs is an atomic value that may be
108  *   modified without taking the list lock).
109  *
110  *   The list_lock is a centralized lock and thus we avoid taking it as
111  *   much as possible. As long as SLUB does not have to handle partial
112  *   slabs, operations can continue without any centralized lock. F.e.
113  *   allocating a long series of objects that fill up slabs does not require
114  *   the list lock.
115  *
116  *   For debug caches, all allocations are forced to go through a list_lock
117  *   protected region to serialize against concurrent validation.
118  *
119  *   cpu_slab->lock local lock
120  *
121  *   This locks protect slowpath manipulation of all kmem_cache_cpu fields
122  *   except the stat counters. This is a percpu structure manipulated only by
123  *   the local cpu, so the lock protects against being preempted or interrupted
124  *   by an irq. Fast path operations rely on lockless operations instead.
125  *
126  *   On PREEMPT_RT, the local lock neither disables interrupts nor preemption
127  *   which means the lockless fastpath cannot be used as it might interfere with
128  *   an in-progress slow path operations. In this case the local lock is always
129  *   taken but it still utilizes the freelist for the common operations.
130  *
131  *   lockless fastpaths
132  *
133  *   The fast path allocation (slab_alloc_node()) and freeing (do_slab_free())
134  *   are fully lockless when satisfied from the percpu slab (and when
135  *   cmpxchg_double is possible to use, otherwise slab_lock is taken).
136  *   They also don't disable preemption or migration or irqs. They rely on
137  *   the transaction id (tid) field to detect being preempted or moved to
138  *   another cpu.
139  *
140  *   irq, preemption, migration considerations
141  *
142  *   Interrupts are disabled as part of list_lock or local_lock operations, or
143  *   around the slab_lock operation, in order to make the slab allocator safe
144  *   to use in the context of an irq.
145  *
146  *   In addition, preemption (or migration on PREEMPT_RT) is disabled in the
147  *   allocation slowpath, bulk allocation, and put_cpu_partial(), so that the
148  *   local cpu doesn't change in the process and e.g. the kmem_cache_cpu pointer
149  *   doesn't have to be revalidated in each section protected by the local lock.
150  *
151  * SLUB assigns one slab for allocation to each processor.
152  * Allocations only occur from these slabs called cpu slabs.
153  *
154  * Slabs with free elements are kept on a partial list and during regular
155  * operations no list for full slabs is used. If an object in a full slab is
156  * freed then the slab will show up again on the partial lists.
157  * We track full slabs for debugging purposes though because otherwise we
158  * cannot scan all objects.
159  *
160  * Slabs are freed when they become empty. Teardown and setup is
161  * minimal so we rely on the page allocators per cpu caches for
162  * fast frees and allocs.
163  *
164  * slab->frozen		The slab is frozen and exempt from list processing.
165  * 			This means that the slab is dedicated to a purpose
166  * 			such as satisfying allocations for a specific
167  * 			processor. Objects may be freed in the slab while
168  * 			it is frozen but slab_free will then skip the usual
169  * 			list operations. It is up to the processor holding
170  * 			the slab to integrate the slab into the slab lists
171  * 			when the slab is no longer needed.
172  *
173  * 			One use of this flag is to mark slabs that are
174  * 			used for allocations. Then such a slab becomes a cpu
175  * 			slab. The cpu slab may be equipped with an additional
176  * 			freelist that allows lockless access to
177  * 			free objects in addition to the regular freelist
178  * 			that requires the slab lock.
179  *
180  * SLAB_DEBUG_FLAGS	Slab requires special handling due to debug
181  * 			options set. This moves	slab handling out of
182  * 			the fast path and disables lockless freelists.
183  */
184 
185 /*
186  * We could simply use migrate_disable()/enable() but as long as it's a
187  * function call even on !PREEMPT_RT, use inline preempt_disable() there.
188  */
189 #ifndef CONFIG_PREEMPT_RT
190 #define slub_get_cpu_ptr(var)		get_cpu_ptr(var)
191 #define slub_put_cpu_ptr(var)		put_cpu_ptr(var)
192 #define USE_LOCKLESS_FAST_PATH()	(true)
193 #else
194 #define slub_get_cpu_ptr(var)		\
195 ({					\
196 	migrate_disable();		\
197 	this_cpu_ptr(var);		\
198 })
199 #define slub_put_cpu_ptr(var)		\
200 do {					\
201 	(void)(var);			\
202 	migrate_enable();		\
203 } while (0)
204 #define USE_LOCKLESS_FAST_PATH()	(false)
205 #endif
206 
207 #ifndef CONFIG_SLUB_TINY
208 #define __fastpath_inline __always_inline
209 #else
210 #define __fastpath_inline
211 #endif
212 
213 #ifdef CONFIG_SLUB_DEBUG
214 #ifdef CONFIG_SLUB_DEBUG_ON
215 DEFINE_STATIC_KEY_TRUE(slub_debug_enabled);
216 #else
217 DEFINE_STATIC_KEY_FALSE(slub_debug_enabled);
218 #endif
219 #endif		/* CONFIG_SLUB_DEBUG */
220 
221 #ifdef CONFIG_NUMA
222 static DEFINE_STATIC_KEY_FALSE(strict_numa);
223 #endif
224 
225 /* Structure holding parameters for get_partial() call chain */
226 struct partial_context {
227 	gfp_t flags;
228 	unsigned int orig_size;
229 	void *object;
230 };
231 
kmem_cache_debug(struct kmem_cache * s)232 static inline bool kmem_cache_debug(struct kmem_cache *s)
233 {
234 	return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS);
235 }
236 
fixup_red_left(struct kmem_cache * s,void * p)237 void *fixup_red_left(struct kmem_cache *s, void *p)
238 {
239 	if (kmem_cache_debug_flags(s, SLAB_RED_ZONE))
240 		p += s->red_left_pad;
241 
242 	return p;
243 }
244 
kmem_cache_has_cpu_partial(struct kmem_cache * s)245 static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
246 {
247 #ifdef CONFIG_SLUB_CPU_PARTIAL
248 	return !kmem_cache_debug(s);
249 #else
250 	return false;
251 #endif
252 }
253 
254 /*
255  * Issues still to be resolved:
256  *
257  * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
258  *
259  * - Variable sizing of the per node arrays
260  */
261 
262 /* Enable to log cmpxchg failures */
263 #undef SLUB_DEBUG_CMPXCHG
264 
265 #ifndef CONFIG_SLUB_TINY
266 /*
267  * Minimum number of partial slabs. These will be left on the partial
268  * lists even if they are empty. kmem_cache_shrink may reclaim them.
269  */
270 #define MIN_PARTIAL 5
271 
272 /*
273  * Maximum number of desirable partial slabs.
274  * The existence of more partial slabs makes kmem_cache_shrink
275  * sort the partial list by the number of objects in use.
276  */
277 #define MAX_PARTIAL 10
278 #else
279 #define MIN_PARTIAL 0
280 #define MAX_PARTIAL 0
281 #endif
282 
283 #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
284 				SLAB_POISON | SLAB_STORE_USER)
285 
286 /*
287  * These debug flags cannot use CMPXCHG because there might be consistency
288  * issues when checking or reading debug information
289  */
290 #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
291 				SLAB_TRACE)
292 
293 
294 /*
295  * Debugging flags that require metadata to be stored in the slab.  These get
296  * disabled when slab_debug=O is used and a cache's min order increases with
297  * metadata.
298  */
299 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
300 
301 #define OO_SHIFT	16
302 #define OO_MASK		((1 << OO_SHIFT) - 1)
303 #define MAX_OBJS_PER_PAGE	32767 /* since slab.objects is u15 */
304 
305 /* Internal SLUB flags */
306 /* Poison object */
307 #define __OBJECT_POISON		__SLAB_FLAG_BIT(_SLAB_OBJECT_POISON)
308 /* Use cmpxchg_double */
309 
310 #ifdef system_has_freelist_aba
311 #define __CMPXCHG_DOUBLE	__SLAB_FLAG_BIT(_SLAB_CMPXCHG_DOUBLE)
312 #else
313 #define __CMPXCHG_DOUBLE	__SLAB_FLAG_UNUSED
314 #endif
315 
316 /*
317  * Tracking user of a slab.
318  */
319 #define TRACK_ADDRS_COUNT 16
320 struct track {
321 	unsigned long addr;	/* Called from address */
322 #ifdef CONFIG_STACKDEPOT
323 	depot_stack_handle_t handle;
324 #endif
325 	int cpu;		/* Was running on cpu */
326 	int pid;		/* Pid context */
327 	unsigned long when;	/* When did the operation occur */
328 };
329 
330 enum track_item { TRACK_ALLOC, TRACK_FREE };
331 
332 #ifdef SLAB_SUPPORTS_SYSFS
333 static int sysfs_slab_add(struct kmem_cache *);
334 static int sysfs_slab_alias(struct kmem_cache *, const char *);
335 #else
sysfs_slab_add(struct kmem_cache * s)336 static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
sysfs_slab_alias(struct kmem_cache * s,const char * p)337 static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
338 							{ return 0; }
339 #endif
340 
341 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
342 static void debugfs_slab_add(struct kmem_cache *);
343 #else
debugfs_slab_add(struct kmem_cache * s)344 static inline void debugfs_slab_add(struct kmem_cache *s) { }
345 #endif
346 
347 enum stat_item {
348 	ALLOC_FASTPATH,		/* Allocation from cpu slab */
349 	ALLOC_SLOWPATH,		/* Allocation by getting a new cpu slab */
350 	FREE_FASTPATH,		/* Free to cpu slab */
351 	FREE_SLOWPATH,		/* Freeing not to cpu slab */
352 	FREE_FROZEN,		/* Freeing to frozen slab */
353 	FREE_ADD_PARTIAL,	/* Freeing moves slab to partial list */
354 	FREE_REMOVE_PARTIAL,	/* Freeing removes last object */
355 	ALLOC_FROM_PARTIAL,	/* Cpu slab acquired from node partial list */
356 	ALLOC_SLAB,		/* Cpu slab acquired from page allocator */
357 	ALLOC_REFILL,		/* Refill cpu slab from slab freelist */
358 	ALLOC_NODE_MISMATCH,	/* Switching cpu slab */
359 	FREE_SLAB,		/* Slab freed to the page allocator */
360 	CPUSLAB_FLUSH,		/* Abandoning of the cpu slab */
361 	DEACTIVATE_FULL,	/* Cpu slab was full when deactivated */
362 	DEACTIVATE_EMPTY,	/* Cpu slab was empty when deactivated */
363 	DEACTIVATE_TO_HEAD,	/* Cpu slab was moved to the head of partials */
364 	DEACTIVATE_TO_TAIL,	/* Cpu slab was moved to the tail of partials */
365 	DEACTIVATE_REMOTE_FREES,/* Slab contained remotely freed objects */
366 	DEACTIVATE_BYPASS,	/* Implicit deactivation */
367 	ORDER_FALLBACK,		/* Number of times fallback was necessary */
368 	CMPXCHG_DOUBLE_CPU_FAIL,/* Failures of this_cpu_cmpxchg_double */
369 	CMPXCHG_DOUBLE_FAIL,	/* Failures of slab freelist update */
370 	CPU_PARTIAL_ALLOC,	/* Used cpu partial on alloc */
371 	CPU_PARTIAL_FREE,	/* Refill cpu partial on free */
372 	CPU_PARTIAL_NODE,	/* Refill cpu partial from node partial */
373 	CPU_PARTIAL_DRAIN,	/* Drain cpu partial to node partial */
374 	NR_SLUB_STAT_ITEMS
375 };
376 
377 #ifndef CONFIG_SLUB_TINY
378 /*
379  * When changing the layout, make sure freelist and tid are still compatible
380  * with this_cpu_cmpxchg_double() alignment requirements.
381  */
382 struct kmem_cache_cpu {
383 	union {
384 		struct {
385 			void **freelist;	/* Pointer to next available object */
386 			unsigned long tid;	/* Globally unique transaction id */
387 		};
388 		freelist_aba_t freelist_tid;
389 	};
390 	struct slab *slab;	/* The slab from which we are allocating */
391 #ifdef CONFIG_SLUB_CPU_PARTIAL
392 	struct slab *partial;	/* Partially allocated slabs */
393 #endif
394 	local_lock_t lock;	/* Protects the fields above */
395 #ifdef CONFIG_SLUB_STATS
396 	unsigned int stat[NR_SLUB_STAT_ITEMS];
397 #endif
398 };
399 #endif /* CONFIG_SLUB_TINY */
400 
stat(const struct kmem_cache * s,enum stat_item si)401 static inline void stat(const struct kmem_cache *s, enum stat_item si)
402 {
403 #ifdef CONFIG_SLUB_STATS
404 	/*
405 	 * The rmw is racy on a preemptible kernel but this is acceptable, so
406 	 * avoid this_cpu_add()'s irq-disable overhead.
407 	 */
408 	raw_cpu_inc(s->cpu_slab->stat[si]);
409 #endif
410 }
411 
412 static inline
stat_add(const struct kmem_cache * s,enum stat_item si,int v)413 void stat_add(const struct kmem_cache *s, enum stat_item si, int v)
414 {
415 #ifdef CONFIG_SLUB_STATS
416 	raw_cpu_add(s->cpu_slab->stat[si], v);
417 #endif
418 }
419 
420 /*
421  * The slab lists for all objects.
422  */
423 struct kmem_cache_node {
424 	spinlock_t list_lock;
425 	unsigned long nr_partial;
426 	struct list_head partial;
427 #ifdef CONFIG_SLUB_DEBUG
428 	atomic_long_t nr_slabs;
429 	atomic_long_t total_objects;
430 	struct list_head full;
431 #endif
432 };
433 
get_node(struct kmem_cache * s,int node)434 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
435 {
436 	return s->node[node];
437 }
438 
439 /*
440  * Iterator over all nodes. The body will be executed for each node that has
441  * a kmem_cache_node structure allocated (which is true for all online nodes)
442  */
443 #define for_each_kmem_cache_node(__s, __node, __n) \
444 	for (__node = 0; __node < nr_node_ids; __node++) \
445 		 if ((__n = get_node(__s, __node)))
446 
447 /*
448  * Tracks for which NUMA nodes we have kmem_cache_nodes allocated.
449  * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily
450  * differ during memory hotplug/hotremove operations.
451  * Protected by slab_mutex.
452  */
453 static nodemask_t slab_nodes;
454 
455 #ifndef CONFIG_SLUB_TINY
456 /*
457  * Workqueue used for flush_cpu_slab().
458  */
459 static struct workqueue_struct *flushwq;
460 #endif
461 
462 /********************************************************************
463  * 			Core slab cache functions
464  *******************************************************************/
465 
466 /*
467  * Returns freelist pointer (ptr). With hardening, this is obfuscated
468  * with an XOR of the address where the pointer is held and a per-cache
469  * random number.
470  */
freelist_ptr_encode(const struct kmem_cache * s,void * ptr,unsigned long ptr_addr)471 static inline freeptr_t freelist_ptr_encode(const struct kmem_cache *s,
472 					    void *ptr, unsigned long ptr_addr)
473 {
474 	unsigned long encoded;
475 
476 #ifdef CONFIG_SLAB_FREELIST_HARDENED
477 	encoded = (unsigned long)ptr ^ s->random ^ swab(ptr_addr);
478 #else
479 	encoded = (unsigned long)ptr;
480 #endif
481 	return (freeptr_t){.v = encoded};
482 }
483 
freelist_ptr_decode(const struct kmem_cache * s,freeptr_t ptr,unsigned long ptr_addr)484 static inline void *freelist_ptr_decode(const struct kmem_cache *s,
485 					freeptr_t ptr, unsigned long ptr_addr)
486 {
487 	void *decoded;
488 
489 #ifdef CONFIG_SLAB_FREELIST_HARDENED
490 	decoded = (void *)(ptr.v ^ s->random ^ swab(ptr_addr));
491 #else
492 	decoded = (void *)ptr.v;
493 #endif
494 	return decoded;
495 }
496 
get_freepointer(struct kmem_cache * s,void * object)497 static inline void *get_freepointer(struct kmem_cache *s, void *object)
498 {
499 	unsigned long ptr_addr;
500 	freeptr_t p;
501 
502 	object = kasan_reset_tag(object);
503 	ptr_addr = (unsigned long)object + s->offset;
504 	p = *(freeptr_t *)(ptr_addr);
505 	return freelist_ptr_decode(s, p, ptr_addr);
506 }
507 
508 #ifndef CONFIG_SLUB_TINY
prefetch_freepointer(const struct kmem_cache * s,void * object)509 static void prefetch_freepointer(const struct kmem_cache *s, void *object)
510 {
511 	prefetchw(object + s->offset);
512 }
513 #endif
514 
515 /*
516  * When running under KMSAN, get_freepointer_safe() may return an uninitialized
517  * pointer value in the case the current thread loses the race for the next
518  * memory chunk in the freelist. In that case this_cpu_cmpxchg_double() in
519  * slab_alloc_node() will fail, so the uninitialized value won't be used, but
520  * KMSAN will still check all arguments of cmpxchg because of imperfect
521  * handling of inline assembly.
522  * To work around this problem, we apply __no_kmsan_checks to ensure that
523  * get_freepointer_safe() returns initialized memory.
524  */
525 __no_kmsan_checks
get_freepointer_safe(struct kmem_cache * s,void * object)526 static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
527 {
528 	unsigned long freepointer_addr;
529 	freeptr_t p;
530 
531 	if (!debug_pagealloc_enabled_static())
532 		return get_freepointer(s, object);
533 
534 	object = kasan_reset_tag(object);
535 	freepointer_addr = (unsigned long)object + s->offset;
536 	copy_from_kernel_nofault(&p, (freeptr_t *)freepointer_addr, sizeof(p));
537 	return freelist_ptr_decode(s, p, freepointer_addr);
538 }
539 
set_freepointer(struct kmem_cache * s,void * object,void * fp)540 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
541 {
542 	unsigned long freeptr_addr = (unsigned long)object + s->offset;
543 
544 #ifdef CONFIG_SLAB_FREELIST_HARDENED
545 	BUG_ON(object == fp); /* naive detection of double free or corruption */
546 #endif
547 
548 	freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr);
549 	*(freeptr_t *)freeptr_addr = freelist_ptr_encode(s, fp, freeptr_addr);
550 }
551 
552 /*
553  * See comment in calculate_sizes().
554  */
freeptr_outside_object(struct kmem_cache * s)555 static inline bool freeptr_outside_object(struct kmem_cache *s)
556 {
557 	return s->offset >= s->inuse;
558 }
559 
560 /*
561  * Return offset of the end of info block which is inuse + free pointer if
562  * not overlapping with object.
563  */
get_info_end(struct kmem_cache * s)564 static inline unsigned int get_info_end(struct kmem_cache *s)
565 {
566 	if (freeptr_outside_object(s))
567 		return s->inuse + sizeof(void *);
568 	else
569 		return s->inuse;
570 }
571 
572 /* Loop over all objects in a slab */
573 #define for_each_object(__p, __s, __addr, __objects) \
574 	for (__p = fixup_red_left(__s, __addr); \
575 		__p < (__addr) + (__objects) * (__s)->size; \
576 		__p += (__s)->size)
577 
order_objects(unsigned int order,unsigned int size)578 static inline unsigned int order_objects(unsigned int order, unsigned int size)
579 {
580 	return ((unsigned int)PAGE_SIZE << order) / size;
581 }
582 
oo_make(unsigned int order,unsigned int size)583 static inline struct kmem_cache_order_objects oo_make(unsigned int order,
584 		unsigned int size)
585 {
586 	struct kmem_cache_order_objects x = {
587 		(order << OO_SHIFT) + order_objects(order, size)
588 	};
589 
590 	return x;
591 }
592 
oo_order(struct kmem_cache_order_objects x)593 static inline unsigned int oo_order(struct kmem_cache_order_objects x)
594 {
595 	return x.x >> OO_SHIFT;
596 }
597 
oo_objects(struct kmem_cache_order_objects x)598 static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
599 {
600 	return x.x & OO_MASK;
601 }
602 
603 #ifdef CONFIG_SLUB_CPU_PARTIAL
slub_set_cpu_partial(struct kmem_cache * s,unsigned int nr_objects)604 static void slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects)
605 {
606 	unsigned int nr_slabs;
607 
608 	s->cpu_partial = nr_objects;
609 
610 	/*
611 	 * We take the number of objects but actually limit the number of
612 	 * slabs on the per cpu partial list, in order to limit excessive
613 	 * growth of the list. For simplicity we assume that the slabs will
614 	 * be half-full.
615 	 */
616 	nr_slabs = DIV_ROUND_UP(nr_objects * 2, oo_objects(s->oo));
617 	s->cpu_partial_slabs = nr_slabs;
618 }
619 
slub_get_cpu_partial(struct kmem_cache * s)620 static inline unsigned int slub_get_cpu_partial(struct kmem_cache *s)
621 {
622 	return s->cpu_partial_slabs;
623 }
624 #else
625 static inline void
slub_set_cpu_partial(struct kmem_cache * s,unsigned int nr_objects)626 slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects)
627 {
628 }
629 
slub_get_cpu_partial(struct kmem_cache * s)630 static inline unsigned int slub_get_cpu_partial(struct kmem_cache *s)
631 {
632 	return 0;
633 }
634 #endif /* CONFIG_SLUB_CPU_PARTIAL */
635 
636 /*
637  * Per slab locking using the pagelock
638  */
slab_lock(struct slab * slab)639 static __always_inline void slab_lock(struct slab *slab)
640 {
641 	bit_spin_lock(PG_locked, &slab->__page_flags);
642 }
643 
slab_unlock(struct slab * slab)644 static __always_inline void slab_unlock(struct slab *slab)
645 {
646 	bit_spin_unlock(PG_locked, &slab->__page_flags);
647 }
648 
649 static inline bool
__update_freelist_fast(struct slab * slab,void * freelist_old,unsigned long counters_old,void * freelist_new,unsigned long counters_new)650 __update_freelist_fast(struct slab *slab,
651 		      void *freelist_old, unsigned long counters_old,
652 		      void *freelist_new, unsigned long counters_new)
653 {
654 #ifdef system_has_freelist_aba
655 	freelist_aba_t old = { .freelist = freelist_old, .counter = counters_old };
656 	freelist_aba_t new = { .freelist = freelist_new, .counter = counters_new };
657 
658 	return try_cmpxchg_freelist(&slab->freelist_counter.full, &old.full, new.full);
659 #else
660 	return false;
661 #endif
662 }
663 
664 static inline bool
__update_freelist_slow(struct slab * slab,void * freelist_old,unsigned long counters_old,void * freelist_new,unsigned long counters_new)665 __update_freelist_slow(struct slab *slab,
666 		      void *freelist_old, unsigned long counters_old,
667 		      void *freelist_new, unsigned long counters_new)
668 {
669 	bool ret = false;
670 
671 	slab_lock(slab);
672 	if (slab->freelist == freelist_old &&
673 	    slab->counters == counters_old) {
674 		slab->freelist = freelist_new;
675 		slab->counters = counters_new;
676 		ret = true;
677 	}
678 	slab_unlock(slab);
679 
680 	return ret;
681 }
682 
683 /*
684  * Interrupts must be disabled (for the fallback code to work right), typically
685  * by an _irqsave() lock variant. On PREEMPT_RT the preempt_disable(), which is
686  * part of bit_spin_lock(), is sufficient because the policy is not to allow any
687  * allocation/ free operation in hardirq context. Therefore nothing can
688  * interrupt the operation.
689  */
__slab_update_freelist(struct kmem_cache * s,struct slab * slab,void * freelist_old,unsigned long counters_old,void * freelist_new,unsigned long counters_new,const char * n)690 static inline bool __slab_update_freelist(struct kmem_cache *s, struct slab *slab,
691 		void *freelist_old, unsigned long counters_old,
692 		void *freelist_new, unsigned long counters_new,
693 		const char *n)
694 {
695 	bool ret;
696 
697 	if (USE_LOCKLESS_FAST_PATH())
698 		lockdep_assert_irqs_disabled();
699 
700 	if (s->flags & __CMPXCHG_DOUBLE) {
701 		ret = __update_freelist_fast(slab, freelist_old, counters_old,
702 				            freelist_new, counters_new);
703 	} else {
704 		ret = __update_freelist_slow(slab, freelist_old, counters_old,
705 				            freelist_new, counters_new);
706 	}
707 	if (likely(ret))
708 		return true;
709 
710 	cpu_relax();
711 	stat(s, CMPXCHG_DOUBLE_FAIL);
712 
713 #ifdef SLUB_DEBUG_CMPXCHG
714 	pr_info("%s %s: cmpxchg double redo ", n, s->name);
715 #endif
716 
717 	return false;
718 }
719 
slab_update_freelist(struct kmem_cache * s,struct slab * slab,void * freelist_old,unsigned long counters_old,void * freelist_new,unsigned long counters_new,const char * n)720 static inline bool slab_update_freelist(struct kmem_cache *s, struct slab *slab,
721 		void *freelist_old, unsigned long counters_old,
722 		void *freelist_new, unsigned long counters_new,
723 		const char *n)
724 {
725 	bool ret;
726 
727 	if (s->flags & __CMPXCHG_DOUBLE) {
728 		ret = __update_freelist_fast(slab, freelist_old, counters_old,
729 				            freelist_new, counters_new);
730 	} else {
731 		unsigned long flags;
732 
733 		local_irq_save(flags);
734 		ret = __update_freelist_slow(slab, freelist_old, counters_old,
735 				            freelist_new, counters_new);
736 		local_irq_restore(flags);
737 	}
738 	if (likely(ret))
739 		return true;
740 
741 	cpu_relax();
742 	stat(s, CMPXCHG_DOUBLE_FAIL);
743 
744 #ifdef SLUB_DEBUG_CMPXCHG
745 	pr_info("%s %s: cmpxchg double redo ", n, s->name);
746 #endif
747 
748 	return false;
749 }
750 
751 /*
752  * kmalloc caches has fixed sizes (mostly power of 2), and kmalloc() API
753  * family will round up the real request size to these fixed ones, so
754  * there could be an extra area than what is requested. Save the original
755  * request size in the meta data area, for better debug and sanity check.
756  */
set_orig_size(struct kmem_cache * s,void * object,unsigned int orig_size)757 static inline void set_orig_size(struct kmem_cache *s,
758 				void *object, unsigned int orig_size)
759 {
760 	void *p = kasan_reset_tag(object);
761 
762 	if (!slub_debug_orig_size(s))
763 		return;
764 
765 	p += get_info_end(s);
766 	p += sizeof(struct track) * 2;
767 
768 	*(unsigned int *)p = orig_size;
769 }
770 
get_orig_size(struct kmem_cache * s,void * object)771 static inline unsigned int get_orig_size(struct kmem_cache *s, void *object)
772 {
773 	void *p = kasan_reset_tag(object);
774 
775 	if (is_kfence_address(object))
776 		return kfence_ksize(object);
777 
778 	if (!slub_debug_orig_size(s))
779 		return s->object_size;
780 
781 	p += get_info_end(s);
782 	p += sizeof(struct track) * 2;
783 
784 	return *(unsigned int *)p;
785 }
786 
787 #ifdef CONFIG_SLUB_DEBUG
788 static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)];
789 static DEFINE_SPINLOCK(object_map_lock);
790 
__fill_map(unsigned long * obj_map,struct kmem_cache * s,struct slab * slab)791 static void __fill_map(unsigned long *obj_map, struct kmem_cache *s,
792 		       struct slab *slab)
793 {
794 	void *addr = slab_address(slab);
795 	void *p;
796 
797 	bitmap_zero(obj_map, slab->objects);
798 
799 	for (p = slab->freelist; p; p = get_freepointer(s, p))
800 		set_bit(__obj_to_index(s, addr, p), obj_map);
801 }
802 
803 #if IS_ENABLED(CONFIG_KUNIT)
slab_add_kunit_errors(void)804 static bool slab_add_kunit_errors(void)
805 {
806 	struct kunit_resource *resource;
807 
808 	if (!kunit_get_current_test())
809 		return false;
810 
811 	resource = kunit_find_named_resource(current->kunit_test, "slab_errors");
812 	if (!resource)
813 		return false;
814 
815 	(*(int *)resource->data)++;
816 	kunit_put_resource(resource);
817 	return true;
818 }
819 
slab_in_kunit_test(void)820 bool slab_in_kunit_test(void)
821 {
822 	struct kunit_resource *resource;
823 
824 	if (!kunit_get_current_test())
825 		return false;
826 
827 	resource = kunit_find_named_resource(current->kunit_test, "slab_errors");
828 	if (!resource)
829 		return false;
830 
831 	kunit_put_resource(resource);
832 	return true;
833 }
834 #else
slab_add_kunit_errors(void)835 static inline bool slab_add_kunit_errors(void) { return false; }
836 #endif
837 
size_from_object(struct kmem_cache * s)838 static inline unsigned int size_from_object(struct kmem_cache *s)
839 {
840 	if (s->flags & SLAB_RED_ZONE)
841 		return s->size - s->red_left_pad;
842 
843 	return s->size;
844 }
845 
restore_red_left(struct kmem_cache * s,void * p)846 static inline void *restore_red_left(struct kmem_cache *s, void *p)
847 {
848 	if (s->flags & SLAB_RED_ZONE)
849 		p -= s->red_left_pad;
850 
851 	return p;
852 }
853 
854 /*
855  * Debug settings:
856  */
857 #if defined(CONFIG_SLUB_DEBUG_ON)
858 static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
859 #else
860 static slab_flags_t slub_debug;
861 #endif
862 
863 static char *slub_debug_string;
864 static int disable_higher_order_debug;
865 
866 /*
867  * slub is about to manipulate internal object metadata.  This memory lies
868  * outside the range of the allocated object, so accessing it would normally
869  * be reported by kasan as a bounds error.  metadata_access_enable() is used
870  * to tell kasan that these accesses are OK.
871  */
metadata_access_enable(void)872 static inline void metadata_access_enable(void)
873 {
874 	kasan_disable_current();
875 	kmsan_disable_current();
876 }
877 
metadata_access_disable(void)878 static inline void metadata_access_disable(void)
879 {
880 	kmsan_enable_current();
881 	kasan_enable_current();
882 }
883 
884 /*
885  * Object debugging
886  */
887 
888 /* Verify that a pointer has an address that is valid within a slab page */
check_valid_pointer(struct kmem_cache * s,struct slab * slab,void * object)889 static inline int check_valid_pointer(struct kmem_cache *s,
890 				struct slab *slab, void *object)
891 {
892 	void *base;
893 
894 	if (!object)
895 		return 1;
896 
897 	base = slab_address(slab);
898 	object = kasan_reset_tag(object);
899 	object = restore_red_left(s, object);
900 	if (object < base || object >= base + slab->objects * s->size ||
901 		(object - base) % s->size) {
902 		return 0;
903 	}
904 
905 	return 1;
906 }
907 
print_section(char * level,char * text,u8 * addr,unsigned int length)908 static void print_section(char *level, char *text, u8 *addr,
909 			  unsigned int length)
910 {
911 	metadata_access_enable();
912 	print_hex_dump(level, text, DUMP_PREFIX_ADDRESS,
913 			16, 1, kasan_reset_tag((void *)addr), length, 1);
914 	metadata_access_disable();
915 }
916 
get_track(struct kmem_cache * s,void * object,enum track_item alloc)917 static struct track *get_track(struct kmem_cache *s, void *object,
918 	enum track_item alloc)
919 {
920 	struct track *p;
921 
922 	p = object + get_info_end(s);
923 
924 	return kasan_reset_tag(p + alloc);
925 }
926 
927 #ifdef CONFIG_STACKDEPOT
set_track_prepare(void)928 static noinline depot_stack_handle_t set_track_prepare(void)
929 {
930 	depot_stack_handle_t handle;
931 	unsigned long entries[TRACK_ADDRS_COUNT];
932 	unsigned int nr_entries;
933 
934 	nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 3);
935 	handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT);
936 
937 	return handle;
938 }
939 #else
set_track_prepare(void)940 static inline depot_stack_handle_t set_track_prepare(void)
941 {
942 	return 0;
943 }
944 #endif
945 
set_track_update(struct kmem_cache * s,void * object,enum track_item alloc,unsigned long addr,depot_stack_handle_t handle)946 static void set_track_update(struct kmem_cache *s, void *object,
947 			     enum track_item alloc, unsigned long addr,
948 			     depot_stack_handle_t handle)
949 {
950 	struct track *p = get_track(s, object, alloc);
951 
952 #ifdef CONFIG_STACKDEPOT
953 	p->handle = handle;
954 #endif
955 	p->addr = addr;
956 	p->cpu = smp_processor_id();
957 	p->pid = current->pid;
958 	p->when = jiffies;
959 }
960 
set_track(struct kmem_cache * s,void * object,enum track_item alloc,unsigned long addr)961 static __always_inline void set_track(struct kmem_cache *s, void *object,
962 				      enum track_item alloc, unsigned long addr)
963 {
964 	depot_stack_handle_t handle = set_track_prepare();
965 
966 	set_track_update(s, object, alloc, addr, handle);
967 }
968 
init_tracking(struct kmem_cache * s,void * object)969 static void init_tracking(struct kmem_cache *s, void *object)
970 {
971 	struct track *p;
972 
973 	if (!(s->flags & SLAB_STORE_USER))
974 		return;
975 
976 	p = get_track(s, object, TRACK_ALLOC);
977 	memset(p, 0, 2*sizeof(struct track));
978 }
979 
print_track(const char * s,struct track * t,unsigned long pr_time)980 static void print_track(const char *s, struct track *t, unsigned long pr_time)
981 {
982 	depot_stack_handle_t handle __maybe_unused;
983 
984 	if (!t->addr)
985 		return;
986 
987 	pr_err("%s in %pS age=%lu cpu=%u pid=%d\n",
988 	       s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
989 #ifdef CONFIG_STACKDEPOT
990 	handle = READ_ONCE(t->handle);
991 	if (handle)
992 		stack_depot_print(handle);
993 	else
994 		pr_err("object allocation/free stack trace missing\n");
995 #endif
996 }
997 
print_tracking(struct kmem_cache * s,void * object)998 void print_tracking(struct kmem_cache *s, void *object)
999 {
1000 	unsigned long pr_time = jiffies;
1001 	if (!(s->flags & SLAB_STORE_USER))
1002 		return;
1003 
1004 	print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
1005 	print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
1006 }
1007 
print_slab_info(const struct slab * slab)1008 static void print_slab_info(const struct slab *slab)
1009 {
1010 	pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%pGp\n",
1011 	       slab, slab->objects, slab->inuse, slab->freelist,
1012 	       &slab->__page_flags);
1013 }
1014 
skip_orig_size_check(struct kmem_cache * s,const void * object)1015 void skip_orig_size_check(struct kmem_cache *s, const void *object)
1016 {
1017 	set_orig_size(s, (void *)object, s->object_size);
1018 }
1019 
slab_bug(struct kmem_cache * s,char * fmt,...)1020 static void slab_bug(struct kmem_cache *s, char *fmt, ...)
1021 {
1022 	struct va_format vaf;
1023 	va_list args;
1024 
1025 	va_start(args, fmt);
1026 	vaf.fmt = fmt;
1027 	vaf.va = &args;
1028 	pr_err("=============================================================================\n");
1029 	pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
1030 	pr_err("-----------------------------------------------------------------------------\n\n");
1031 	va_end(args);
1032 }
1033 
1034 __printf(2, 3)
slab_fix(struct kmem_cache * s,char * fmt,...)1035 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
1036 {
1037 	struct va_format vaf;
1038 	va_list args;
1039 
1040 	if (slab_add_kunit_errors())
1041 		return;
1042 
1043 	va_start(args, fmt);
1044 	vaf.fmt = fmt;
1045 	vaf.va = &args;
1046 	pr_err("FIX %s: %pV\n", s->name, &vaf);
1047 	va_end(args);
1048 }
1049 
print_trailer(struct kmem_cache * s,struct slab * slab,u8 * p)1050 static void print_trailer(struct kmem_cache *s, struct slab *slab, u8 *p)
1051 {
1052 	unsigned int off;	/* Offset of last byte */
1053 	u8 *addr = slab_address(slab);
1054 
1055 	print_tracking(s, p);
1056 
1057 	print_slab_info(slab);
1058 
1059 	pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n",
1060 	       p, p - addr, get_freepointer(s, p));
1061 
1062 	if (s->flags & SLAB_RED_ZONE)
1063 		print_section(KERN_ERR, "Redzone  ", p - s->red_left_pad,
1064 			      s->red_left_pad);
1065 	else if (p > addr + 16)
1066 		print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
1067 
1068 	print_section(KERN_ERR,         "Object   ", p,
1069 		      min_t(unsigned int, s->object_size, PAGE_SIZE));
1070 	if (s->flags & SLAB_RED_ZONE)
1071 		print_section(KERN_ERR, "Redzone  ", p + s->object_size,
1072 			s->inuse - s->object_size);
1073 
1074 	off = get_info_end(s);
1075 
1076 	if (s->flags & SLAB_STORE_USER)
1077 		off += 2 * sizeof(struct track);
1078 
1079 	if (slub_debug_orig_size(s))
1080 		off += sizeof(unsigned int);
1081 
1082 	off += kasan_metadata_size(s, false);
1083 
1084 	if (off != size_from_object(s))
1085 		/* Beginning of the filler is the free pointer */
1086 		print_section(KERN_ERR, "Padding  ", p + off,
1087 			      size_from_object(s) - off);
1088 
1089 	dump_stack();
1090 }
1091 
object_err(struct kmem_cache * s,struct slab * slab,u8 * object,char * reason)1092 static void object_err(struct kmem_cache *s, struct slab *slab,
1093 			u8 *object, char *reason)
1094 {
1095 	if (slab_add_kunit_errors())
1096 		return;
1097 
1098 	slab_bug(s, "%s", reason);
1099 	print_trailer(s, slab, object);
1100 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
1101 }
1102 
freelist_corrupted(struct kmem_cache * s,struct slab * slab,void ** freelist,void * nextfree)1103 static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab,
1104 			       void **freelist, void *nextfree)
1105 {
1106 	if ((s->flags & SLAB_CONSISTENCY_CHECKS) &&
1107 	    !check_valid_pointer(s, slab, nextfree) && freelist) {
1108 		object_err(s, slab, *freelist, "Freechain corrupt");
1109 		*freelist = NULL;
1110 		slab_fix(s, "Isolate corrupted freechain");
1111 		return true;
1112 	}
1113 
1114 	return false;
1115 }
1116 
slab_err(struct kmem_cache * s,struct slab * slab,const char * fmt,...)1117 static __printf(3, 4) void slab_err(struct kmem_cache *s, struct slab *slab,
1118 			const char *fmt, ...)
1119 {
1120 	va_list args;
1121 	char buf[100];
1122 
1123 	if (slab_add_kunit_errors())
1124 		return;
1125 
1126 	va_start(args, fmt);
1127 	vsnprintf(buf, sizeof(buf), fmt, args);
1128 	va_end(args);
1129 	slab_bug(s, "%s", buf);
1130 	print_slab_info(slab);
1131 	dump_stack();
1132 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
1133 }
1134 
init_object(struct kmem_cache * s,void * object,u8 val)1135 static void init_object(struct kmem_cache *s, void *object, u8 val)
1136 {
1137 	u8 *p = kasan_reset_tag(object);
1138 	unsigned int poison_size = s->object_size;
1139 
1140 	if (s->flags & SLAB_RED_ZONE) {
1141 		/*
1142 		 * Here and below, avoid overwriting the KMSAN shadow. Keeping
1143 		 * the shadow makes it possible to distinguish uninit-value
1144 		 * from use-after-free.
1145 		 */
1146 		memset_no_sanitize_memory(p - s->red_left_pad, val,
1147 					  s->red_left_pad);
1148 
1149 		if (slub_debug_orig_size(s) && val == SLUB_RED_ACTIVE) {
1150 			/*
1151 			 * Redzone the extra allocated space by kmalloc than
1152 			 * requested, and the poison size will be limited to
1153 			 * the original request size accordingly.
1154 			 */
1155 			poison_size = get_orig_size(s, object);
1156 		}
1157 	}
1158 
1159 	if (s->flags & __OBJECT_POISON) {
1160 		memset_no_sanitize_memory(p, POISON_FREE, poison_size - 1);
1161 		memset_no_sanitize_memory(p + poison_size - 1, POISON_END, 1);
1162 	}
1163 
1164 	if (s->flags & SLAB_RED_ZONE)
1165 		memset_no_sanitize_memory(p + poison_size, val,
1166 					  s->inuse - poison_size);
1167 }
1168 
restore_bytes(struct kmem_cache * s,char * message,u8 data,void * from,void * to)1169 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
1170 						void *from, void *to)
1171 {
1172 	slab_fix(s, "Restoring %s 0x%p-0x%p=0x%x", message, from, to - 1, data);
1173 	memset(from, data, to - from);
1174 }
1175 
1176 #ifdef CONFIG_KMSAN
1177 #define pad_check_attributes noinline __no_kmsan_checks
1178 #else
1179 #define pad_check_attributes
1180 #endif
1181 
1182 static pad_check_attributes int
check_bytes_and_report(struct kmem_cache * s,struct slab * slab,u8 * object,char * what,u8 * start,unsigned int value,unsigned int bytes)1183 check_bytes_and_report(struct kmem_cache *s, struct slab *slab,
1184 		       u8 *object, char *what,
1185 		       u8 *start, unsigned int value, unsigned int bytes)
1186 {
1187 	u8 *fault;
1188 	u8 *end;
1189 	u8 *addr = slab_address(slab);
1190 
1191 	metadata_access_enable();
1192 	fault = memchr_inv(kasan_reset_tag(start), value, bytes);
1193 	metadata_access_disable();
1194 	if (!fault)
1195 		return 1;
1196 
1197 	end = start + bytes;
1198 	while (end > fault && end[-1] == value)
1199 		end--;
1200 
1201 	if (slab_add_kunit_errors())
1202 		goto skip_bug_print;
1203 
1204 	slab_bug(s, "%s overwritten", what);
1205 	pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n",
1206 					fault, end - 1, fault - addr,
1207 					fault[0], value);
1208 
1209 skip_bug_print:
1210 	restore_bytes(s, what, value, fault, end);
1211 	return 0;
1212 }
1213 
1214 /*
1215  * Object layout:
1216  *
1217  * object address
1218  * 	Bytes of the object to be managed.
1219  * 	If the freepointer may overlay the object then the free
1220  *	pointer is at the middle of the object.
1221  *
1222  * 	Poisoning uses 0x6b (POISON_FREE) and the last byte is
1223  * 	0xa5 (POISON_END)
1224  *
1225  * object + s->object_size
1226  * 	Padding to reach word boundary. This is also used for Redzoning.
1227  * 	Padding is extended by another word if Redzoning is enabled and
1228  * 	object_size == inuse.
1229  *
1230  * 	We fill with 0xbb (SLUB_RED_INACTIVE) for inactive objects and with
1231  * 	0xcc (SLUB_RED_ACTIVE) for objects in use.
1232  *
1233  * object + s->inuse
1234  * 	Meta data starts here.
1235  *
1236  * 	A. Free pointer (if we cannot overwrite object on free)
1237  * 	B. Tracking data for SLAB_STORE_USER
1238  *	C. Original request size for kmalloc object (SLAB_STORE_USER enabled)
1239  *	D. Padding to reach required alignment boundary or at minimum
1240  * 		one word if debugging is on to be able to detect writes
1241  * 		before the word boundary.
1242  *
1243  *	Padding is done using 0x5a (POISON_INUSE)
1244  *
1245  * object + s->size
1246  * 	Nothing is used beyond s->size.
1247  *
1248  * If slabcaches are merged then the object_size and inuse boundaries are mostly
1249  * ignored. And therefore no slab options that rely on these boundaries
1250  * may be used with merged slabcaches.
1251  */
1252 
check_pad_bytes(struct kmem_cache * s,struct slab * slab,u8 * p)1253 static int check_pad_bytes(struct kmem_cache *s, struct slab *slab, u8 *p)
1254 {
1255 	unsigned long off = get_info_end(s);	/* The end of info */
1256 
1257 	if (s->flags & SLAB_STORE_USER) {
1258 		/* We also have user information there */
1259 		off += 2 * sizeof(struct track);
1260 
1261 		if (s->flags & SLAB_KMALLOC)
1262 			off += sizeof(unsigned int);
1263 	}
1264 
1265 	off += kasan_metadata_size(s, false);
1266 
1267 	if (size_from_object(s) == off)
1268 		return 1;
1269 
1270 	return check_bytes_and_report(s, slab, p, "Object padding",
1271 			p + off, POISON_INUSE, size_from_object(s) - off);
1272 }
1273 
1274 /* Check the pad bytes at the end of a slab page */
1275 static pad_check_attributes void
slab_pad_check(struct kmem_cache * s,struct slab * slab)1276 slab_pad_check(struct kmem_cache *s, struct slab *slab)
1277 {
1278 	u8 *start;
1279 	u8 *fault;
1280 	u8 *end;
1281 	u8 *pad;
1282 	int length;
1283 	int remainder;
1284 
1285 	if (!(s->flags & SLAB_POISON))
1286 		return;
1287 
1288 	start = slab_address(slab);
1289 	length = slab_size(slab);
1290 	end = start + length;
1291 	remainder = length % s->size;
1292 	if (!remainder)
1293 		return;
1294 
1295 	pad = end - remainder;
1296 	metadata_access_enable();
1297 	fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder);
1298 	metadata_access_disable();
1299 	if (!fault)
1300 		return;
1301 	while (end > fault && end[-1] == POISON_INUSE)
1302 		end--;
1303 
1304 	slab_err(s, slab, "Padding overwritten. 0x%p-0x%p @offset=%tu",
1305 			fault, end - 1, fault - start);
1306 	print_section(KERN_ERR, "Padding ", pad, remainder);
1307 
1308 	restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
1309 }
1310 
check_object(struct kmem_cache * s,struct slab * slab,void * object,u8 val)1311 static int check_object(struct kmem_cache *s, struct slab *slab,
1312 					void *object, u8 val)
1313 {
1314 	u8 *p = object;
1315 	u8 *endobject = object + s->object_size;
1316 	unsigned int orig_size, kasan_meta_size;
1317 	int ret = 1;
1318 
1319 	if (s->flags & SLAB_RED_ZONE) {
1320 		if (!check_bytes_and_report(s, slab, object, "Left Redzone",
1321 			object - s->red_left_pad, val, s->red_left_pad))
1322 			ret = 0;
1323 
1324 		if (!check_bytes_and_report(s, slab, object, "Right Redzone",
1325 			endobject, val, s->inuse - s->object_size))
1326 			ret = 0;
1327 
1328 		if (slub_debug_orig_size(s) && val == SLUB_RED_ACTIVE) {
1329 			orig_size = get_orig_size(s, object);
1330 
1331 			if (s->object_size > orig_size  &&
1332 				!check_bytes_and_report(s, slab, object,
1333 					"kmalloc Redzone", p + orig_size,
1334 					val, s->object_size - orig_size)) {
1335 				ret = 0;
1336 			}
1337 		}
1338 	} else {
1339 		if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
1340 			if (!check_bytes_and_report(s, slab, p, "Alignment padding",
1341 				endobject, POISON_INUSE,
1342 				s->inuse - s->object_size))
1343 				ret = 0;
1344 		}
1345 	}
1346 
1347 	if (s->flags & SLAB_POISON) {
1348 		if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON)) {
1349 			/*
1350 			 * KASAN can save its free meta data inside of the
1351 			 * object at offset 0. Thus, skip checking the part of
1352 			 * the redzone that overlaps with the meta data.
1353 			 */
1354 			kasan_meta_size = kasan_metadata_size(s, true);
1355 			if (kasan_meta_size < s->object_size - 1 &&
1356 			    !check_bytes_and_report(s, slab, p, "Poison",
1357 					p + kasan_meta_size, POISON_FREE,
1358 					s->object_size - kasan_meta_size - 1))
1359 				ret = 0;
1360 			if (kasan_meta_size < s->object_size &&
1361 			    !check_bytes_and_report(s, slab, p, "End Poison",
1362 					p + s->object_size - 1, POISON_END, 1))
1363 				ret = 0;
1364 		}
1365 		/*
1366 		 * check_pad_bytes cleans up on its own.
1367 		 */
1368 		if (!check_pad_bytes(s, slab, p))
1369 			ret = 0;
1370 	}
1371 
1372 	/*
1373 	 * Cannot check freepointer while object is allocated if
1374 	 * object and freepointer overlap.
1375 	 */
1376 	if ((freeptr_outside_object(s) || val != SLUB_RED_ACTIVE) &&
1377 	    !check_valid_pointer(s, slab, get_freepointer(s, p))) {
1378 		object_err(s, slab, p, "Freepointer corrupt");
1379 		/*
1380 		 * No choice but to zap it and thus lose the remainder
1381 		 * of the free objects in this slab. May cause
1382 		 * another error because the object count is now wrong.
1383 		 */
1384 		set_freepointer(s, p, NULL);
1385 		ret = 0;
1386 	}
1387 
1388 	if (!ret && !slab_in_kunit_test()) {
1389 		print_trailer(s, slab, object);
1390 		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
1391 	}
1392 
1393 	return ret;
1394 }
1395 
check_slab(struct kmem_cache * s,struct slab * slab)1396 static int check_slab(struct kmem_cache *s, struct slab *slab)
1397 {
1398 	int maxobj;
1399 
1400 	if (!folio_test_slab(slab_folio(slab))) {
1401 		slab_err(s, slab, "Not a valid slab page");
1402 		return 0;
1403 	}
1404 
1405 	maxobj = order_objects(slab_order(slab), s->size);
1406 	if (slab->objects > maxobj) {
1407 		slab_err(s, slab, "objects %u > max %u",
1408 			slab->objects, maxobj);
1409 		return 0;
1410 	}
1411 	if (slab->inuse > slab->objects) {
1412 		slab_err(s, slab, "inuse %u > max %u",
1413 			slab->inuse, slab->objects);
1414 		return 0;
1415 	}
1416 	if (slab->frozen) {
1417 		slab_err(s, slab, "Slab disabled since SLUB metadata consistency check failed");
1418 		return 0;
1419 	}
1420 
1421 	/* Slab_pad_check fixes things up after itself */
1422 	slab_pad_check(s, slab);
1423 	return 1;
1424 }
1425 
1426 /*
1427  * Determine if a certain object in a slab is on the freelist. Must hold the
1428  * slab lock to guarantee that the chains are in a consistent state.
1429  */
on_freelist(struct kmem_cache * s,struct slab * slab,void * search)1430 static int on_freelist(struct kmem_cache *s, struct slab *slab, void *search)
1431 {
1432 	int nr = 0;
1433 	void *fp;
1434 	void *object = NULL;
1435 	int max_objects;
1436 
1437 	fp = slab->freelist;
1438 	while (fp && nr <= slab->objects) {
1439 		if (fp == search)
1440 			return 1;
1441 		if (!check_valid_pointer(s, slab, fp)) {
1442 			if (object) {
1443 				object_err(s, slab, object,
1444 					"Freechain corrupt");
1445 				set_freepointer(s, object, NULL);
1446 			} else {
1447 				slab_err(s, slab, "Freepointer corrupt");
1448 				slab->freelist = NULL;
1449 				slab->inuse = slab->objects;
1450 				slab_fix(s, "Freelist cleared");
1451 				return 0;
1452 			}
1453 			break;
1454 		}
1455 		object = fp;
1456 		fp = get_freepointer(s, object);
1457 		nr++;
1458 	}
1459 
1460 	max_objects = order_objects(slab_order(slab), s->size);
1461 	if (max_objects > MAX_OBJS_PER_PAGE)
1462 		max_objects = MAX_OBJS_PER_PAGE;
1463 
1464 	if (slab->objects != max_objects) {
1465 		slab_err(s, slab, "Wrong number of objects. Found %d but should be %d",
1466 			 slab->objects, max_objects);
1467 		slab->objects = max_objects;
1468 		slab_fix(s, "Number of objects adjusted");
1469 	}
1470 	if (slab->inuse != slab->objects - nr) {
1471 		slab_err(s, slab, "Wrong object count. Counter is %d but counted were %d",
1472 			 slab->inuse, slab->objects - nr);
1473 		slab->inuse = slab->objects - nr;
1474 		slab_fix(s, "Object count adjusted");
1475 	}
1476 	return search == NULL;
1477 }
1478 
trace(struct kmem_cache * s,struct slab * slab,void * object,int alloc)1479 static void trace(struct kmem_cache *s, struct slab *slab, void *object,
1480 								int alloc)
1481 {
1482 	if (s->flags & SLAB_TRACE) {
1483 		pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
1484 			s->name,
1485 			alloc ? "alloc" : "free",
1486 			object, slab->inuse,
1487 			slab->freelist);
1488 
1489 		if (!alloc)
1490 			print_section(KERN_INFO, "Object ", (void *)object,
1491 					s->object_size);
1492 
1493 		dump_stack();
1494 	}
1495 }
1496 
1497 /*
1498  * Tracking of fully allocated slabs for debugging purposes.
1499  */
add_full(struct kmem_cache * s,struct kmem_cache_node * n,struct slab * slab)1500 static void add_full(struct kmem_cache *s,
1501 	struct kmem_cache_node *n, struct slab *slab)
1502 {
1503 	if (!(s->flags & SLAB_STORE_USER))
1504 		return;
1505 
1506 	lockdep_assert_held(&n->list_lock);
1507 	list_add(&slab->slab_list, &n->full);
1508 }
1509 
remove_full(struct kmem_cache * s,struct kmem_cache_node * n,struct slab * slab)1510 static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct slab *slab)
1511 {
1512 	if (!(s->flags & SLAB_STORE_USER))
1513 		return;
1514 
1515 	lockdep_assert_held(&n->list_lock);
1516 	list_del(&slab->slab_list);
1517 }
1518 
node_nr_slabs(struct kmem_cache_node * n)1519 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1520 {
1521 	return atomic_long_read(&n->nr_slabs);
1522 }
1523 
inc_slabs_node(struct kmem_cache * s,int node,int objects)1524 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
1525 {
1526 	struct kmem_cache_node *n = get_node(s, node);
1527 
1528 	atomic_long_inc(&n->nr_slabs);
1529 	atomic_long_add(objects, &n->total_objects);
1530 }
dec_slabs_node(struct kmem_cache * s,int node,int objects)1531 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
1532 {
1533 	struct kmem_cache_node *n = get_node(s, node);
1534 
1535 	atomic_long_dec(&n->nr_slabs);
1536 	atomic_long_sub(objects, &n->total_objects);
1537 }
1538 
1539 /* Object debug checks for alloc/free paths */
setup_object_debug(struct kmem_cache * s,void * object)1540 static void setup_object_debug(struct kmem_cache *s, void *object)
1541 {
1542 	if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))
1543 		return;
1544 
1545 	init_object(s, object, SLUB_RED_INACTIVE);
1546 	init_tracking(s, object);
1547 }
1548 
1549 static
setup_slab_debug(struct kmem_cache * s,struct slab * slab,void * addr)1550 void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr)
1551 {
1552 	if (!kmem_cache_debug_flags(s, SLAB_POISON))
1553 		return;
1554 
1555 	metadata_access_enable();
1556 	memset(kasan_reset_tag(addr), POISON_INUSE, slab_size(slab));
1557 	metadata_access_disable();
1558 }
1559 
alloc_consistency_checks(struct kmem_cache * s,struct slab * slab,void * object)1560 static inline int alloc_consistency_checks(struct kmem_cache *s,
1561 					struct slab *slab, void *object)
1562 {
1563 	if (!check_slab(s, slab))
1564 		return 0;
1565 
1566 	if (!check_valid_pointer(s, slab, object)) {
1567 		object_err(s, slab, object, "Freelist Pointer check fails");
1568 		return 0;
1569 	}
1570 
1571 	if (!check_object(s, slab, object, SLUB_RED_INACTIVE))
1572 		return 0;
1573 
1574 	return 1;
1575 }
1576 
alloc_debug_processing(struct kmem_cache * s,struct slab * slab,void * object,int orig_size)1577 static noinline bool alloc_debug_processing(struct kmem_cache *s,
1578 			struct slab *slab, void *object, int orig_size)
1579 {
1580 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1581 		if (!alloc_consistency_checks(s, slab, object))
1582 			goto bad;
1583 	}
1584 
1585 	/* Success. Perform special debug activities for allocs */
1586 	trace(s, slab, object, 1);
1587 	set_orig_size(s, object, orig_size);
1588 	init_object(s, object, SLUB_RED_ACTIVE);
1589 	return true;
1590 
1591 bad:
1592 	if (folio_test_slab(slab_folio(slab))) {
1593 		/*
1594 		 * If this is a slab page then lets do the best we can
1595 		 * to avoid issues in the future. Marking all objects
1596 		 * as used avoids touching the remaining objects.
1597 		 */
1598 		slab_fix(s, "Marking all objects used");
1599 		slab->inuse = slab->objects;
1600 		slab->freelist = NULL;
1601 		slab->frozen = 1; /* mark consistency-failed slab as frozen */
1602 	}
1603 	return false;
1604 }
1605 
free_consistency_checks(struct kmem_cache * s,struct slab * slab,void * object,unsigned long addr)1606 static inline int free_consistency_checks(struct kmem_cache *s,
1607 		struct slab *slab, void *object, unsigned long addr)
1608 {
1609 	if (!check_valid_pointer(s, slab, object)) {
1610 		slab_err(s, slab, "Invalid object pointer 0x%p", object);
1611 		return 0;
1612 	}
1613 
1614 	if (on_freelist(s, slab, object)) {
1615 		object_err(s, slab, object, "Object already free");
1616 		return 0;
1617 	}
1618 
1619 	if (!check_object(s, slab, object, SLUB_RED_ACTIVE))
1620 		return 0;
1621 
1622 	if (unlikely(s != slab->slab_cache)) {
1623 		if (!folio_test_slab(slab_folio(slab))) {
1624 			slab_err(s, slab, "Attempt to free object(0x%p) outside of slab",
1625 				 object);
1626 		} else if (!slab->slab_cache) {
1627 			pr_err("SLUB <none>: no slab for object 0x%p.\n",
1628 			       object);
1629 			dump_stack();
1630 		} else
1631 			object_err(s, slab, object,
1632 					"page slab pointer corrupt.");
1633 		return 0;
1634 	}
1635 	return 1;
1636 }
1637 
1638 /*
1639  * Parse a block of slab_debug options. Blocks are delimited by ';'
1640  *
1641  * @str:    start of block
1642  * @flags:  returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified
1643  * @slabs:  return start of list of slabs, or NULL when there's no list
1644  * @init:   assume this is initial parsing and not per-kmem-create parsing
1645  *
1646  * returns the start of next block if there's any, or NULL
1647  */
1648 static char *
parse_slub_debug_flags(char * str,slab_flags_t * flags,char ** slabs,bool init)1649 parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init)
1650 {
1651 	bool higher_order_disable = false;
1652 
1653 	/* Skip any completely empty blocks */
1654 	while (*str && *str == ';')
1655 		str++;
1656 
1657 	if (*str == ',') {
1658 		/*
1659 		 * No options but restriction on slabs. This means full
1660 		 * debugging for slabs matching a pattern.
1661 		 */
1662 		*flags = DEBUG_DEFAULT_FLAGS;
1663 		goto check_slabs;
1664 	}
1665 	*flags = 0;
1666 
1667 	/* Determine which debug features should be switched on */
1668 	for (; *str && *str != ',' && *str != ';'; str++) {
1669 		switch (tolower(*str)) {
1670 		case '-':
1671 			*flags = 0;
1672 			break;
1673 		case 'f':
1674 			*flags |= SLAB_CONSISTENCY_CHECKS;
1675 			break;
1676 		case 'z':
1677 			*flags |= SLAB_RED_ZONE;
1678 			break;
1679 		case 'p':
1680 			*flags |= SLAB_POISON;
1681 			break;
1682 		case 'u':
1683 			*flags |= SLAB_STORE_USER;
1684 			break;
1685 		case 't':
1686 			*flags |= SLAB_TRACE;
1687 			break;
1688 		case 'a':
1689 			*flags |= SLAB_FAILSLAB;
1690 			break;
1691 		case 'o':
1692 			/*
1693 			 * Avoid enabling debugging on caches if its minimum
1694 			 * order would increase as a result.
1695 			 */
1696 			higher_order_disable = true;
1697 			break;
1698 		default:
1699 			if (init)
1700 				pr_err("slab_debug option '%c' unknown. skipped\n", *str);
1701 		}
1702 	}
1703 check_slabs:
1704 	if (*str == ',')
1705 		*slabs = ++str;
1706 	else
1707 		*slabs = NULL;
1708 
1709 	/* Skip over the slab list */
1710 	while (*str && *str != ';')
1711 		str++;
1712 
1713 	/* Skip any completely empty blocks */
1714 	while (*str && *str == ';')
1715 		str++;
1716 
1717 	if (init && higher_order_disable)
1718 		disable_higher_order_debug = 1;
1719 
1720 	if (*str)
1721 		return str;
1722 	else
1723 		return NULL;
1724 }
1725 
setup_slub_debug(char * str)1726 static int __init setup_slub_debug(char *str)
1727 {
1728 	slab_flags_t flags;
1729 	slab_flags_t global_flags;
1730 	char *saved_str;
1731 	char *slab_list;
1732 	bool global_slub_debug_changed = false;
1733 	bool slab_list_specified = false;
1734 
1735 	global_flags = DEBUG_DEFAULT_FLAGS;
1736 	if (*str++ != '=' || !*str)
1737 		/*
1738 		 * No options specified. Switch on full debugging.
1739 		 */
1740 		goto out;
1741 
1742 	saved_str = str;
1743 	while (str) {
1744 		str = parse_slub_debug_flags(str, &flags, &slab_list, true);
1745 
1746 		if (!slab_list) {
1747 			global_flags = flags;
1748 			global_slub_debug_changed = true;
1749 		} else {
1750 			slab_list_specified = true;
1751 			if (flags & SLAB_STORE_USER)
1752 				stack_depot_request_early_init();
1753 		}
1754 	}
1755 
1756 	/*
1757 	 * For backwards compatibility, a single list of flags with list of
1758 	 * slabs means debugging is only changed for those slabs, so the global
1759 	 * slab_debug should be unchanged (0 or DEBUG_DEFAULT_FLAGS, depending
1760 	 * on CONFIG_SLUB_DEBUG_ON). We can extended that to multiple lists as
1761 	 * long as there is no option specifying flags without a slab list.
1762 	 */
1763 	if (slab_list_specified) {
1764 		if (!global_slub_debug_changed)
1765 			global_flags = slub_debug;
1766 		slub_debug_string = saved_str;
1767 	}
1768 out:
1769 	slub_debug = global_flags;
1770 	if (slub_debug & SLAB_STORE_USER)
1771 		stack_depot_request_early_init();
1772 	if (slub_debug != 0 || slub_debug_string)
1773 		static_branch_enable(&slub_debug_enabled);
1774 	else
1775 		static_branch_disable(&slub_debug_enabled);
1776 	if ((static_branch_unlikely(&init_on_alloc) ||
1777 	     static_branch_unlikely(&init_on_free)) &&
1778 	    (slub_debug & SLAB_POISON))
1779 		pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n");
1780 	return 1;
1781 }
1782 
1783 __setup("slab_debug", setup_slub_debug);
1784 __setup_param("slub_debug", slub_debug, setup_slub_debug, 0);
1785 
1786 /*
1787  * kmem_cache_flags - apply debugging options to the cache
1788  * @flags:		flags to set
1789  * @name:		name of the cache
1790  *
1791  * Debug option(s) are applied to @flags. In addition to the debug
1792  * option(s), if a slab name (or multiple) is specified i.e.
1793  * slab_debug=<Debug-Options>,<slab name1>,<slab name2> ...
1794  * then only the select slabs will receive the debug option(s).
1795  */
kmem_cache_flags(slab_flags_t flags,const char * name)1796 slab_flags_t kmem_cache_flags(slab_flags_t flags, const char *name)
1797 {
1798 	char *iter;
1799 	size_t len;
1800 	char *next_block;
1801 	slab_flags_t block_flags;
1802 	slab_flags_t slub_debug_local = slub_debug;
1803 
1804 	if (flags & SLAB_NO_USER_FLAGS)
1805 		return flags;
1806 
1807 	/*
1808 	 * If the slab cache is for debugging (e.g. kmemleak) then
1809 	 * don't store user (stack trace) information by default,
1810 	 * but let the user enable it via the command line below.
1811 	 */
1812 	if (flags & SLAB_NOLEAKTRACE)
1813 		slub_debug_local &= ~SLAB_STORE_USER;
1814 
1815 	len = strlen(name);
1816 	next_block = slub_debug_string;
1817 	/* Go through all blocks of debug options, see if any matches our slab's name */
1818 	while (next_block) {
1819 		next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false);
1820 		if (!iter)
1821 			continue;
1822 		/* Found a block that has a slab list, search it */
1823 		while (*iter) {
1824 			char *end, *glob;
1825 			size_t cmplen;
1826 
1827 			end = strchrnul(iter, ',');
1828 			if (next_block && next_block < end)
1829 				end = next_block - 1;
1830 
1831 			glob = strnchr(iter, end - iter, '*');
1832 			if (glob)
1833 				cmplen = glob - iter;
1834 			else
1835 				cmplen = max_t(size_t, len, (end - iter));
1836 
1837 			if (!strncmp(name, iter, cmplen)) {
1838 				flags |= block_flags;
1839 				return flags;
1840 			}
1841 
1842 			if (!*end || *end == ';')
1843 				break;
1844 			iter = end + 1;
1845 		}
1846 	}
1847 
1848 	return flags | slub_debug_local;
1849 }
1850 #else /* !CONFIG_SLUB_DEBUG */
setup_object_debug(struct kmem_cache * s,void * object)1851 static inline void setup_object_debug(struct kmem_cache *s, void *object) {}
1852 static inline
setup_slab_debug(struct kmem_cache * s,struct slab * slab,void * addr)1853 void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr) {}
1854 
alloc_debug_processing(struct kmem_cache * s,struct slab * slab,void * object,int orig_size)1855 static inline bool alloc_debug_processing(struct kmem_cache *s,
1856 	struct slab *slab, void *object, int orig_size) { return true; }
1857 
free_debug_processing(struct kmem_cache * s,struct slab * slab,void * head,void * tail,int * bulk_cnt,unsigned long addr,depot_stack_handle_t handle)1858 static inline bool free_debug_processing(struct kmem_cache *s,
1859 	struct slab *slab, void *head, void *tail, int *bulk_cnt,
1860 	unsigned long addr, depot_stack_handle_t handle) { return true; }
1861 
slab_pad_check(struct kmem_cache * s,struct slab * slab)1862 static inline void slab_pad_check(struct kmem_cache *s, struct slab *slab) {}
check_object(struct kmem_cache * s,struct slab * slab,void * object,u8 val)1863 static inline int check_object(struct kmem_cache *s, struct slab *slab,
1864 			void *object, u8 val) { return 1; }
set_track_prepare(void)1865 static inline depot_stack_handle_t set_track_prepare(void) { return 0; }
set_track(struct kmem_cache * s,void * object,enum track_item alloc,unsigned long addr)1866 static inline void set_track(struct kmem_cache *s, void *object,
1867 			     enum track_item alloc, unsigned long addr) {}
add_full(struct kmem_cache * s,struct kmem_cache_node * n,struct slab * slab)1868 static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1869 					struct slab *slab) {}
remove_full(struct kmem_cache * s,struct kmem_cache_node * n,struct slab * slab)1870 static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1871 					struct slab *slab) {}
kmem_cache_flags(slab_flags_t flags,const char * name)1872 slab_flags_t kmem_cache_flags(slab_flags_t flags, const char *name)
1873 {
1874 	return flags;
1875 }
1876 #define slub_debug 0
1877 
1878 #define disable_higher_order_debug 0
1879 
node_nr_slabs(struct kmem_cache_node * n)1880 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1881 							{ return 0; }
inc_slabs_node(struct kmem_cache * s,int node,int objects)1882 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1883 							int objects) {}
dec_slabs_node(struct kmem_cache * s,int node,int objects)1884 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1885 							int objects) {}
1886 #ifndef CONFIG_SLUB_TINY
freelist_corrupted(struct kmem_cache * s,struct slab * slab,void ** freelist,void * nextfree)1887 static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab,
1888 			       void **freelist, void *nextfree)
1889 {
1890 	return false;
1891 }
1892 #endif
1893 #endif /* CONFIG_SLUB_DEBUG */
1894 
1895 #ifdef CONFIG_SLAB_OBJ_EXT
1896 
1897 #ifdef CONFIG_MEM_ALLOC_PROFILING_DEBUG
1898 
mark_objexts_empty(struct slabobj_ext * obj_exts)1899 static inline void mark_objexts_empty(struct slabobj_ext *obj_exts)
1900 {
1901 	struct slabobj_ext *slab_exts;
1902 	struct slab *obj_exts_slab;
1903 
1904 	obj_exts_slab = virt_to_slab(obj_exts);
1905 	slab_exts = slab_obj_exts(obj_exts_slab);
1906 	if (slab_exts) {
1907 		unsigned int offs = obj_to_index(obj_exts_slab->slab_cache,
1908 						 obj_exts_slab, obj_exts);
1909 		/* codetag should be NULL */
1910 		WARN_ON(slab_exts[offs].ref.ct);
1911 		set_codetag_empty(&slab_exts[offs].ref);
1912 	}
1913 }
1914 
mark_failed_objexts_alloc(struct slab * slab)1915 static inline void mark_failed_objexts_alloc(struct slab *slab)
1916 {
1917 	slab->obj_exts = OBJEXTS_ALLOC_FAIL;
1918 }
1919 
handle_failed_objexts_alloc(unsigned long obj_exts,struct slabobj_ext * vec,unsigned int objects)1920 static inline void handle_failed_objexts_alloc(unsigned long obj_exts,
1921 			struct slabobj_ext *vec, unsigned int objects)
1922 {
1923 	/*
1924 	 * If vector previously failed to allocate then we have live
1925 	 * objects with no tag reference. Mark all references in this
1926 	 * vector as empty to avoid warnings later on.
1927 	 */
1928 	if (obj_exts & OBJEXTS_ALLOC_FAIL) {
1929 		unsigned int i;
1930 
1931 		for (i = 0; i < objects; i++)
1932 			set_codetag_empty(&vec[i].ref);
1933 	}
1934 }
1935 
1936 #else /* CONFIG_MEM_ALLOC_PROFILING_DEBUG */
1937 
mark_objexts_empty(struct slabobj_ext * obj_exts)1938 static inline void mark_objexts_empty(struct slabobj_ext *obj_exts) {}
mark_failed_objexts_alloc(struct slab * slab)1939 static inline void mark_failed_objexts_alloc(struct slab *slab) {}
handle_failed_objexts_alloc(unsigned long obj_exts,struct slabobj_ext * vec,unsigned int objects)1940 static inline void handle_failed_objexts_alloc(unsigned long obj_exts,
1941 			struct slabobj_ext *vec, unsigned int objects) {}
1942 
1943 #endif /* CONFIG_MEM_ALLOC_PROFILING_DEBUG */
1944 
1945 /*
1946  * The allocated objcg pointers array is not accounted directly.
1947  * Moreover, it should not come from DMA buffer and is not readily
1948  * reclaimable. So those GFP bits should be masked off.
1949  */
1950 #define OBJCGS_CLEAR_MASK	(__GFP_DMA | __GFP_RECLAIMABLE | \
1951 				__GFP_ACCOUNT | __GFP_NOFAIL)
1952 
alloc_slab_obj_exts(struct slab * slab,struct kmem_cache * s,gfp_t gfp,bool new_slab)1953 int alloc_slab_obj_exts(struct slab *slab, struct kmem_cache *s,
1954 		        gfp_t gfp, bool new_slab)
1955 {
1956 	unsigned int objects = objs_per_slab(s, slab);
1957 	unsigned long new_exts;
1958 	unsigned long old_exts;
1959 	struct slabobj_ext *vec;
1960 
1961 	gfp &= ~OBJCGS_CLEAR_MASK;
1962 	/* Prevent recursive extension vector allocation */
1963 	gfp |= __GFP_NO_OBJ_EXT;
1964 	vec = kcalloc_node(objects, sizeof(struct slabobj_ext), gfp,
1965 			   slab_nid(slab));
1966 	if (!vec) {
1967 		/* Mark vectors which failed to allocate */
1968 		if (new_slab)
1969 			mark_failed_objexts_alloc(slab);
1970 
1971 		return -ENOMEM;
1972 	}
1973 
1974 	new_exts = (unsigned long)vec;
1975 #ifdef CONFIG_MEMCG
1976 	new_exts |= MEMCG_DATA_OBJEXTS;
1977 #endif
1978 	old_exts = READ_ONCE(slab->obj_exts);
1979 	handle_failed_objexts_alloc(old_exts, vec, objects);
1980 	if (new_slab) {
1981 		/*
1982 		 * If the slab is brand new and nobody can yet access its
1983 		 * obj_exts, no synchronization is required and obj_exts can
1984 		 * be simply assigned.
1985 		 */
1986 		slab->obj_exts = new_exts;
1987 	} else if ((old_exts & ~OBJEXTS_FLAGS_MASK) ||
1988 		   cmpxchg(&slab->obj_exts, old_exts, new_exts) != old_exts) {
1989 		/*
1990 		 * If the slab is already in use, somebody can allocate and
1991 		 * assign slabobj_exts in parallel. In this case the existing
1992 		 * objcg vector should be reused.
1993 		 */
1994 		mark_objexts_empty(vec);
1995 		kfree(vec);
1996 		return 0;
1997 	}
1998 
1999 	kmemleak_not_leak(vec);
2000 	return 0;
2001 }
2002 
free_slab_obj_exts(struct slab * slab)2003 static inline void free_slab_obj_exts(struct slab *slab)
2004 {
2005 	struct slabobj_ext *obj_exts;
2006 
2007 	obj_exts = slab_obj_exts(slab);
2008 	if (!obj_exts)
2009 		return;
2010 
2011 	/*
2012 	 * obj_exts was created with __GFP_NO_OBJ_EXT flag, therefore its
2013 	 * corresponding extension will be NULL. alloc_tag_sub() will throw a
2014 	 * warning if slab has extensions but the extension of an object is
2015 	 * NULL, therefore replace NULL with CODETAG_EMPTY to indicate that
2016 	 * the extension for obj_exts is expected to be NULL.
2017 	 */
2018 	mark_objexts_empty(obj_exts);
2019 	kfree(obj_exts);
2020 	slab->obj_exts = 0;
2021 }
2022 
need_slab_obj_ext(void)2023 static inline bool need_slab_obj_ext(void)
2024 {
2025 	if (mem_alloc_profiling_enabled())
2026 		return true;
2027 
2028 	/*
2029 	 * CONFIG_MEMCG creates vector of obj_cgroup objects conditionally
2030 	 * inside memcg_slab_post_alloc_hook. No other users for now.
2031 	 */
2032 	return false;
2033 }
2034 
2035 #else /* CONFIG_SLAB_OBJ_EXT */
2036 
alloc_slab_obj_exts(struct slab * slab,struct kmem_cache * s,gfp_t gfp,bool new_slab)2037 static int alloc_slab_obj_exts(struct slab *slab, struct kmem_cache *s,
2038 			       gfp_t gfp, bool new_slab)
2039 {
2040 	return 0;
2041 }
2042 
free_slab_obj_exts(struct slab * slab)2043 static inline void free_slab_obj_exts(struct slab *slab)
2044 {
2045 }
2046 
need_slab_obj_ext(void)2047 static inline bool need_slab_obj_ext(void)
2048 {
2049 	return false;
2050 }
2051 
2052 #endif /* CONFIG_SLAB_OBJ_EXT */
2053 
2054 #ifdef CONFIG_MEM_ALLOC_PROFILING
2055 
2056 static inline struct slabobj_ext *
prepare_slab_obj_exts_hook(struct kmem_cache * s,gfp_t flags,void * p)2057 prepare_slab_obj_exts_hook(struct kmem_cache *s, gfp_t flags, void *p)
2058 {
2059 	struct slab *slab;
2060 
2061 	if (!p)
2062 		return NULL;
2063 
2064 	if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
2065 		return NULL;
2066 
2067 	if (flags & __GFP_NO_OBJ_EXT)
2068 		return NULL;
2069 
2070 	slab = virt_to_slab(p);
2071 	if (!slab_obj_exts(slab) &&
2072 	    WARN(alloc_slab_obj_exts(slab, s, flags, false),
2073 		 "%s, %s: Failed to create slab extension vector!\n",
2074 		 __func__, s->name))
2075 		return NULL;
2076 
2077 	return slab_obj_exts(slab) + obj_to_index(s, slab, p);
2078 }
2079 
2080 static inline void
alloc_tagging_slab_alloc_hook(struct kmem_cache * s,void * object,gfp_t flags)2081 alloc_tagging_slab_alloc_hook(struct kmem_cache *s, void *object, gfp_t flags)
2082 {
2083 	if (need_slab_obj_ext()) {
2084 		struct slabobj_ext *obj_exts;
2085 
2086 		obj_exts = prepare_slab_obj_exts_hook(s, flags, object);
2087 		/*
2088 		 * Currently obj_exts is used only for allocation profiling.
2089 		 * If other users appear then mem_alloc_profiling_enabled()
2090 		 * check should be added before alloc_tag_add().
2091 		 */
2092 		if (likely(obj_exts))
2093 			alloc_tag_add(&obj_exts->ref, current->alloc_tag, s->size);
2094 	}
2095 }
2096 
2097 static inline void
alloc_tagging_slab_free_hook(struct kmem_cache * s,struct slab * slab,void ** p,int objects)2098 alloc_tagging_slab_free_hook(struct kmem_cache *s, struct slab *slab, void **p,
2099 			     int objects)
2100 {
2101 	struct slabobj_ext *obj_exts;
2102 	int i;
2103 
2104 	if (!mem_alloc_profiling_enabled())
2105 		return;
2106 
2107 	/* slab->obj_exts might not be NULL if it was created for MEMCG accounting. */
2108 	if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE))
2109 		return;
2110 
2111 	obj_exts = slab_obj_exts(slab);
2112 	if (!obj_exts)
2113 		return;
2114 
2115 	for (i = 0; i < objects; i++) {
2116 		unsigned int off = obj_to_index(s, slab, p[i]);
2117 
2118 		alloc_tag_sub(&obj_exts[off].ref, s->size);
2119 	}
2120 }
2121 
2122 #else /* CONFIG_MEM_ALLOC_PROFILING */
2123 
2124 static inline void
alloc_tagging_slab_alloc_hook(struct kmem_cache * s,void * object,gfp_t flags)2125 alloc_tagging_slab_alloc_hook(struct kmem_cache *s, void *object, gfp_t flags)
2126 {
2127 }
2128 
2129 static inline void
alloc_tagging_slab_free_hook(struct kmem_cache * s,struct slab * slab,void ** p,int objects)2130 alloc_tagging_slab_free_hook(struct kmem_cache *s, struct slab *slab, void **p,
2131 			     int objects)
2132 {
2133 }
2134 
2135 #endif /* CONFIG_MEM_ALLOC_PROFILING */
2136 
2137 
2138 #ifdef CONFIG_MEMCG
2139 
2140 static void memcg_alloc_abort_single(struct kmem_cache *s, void *object);
2141 
2142 static __fastpath_inline
memcg_slab_post_alloc_hook(struct kmem_cache * s,struct list_lru * lru,gfp_t flags,size_t size,void ** p)2143 bool memcg_slab_post_alloc_hook(struct kmem_cache *s, struct list_lru *lru,
2144 				gfp_t flags, size_t size, void **p)
2145 {
2146 	if (likely(!memcg_kmem_online()))
2147 		return true;
2148 
2149 	if (likely(!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT)))
2150 		return true;
2151 
2152 	if (likely(__memcg_slab_post_alloc_hook(s, lru, flags, size, p)))
2153 		return true;
2154 
2155 	if (likely(size == 1)) {
2156 		memcg_alloc_abort_single(s, *p);
2157 		*p = NULL;
2158 	} else {
2159 		kmem_cache_free_bulk(s, size, p);
2160 	}
2161 
2162 	return false;
2163 }
2164 
2165 static __fastpath_inline
memcg_slab_free_hook(struct kmem_cache * s,struct slab * slab,void ** p,int objects)2166 void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab, void **p,
2167 			  int objects)
2168 {
2169 	struct slabobj_ext *obj_exts;
2170 
2171 	if (!memcg_kmem_online())
2172 		return;
2173 
2174 	obj_exts = slab_obj_exts(slab);
2175 	if (likely(!obj_exts))
2176 		return;
2177 
2178 	__memcg_slab_free_hook(s, slab, p, objects, obj_exts);
2179 }
2180 
2181 static __fastpath_inline
memcg_slab_post_charge(void * p,gfp_t flags)2182 bool memcg_slab_post_charge(void *p, gfp_t flags)
2183 {
2184 	struct slabobj_ext *slab_exts;
2185 	struct kmem_cache *s;
2186 	struct folio *folio;
2187 	struct slab *slab;
2188 	unsigned long off;
2189 
2190 	folio = virt_to_folio(p);
2191 	if (!folio_test_slab(folio)) {
2192 		int size;
2193 
2194 		if (folio_memcg_kmem(folio))
2195 			return true;
2196 
2197 		if (__memcg_kmem_charge_page(folio_page(folio, 0), flags,
2198 					     folio_order(folio)))
2199 			return false;
2200 
2201 		/*
2202 		 * This folio has already been accounted in the global stats but
2203 		 * not in the memcg stats. So, subtract from the global and use
2204 		 * the interface which adds to both global and memcg stats.
2205 		 */
2206 		size = folio_size(folio);
2207 		node_stat_mod_folio(folio, NR_SLAB_UNRECLAIMABLE_B, -size);
2208 		lruvec_stat_mod_folio(folio, NR_SLAB_UNRECLAIMABLE_B, size);
2209 		return true;
2210 	}
2211 
2212 	slab = folio_slab(folio);
2213 	s = slab->slab_cache;
2214 
2215 	/*
2216 	 * Ignore KMALLOC_NORMAL cache to avoid possible circular dependency
2217 	 * of slab_obj_exts being allocated from the same slab and thus the slab
2218 	 * becoming effectively unfreeable.
2219 	 */
2220 	if (is_kmalloc_normal(s))
2221 		return true;
2222 
2223 	/* Ignore already charged objects. */
2224 	slab_exts = slab_obj_exts(slab);
2225 	if (slab_exts) {
2226 		off = obj_to_index(s, slab, p);
2227 		if (unlikely(slab_exts[off].objcg))
2228 			return true;
2229 	}
2230 
2231 	return __memcg_slab_post_alloc_hook(s, NULL, flags, 1, &p);
2232 }
2233 
2234 #else /* CONFIG_MEMCG */
memcg_slab_post_alloc_hook(struct kmem_cache * s,struct list_lru * lru,gfp_t flags,size_t size,void ** p)2235 static inline bool memcg_slab_post_alloc_hook(struct kmem_cache *s,
2236 					      struct list_lru *lru,
2237 					      gfp_t flags, size_t size,
2238 					      void **p)
2239 {
2240 	return true;
2241 }
2242 
memcg_slab_free_hook(struct kmem_cache * s,struct slab * slab,void ** p,int objects)2243 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
2244 					void **p, int objects)
2245 {
2246 }
2247 
memcg_slab_post_charge(void * p,gfp_t flags)2248 static inline bool memcg_slab_post_charge(void *p, gfp_t flags)
2249 {
2250 	return true;
2251 }
2252 #endif /* CONFIG_MEMCG */
2253 
2254 #ifdef CONFIG_SLUB_RCU_DEBUG
2255 static void slab_free_after_rcu_debug(struct rcu_head *rcu_head);
2256 
2257 struct rcu_delayed_free {
2258 	struct rcu_head head;
2259 	void *object;
2260 };
2261 #endif
2262 
2263 /*
2264  * Hooks for other subsystems that check memory allocations. In a typical
2265  * production configuration these hooks all should produce no code at all.
2266  *
2267  * Returns true if freeing of the object can proceed, false if its reuse
2268  * was delayed by CONFIG_SLUB_RCU_DEBUG or KASAN quarantine, or it was returned
2269  * to KFENCE.
2270  */
2271 static __always_inline
slab_free_hook(struct kmem_cache * s,void * x,bool init,bool after_rcu_delay)2272 bool slab_free_hook(struct kmem_cache *s, void *x, bool init,
2273 		    bool after_rcu_delay)
2274 {
2275 	/* Are the object contents still accessible? */
2276 	bool still_accessible = (s->flags & SLAB_TYPESAFE_BY_RCU) && !after_rcu_delay;
2277 
2278 	kmemleak_free_recursive(x, s->flags);
2279 	kmsan_slab_free(s, x);
2280 
2281 	debug_check_no_locks_freed(x, s->object_size);
2282 
2283 	if (!(s->flags & SLAB_DEBUG_OBJECTS))
2284 		debug_check_no_obj_freed(x, s->object_size);
2285 
2286 	/* Use KCSAN to help debug racy use-after-free. */
2287 	if (!still_accessible)
2288 		__kcsan_check_access(x, s->object_size,
2289 				     KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT);
2290 
2291 	if (kfence_free(x))
2292 		return false;
2293 
2294 	/*
2295 	 * Give KASAN a chance to notice an invalid free operation before we
2296 	 * modify the object.
2297 	 */
2298 	if (kasan_slab_pre_free(s, x))
2299 		return false;
2300 
2301 #ifdef CONFIG_SLUB_RCU_DEBUG
2302 	if (still_accessible) {
2303 		struct rcu_delayed_free *delayed_free;
2304 
2305 		delayed_free = kmalloc(sizeof(*delayed_free), GFP_NOWAIT);
2306 		if (delayed_free) {
2307 			/*
2308 			 * Let KASAN track our call stack as a "related work
2309 			 * creation", just like if the object had been freed
2310 			 * normally via kfree_rcu().
2311 			 * We have to do this manually because the rcu_head is
2312 			 * not located inside the object.
2313 			 */
2314 			kasan_record_aux_stack(x);
2315 
2316 			delayed_free->object = x;
2317 			call_rcu(&delayed_free->head, slab_free_after_rcu_debug);
2318 			return false;
2319 		}
2320 	}
2321 #endif /* CONFIG_SLUB_RCU_DEBUG */
2322 
2323 	/*
2324 	 * As memory initialization might be integrated into KASAN,
2325 	 * kasan_slab_free and initialization memset's must be
2326 	 * kept together to avoid discrepancies in behavior.
2327 	 *
2328 	 * The initialization memset's clear the object and the metadata,
2329 	 * but don't touch the SLAB redzone.
2330 	 *
2331 	 * The object's freepointer is also avoided if stored outside the
2332 	 * object.
2333 	 */
2334 	if (unlikely(init)) {
2335 		int rsize;
2336 		unsigned int inuse, orig_size;
2337 
2338 		inuse = get_info_end(s);
2339 		orig_size = get_orig_size(s, x);
2340 		if (!kasan_has_integrated_init())
2341 			memset(kasan_reset_tag(x), 0, orig_size);
2342 		rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0;
2343 		memset((char *)kasan_reset_tag(x) + inuse, 0,
2344 		       s->size - inuse - rsize);
2345 		/*
2346 		 * Restore orig_size, otherwize kmalloc redzone overwritten
2347 		 * would be reported
2348 		 */
2349 		set_orig_size(s, x, orig_size);
2350 
2351 	}
2352 	/* KASAN might put x into memory quarantine, delaying its reuse. */
2353 	return !kasan_slab_free(s, x, init, still_accessible);
2354 }
2355 
2356 static __fastpath_inline
slab_free_freelist_hook(struct kmem_cache * s,void ** head,void ** tail,int * cnt)2357 bool slab_free_freelist_hook(struct kmem_cache *s, void **head, void **tail,
2358 			     int *cnt)
2359 {
2360 
2361 	void *object;
2362 	void *next = *head;
2363 	void *old_tail = *tail;
2364 	bool init;
2365 
2366 	if (is_kfence_address(next)) {
2367 		slab_free_hook(s, next, false, false);
2368 		return false;
2369 	}
2370 
2371 	/* Head and tail of the reconstructed freelist */
2372 	*head = NULL;
2373 	*tail = NULL;
2374 
2375 	init = slab_want_init_on_free(s);
2376 
2377 	do {
2378 		object = next;
2379 		next = get_freepointer(s, object);
2380 
2381 		/* If object's reuse doesn't have to be delayed */
2382 		if (likely(slab_free_hook(s, object, init, false))) {
2383 			/* Move object to the new freelist */
2384 			set_freepointer(s, object, *head);
2385 			*head = object;
2386 			if (!*tail)
2387 				*tail = object;
2388 		} else {
2389 			/*
2390 			 * Adjust the reconstructed freelist depth
2391 			 * accordingly if object's reuse is delayed.
2392 			 */
2393 			--(*cnt);
2394 		}
2395 	} while (object != old_tail);
2396 
2397 	return *head != NULL;
2398 }
2399 
setup_object(struct kmem_cache * s,void * object)2400 static void *setup_object(struct kmem_cache *s, void *object)
2401 {
2402 	setup_object_debug(s, object);
2403 	object = kasan_init_slab_obj(s, object);
2404 	if (unlikely(s->ctor)) {
2405 		kasan_unpoison_new_object(s, object);
2406 		s->ctor(object);
2407 		kasan_poison_new_object(s, object);
2408 	}
2409 	return object;
2410 }
2411 
2412 /*
2413  * Slab allocation and freeing
2414  */
alloc_slab_page(gfp_t flags,int node,struct kmem_cache_order_objects oo)2415 static inline struct slab *alloc_slab_page(gfp_t flags, int node,
2416 		struct kmem_cache_order_objects oo)
2417 {
2418 	struct folio *folio;
2419 	struct slab *slab;
2420 	unsigned int order = oo_order(oo);
2421 
2422 	if (node == NUMA_NO_NODE)
2423 		folio = (struct folio *)alloc_frozen_pages(flags, order);
2424 	else
2425 		folio = (struct folio *)__alloc_frozen_pages(flags, order, node, NULL);
2426 
2427 	if (!folio)
2428 		return NULL;
2429 
2430 	slab = folio_slab(folio);
2431 	__folio_set_slab(folio);
2432 	if (folio_is_pfmemalloc(folio))
2433 		slab_set_pfmemalloc(slab);
2434 
2435 	return slab;
2436 }
2437 
2438 #ifdef CONFIG_SLAB_FREELIST_RANDOM
2439 /* Pre-initialize the random sequence cache */
init_cache_random_seq(struct kmem_cache * s)2440 static int init_cache_random_seq(struct kmem_cache *s)
2441 {
2442 	unsigned int count = oo_objects(s->oo);
2443 	int err;
2444 
2445 	/* Bailout if already initialised */
2446 	if (s->random_seq)
2447 		return 0;
2448 
2449 	err = cache_random_seq_create(s, count, GFP_KERNEL);
2450 	if (err) {
2451 		pr_err("SLUB: Unable to initialize free list for %s\n",
2452 			s->name);
2453 		return err;
2454 	}
2455 
2456 	/* Transform to an offset on the set of pages */
2457 	if (s->random_seq) {
2458 		unsigned int i;
2459 
2460 		for (i = 0; i < count; i++)
2461 			s->random_seq[i] *= s->size;
2462 	}
2463 	return 0;
2464 }
2465 
2466 /* Initialize each random sequence freelist per cache */
init_freelist_randomization(void)2467 static void __init init_freelist_randomization(void)
2468 {
2469 	struct kmem_cache *s;
2470 
2471 	mutex_lock(&slab_mutex);
2472 
2473 	list_for_each_entry(s, &slab_caches, list)
2474 		init_cache_random_seq(s);
2475 
2476 	mutex_unlock(&slab_mutex);
2477 }
2478 
2479 /* Get the next entry on the pre-computed freelist randomized */
next_freelist_entry(struct kmem_cache * s,unsigned long * pos,void * start,unsigned long page_limit,unsigned long freelist_count)2480 static void *next_freelist_entry(struct kmem_cache *s,
2481 				unsigned long *pos, void *start,
2482 				unsigned long page_limit,
2483 				unsigned long freelist_count)
2484 {
2485 	unsigned int idx;
2486 
2487 	/*
2488 	 * If the target page allocation failed, the number of objects on the
2489 	 * page might be smaller than the usual size defined by the cache.
2490 	 */
2491 	do {
2492 		idx = s->random_seq[*pos];
2493 		*pos += 1;
2494 		if (*pos >= freelist_count)
2495 			*pos = 0;
2496 	} while (unlikely(idx >= page_limit));
2497 
2498 	return (char *)start + idx;
2499 }
2500 
2501 /* Shuffle the single linked freelist based on a random pre-computed sequence */
shuffle_freelist(struct kmem_cache * s,struct slab * slab)2502 static bool shuffle_freelist(struct kmem_cache *s, struct slab *slab)
2503 {
2504 	void *start;
2505 	void *cur;
2506 	void *next;
2507 	unsigned long idx, pos, page_limit, freelist_count;
2508 
2509 	if (slab->objects < 2 || !s->random_seq)
2510 		return false;
2511 
2512 	freelist_count = oo_objects(s->oo);
2513 	pos = get_random_u32_below(freelist_count);
2514 
2515 	page_limit = slab->objects * s->size;
2516 	start = fixup_red_left(s, slab_address(slab));
2517 
2518 	/* First entry is used as the base of the freelist */
2519 	cur = next_freelist_entry(s, &pos, start, page_limit, freelist_count);
2520 	cur = setup_object(s, cur);
2521 	slab->freelist = cur;
2522 
2523 	for (idx = 1; idx < slab->objects; idx++) {
2524 		next = next_freelist_entry(s, &pos, start, page_limit,
2525 			freelist_count);
2526 		next = setup_object(s, next);
2527 		set_freepointer(s, cur, next);
2528 		cur = next;
2529 	}
2530 	set_freepointer(s, cur, NULL);
2531 
2532 	return true;
2533 }
2534 #else
init_cache_random_seq(struct kmem_cache * s)2535 static inline int init_cache_random_seq(struct kmem_cache *s)
2536 {
2537 	return 0;
2538 }
init_freelist_randomization(void)2539 static inline void init_freelist_randomization(void) { }
shuffle_freelist(struct kmem_cache * s,struct slab * slab)2540 static inline bool shuffle_freelist(struct kmem_cache *s, struct slab *slab)
2541 {
2542 	return false;
2543 }
2544 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
2545 
account_slab(struct slab * slab,int order,struct kmem_cache * s,gfp_t gfp)2546 static __always_inline void account_slab(struct slab *slab, int order,
2547 					 struct kmem_cache *s, gfp_t gfp)
2548 {
2549 	if (memcg_kmem_online() && (s->flags & SLAB_ACCOUNT))
2550 		alloc_slab_obj_exts(slab, s, gfp, true);
2551 
2552 	mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
2553 			    PAGE_SIZE << order);
2554 }
2555 
unaccount_slab(struct slab * slab,int order,struct kmem_cache * s)2556 static __always_inline void unaccount_slab(struct slab *slab, int order,
2557 					   struct kmem_cache *s)
2558 {
2559 	if (memcg_kmem_online() || need_slab_obj_ext())
2560 		free_slab_obj_exts(slab);
2561 
2562 	mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
2563 			    -(PAGE_SIZE << order));
2564 }
2565 
allocate_slab(struct kmem_cache * s,gfp_t flags,int node)2566 static struct slab *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
2567 {
2568 	struct slab *slab;
2569 	struct kmem_cache_order_objects oo = s->oo;
2570 	gfp_t alloc_gfp;
2571 	void *start, *p, *next;
2572 	int idx;
2573 	bool shuffle;
2574 
2575 	flags &= gfp_allowed_mask;
2576 
2577 	flags |= s->allocflags;
2578 
2579 	/*
2580 	 * Let the initial higher-order allocation fail under memory pressure
2581 	 * so we fall-back to the minimum order allocation.
2582 	 */
2583 	alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
2584 	if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
2585 		alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_RECLAIM;
2586 
2587 	slab = alloc_slab_page(alloc_gfp, node, oo);
2588 	if (unlikely(!slab)) {
2589 		oo = s->min;
2590 		alloc_gfp = flags;
2591 		/*
2592 		 * Allocation may have failed due to fragmentation.
2593 		 * Try a lower order alloc if possible
2594 		 */
2595 		slab = alloc_slab_page(alloc_gfp, node, oo);
2596 		if (unlikely(!slab))
2597 			return NULL;
2598 		stat(s, ORDER_FALLBACK);
2599 	}
2600 
2601 	slab->objects = oo_objects(oo);
2602 	slab->inuse = 0;
2603 	slab->frozen = 0;
2604 
2605 	account_slab(slab, oo_order(oo), s, flags);
2606 
2607 	slab->slab_cache = s;
2608 
2609 	kasan_poison_slab(slab);
2610 
2611 	start = slab_address(slab);
2612 
2613 	setup_slab_debug(s, slab, start);
2614 
2615 	shuffle = shuffle_freelist(s, slab);
2616 
2617 	if (!shuffle) {
2618 		start = fixup_red_left(s, start);
2619 		start = setup_object(s, start);
2620 		slab->freelist = start;
2621 		for (idx = 0, p = start; idx < slab->objects - 1; idx++) {
2622 			next = p + s->size;
2623 			next = setup_object(s, next);
2624 			set_freepointer(s, p, next);
2625 			p = next;
2626 		}
2627 		set_freepointer(s, p, NULL);
2628 	}
2629 
2630 	return slab;
2631 }
2632 
new_slab(struct kmem_cache * s,gfp_t flags,int node)2633 static struct slab *new_slab(struct kmem_cache *s, gfp_t flags, int node)
2634 {
2635 	if (unlikely(flags & GFP_SLAB_BUG_MASK))
2636 		flags = kmalloc_fix_flags(flags);
2637 
2638 	WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2639 
2640 	return allocate_slab(s,
2641 		flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
2642 }
2643 
__free_slab(struct kmem_cache * s,struct slab * slab)2644 static void __free_slab(struct kmem_cache *s, struct slab *slab)
2645 {
2646 	struct folio *folio = slab_folio(slab);
2647 	int order = folio_order(folio);
2648 	int pages = 1 << order;
2649 
2650 	__slab_clear_pfmemalloc(slab);
2651 	folio->mapping = NULL;
2652 	__folio_clear_slab(folio);
2653 	mm_account_reclaimed_pages(pages);
2654 	unaccount_slab(slab, order, s);
2655 	free_frozen_pages(&folio->page, order);
2656 }
2657 
rcu_free_slab(struct rcu_head * h)2658 static void rcu_free_slab(struct rcu_head *h)
2659 {
2660 	struct slab *slab = container_of(h, struct slab, rcu_head);
2661 
2662 	__free_slab(slab->slab_cache, slab);
2663 }
2664 
free_slab(struct kmem_cache * s,struct slab * slab)2665 static void free_slab(struct kmem_cache *s, struct slab *slab)
2666 {
2667 	if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) {
2668 		void *p;
2669 
2670 		slab_pad_check(s, slab);
2671 		for_each_object(p, s, slab_address(slab), slab->objects)
2672 			check_object(s, slab, p, SLUB_RED_INACTIVE);
2673 	}
2674 
2675 	if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU))
2676 		call_rcu(&slab->rcu_head, rcu_free_slab);
2677 	else
2678 		__free_slab(s, slab);
2679 }
2680 
discard_slab(struct kmem_cache * s,struct slab * slab)2681 static void discard_slab(struct kmem_cache *s, struct slab *slab)
2682 {
2683 	dec_slabs_node(s, slab_nid(slab), slab->objects);
2684 	free_slab(s, slab);
2685 }
2686 
2687 /*
2688  * SLUB reuses PG_workingset bit to keep track of whether it's on
2689  * the per-node partial list.
2690  */
slab_test_node_partial(const struct slab * slab)2691 static inline bool slab_test_node_partial(const struct slab *slab)
2692 {
2693 	return folio_test_workingset(slab_folio(slab));
2694 }
2695 
slab_set_node_partial(struct slab * slab)2696 static inline void slab_set_node_partial(struct slab *slab)
2697 {
2698 	set_bit(PG_workingset, folio_flags(slab_folio(slab), 0));
2699 }
2700 
slab_clear_node_partial(struct slab * slab)2701 static inline void slab_clear_node_partial(struct slab *slab)
2702 {
2703 	clear_bit(PG_workingset, folio_flags(slab_folio(slab), 0));
2704 }
2705 
2706 /*
2707  * Management of partially allocated slabs.
2708  */
2709 static inline void
__add_partial(struct kmem_cache_node * n,struct slab * slab,int tail)2710 __add_partial(struct kmem_cache_node *n, struct slab *slab, int tail)
2711 {
2712 	n->nr_partial++;
2713 	if (tail == DEACTIVATE_TO_TAIL)
2714 		list_add_tail(&slab->slab_list, &n->partial);
2715 	else
2716 		list_add(&slab->slab_list, &n->partial);
2717 	slab_set_node_partial(slab);
2718 }
2719 
add_partial(struct kmem_cache_node * n,struct slab * slab,int tail)2720 static inline void add_partial(struct kmem_cache_node *n,
2721 				struct slab *slab, int tail)
2722 {
2723 	lockdep_assert_held(&n->list_lock);
2724 	__add_partial(n, slab, tail);
2725 }
2726 
remove_partial(struct kmem_cache_node * n,struct slab * slab)2727 static inline void remove_partial(struct kmem_cache_node *n,
2728 					struct slab *slab)
2729 {
2730 	lockdep_assert_held(&n->list_lock);
2731 	list_del(&slab->slab_list);
2732 	slab_clear_node_partial(slab);
2733 	n->nr_partial--;
2734 }
2735 
2736 /*
2737  * Called only for kmem_cache_debug() caches instead of remove_partial(), with a
2738  * slab from the n->partial list. Remove only a single object from the slab, do
2739  * the alloc_debug_processing() checks and leave the slab on the list, or move
2740  * it to full list if it was the last free object.
2741  */
alloc_single_from_partial(struct kmem_cache * s,struct kmem_cache_node * n,struct slab * slab,int orig_size)2742 static void *alloc_single_from_partial(struct kmem_cache *s,
2743 		struct kmem_cache_node *n, struct slab *slab, int orig_size)
2744 {
2745 	void *object;
2746 
2747 	lockdep_assert_held(&n->list_lock);
2748 
2749 	object = slab->freelist;
2750 	slab->freelist = get_freepointer(s, object);
2751 	slab->inuse++;
2752 
2753 	if (!alloc_debug_processing(s, slab, object, orig_size)) {
2754 		if (folio_test_slab(slab_folio(slab)))
2755 			remove_partial(n, slab);
2756 		return NULL;
2757 	}
2758 
2759 	if (slab->inuse == slab->objects) {
2760 		remove_partial(n, slab);
2761 		add_full(s, n, slab);
2762 	}
2763 
2764 	return object;
2765 }
2766 
2767 /*
2768  * Called only for kmem_cache_debug() caches to allocate from a freshly
2769  * allocated slab. Allocate a single object instead of whole freelist
2770  * and put the slab to the partial (or full) list.
2771  */
alloc_single_from_new_slab(struct kmem_cache * s,struct slab * slab,int orig_size)2772 static void *alloc_single_from_new_slab(struct kmem_cache *s,
2773 					struct slab *slab, int orig_size)
2774 {
2775 	int nid = slab_nid(slab);
2776 	struct kmem_cache_node *n = get_node(s, nid);
2777 	unsigned long flags;
2778 	void *object;
2779 
2780 
2781 	object = slab->freelist;
2782 	slab->freelist = get_freepointer(s, object);
2783 	slab->inuse = 1;
2784 
2785 	if (!alloc_debug_processing(s, slab, object, orig_size))
2786 		/*
2787 		 * It's not really expected that this would fail on a
2788 		 * freshly allocated slab, but a concurrent memory
2789 		 * corruption in theory could cause that.
2790 		 */
2791 		return NULL;
2792 
2793 	spin_lock_irqsave(&n->list_lock, flags);
2794 
2795 	if (slab->inuse == slab->objects)
2796 		add_full(s, n, slab);
2797 	else
2798 		add_partial(n, slab, DEACTIVATE_TO_HEAD);
2799 
2800 	inc_slabs_node(s, nid, slab->objects);
2801 	spin_unlock_irqrestore(&n->list_lock, flags);
2802 
2803 	return object;
2804 }
2805 
2806 #ifdef CONFIG_SLUB_CPU_PARTIAL
2807 static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain);
2808 #else
put_cpu_partial(struct kmem_cache * s,struct slab * slab,int drain)2809 static inline void put_cpu_partial(struct kmem_cache *s, struct slab *slab,
2810 				   int drain) { }
2811 #endif
2812 static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags);
2813 
2814 /*
2815  * Try to allocate a partial slab from a specific node.
2816  */
get_partial_node(struct kmem_cache * s,struct kmem_cache_node * n,struct partial_context * pc)2817 static struct slab *get_partial_node(struct kmem_cache *s,
2818 				     struct kmem_cache_node *n,
2819 				     struct partial_context *pc)
2820 {
2821 	struct slab *slab, *slab2, *partial = NULL;
2822 	unsigned long flags;
2823 	unsigned int partial_slabs = 0;
2824 
2825 	/*
2826 	 * Racy check. If we mistakenly see no partial slabs then we
2827 	 * just allocate an empty slab. If we mistakenly try to get a
2828 	 * partial slab and there is none available then get_partial()
2829 	 * will return NULL.
2830 	 */
2831 	if (!n || !n->nr_partial)
2832 		return NULL;
2833 
2834 	spin_lock_irqsave(&n->list_lock, flags);
2835 	list_for_each_entry_safe(slab, slab2, &n->partial, slab_list) {
2836 		if (!pfmemalloc_match(slab, pc->flags))
2837 			continue;
2838 
2839 		if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) {
2840 			void *object = alloc_single_from_partial(s, n, slab,
2841 							pc->orig_size);
2842 			if (object) {
2843 				partial = slab;
2844 				pc->object = object;
2845 				break;
2846 			}
2847 			continue;
2848 		}
2849 
2850 		remove_partial(n, slab);
2851 
2852 		if (!partial) {
2853 			partial = slab;
2854 			stat(s, ALLOC_FROM_PARTIAL);
2855 
2856 			if ((slub_get_cpu_partial(s) == 0)) {
2857 				break;
2858 			}
2859 		} else {
2860 			put_cpu_partial(s, slab, 0);
2861 			stat(s, CPU_PARTIAL_NODE);
2862 
2863 			if (++partial_slabs > slub_get_cpu_partial(s) / 2) {
2864 				break;
2865 			}
2866 		}
2867 	}
2868 	spin_unlock_irqrestore(&n->list_lock, flags);
2869 	return partial;
2870 }
2871 
2872 /*
2873  * Get a slab from somewhere. Search in increasing NUMA distances.
2874  */
get_any_partial(struct kmem_cache * s,struct partial_context * pc)2875 static struct slab *get_any_partial(struct kmem_cache *s,
2876 				    struct partial_context *pc)
2877 {
2878 #ifdef CONFIG_NUMA
2879 	struct zonelist *zonelist;
2880 	struct zoneref *z;
2881 	struct zone *zone;
2882 	enum zone_type highest_zoneidx = gfp_zone(pc->flags);
2883 	struct slab *slab;
2884 	unsigned int cpuset_mems_cookie;
2885 
2886 	/*
2887 	 * The defrag ratio allows a configuration of the tradeoffs between
2888 	 * inter node defragmentation and node local allocations. A lower
2889 	 * defrag_ratio increases the tendency to do local allocations
2890 	 * instead of attempting to obtain partial slabs from other nodes.
2891 	 *
2892 	 * If the defrag_ratio is set to 0 then kmalloc() always
2893 	 * returns node local objects. If the ratio is higher then kmalloc()
2894 	 * may return off node objects because partial slabs are obtained
2895 	 * from other nodes and filled up.
2896 	 *
2897 	 * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
2898 	 * (which makes defrag_ratio = 1000) then every (well almost)
2899 	 * allocation will first attempt to defrag slab caches on other nodes.
2900 	 * This means scanning over all nodes to look for partial slabs which
2901 	 * may be expensive if we do it every time we are trying to find a slab
2902 	 * with available objects.
2903 	 */
2904 	if (!s->remote_node_defrag_ratio ||
2905 			get_cycles() % 1024 > s->remote_node_defrag_ratio)
2906 		return NULL;
2907 
2908 	do {
2909 		cpuset_mems_cookie = read_mems_allowed_begin();
2910 		zonelist = node_zonelist(mempolicy_slab_node(), pc->flags);
2911 		for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) {
2912 			struct kmem_cache_node *n;
2913 
2914 			n = get_node(s, zone_to_nid(zone));
2915 
2916 			if (n && cpuset_zone_allowed(zone, pc->flags) &&
2917 					n->nr_partial > s->min_partial) {
2918 				slab = get_partial_node(s, n, pc);
2919 				if (slab) {
2920 					/*
2921 					 * Don't check read_mems_allowed_retry()
2922 					 * here - if mems_allowed was updated in
2923 					 * parallel, that was a harmless race
2924 					 * between allocation and the cpuset
2925 					 * update
2926 					 */
2927 					return slab;
2928 				}
2929 			}
2930 		}
2931 	} while (read_mems_allowed_retry(cpuset_mems_cookie));
2932 #endif	/* CONFIG_NUMA */
2933 	return NULL;
2934 }
2935 
2936 /*
2937  * Get a partial slab, lock it and return it.
2938  */
get_partial(struct kmem_cache * s,int node,struct partial_context * pc)2939 static struct slab *get_partial(struct kmem_cache *s, int node,
2940 				struct partial_context *pc)
2941 {
2942 	struct slab *slab;
2943 	int searchnode = node;
2944 
2945 	if (node == NUMA_NO_NODE)
2946 		searchnode = numa_mem_id();
2947 
2948 	slab = get_partial_node(s, get_node(s, searchnode), pc);
2949 	if (slab || (node != NUMA_NO_NODE && (pc->flags & __GFP_THISNODE)))
2950 		return slab;
2951 
2952 	return get_any_partial(s, pc);
2953 }
2954 
2955 #ifndef CONFIG_SLUB_TINY
2956 
2957 #ifdef CONFIG_PREEMPTION
2958 /*
2959  * Calculate the next globally unique transaction for disambiguation
2960  * during cmpxchg. The transactions start with the cpu number and are then
2961  * incremented by CONFIG_NR_CPUS.
2962  */
2963 #define TID_STEP  roundup_pow_of_two(CONFIG_NR_CPUS)
2964 #else
2965 /*
2966  * No preemption supported therefore also no need to check for
2967  * different cpus.
2968  */
2969 #define TID_STEP 1
2970 #endif /* CONFIG_PREEMPTION */
2971 
next_tid(unsigned long tid)2972 static inline unsigned long next_tid(unsigned long tid)
2973 {
2974 	return tid + TID_STEP;
2975 }
2976 
2977 #ifdef SLUB_DEBUG_CMPXCHG
tid_to_cpu(unsigned long tid)2978 static inline unsigned int tid_to_cpu(unsigned long tid)
2979 {
2980 	return tid % TID_STEP;
2981 }
2982 
tid_to_event(unsigned long tid)2983 static inline unsigned long tid_to_event(unsigned long tid)
2984 {
2985 	return tid / TID_STEP;
2986 }
2987 #endif
2988 
init_tid(int cpu)2989 static inline unsigned int init_tid(int cpu)
2990 {
2991 	return cpu;
2992 }
2993 
note_cmpxchg_failure(const char * n,const struct kmem_cache * s,unsigned long tid)2994 static inline void note_cmpxchg_failure(const char *n,
2995 		const struct kmem_cache *s, unsigned long tid)
2996 {
2997 #ifdef SLUB_DEBUG_CMPXCHG
2998 	unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
2999 
3000 	pr_info("%s %s: cmpxchg redo ", n, s->name);
3001 
3002 #ifdef CONFIG_PREEMPTION
3003 	if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
3004 		pr_warn("due to cpu change %d -> %d\n",
3005 			tid_to_cpu(tid), tid_to_cpu(actual_tid));
3006 	else
3007 #endif
3008 	if (tid_to_event(tid) != tid_to_event(actual_tid))
3009 		pr_warn("due to cpu running other code. Event %ld->%ld\n",
3010 			tid_to_event(tid), tid_to_event(actual_tid));
3011 	else
3012 		pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
3013 			actual_tid, tid, next_tid(tid));
3014 #endif
3015 	stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
3016 }
3017 
init_kmem_cache_cpus(struct kmem_cache * s)3018 static void init_kmem_cache_cpus(struct kmem_cache *s)
3019 {
3020 	int cpu;
3021 	struct kmem_cache_cpu *c;
3022 
3023 	for_each_possible_cpu(cpu) {
3024 		c = per_cpu_ptr(s->cpu_slab, cpu);
3025 		local_lock_init(&c->lock);
3026 		c->tid = init_tid(cpu);
3027 	}
3028 }
3029 
3030 /*
3031  * Finishes removing the cpu slab. Merges cpu's freelist with slab's freelist,
3032  * unfreezes the slabs and puts it on the proper list.
3033  * Assumes the slab has been already safely taken away from kmem_cache_cpu
3034  * by the caller.
3035  */
deactivate_slab(struct kmem_cache * s,struct slab * slab,void * freelist)3036 static void deactivate_slab(struct kmem_cache *s, struct slab *slab,
3037 			    void *freelist)
3038 {
3039 	struct kmem_cache_node *n = get_node(s, slab_nid(slab));
3040 	int free_delta = 0;
3041 	void *nextfree, *freelist_iter, *freelist_tail;
3042 	int tail = DEACTIVATE_TO_HEAD;
3043 	unsigned long flags = 0;
3044 	struct slab new;
3045 	struct slab old;
3046 
3047 	if (READ_ONCE(slab->freelist)) {
3048 		stat(s, DEACTIVATE_REMOTE_FREES);
3049 		tail = DEACTIVATE_TO_TAIL;
3050 	}
3051 
3052 	/*
3053 	 * Stage one: Count the objects on cpu's freelist as free_delta and
3054 	 * remember the last object in freelist_tail for later splicing.
3055 	 */
3056 	freelist_tail = NULL;
3057 	freelist_iter = freelist;
3058 	while (freelist_iter) {
3059 		nextfree = get_freepointer(s, freelist_iter);
3060 
3061 		/*
3062 		 * If 'nextfree' is invalid, it is possible that the object at
3063 		 * 'freelist_iter' is already corrupted.  So isolate all objects
3064 		 * starting at 'freelist_iter' by skipping them.
3065 		 */
3066 		if (freelist_corrupted(s, slab, &freelist_iter, nextfree))
3067 			break;
3068 
3069 		freelist_tail = freelist_iter;
3070 		free_delta++;
3071 
3072 		freelist_iter = nextfree;
3073 	}
3074 
3075 	/*
3076 	 * Stage two: Unfreeze the slab while splicing the per-cpu
3077 	 * freelist to the head of slab's freelist.
3078 	 */
3079 	do {
3080 		old.freelist = READ_ONCE(slab->freelist);
3081 		old.counters = READ_ONCE(slab->counters);
3082 		VM_BUG_ON(!old.frozen);
3083 
3084 		/* Determine target state of the slab */
3085 		new.counters = old.counters;
3086 		new.frozen = 0;
3087 		if (freelist_tail) {
3088 			new.inuse -= free_delta;
3089 			set_freepointer(s, freelist_tail, old.freelist);
3090 			new.freelist = freelist;
3091 		} else {
3092 			new.freelist = old.freelist;
3093 		}
3094 	} while (!slab_update_freelist(s, slab,
3095 		old.freelist, old.counters,
3096 		new.freelist, new.counters,
3097 		"unfreezing slab"));
3098 
3099 	/*
3100 	 * Stage three: Manipulate the slab list based on the updated state.
3101 	 */
3102 	if (!new.inuse && n->nr_partial >= s->min_partial) {
3103 		stat(s, DEACTIVATE_EMPTY);
3104 		discard_slab(s, slab);
3105 		stat(s, FREE_SLAB);
3106 	} else if (new.freelist) {
3107 		spin_lock_irqsave(&n->list_lock, flags);
3108 		add_partial(n, slab, tail);
3109 		spin_unlock_irqrestore(&n->list_lock, flags);
3110 		stat(s, tail);
3111 	} else {
3112 		stat(s, DEACTIVATE_FULL);
3113 	}
3114 }
3115 
3116 #ifdef CONFIG_SLUB_CPU_PARTIAL
__put_partials(struct kmem_cache * s,struct slab * partial_slab)3117 static void __put_partials(struct kmem_cache *s, struct slab *partial_slab)
3118 {
3119 	struct kmem_cache_node *n = NULL, *n2 = NULL;
3120 	struct slab *slab, *slab_to_discard = NULL;
3121 	unsigned long flags = 0;
3122 
3123 	while (partial_slab) {
3124 		slab = partial_slab;
3125 		partial_slab = slab->next;
3126 
3127 		n2 = get_node(s, slab_nid(slab));
3128 		if (n != n2) {
3129 			if (n)
3130 				spin_unlock_irqrestore(&n->list_lock, flags);
3131 
3132 			n = n2;
3133 			spin_lock_irqsave(&n->list_lock, flags);
3134 		}
3135 
3136 		if (unlikely(!slab->inuse && n->nr_partial >= s->min_partial)) {
3137 			slab->next = slab_to_discard;
3138 			slab_to_discard = slab;
3139 		} else {
3140 			add_partial(n, slab, DEACTIVATE_TO_TAIL);
3141 			stat(s, FREE_ADD_PARTIAL);
3142 		}
3143 	}
3144 
3145 	if (n)
3146 		spin_unlock_irqrestore(&n->list_lock, flags);
3147 
3148 	while (slab_to_discard) {
3149 		slab = slab_to_discard;
3150 		slab_to_discard = slab_to_discard->next;
3151 
3152 		stat(s, DEACTIVATE_EMPTY);
3153 		discard_slab(s, slab);
3154 		stat(s, FREE_SLAB);
3155 	}
3156 }
3157 
3158 /*
3159  * Put all the cpu partial slabs to the node partial list.
3160  */
put_partials(struct kmem_cache * s)3161 static void put_partials(struct kmem_cache *s)
3162 {
3163 	struct slab *partial_slab;
3164 	unsigned long flags;
3165 
3166 	local_lock_irqsave(&s->cpu_slab->lock, flags);
3167 	partial_slab = this_cpu_read(s->cpu_slab->partial);
3168 	this_cpu_write(s->cpu_slab->partial, NULL);
3169 	local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3170 
3171 	if (partial_slab)
3172 		__put_partials(s, partial_slab);
3173 }
3174 
put_partials_cpu(struct kmem_cache * s,struct kmem_cache_cpu * c)3175 static void put_partials_cpu(struct kmem_cache *s,
3176 			     struct kmem_cache_cpu *c)
3177 {
3178 	struct slab *partial_slab;
3179 
3180 	partial_slab = slub_percpu_partial(c);
3181 	c->partial = NULL;
3182 
3183 	if (partial_slab)
3184 		__put_partials(s, partial_slab);
3185 }
3186 
3187 /*
3188  * Put a slab into a partial slab slot if available.
3189  *
3190  * If we did not find a slot then simply move all the partials to the
3191  * per node partial list.
3192  */
put_cpu_partial(struct kmem_cache * s,struct slab * slab,int drain)3193 static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain)
3194 {
3195 	struct slab *oldslab;
3196 	struct slab *slab_to_put = NULL;
3197 	unsigned long flags;
3198 	int slabs = 0;
3199 
3200 	local_lock_irqsave(&s->cpu_slab->lock, flags);
3201 
3202 	oldslab = this_cpu_read(s->cpu_slab->partial);
3203 
3204 	if (oldslab) {
3205 		if (drain && oldslab->slabs >= s->cpu_partial_slabs) {
3206 			/*
3207 			 * Partial array is full. Move the existing set to the
3208 			 * per node partial list. Postpone the actual unfreezing
3209 			 * outside of the critical section.
3210 			 */
3211 			slab_to_put = oldslab;
3212 			oldslab = NULL;
3213 		} else {
3214 			slabs = oldslab->slabs;
3215 		}
3216 	}
3217 
3218 	slabs++;
3219 
3220 	slab->slabs = slabs;
3221 	slab->next = oldslab;
3222 
3223 	this_cpu_write(s->cpu_slab->partial, slab);
3224 
3225 	local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3226 
3227 	if (slab_to_put) {
3228 		__put_partials(s, slab_to_put);
3229 		stat(s, CPU_PARTIAL_DRAIN);
3230 	}
3231 }
3232 
3233 #else	/* CONFIG_SLUB_CPU_PARTIAL */
3234 
put_partials(struct kmem_cache * s)3235 static inline void put_partials(struct kmem_cache *s) { }
put_partials_cpu(struct kmem_cache * s,struct kmem_cache_cpu * c)3236 static inline void put_partials_cpu(struct kmem_cache *s,
3237 				    struct kmem_cache_cpu *c) { }
3238 
3239 #endif	/* CONFIG_SLUB_CPU_PARTIAL */
3240 
flush_slab(struct kmem_cache * s,struct kmem_cache_cpu * c)3241 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
3242 {
3243 	unsigned long flags;
3244 	struct slab *slab;
3245 	void *freelist;
3246 
3247 	local_lock_irqsave(&s->cpu_slab->lock, flags);
3248 
3249 	slab = c->slab;
3250 	freelist = c->freelist;
3251 
3252 	c->slab = NULL;
3253 	c->freelist = NULL;
3254 	c->tid = next_tid(c->tid);
3255 
3256 	local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3257 
3258 	if (slab) {
3259 		deactivate_slab(s, slab, freelist);
3260 		stat(s, CPUSLAB_FLUSH);
3261 	}
3262 }
3263 
__flush_cpu_slab(struct kmem_cache * s,int cpu)3264 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
3265 {
3266 	struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
3267 	void *freelist = c->freelist;
3268 	struct slab *slab = c->slab;
3269 
3270 	c->slab = NULL;
3271 	c->freelist = NULL;
3272 	c->tid = next_tid(c->tid);
3273 
3274 	if (slab) {
3275 		deactivate_slab(s, slab, freelist);
3276 		stat(s, CPUSLAB_FLUSH);
3277 	}
3278 
3279 	put_partials_cpu(s, c);
3280 }
3281 
3282 struct slub_flush_work {
3283 	struct work_struct work;
3284 	struct kmem_cache *s;
3285 	bool skip;
3286 };
3287 
3288 /*
3289  * Flush cpu slab.
3290  *
3291  * Called from CPU work handler with migration disabled.
3292  */
flush_cpu_slab(struct work_struct * w)3293 static void flush_cpu_slab(struct work_struct *w)
3294 {
3295 	struct kmem_cache *s;
3296 	struct kmem_cache_cpu *c;
3297 	struct slub_flush_work *sfw;
3298 
3299 	sfw = container_of(w, struct slub_flush_work, work);
3300 
3301 	s = sfw->s;
3302 	c = this_cpu_ptr(s->cpu_slab);
3303 
3304 	if (c->slab)
3305 		flush_slab(s, c);
3306 
3307 	put_partials(s);
3308 }
3309 
has_cpu_slab(int cpu,struct kmem_cache * s)3310 static bool has_cpu_slab(int cpu, struct kmem_cache *s)
3311 {
3312 	struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
3313 
3314 	return c->slab || slub_percpu_partial(c);
3315 }
3316 
3317 static DEFINE_MUTEX(flush_lock);
3318 static DEFINE_PER_CPU(struct slub_flush_work, slub_flush);
3319 
flush_all_cpus_locked(struct kmem_cache * s)3320 static void flush_all_cpus_locked(struct kmem_cache *s)
3321 {
3322 	struct slub_flush_work *sfw;
3323 	unsigned int cpu;
3324 
3325 	lockdep_assert_cpus_held();
3326 	mutex_lock(&flush_lock);
3327 
3328 	for_each_online_cpu(cpu) {
3329 		sfw = &per_cpu(slub_flush, cpu);
3330 		if (!has_cpu_slab(cpu, s)) {
3331 			sfw->skip = true;
3332 			continue;
3333 		}
3334 		INIT_WORK(&sfw->work, flush_cpu_slab);
3335 		sfw->skip = false;
3336 		sfw->s = s;
3337 		queue_work_on(cpu, flushwq, &sfw->work);
3338 	}
3339 
3340 	for_each_online_cpu(cpu) {
3341 		sfw = &per_cpu(slub_flush, cpu);
3342 		if (sfw->skip)
3343 			continue;
3344 		flush_work(&sfw->work);
3345 	}
3346 
3347 	mutex_unlock(&flush_lock);
3348 }
3349 
flush_all(struct kmem_cache * s)3350 static void flush_all(struct kmem_cache *s)
3351 {
3352 	cpus_read_lock();
3353 	flush_all_cpus_locked(s);
3354 	cpus_read_unlock();
3355 }
3356 
3357 /*
3358  * Use the cpu notifier to insure that the cpu slabs are flushed when
3359  * necessary.
3360  */
slub_cpu_dead(unsigned int cpu)3361 static int slub_cpu_dead(unsigned int cpu)
3362 {
3363 	struct kmem_cache *s;
3364 
3365 	mutex_lock(&slab_mutex);
3366 	list_for_each_entry(s, &slab_caches, list)
3367 		__flush_cpu_slab(s, cpu);
3368 	mutex_unlock(&slab_mutex);
3369 	return 0;
3370 }
3371 
3372 #else /* CONFIG_SLUB_TINY */
flush_all_cpus_locked(struct kmem_cache * s)3373 static inline void flush_all_cpus_locked(struct kmem_cache *s) { }
flush_all(struct kmem_cache * s)3374 static inline void flush_all(struct kmem_cache *s) { }
__flush_cpu_slab(struct kmem_cache * s,int cpu)3375 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) { }
slub_cpu_dead(unsigned int cpu)3376 static inline int slub_cpu_dead(unsigned int cpu) { return 0; }
3377 #endif /* CONFIG_SLUB_TINY */
3378 
3379 /*
3380  * Check if the objects in a per cpu structure fit numa
3381  * locality expectations.
3382  */
node_match(struct slab * slab,int node)3383 static inline int node_match(struct slab *slab, int node)
3384 {
3385 #ifdef CONFIG_NUMA
3386 	if (node != NUMA_NO_NODE && slab_nid(slab) != node)
3387 		return 0;
3388 #endif
3389 	return 1;
3390 }
3391 
3392 #ifdef CONFIG_SLUB_DEBUG
count_free(struct slab * slab)3393 static int count_free(struct slab *slab)
3394 {
3395 	return slab->objects - slab->inuse;
3396 }
3397 
node_nr_objs(struct kmem_cache_node * n)3398 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
3399 {
3400 	return atomic_long_read(&n->total_objects);
3401 }
3402 
3403 /* Supports checking bulk free of a constructed freelist */
free_debug_processing(struct kmem_cache * s,struct slab * slab,void * head,void * tail,int * bulk_cnt,unsigned long addr,depot_stack_handle_t handle)3404 static inline bool free_debug_processing(struct kmem_cache *s,
3405 	struct slab *slab, void *head, void *tail, int *bulk_cnt,
3406 	unsigned long addr, depot_stack_handle_t handle)
3407 {
3408 	bool checks_ok = false;
3409 	void *object = head;
3410 	int cnt = 0;
3411 
3412 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
3413 		if (!check_slab(s, slab))
3414 			goto out;
3415 	}
3416 
3417 	if (slab->inuse < *bulk_cnt) {
3418 		slab_err(s, slab, "Slab has %d allocated objects but %d are to be freed\n",
3419 			 slab->inuse, *bulk_cnt);
3420 		goto out;
3421 	}
3422 
3423 next_object:
3424 
3425 	if (++cnt > *bulk_cnt)
3426 		goto out_cnt;
3427 
3428 	if (s->flags & SLAB_CONSISTENCY_CHECKS) {
3429 		if (!free_consistency_checks(s, slab, object, addr))
3430 			goto out;
3431 	}
3432 
3433 	if (s->flags & SLAB_STORE_USER)
3434 		set_track_update(s, object, TRACK_FREE, addr, handle);
3435 	trace(s, slab, object, 0);
3436 	/* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
3437 	init_object(s, object, SLUB_RED_INACTIVE);
3438 
3439 	/* Reached end of constructed freelist yet? */
3440 	if (object != tail) {
3441 		object = get_freepointer(s, object);
3442 		goto next_object;
3443 	}
3444 	checks_ok = true;
3445 
3446 out_cnt:
3447 	if (cnt != *bulk_cnt) {
3448 		slab_err(s, slab, "Bulk free expected %d objects but found %d\n",
3449 			 *bulk_cnt, cnt);
3450 		*bulk_cnt = cnt;
3451 	}
3452 
3453 out:
3454 
3455 	if (!checks_ok)
3456 		slab_fix(s, "Object at 0x%p not freed", object);
3457 
3458 	return checks_ok;
3459 }
3460 #endif /* CONFIG_SLUB_DEBUG */
3461 
3462 #if defined(CONFIG_SLUB_DEBUG) || defined(SLAB_SUPPORTS_SYSFS)
count_partial(struct kmem_cache_node * n,int (* get_count)(struct slab *))3463 static unsigned long count_partial(struct kmem_cache_node *n,
3464 					int (*get_count)(struct slab *))
3465 {
3466 	unsigned long flags;
3467 	unsigned long x = 0;
3468 	struct slab *slab;
3469 
3470 	spin_lock_irqsave(&n->list_lock, flags);
3471 	list_for_each_entry(slab, &n->partial, slab_list)
3472 		x += get_count(slab);
3473 	spin_unlock_irqrestore(&n->list_lock, flags);
3474 	return x;
3475 }
3476 #endif /* CONFIG_SLUB_DEBUG || SLAB_SUPPORTS_SYSFS */
3477 
3478 #ifdef CONFIG_SLUB_DEBUG
3479 #define MAX_PARTIAL_TO_SCAN 10000
3480 
count_partial_free_approx(struct kmem_cache_node * n)3481 static unsigned long count_partial_free_approx(struct kmem_cache_node *n)
3482 {
3483 	unsigned long flags;
3484 	unsigned long x = 0;
3485 	struct slab *slab;
3486 
3487 	spin_lock_irqsave(&n->list_lock, flags);
3488 	if (n->nr_partial <= MAX_PARTIAL_TO_SCAN) {
3489 		list_for_each_entry(slab, &n->partial, slab_list)
3490 			x += slab->objects - slab->inuse;
3491 	} else {
3492 		/*
3493 		 * For a long list, approximate the total count of objects in
3494 		 * it to meet the limit on the number of slabs to scan.
3495 		 * Scan from both the list's head and tail for better accuracy.
3496 		 */
3497 		unsigned long scanned = 0;
3498 
3499 		list_for_each_entry(slab, &n->partial, slab_list) {
3500 			x += slab->objects - slab->inuse;
3501 			if (++scanned == MAX_PARTIAL_TO_SCAN / 2)
3502 				break;
3503 		}
3504 		list_for_each_entry_reverse(slab, &n->partial, slab_list) {
3505 			x += slab->objects - slab->inuse;
3506 			if (++scanned == MAX_PARTIAL_TO_SCAN)
3507 				break;
3508 		}
3509 		x = mult_frac(x, n->nr_partial, scanned);
3510 		x = min(x, node_nr_objs(n));
3511 	}
3512 	spin_unlock_irqrestore(&n->list_lock, flags);
3513 	return x;
3514 }
3515 
3516 static noinline void
slab_out_of_memory(struct kmem_cache * s,gfp_t gfpflags,int nid)3517 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
3518 {
3519 	static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
3520 				      DEFAULT_RATELIMIT_BURST);
3521 	int cpu = raw_smp_processor_id();
3522 	int node;
3523 	struct kmem_cache_node *n;
3524 
3525 	if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
3526 		return;
3527 
3528 	pr_warn("SLUB: Unable to allocate memory on CPU %u (of node %d) on node %d, gfp=%#x(%pGg)\n",
3529 		cpu, cpu_to_node(cpu), nid, gfpflags, &gfpflags);
3530 	pr_warn("  cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
3531 		s->name, s->object_size, s->size, oo_order(s->oo),
3532 		oo_order(s->min));
3533 
3534 	if (oo_order(s->min) > get_order(s->object_size))
3535 		pr_warn("  %s debugging increased min order, use slab_debug=O to disable.\n",
3536 			s->name);
3537 
3538 	for_each_kmem_cache_node(s, node, n) {
3539 		unsigned long nr_slabs;
3540 		unsigned long nr_objs;
3541 		unsigned long nr_free;
3542 
3543 		nr_free  = count_partial_free_approx(n);
3544 		nr_slabs = node_nr_slabs(n);
3545 		nr_objs  = node_nr_objs(n);
3546 
3547 		pr_warn("  node %d: slabs: %ld, objs: %ld, free: %ld\n",
3548 			node, nr_slabs, nr_objs, nr_free);
3549 	}
3550 }
3551 #else /* CONFIG_SLUB_DEBUG */
3552 static inline void
slab_out_of_memory(struct kmem_cache * s,gfp_t gfpflags,int nid)3553 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) { }
3554 #endif
3555 
pfmemalloc_match(struct slab * slab,gfp_t gfpflags)3556 static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags)
3557 {
3558 	if (unlikely(slab_test_pfmemalloc(slab)))
3559 		return gfp_pfmemalloc_allowed(gfpflags);
3560 
3561 	return true;
3562 }
3563 
3564 #ifndef CONFIG_SLUB_TINY
3565 static inline bool
__update_cpu_freelist_fast(struct kmem_cache * s,void * freelist_old,void * freelist_new,unsigned long tid)3566 __update_cpu_freelist_fast(struct kmem_cache *s,
3567 			   void *freelist_old, void *freelist_new,
3568 			   unsigned long tid)
3569 {
3570 	freelist_aba_t old = { .freelist = freelist_old, .counter = tid };
3571 	freelist_aba_t new = { .freelist = freelist_new, .counter = next_tid(tid) };
3572 
3573 	return this_cpu_try_cmpxchg_freelist(s->cpu_slab->freelist_tid.full,
3574 					     &old.full, new.full);
3575 }
3576 
3577 /*
3578  * Check the slab->freelist and either transfer the freelist to the
3579  * per cpu freelist or deactivate the slab.
3580  *
3581  * The slab is still frozen if the return value is not NULL.
3582  *
3583  * If this function returns NULL then the slab has been unfrozen.
3584  */
get_freelist(struct kmem_cache * s,struct slab * slab)3585 static inline void *get_freelist(struct kmem_cache *s, struct slab *slab)
3586 {
3587 	struct slab new;
3588 	unsigned long counters;
3589 	void *freelist;
3590 
3591 	lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock));
3592 
3593 	do {
3594 		freelist = slab->freelist;
3595 		counters = slab->counters;
3596 
3597 		new.counters = counters;
3598 
3599 		new.inuse = slab->objects;
3600 		new.frozen = freelist != NULL;
3601 
3602 	} while (!__slab_update_freelist(s, slab,
3603 		freelist, counters,
3604 		NULL, new.counters,
3605 		"get_freelist"));
3606 
3607 	return freelist;
3608 }
3609 
3610 /*
3611  * Freeze the partial slab and return the pointer to the freelist.
3612  */
freeze_slab(struct kmem_cache * s,struct slab * slab)3613 static inline void *freeze_slab(struct kmem_cache *s, struct slab *slab)
3614 {
3615 	struct slab new;
3616 	unsigned long counters;
3617 	void *freelist;
3618 
3619 	do {
3620 		freelist = slab->freelist;
3621 		counters = slab->counters;
3622 
3623 		new.counters = counters;
3624 		VM_BUG_ON(new.frozen);
3625 
3626 		new.inuse = slab->objects;
3627 		new.frozen = 1;
3628 
3629 	} while (!slab_update_freelist(s, slab,
3630 		freelist, counters,
3631 		NULL, new.counters,
3632 		"freeze_slab"));
3633 
3634 	return freelist;
3635 }
3636 
3637 /*
3638  * Slow path. The lockless freelist is empty or we need to perform
3639  * debugging duties.
3640  *
3641  * Processing is still very fast if new objects have been freed to the
3642  * regular freelist. In that case we simply take over the regular freelist
3643  * as the lockless freelist and zap the regular freelist.
3644  *
3645  * If that is not working then we fall back to the partial lists. We take the
3646  * first element of the freelist as the object to allocate now and move the
3647  * rest of the freelist to the lockless freelist.
3648  *
3649  * And if we were unable to get a new slab from the partial slab lists then
3650  * we need to allocate a new slab. This is the slowest path since it involves
3651  * a call to the page allocator and the setup of a new slab.
3652  *
3653  * Version of __slab_alloc to use when we know that preemption is
3654  * already disabled (which is the case for bulk allocation).
3655  */
___slab_alloc(struct kmem_cache * s,gfp_t gfpflags,int node,unsigned long addr,struct kmem_cache_cpu * c,unsigned int orig_size)3656 static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
3657 			  unsigned long addr, struct kmem_cache_cpu *c, unsigned int orig_size)
3658 {
3659 	void *freelist;
3660 	struct slab *slab;
3661 	unsigned long flags;
3662 	struct partial_context pc;
3663 	bool try_thisnode = true;
3664 
3665 	stat(s, ALLOC_SLOWPATH);
3666 
3667 reread_slab:
3668 
3669 	slab = READ_ONCE(c->slab);
3670 	if (!slab) {
3671 		/*
3672 		 * if the node is not online or has no normal memory, just
3673 		 * ignore the node constraint
3674 		 */
3675 		if (unlikely(node != NUMA_NO_NODE &&
3676 			     !node_isset(node, slab_nodes)))
3677 			node = NUMA_NO_NODE;
3678 		goto new_slab;
3679 	}
3680 
3681 	if (unlikely(!node_match(slab, node))) {
3682 		/*
3683 		 * same as above but node_match() being false already
3684 		 * implies node != NUMA_NO_NODE
3685 		 */
3686 		if (!node_isset(node, slab_nodes)) {
3687 			node = NUMA_NO_NODE;
3688 		} else {
3689 			stat(s, ALLOC_NODE_MISMATCH);
3690 			goto deactivate_slab;
3691 		}
3692 	}
3693 
3694 	/*
3695 	 * By rights, we should be searching for a slab page that was
3696 	 * PFMEMALLOC but right now, we are losing the pfmemalloc
3697 	 * information when the page leaves the per-cpu allocator
3698 	 */
3699 	if (unlikely(!pfmemalloc_match(slab, gfpflags)))
3700 		goto deactivate_slab;
3701 
3702 	/* must check again c->slab in case we got preempted and it changed */
3703 	local_lock_irqsave(&s->cpu_slab->lock, flags);
3704 	if (unlikely(slab != c->slab)) {
3705 		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3706 		goto reread_slab;
3707 	}
3708 	freelist = c->freelist;
3709 	if (freelist)
3710 		goto load_freelist;
3711 
3712 	freelist = get_freelist(s, slab);
3713 
3714 	if (!freelist) {
3715 		c->slab = NULL;
3716 		c->tid = next_tid(c->tid);
3717 		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3718 		stat(s, DEACTIVATE_BYPASS);
3719 		goto new_slab;
3720 	}
3721 
3722 	stat(s, ALLOC_REFILL);
3723 
3724 load_freelist:
3725 
3726 	lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock));
3727 
3728 	/*
3729 	 * freelist is pointing to the list of objects to be used.
3730 	 * slab is pointing to the slab from which the objects are obtained.
3731 	 * That slab must be frozen for per cpu allocations to work.
3732 	 */
3733 	VM_BUG_ON(!c->slab->frozen);
3734 	c->freelist = get_freepointer(s, freelist);
3735 	c->tid = next_tid(c->tid);
3736 	local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3737 	return freelist;
3738 
3739 deactivate_slab:
3740 
3741 	local_lock_irqsave(&s->cpu_slab->lock, flags);
3742 	if (slab != c->slab) {
3743 		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3744 		goto reread_slab;
3745 	}
3746 	freelist = c->freelist;
3747 	c->slab = NULL;
3748 	c->freelist = NULL;
3749 	c->tid = next_tid(c->tid);
3750 	local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3751 	deactivate_slab(s, slab, freelist);
3752 
3753 new_slab:
3754 
3755 #ifdef CONFIG_SLUB_CPU_PARTIAL
3756 	while (slub_percpu_partial(c)) {
3757 		local_lock_irqsave(&s->cpu_slab->lock, flags);
3758 		if (unlikely(c->slab)) {
3759 			local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3760 			goto reread_slab;
3761 		}
3762 		if (unlikely(!slub_percpu_partial(c))) {
3763 			local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3764 			/* we were preempted and partial list got empty */
3765 			goto new_objects;
3766 		}
3767 
3768 		slab = slub_percpu_partial(c);
3769 		slub_set_percpu_partial(c, slab);
3770 
3771 		if (likely(node_match(slab, node) &&
3772 			   pfmemalloc_match(slab, gfpflags))) {
3773 			c->slab = slab;
3774 			freelist = get_freelist(s, slab);
3775 			VM_BUG_ON(!freelist);
3776 			stat(s, CPU_PARTIAL_ALLOC);
3777 			goto load_freelist;
3778 		}
3779 
3780 		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3781 
3782 		slab->next = NULL;
3783 		__put_partials(s, slab);
3784 	}
3785 #endif
3786 
3787 new_objects:
3788 
3789 	pc.flags = gfpflags;
3790 	/*
3791 	 * When a preferred node is indicated but no __GFP_THISNODE
3792 	 *
3793 	 * 1) try to get a partial slab from target node only by having
3794 	 *    __GFP_THISNODE in pc.flags for get_partial()
3795 	 * 2) if 1) failed, try to allocate a new slab from target node with
3796 	 *    GPF_NOWAIT | __GFP_THISNODE opportunistically
3797 	 * 3) if 2) failed, retry with original gfpflags which will allow
3798 	 *    get_partial() try partial lists of other nodes before potentially
3799 	 *    allocating new page from other nodes
3800 	 */
3801 	if (unlikely(node != NUMA_NO_NODE && !(gfpflags & __GFP_THISNODE)
3802 		     && try_thisnode))
3803 		pc.flags = GFP_NOWAIT | __GFP_THISNODE;
3804 
3805 	pc.orig_size = orig_size;
3806 	slab = get_partial(s, node, &pc);
3807 	if (slab) {
3808 		if (kmem_cache_debug(s)) {
3809 			freelist = pc.object;
3810 			/*
3811 			 * For debug caches here we had to go through
3812 			 * alloc_single_from_partial() so just store the
3813 			 * tracking info and return the object.
3814 			 */
3815 			if (s->flags & SLAB_STORE_USER)
3816 				set_track(s, freelist, TRACK_ALLOC, addr);
3817 
3818 			return freelist;
3819 		}
3820 
3821 		freelist = freeze_slab(s, slab);
3822 		goto retry_load_slab;
3823 	}
3824 
3825 	slub_put_cpu_ptr(s->cpu_slab);
3826 	slab = new_slab(s, pc.flags, node);
3827 	c = slub_get_cpu_ptr(s->cpu_slab);
3828 
3829 	if (unlikely(!slab)) {
3830 		if (node != NUMA_NO_NODE && !(gfpflags & __GFP_THISNODE)
3831 		    && try_thisnode) {
3832 			try_thisnode = false;
3833 			goto new_objects;
3834 		}
3835 		slab_out_of_memory(s, gfpflags, node);
3836 		return NULL;
3837 	}
3838 
3839 	stat(s, ALLOC_SLAB);
3840 
3841 	if (kmem_cache_debug(s)) {
3842 		freelist = alloc_single_from_new_slab(s, slab, orig_size);
3843 
3844 		if (unlikely(!freelist))
3845 			goto new_objects;
3846 
3847 		if (s->flags & SLAB_STORE_USER)
3848 			set_track(s, freelist, TRACK_ALLOC, addr);
3849 
3850 		return freelist;
3851 	}
3852 
3853 	/*
3854 	 * No other reference to the slab yet so we can
3855 	 * muck around with it freely without cmpxchg
3856 	 */
3857 	freelist = slab->freelist;
3858 	slab->freelist = NULL;
3859 	slab->inuse = slab->objects;
3860 	slab->frozen = 1;
3861 
3862 	inc_slabs_node(s, slab_nid(slab), slab->objects);
3863 
3864 	if (unlikely(!pfmemalloc_match(slab, gfpflags))) {
3865 		/*
3866 		 * For !pfmemalloc_match() case we don't load freelist so that
3867 		 * we don't make further mismatched allocations easier.
3868 		 */
3869 		deactivate_slab(s, slab, get_freepointer(s, freelist));
3870 		return freelist;
3871 	}
3872 
3873 retry_load_slab:
3874 
3875 	local_lock_irqsave(&s->cpu_slab->lock, flags);
3876 	if (unlikely(c->slab)) {
3877 		void *flush_freelist = c->freelist;
3878 		struct slab *flush_slab = c->slab;
3879 
3880 		c->slab = NULL;
3881 		c->freelist = NULL;
3882 		c->tid = next_tid(c->tid);
3883 
3884 		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3885 
3886 		deactivate_slab(s, flush_slab, flush_freelist);
3887 
3888 		stat(s, CPUSLAB_FLUSH);
3889 
3890 		goto retry_load_slab;
3891 	}
3892 	c->slab = slab;
3893 
3894 	goto load_freelist;
3895 }
3896 
3897 /*
3898  * A wrapper for ___slab_alloc() for contexts where preemption is not yet
3899  * disabled. Compensates for possible cpu changes by refetching the per cpu area
3900  * pointer.
3901  */
__slab_alloc(struct kmem_cache * s,gfp_t gfpflags,int node,unsigned long addr,struct kmem_cache_cpu * c,unsigned int orig_size)3902 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
3903 			  unsigned long addr, struct kmem_cache_cpu *c, unsigned int orig_size)
3904 {
3905 	void *p;
3906 
3907 #ifdef CONFIG_PREEMPT_COUNT
3908 	/*
3909 	 * We may have been preempted and rescheduled on a different
3910 	 * cpu before disabling preemption. Need to reload cpu area
3911 	 * pointer.
3912 	 */
3913 	c = slub_get_cpu_ptr(s->cpu_slab);
3914 #endif
3915 
3916 	p = ___slab_alloc(s, gfpflags, node, addr, c, orig_size);
3917 #ifdef CONFIG_PREEMPT_COUNT
3918 	slub_put_cpu_ptr(s->cpu_slab);
3919 #endif
3920 	return p;
3921 }
3922 
__slab_alloc_node(struct kmem_cache * s,gfp_t gfpflags,int node,unsigned long addr,size_t orig_size)3923 static __always_inline void *__slab_alloc_node(struct kmem_cache *s,
3924 		gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
3925 {
3926 	struct kmem_cache_cpu *c;
3927 	struct slab *slab;
3928 	unsigned long tid;
3929 	void *object;
3930 
3931 redo:
3932 	/*
3933 	 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
3934 	 * enabled. We may switch back and forth between cpus while
3935 	 * reading from one cpu area. That does not matter as long
3936 	 * as we end up on the original cpu again when doing the cmpxchg.
3937 	 *
3938 	 * We must guarantee that tid and kmem_cache_cpu are retrieved on the
3939 	 * same cpu. We read first the kmem_cache_cpu pointer and use it to read
3940 	 * the tid. If we are preempted and switched to another cpu between the
3941 	 * two reads, it's OK as the two are still associated with the same cpu
3942 	 * and cmpxchg later will validate the cpu.
3943 	 */
3944 	c = raw_cpu_ptr(s->cpu_slab);
3945 	tid = READ_ONCE(c->tid);
3946 
3947 	/*
3948 	 * Irqless object alloc/free algorithm used here depends on sequence
3949 	 * of fetching cpu_slab's data. tid should be fetched before anything
3950 	 * on c to guarantee that object and slab associated with previous tid
3951 	 * won't be used with current tid. If we fetch tid first, object and
3952 	 * slab could be one associated with next tid and our alloc/free
3953 	 * request will be failed. In this case, we will retry. So, no problem.
3954 	 */
3955 	barrier();
3956 
3957 	/*
3958 	 * The transaction ids are globally unique per cpu and per operation on
3959 	 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
3960 	 * occurs on the right processor and that there was no operation on the
3961 	 * linked list in between.
3962 	 */
3963 
3964 	object = c->freelist;
3965 	slab = c->slab;
3966 
3967 #ifdef CONFIG_NUMA
3968 	if (static_branch_unlikely(&strict_numa) &&
3969 			node == NUMA_NO_NODE) {
3970 
3971 		struct mempolicy *mpol = current->mempolicy;
3972 
3973 		if (mpol) {
3974 			/*
3975 			 * Special BIND rule support. If existing slab
3976 			 * is in permitted set then do not redirect
3977 			 * to a particular node.
3978 			 * Otherwise we apply the memory policy to get
3979 			 * the node we need to allocate on.
3980 			 */
3981 			if (mpol->mode != MPOL_BIND || !slab ||
3982 					!node_isset(slab_nid(slab), mpol->nodes))
3983 
3984 				node = mempolicy_slab_node();
3985 		}
3986 	}
3987 #endif
3988 
3989 	if (!USE_LOCKLESS_FAST_PATH() ||
3990 	    unlikely(!object || !slab || !node_match(slab, node))) {
3991 		object = __slab_alloc(s, gfpflags, node, addr, c, orig_size);
3992 	} else {
3993 		void *next_object = get_freepointer_safe(s, object);
3994 
3995 		/*
3996 		 * The cmpxchg will only match if there was no additional
3997 		 * operation and if we are on the right processor.
3998 		 *
3999 		 * The cmpxchg does the following atomically (without lock
4000 		 * semantics!)
4001 		 * 1. Relocate first pointer to the current per cpu area.
4002 		 * 2. Verify that tid and freelist have not been changed
4003 		 * 3. If they were not changed replace tid and freelist
4004 		 *
4005 		 * Since this is without lock semantics the protection is only
4006 		 * against code executing on this cpu *not* from access by
4007 		 * other cpus.
4008 		 */
4009 		if (unlikely(!__update_cpu_freelist_fast(s, object, next_object, tid))) {
4010 			note_cmpxchg_failure("slab_alloc", s, tid);
4011 			goto redo;
4012 		}
4013 		prefetch_freepointer(s, next_object);
4014 		stat(s, ALLOC_FASTPATH);
4015 	}
4016 
4017 	return object;
4018 }
4019 #else /* CONFIG_SLUB_TINY */
__slab_alloc_node(struct kmem_cache * s,gfp_t gfpflags,int node,unsigned long addr,size_t orig_size)4020 static void *__slab_alloc_node(struct kmem_cache *s,
4021 		gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
4022 {
4023 	struct partial_context pc;
4024 	struct slab *slab;
4025 	void *object;
4026 
4027 	pc.flags = gfpflags;
4028 	pc.orig_size = orig_size;
4029 	slab = get_partial(s, node, &pc);
4030 
4031 	if (slab)
4032 		return pc.object;
4033 
4034 	slab = new_slab(s, gfpflags, node);
4035 	if (unlikely(!slab)) {
4036 		slab_out_of_memory(s, gfpflags, node);
4037 		return NULL;
4038 	}
4039 
4040 	object = alloc_single_from_new_slab(s, slab, orig_size);
4041 
4042 	return object;
4043 }
4044 #endif /* CONFIG_SLUB_TINY */
4045 
4046 /*
4047  * If the object has been wiped upon free, make sure it's fully initialized by
4048  * zeroing out freelist pointer.
4049  *
4050  * Note that we also wipe custom freelist pointers.
4051  */
maybe_wipe_obj_freeptr(struct kmem_cache * s,void * obj)4052 static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s,
4053 						   void *obj)
4054 {
4055 	if (unlikely(slab_want_init_on_free(s)) && obj &&
4056 	    !freeptr_outside_object(s))
4057 		memset((void *)((char *)kasan_reset_tag(obj) + s->offset),
4058 			0, sizeof(void *));
4059 }
4060 
4061 static __fastpath_inline
slab_pre_alloc_hook(struct kmem_cache * s,gfp_t flags)4062 struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
4063 {
4064 	flags &= gfp_allowed_mask;
4065 
4066 	might_alloc(flags);
4067 
4068 	if (unlikely(should_failslab(s, flags)))
4069 		return NULL;
4070 
4071 	return s;
4072 }
4073 
4074 static __fastpath_inline
slab_post_alloc_hook(struct kmem_cache * s,struct list_lru * lru,gfp_t flags,size_t size,void ** p,bool init,unsigned int orig_size)4075 bool slab_post_alloc_hook(struct kmem_cache *s, struct list_lru *lru,
4076 			  gfp_t flags, size_t size, void **p, bool init,
4077 			  unsigned int orig_size)
4078 {
4079 	unsigned int zero_size = s->object_size;
4080 	bool kasan_init = init;
4081 	size_t i;
4082 	gfp_t init_flags = flags & gfp_allowed_mask;
4083 
4084 	/*
4085 	 * For kmalloc object, the allocated memory size(object_size) is likely
4086 	 * larger than the requested size(orig_size). If redzone check is
4087 	 * enabled for the extra space, don't zero it, as it will be redzoned
4088 	 * soon. The redzone operation for this extra space could be seen as a
4089 	 * replacement of current poisoning under certain debug option, and
4090 	 * won't break other sanity checks.
4091 	 */
4092 	if (kmem_cache_debug_flags(s, SLAB_STORE_USER | SLAB_RED_ZONE) &&
4093 	    (s->flags & SLAB_KMALLOC))
4094 		zero_size = orig_size;
4095 
4096 	/*
4097 	 * When slab_debug is enabled, avoid memory initialization integrated
4098 	 * into KASAN and instead zero out the memory via the memset below with
4099 	 * the proper size. Otherwise, KASAN might overwrite SLUB redzones and
4100 	 * cause false-positive reports. This does not lead to a performance
4101 	 * penalty on production builds, as slab_debug is not intended to be
4102 	 * enabled there.
4103 	 */
4104 	if (__slub_debug_enabled())
4105 		kasan_init = false;
4106 
4107 	/*
4108 	 * As memory initialization might be integrated into KASAN,
4109 	 * kasan_slab_alloc and initialization memset must be
4110 	 * kept together to avoid discrepancies in behavior.
4111 	 *
4112 	 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
4113 	 */
4114 	for (i = 0; i < size; i++) {
4115 		p[i] = kasan_slab_alloc(s, p[i], init_flags, kasan_init);
4116 		if (p[i] && init && (!kasan_init ||
4117 				     !kasan_has_integrated_init()))
4118 			memset(p[i], 0, zero_size);
4119 		kmemleak_alloc_recursive(p[i], s->object_size, 1,
4120 					 s->flags, init_flags);
4121 		kmsan_slab_alloc(s, p[i], init_flags);
4122 		alloc_tagging_slab_alloc_hook(s, p[i], flags);
4123 	}
4124 
4125 	return memcg_slab_post_alloc_hook(s, lru, flags, size, p);
4126 }
4127 
4128 /*
4129  * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
4130  * have the fastpath folded into their functions. So no function call
4131  * overhead for requests that can be satisfied on the fastpath.
4132  *
4133  * The fastpath works by first checking if the lockless freelist can be used.
4134  * If not then __slab_alloc is called for slow processing.
4135  *
4136  * Otherwise we can simply pick the next object from the lockless free list.
4137  */
slab_alloc_node(struct kmem_cache * s,struct list_lru * lru,gfp_t gfpflags,int node,unsigned long addr,size_t orig_size)4138 static __fastpath_inline void *slab_alloc_node(struct kmem_cache *s, struct list_lru *lru,
4139 		gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
4140 {
4141 	void *object;
4142 	bool init = false;
4143 
4144 	s = slab_pre_alloc_hook(s, gfpflags);
4145 	if (unlikely(!s))
4146 		return NULL;
4147 
4148 	object = kfence_alloc(s, orig_size, gfpflags);
4149 	if (unlikely(object))
4150 		goto out;
4151 
4152 	object = __slab_alloc_node(s, gfpflags, node, addr, orig_size);
4153 
4154 	maybe_wipe_obj_freeptr(s, object);
4155 	init = slab_want_init_on_alloc(gfpflags, s);
4156 
4157 out:
4158 	/*
4159 	 * When init equals 'true', like for kzalloc() family, only
4160 	 * @orig_size bytes might be zeroed instead of s->object_size
4161 	 * In case this fails due to memcg_slab_post_alloc_hook(),
4162 	 * object is set to NULL
4163 	 */
4164 	slab_post_alloc_hook(s, lru, gfpflags, 1, &object, init, orig_size);
4165 
4166 	return object;
4167 }
4168 
kmem_cache_alloc_noprof(struct kmem_cache * s,gfp_t gfpflags)4169 void *kmem_cache_alloc_noprof(struct kmem_cache *s, gfp_t gfpflags)
4170 {
4171 	void *ret = slab_alloc_node(s, NULL, gfpflags, NUMA_NO_NODE, _RET_IP_,
4172 				    s->object_size);
4173 
4174 	trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, NUMA_NO_NODE);
4175 
4176 	return ret;
4177 }
4178 EXPORT_SYMBOL(kmem_cache_alloc_noprof);
4179 
kmem_cache_alloc_lru_noprof(struct kmem_cache * s,struct list_lru * lru,gfp_t gfpflags)4180 void *kmem_cache_alloc_lru_noprof(struct kmem_cache *s, struct list_lru *lru,
4181 			   gfp_t gfpflags)
4182 {
4183 	void *ret = slab_alloc_node(s, lru, gfpflags, NUMA_NO_NODE, _RET_IP_,
4184 				    s->object_size);
4185 
4186 	trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, NUMA_NO_NODE);
4187 
4188 	return ret;
4189 }
4190 EXPORT_SYMBOL(kmem_cache_alloc_lru_noprof);
4191 
kmem_cache_charge(void * objp,gfp_t gfpflags)4192 bool kmem_cache_charge(void *objp, gfp_t gfpflags)
4193 {
4194 	if (!memcg_kmem_online())
4195 		return true;
4196 
4197 	return memcg_slab_post_charge(objp, gfpflags);
4198 }
4199 EXPORT_SYMBOL(kmem_cache_charge);
4200 
4201 /**
4202  * kmem_cache_alloc_node - Allocate an object on the specified node
4203  * @s: The cache to allocate from.
4204  * @gfpflags: See kmalloc().
4205  * @node: node number of the target node.
4206  *
4207  * Identical to kmem_cache_alloc but it will allocate memory on the given
4208  * node, which can improve the performance for cpu bound structures.
4209  *
4210  * Fallback to other node is possible if __GFP_THISNODE is not set.
4211  *
4212  * Return: pointer to the new object or %NULL in case of error
4213  */
kmem_cache_alloc_node_noprof(struct kmem_cache * s,gfp_t gfpflags,int node)4214 void *kmem_cache_alloc_node_noprof(struct kmem_cache *s, gfp_t gfpflags, int node)
4215 {
4216 	void *ret = slab_alloc_node(s, NULL, gfpflags, node, _RET_IP_, s->object_size);
4217 
4218 	trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, node);
4219 
4220 	return ret;
4221 }
4222 EXPORT_SYMBOL(kmem_cache_alloc_node_noprof);
4223 
4224 /*
4225  * To avoid unnecessary overhead, we pass through large allocation requests
4226  * directly to the page allocator. We use __GFP_COMP, because we will need to
4227  * know the allocation order to free the pages properly in kfree.
4228  */
___kmalloc_large_node(size_t size,gfp_t flags,int node)4229 static void *___kmalloc_large_node(size_t size, gfp_t flags, int node)
4230 {
4231 	struct folio *folio;
4232 	void *ptr = NULL;
4233 	unsigned int order = get_order(size);
4234 
4235 	if (unlikely(flags & GFP_SLAB_BUG_MASK))
4236 		flags = kmalloc_fix_flags(flags);
4237 
4238 	flags |= __GFP_COMP;
4239 	folio = (struct folio *)alloc_pages_node_noprof(node, flags, order);
4240 	if (folio) {
4241 		ptr = folio_address(folio);
4242 		lruvec_stat_mod_folio(folio, NR_SLAB_UNRECLAIMABLE_B,
4243 				      PAGE_SIZE << order);
4244 	}
4245 
4246 	ptr = kasan_kmalloc_large(ptr, size, flags);
4247 	/* As ptr might get tagged, call kmemleak hook after KASAN. */
4248 	kmemleak_alloc(ptr, size, 1, flags);
4249 	kmsan_kmalloc_large(ptr, size, flags);
4250 
4251 	return ptr;
4252 }
4253 
__kmalloc_large_noprof(size_t size,gfp_t flags)4254 void *__kmalloc_large_noprof(size_t size, gfp_t flags)
4255 {
4256 	void *ret = ___kmalloc_large_node(size, flags, NUMA_NO_NODE);
4257 
4258 	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
4259 		      flags, NUMA_NO_NODE);
4260 	return ret;
4261 }
4262 EXPORT_SYMBOL(__kmalloc_large_noprof);
4263 
__kmalloc_large_node_noprof(size_t size,gfp_t flags,int node)4264 void *__kmalloc_large_node_noprof(size_t size, gfp_t flags, int node)
4265 {
4266 	void *ret = ___kmalloc_large_node(size, flags, node);
4267 
4268 	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
4269 		      flags, node);
4270 	return ret;
4271 }
4272 EXPORT_SYMBOL(__kmalloc_large_node_noprof);
4273 
4274 static __always_inline
__do_kmalloc_node(size_t size,kmem_buckets * b,gfp_t flags,int node,unsigned long caller)4275 void *__do_kmalloc_node(size_t size, kmem_buckets *b, gfp_t flags, int node,
4276 			unsigned long caller)
4277 {
4278 	struct kmem_cache *s;
4279 	void *ret;
4280 
4281 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
4282 		ret = __kmalloc_large_node_noprof(size, flags, node);
4283 		trace_kmalloc(caller, ret, size,
4284 			      PAGE_SIZE << get_order(size), flags, node);
4285 		return ret;
4286 	}
4287 
4288 	if (unlikely(!size))
4289 		return ZERO_SIZE_PTR;
4290 
4291 	s = kmalloc_slab(size, b, flags, caller);
4292 
4293 	ret = slab_alloc_node(s, NULL, flags, node, caller, size);
4294 	ret = kasan_kmalloc(s, ret, size, flags);
4295 	trace_kmalloc(caller, ret, size, s->size, flags, node);
4296 	return ret;
4297 }
__kmalloc_node_noprof(DECL_BUCKET_PARAMS (size,b),gfp_t flags,int node)4298 void *__kmalloc_node_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node)
4299 {
4300 	return __do_kmalloc_node(size, PASS_BUCKET_PARAM(b), flags, node, _RET_IP_);
4301 }
4302 EXPORT_SYMBOL(__kmalloc_node_noprof);
4303 
__kmalloc_noprof(size_t size,gfp_t flags)4304 void *__kmalloc_noprof(size_t size, gfp_t flags)
4305 {
4306 	return __do_kmalloc_node(size, NULL, flags, NUMA_NO_NODE, _RET_IP_);
4307 }
4308 EXPORT_SYMBOL(__kmalloc_noprof);
4309 
__kmalloc_node_track_caller_noprof(DECL_BUCKET_PARAMS (size,b),gfp_t flags,int node,unsigned long caller)4310 void *__kmalloc_node_track_caller_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags,
4311 					 int node, unsigned long caller)
4312 {
4313 	return __do_kmalloc_node(size, PASS_BUCKET_PARAM(b), flags, node, caller);
4314 
4315 }
4316 EXPORT_SYMBOL(__kmalloc_node_track_caller_noprof);
4317 
__kmalloc_cache_noprof(struct kmem_cache * s,gfp_t gfpflags,size_t size)4318 void *__kmalloc_cache_noprof(struct kmem_cache *s, gfp_t gfpflags, size_t size)
4319 {
4320 	void *ret = slab_alloc_node(s, NULL, gfpflags, NUMA_NO_NODE,
4321 					    _RET_IP_, size);
4322 
4323 	trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, NUMA_NO_NODE);
4324 
4325 	ret = kasan_kmalloc(s, ret, size, gfpflags);
4326 	return ret;
4327 }
4328 EXPORT_SYMBOL(__kmalloc_cache_noprof);
4329 
__kmalloc_cache_node_noprof(struct kmem_cache * s,gfp_t gfpflags,int node,size_t size)4330 void *__kmalloc_cache_node_noprof(struct kmem_cache *s, gfp_t gfpflags,
4331 				  int node, size_t size)
4332 {
4333 	void *ret = slab_alloc_node(s, NULL, gfpflags, node, _RET_IP_, size);
4334 
4335 	trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, node);
4336 
4337 	ret = kasan_kmalloc(s, ret, size, gfpflags);
4338 	return ret;
4339 }
4340 EXPORT_SYMBOL(__kmalloc_cache_node_noprof);
4341 
free_to_partial_list(struct kmem_cache * s,struct slab * slab,void * head,void * tail,int bulk_cnt,unsigned long addr)4342 static noinline void free_to_partial_list(
4343 	struct kmem_cache *s, struct slab *slab,
4344 	void *head, void *tail, int bulk_cnt,
4345 	unsigned long addr)
4346 {
4347 	struct kmem_cache_node *n = get_node(s, slab_nid(slab));
4348 	struct slab *slab_free = NULL;
4349 	int cnt = bulk_cnt;
4350 	unsigned long flags;
4351 	depot_stack_handle_t handle = 0;
4352 
4353 	if (s->flags & SLAB_STORE_USER)
4354 		handle = set_track_prepare();
4355 
4356 	spin_lock_irqsave(&n->list_lock, flags);
4357 
4358 	if (free_debug_processing(s, slab, head, tail, &cnt, addr, handle)) {
4359 		void *prior = slab->freelist;
4360 
4361 		/* Perform the actual freeing while we still hold the locks */
4362 		slab->inuse -= cnt;
4363 		set_freepointer(s, tail, prior);
4364 		slab->freelist = head;
4365 
4366 		/*
4367 		 * If the slab is empty, and node's partial list is full,
4368 		 * it should be discarded anyway no matter it's on full or
4369 		 * partial list.
4370 		 */
4371 		if (slab->inuse == 0 && n->nr_partial >= s->min_partial)
4372 			slab_free = slab;
4373 
4374 		if (!prior) {
4375 			/* was on full list */
4376 			remove_full(s, n, slab);
4377 			if (!slab_free) {
4378 				add_partial(n, slab, DEACTIVATE_TO_TAIL);
4379 				stat(s, FREE_ADD_PARTIAL);
4380 			}
4381 		} else if (slab_free) {
4382 			remove_partial(n, slab);
4383 			stat(s, FREE_REMOVE_PARTIAL);
4384 		}
4385 	}
4386 
4387 	if (slab_free) {
4388 		/*
4389 		 * Update the counters while still holding n->list_lock to
4390 		 * prevent spurious validation warnings
4391 		 */
4392 		dec_slabs_node(s, slab_nid(slab_free), slab_free->objects);
4393 	}
4394 
4395 	spin_unlock_irqrestore(&n->list_lock, flags);
4396 
4397 	if (slab_free) {
4398 		stat(s, FREE_SLAB);
4399 		free_slab(s, slab_free);
4400 	}
4401 }
4402 
4403 /*
4404  * Slow path handling. This may still be called frequently since objects
4405  * have a longer lifetime than the cpu slabs in most processing loads.
4406  *
4407  * So we still attempt to reduce cache line usage. Just take the slab
4408  * lock and free the item. If there is no additional partial slab
4409  * handling required then we can return immediately.
4410  */
__slab_free(struct kmem_cache * s,struct slab * slab,void * head,void * tail,int cnt,unsigned long addr)4411 static void __slab_free(struct kmem_cache *s, struct slab *slab,
4412 			void *head, void *tail, int cnt,
4413 			unsigned long addr)
4414 
4415 {
4416 	void *prior;
4417 	int was_frozen;
4418 	struct slab new;
4419 	unsigned long counters;
4420 	struct kmem_cache_node *n = NULL;
4421 	unsigned long flags;
4422 	bool on_node_partial;
4423 
4424 	stat(s, FREE_SLOWPATH);
4425 
4426 	if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) {
4427 		free_to_partial_list(s, slab, head, tail, cnt, addr);
4428 		return;
4429 	}
4430 
4431 	do {
4432 		if (unlikely(n)) {
4433 			spin_unlock_irqrestore(&n->list_lock, flags);
4434 			n = NULL;
4435 		}
4436 		prior = slab->freelist;
4437 		counters = slab->counters;
4438 		set_freepointer(s, tail, prior);
4439 		new.counters = counters;
4440 		was_frozen = new.frozen;
4441 		new.inuse -= cnt;
4442 		if ((!new.inuse || !prior) && !was_frozen) {
4443 			/* Needs to be taken off a list */
4444 			if (!kmem_cache_has_cpu_partial(s) || prior) {
4445 
4446 				n = get_node(s, slab_nid(slab));
4447 				/*
4448 				 * Speculatively acquire the list_lock.
4449 				 * If the cmpxchg does not succeed then we may
4450 				 * drop the list_lock without any processing.
4451 				 *
4452 				 * Otherwise the list_lock will synchronize with
4453 				 * other processors updating the list of slabs.
4454 				 */
4455 				spin_lock_irqsave(&n->list_lock, flags);
4456 
4457 				on_node_partial = slab_test_node_partial(slab);
4458 			}
4459 		}
4460 
4461 	} while (!slab_update_freelist(s, slab,
4462 		prior, counters,
4463 		head, new.counters,
4464 		"__slab_free"));
4465 
4466 	if (likely(!n)) {
4467 
4468 		if (likely(was_frozen)) {
4469 			/*
4470 			 * The list lock was not taken therefore no list
4471 			 * activity can be necessary.
4472 			 */
4473 			stat(s, FREE_FROZEN);
4474 		} else if (kmem_cache_has_cpu_partial(s) && !prior) {
4475 			/*
4476 			 * If we started with a full slab then put it onto the
4477 			 * per cpu partial list.
4478 			 */
4479 			put_cpu_partial(s, slab, 1);
4480 			stat(s, CPU_PARTIAL_FREE);
4481 		}
4482 
4483 		return;
4484 	}
4485 
4486 	/*
4487 	 * This slab was partially empty but not on the per-node partial list,
4488 	 * in which case we shouldn't manipulate its list, just return.
4489 	 */
4490 	if (prior && !on_node_partial) {
4491 		spin_unlock_irqrestore(&n->list_lock, flags);
4492 		return;
4493 	}
4494 
4495 	if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
4496 		goto slab_empty;
4497 
4498 	/*
4499 	 * Objects left in the slab. If it was not on the partial list before
4500 	 * then add it.
4501 	 */
4502 	if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
4503 		add_partial(n, slab, DEACTIVATE_TO_TAIL);
4504 		stat(s, FREE_ADD_PARTIAL);
4505 	}
4506 	spin_unlock_irqrestore(&n->list_lock, flags);
4507 	return;
4508 
4509 slab_empty:
4510 	if (prior) {
4511 		/*
4512 		 * Slab on the partial list.
4513 		 */
4514 		remove_partial(n, slab);
4515 		stat(s, FREE_REMOVE_PARTIAL);
4516 	}
4517 
4518 	spin_unlock_irqrestore(&n->list_lock, flags);
4519 	stat(s, FREE_SLAB);
4520 	discard_slab(s, slab);
4521 }
4522 
4523 #ifndef CONFIG_SLUB_TINY
4524 /*
4525  * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
4526  * can perform fastpath freeing without additional function calls.
4527  *
4528  * The fastpath is only possible if we are freeing to the current cpu slab
4529  * of this processor. This typically the case if we have just allocated
4530  * the item before.
4531  *
4532  * If fastpath is not possible then fall back to __slab_free where we deal
4533  * with all sorts of special processing.
4534  *
4535  * Bulk free of a freelist with several objects (all pointing to the
4536  * same slab) possible by specifying head and tail ptr, plus objects
4537  * count (cnt). Bulk free indicated by tail pointer being set.
4538  */
do_slab_free(struct kmem_cache * s,struct slab * slab,void * head,void * tail,int cnt,unsigned long addr)4539 static __always_inline void do_slab_free(struct kmem_cache *s,
4540 				struct slab *slab, void *head, void *tail,
4541 				int cnt, unsigned long addr)
4542 {
4543 	struct kmem_cache_cpu *c;
4544 	unsigned long tid;
4545 	void **freelist;
4546 
4547 redo:
4548 	/*
4549 	 * Determine the currently cpus per cpu slab.
4550 	 * The cpu may change afterward. However that does not matter since
4551 	 * data is retrieved via this pointer. If we are on the same cpu
4552 	 * during the cmpxchg then the free will succeed.
4553 	 */
4554 	c = raw_cpu_ptr(s->cpu_slab);
4555 	tid = READ_ONCE(c->tid);
4556 
4557 	/* Same with comment on barrier() in __slab_alloc_node() */
4558 	barrier();
4559 
4560 	if (unlikely(slab != c->slab)) {
4561 		__slab_free(s, slab, head, tail, cnt, addr);
4562 		return;
4563 	}
4564 
4565 	if (USE_LOCKLESS_FAST_PATH()) {
4566 		freelist = READ_ONCE(c->freelist);
4567 
4568 		set_freepointer(s, tail, freelist);
4569 
4570 		if (unlikely(!__update_cpu_freelist_fast(s, freelist, head, tid))) {
4571 			note_cmpxchg_failure("slab_free", s, tid);
4572 			goto redo;
4573 		}
4574 	} else {
4575 		/* Update the free list under the local lock */
4576 		local_lock(&s->cpu_slab->lock);
4577 		c = this_cpu_ptr(s->cpu_slab);
4578 		if (unlikely(slab != c->slab)) {
4579 			local_unlock(&s->cpu_slab->lock);
4580 			goto redo;
4581 		}
4582 		tid = c->tid;
4583 		freelist = c->freelist;
4584 
4585 		set_freepointer(s, tail, freelist);
4586 		c->freelist = head;
4587 		c->tid = next_tid(tid);
4588 
4589 		local_unlock(&s->cpu_slab->lock);
4590 	}
4591 	stat_add(s, FREE_FASTPATH, cnt);
4592 }
4593 #else /* CONFIG_SLUB_TINY */
do_slab_free(struct kmem_cache * s,struct slab * slab,void * head,void * tail,int cnt,unsigned long addr)4594 static void do_slab_free(struct kmem_cache *s,
4595 				struct slab *slab, void *head, void *tail,
4596 				int cnt, unsigned long addr)
4597 {
4598 	__slab_free(s, slab, head, tail, cnt, addr);
4599 }
4600 #endif /* CONFIG_SLUB_TINY */
4601 
4602 static __fastpath_inline
slab_free(struct kmem_cache * s,struct slab * slab,void * object,unsigned long addr)4603 void slab_free(struct kmem_cache *s, struct slab *slab, void *object,
4604 	       unsigned long addr)
4605 {
4606 	memcg_slab_free_hook(s, slab, &object, 1);
4607 	alloc_tagging_slab_free_hook(s, slab, &object, 1);
4608 
4609 	if (likely(slab_free_hook(s, object, slab_want_init_on_free(s), false)))
4610 		do_slab_free(s, slab, object, object, 1, addr);
4611 }
4612 
4613 #ifdef CONFIG_MEMCG
4614 /* Do not inline the rare memcg charging failed path into the allocation path */
4615 static noinline
memcg_alloc_abort_single(struct kmem_cache * s,void * object)4616 void memcg_alloc_abort_single(struct kmem_cache *s, void *object)
4617 {
4618 	if (likely(slab_free_hook(s, object, slab_want_init_on_free(s), false)))
4619 		do_slab_free(s, virt_to_slab(object), object, object, 1, _RET_IP_);
4620 }
4621 #endif
4622 
4623 static __fastpath_inline
slab_free_bulk(struct kmem_cache * s,struct slab * slab,void * head,void * tail,void ** p,int cnt,unsigned long addr)4624 void slab_free_bulk(struct kmem_cache *s, struct slab *slab, void *head,
4625 		    void *tail, void **p, int cnt, unsigned long addr)
4626 {
4627 	memcg_slab_free_hook(s, slab, p, cnt);
4628 	alloc_tagging_slab_free_hook(s, slab, p, cnt);
4629 	/*
4630 	 * With KASAN enabled slab_free_freelist_hook modifies the freelist
4631 	 * to remove objects, whose reuse must be delayed.
4632 	 */
4633 	if (likely(slab_free_freelist_hook(s, &head, &tail, &cnt)))
4634 		do_slab_free(s, slab, head, tail, cnt, addr);
4635 }
4636 
4637 #ifdef CONFIG_SLUB_RCU_DEBUG
slab_free_after_rcu_debug(struct rcu_head * rcu_head)4638 static void slab_free_after_rcu_debug(struct rcu_head *rcu_head)
4639 {
4640 	struct rcu_delayed_free *delayed_free =
4641 			container_of(rcu_head, struct rcu_delayed_free, head);
4642 	void *object = delayed_free->object;
4643 	struct slab *slab = virt_to_slab(object);
4644 	struct kmem_cache *s;
4645 
4646 	kfree(delayed_free);
4647 
4648 	if (WARN_ON(is_kfence_address(object)))
4649 		return;
4650 
4651 	/* find the object and the cache again */
4652 	if (WARN_ON(!slab))
4653 		return;
4654 	s = slab->slab_cache;
4655 	if (WARN_ON(!(s->flags & SLAB_TYPESAFE_BY_RCU)))
4656 		return;
4657 
4658 	/* resume freeing */
4659 	if (slab_free_hook(s, object, slab_want_init_on_free(s), true))
4660 		do_slab_free(s, slab, object, object, 1, _THIS_IP_);
4661 }
4662 #endif /* CONFIG_SLUB_RCU_DEBUG */
4663 
4664 #ifdef CONFIG_KASAN_GENERIC
___cache_free(struct kmem_cache * cache,void * x,unsigned long addr)4665 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
4666 {
4667 	do_slab_free(cache, virt_to_slab(x), x, x, 1, addr);
4668 }
4669 #endif
4670 
virt_to_cache(const void * obj)4671 static inline struct kmem_cache *virt_to_cache(const void *obj)
4672 {
4673 	struct slab *slab;
4674 
4675 	slab = virt_to_slab(obj);
4676 	if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n", __func__))
4677 		return NULL;
4678 	return slab->slab_cache;
4679 }
4680 
cache_from_obj(struct kmem_cache * s,void * x)4681 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
4682 {
4683 	struct kmem_cache *cachep;
4684 
4685 	if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
4686 	    !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
4687 		return s;
4688 
4689 	cachep = virt_to_cache(x);
4690 	if (WARN(cachep && cachep != s,
4691 		 "%s: Wrong slab cache. %s but object is from %s\n",
4692 		 __func__, s->name, cachep->name))
4693 		print_tracking(cachep, x);
4694 	return cachep;
4695 }
4696 
4697 /**
4698  * kmem_cache_free - Deallocate an object
4699  * @s: The cache the allocation was from.
4700  * @x: The previously allocated object.
4701  *
4702  * Free an object which was previously allocated from this
4703  * cache.
4704  */
kmem_cache_free(struct kmem_cache * s,void * x)4705 void kmem_cache_free(struct kmem_cache *s, void *x)
4706 {
4707 	s = cache_from_obj(s, x);
4708 	if (!s)
4709 		return;
4710 	trace_kmem_cache_free(_RET_IP_, x, s);
4711 	slab_free(s, virt_to_slab(x), x, _RET_IP_);
4712 }
4713 EXPORT_SYMBOL(kmem_cache_free);
4714 
free_large_kmalloc(struct folio * folio,void * object)4715 static void free_large_kmalloc(struct folio *folio, void *object)
4716 {
4717 	unsigned int order = folio_order(folio);
4718 
4719 	if (WARN_ON_ONCE(order == 0))
4720 		pr_warn_once("object pointer: 0x%p\n", object);
4721 
4722 	kmemleak_free(object);
4723 	kasan_kfree_large(object);
4724 	kmsan_kfree_large(object);
4725 
4726 	lruvec_stat_mod_folio(folio, NR_SLAB_UNRECLAIMABLE_B,
4727 			      -(PAGE_SIZE << order));
4728 	folio_put(folio);
4729 }
4730 
4731 /**
4732  * kfree - free previously allocated memory
4733  * @object: pointer returned by kmalloc() or kmem_cache_alloc()
4734  *
4735  * If @object is NULL, no operation is performed.
4736  */
kfree(const void * object)4737 void kfree(const void *object)
4738 {
4739 	struct folio *folio;
4740 	struct slab *slab;
4741 	struct kmem_cache *s;
4742 	void *x = (void *)object;
4743 
4744 	trace_kfree(_RET_IP_, object);
4745 
4746 	if (unlikely(ZERO_OR_NULL_PTR(object)))
4747 		return;
4748 
4749 	folio = virt_to_folio(object);
4750 	if (unlikely(!folio_test_slab(folio))) {
4751 		free_large_kmalloc(folio, (void *)object);
4752 		return;
4753 	}
4754 
4755 	slab = folio_slab(folio);
4756 	s = slab->slab_cache;
4757 	slab_free(s, slab, x, _RET_IP_);
4758 }
4759 EXPORT_SYMBOL(kfree);
4760 
4761 static __always_inline __realloc_size(2) void *
__do_krealloc(const void * p,size_t new_size,gfp_t flags)4762 __do_krealloc(const void *p, size_t new_size, gfp_t flags)
4763 {
4764 	void *ret;
4765 	size_t ks = 0;
4766 	int orig_size = 0;
4767 	struct kmem_cache *s = NULL;
4768 
4769 	if (unlikely(ZERO_OR_NULL_PTR(p)))
4770 		goto alloc_new;
4771 
4772 	/* Check for double-free. */
4773 	if (!kasan_check_byte(p))
4774 		return NULL;
4775 
4776 	if (is_kfence_address(p)) {
4777 		ks = orig_size = kfence_ksize(p);
4778 	} else {
4779 		struct folio *folio;
4780 
4781 		folio = virt_to_folio(p);
4782 		if (unlikely(!folio_test_slab(folio))) {
4783 			/* Big kmalloc object */
4784 			WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE);
4785 			WARN_ON(p != folio_address(folio));
4786 			ks = folio_size(folio);
4787 		} else {
4788 			s = folio_slab(folio)->slab_cache;
4789 			orig_size = get_orig_size(s, (void *)p);
4790 			ks = s->object_size;
4791 		}
4792 	}
4793 
4794 	/* If the old object doesn't fit, allocate a bigger one */
4795 	if (new_size > ks)
4796 		goto alloc_new;
4797 
4798 	/* Zero out spare memory. */
4799 	if (want_init_on_alloc(flags)) {
4800 		kasan_disable_current();
4801 		if (orig_size && orig_size < new_size)
4802 			memset(kasan_reset_tag(p) + orig_size, 0, new_size - orig_size);
4803 		else
4804 			memset(kasan_reset_tag(p) + new_size, 0, ks - new_size);
4805 		kasan_enable_current();
4806 	}
4807 
4808 	/* Setup kmalloc redzone when needed */
4809 	if (s && slub_debug_orig_size(s)) {
4810 		set_orig_size(s, (void *)p, new_size);
4811 		if (s->flags & SLAB_RED_ZONE && new_size < ks)
4812 			memset_no_sanitize_memory(kasan_reset_tag(p) + new_size,
4813 						SLUB_RED_ACTIVE, ks - new_size);
4814 	}
4815 
4816 	p = kasan_krealloc(p, new_size, flags);
4817 	return (void *)p;
4818 
4819 alloc_new:
4820 	ret = kmalloc_node_track_caller_noprof(new_size, flags, NUMA_NO_NODE, _RET_IP_);
4821 	if (ret && p) {
4822 		/* Disable KASAN checks as the object's redzone is accessed. */
4823 		kasan_disable_current();
4824 		memcpy(ret, kasan_reset_tag(p), orig_size ?: ks);
4825 		kasan_enable_current();
4826 	}
4827 
4828 	return ret;
4829 }
4830 
4831 /**
4832  * krealloc - reallocate memory. The contents will remain unchanged.
4833  * @p: object to reallocate memory for.
4834  * @new_size: how many bytes of memory are required.
4835  * @flags: the type of memory to allocate.
4836  *
4837  * If @p is %NULL, krealloc() behaves exactly like kmalloc().  If @new_size
4838  * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
4839  *
4840  * If __GFP_ZERO logic is requested, callers must ensure that, starting with the
4841  * initial memory allocation, every subsequent call to this API for the same
4842  * memory allocation is flagged with __GFP_ZERO. Otherwise, it is possible that
4843  * __GFP_ZERO is not fully honored by this API.
4844  *
4845  * When slub_debug_orig_size() is off, krealloc() only knows about the bucket
4846  * size of an allocation (but not the exact size it was allocated with) and
4847  * hence implements the following semantics for shrinking and growing buffers
4848  * with __GFP_ZERO.
4849  *
4850  *         new             bucket
4851  * 0       size             size
4852  * |--------|----------------|
4853  * |  keep  |      zero      |
4854  *
4855  * Otherwise, the original allocation size 'orig_size' could be used to
4856  * precisely clear the requested size, and the new size will also be stored
4857  * as the new 'orig_size'.
4858  *
4859  * In any case, the contents of the object pointed to are preserved up to the
4860  * lesser of the new and old sizes.
4861  *
4862  * Return: pointer to the allocated memory or %NULL in case of error
4863  */
krealloc_noprof(const void * p,size_t new_size,gfp_t flags)4864 void *krealloc_noprof(const void *p, size_t new_size, gfp_t flags)
4865 {
4866 	void *ret;
4867 
4868 	if (unlikely(!new_size)) {
4869 		kfree(p);
4870 		return ZERO_SIZE_PTR;
4871 	}
4872 
4873 	ret = __do_krealloc(p, new_size, flags);
4874 	if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
4875 		kfree(p);
4876 
4877 	return ret;
4878 }
4879 EXPORT_SYMBOL(krealloc_noprof);
4880 
4881 struct detached_freelist {
4882 	struct slab *slab;
4883 	void *tail;
4884 	void *freelist;
4885 	int cnt;
4886 	struct kmem_cache *s;
4887 };
4888 
4889 /*
4890  * This function progressively scans the array with free objects (with
4891  * a limited look ahead) and extract objects belonging to the same
4892  * slab.  It builds a detached freelist directly within the given
4893  * slab/objects.  This can happen without any need for
4894  * synchronization, because the objects are owned by running process.
4895  * The freelist is build up as a single linked list in the objects.
4896  * The idea is, that this detached freelist can then be bulk
4897  * transferred to the real freelist(s), but only requiring a single
4898  * synchronization primitive.  Look ahead in the array is limited due
4899  * to performance reasons.
4900  */
4901 static inline
build_detached_freelist(struct kmem_cache * s,size_t size,void ** p,struct detached_freelist * df)4902 int build_detached_freelist(struct kmem_cache *s, size_t size,
4903 			    void **p, struct detached_freelist *df)
4904 {
4905 	int lookahead = 3;
4906 	void *object;
4907 	struct folio *folio;
4908 	size_t same;
4909 
4910 	object = p[--size];
4911 	folio = virt_to_folio(object);
4912 	if (!s) {
4913 		/* Handle kalloc'ed objects */
4914 		if (unlikely(!folio_test_slab(folio))) {
4915 			free_large_kmalloc(folio, object);
4916 			df->slab = NULL;
4917 			return size;
4918 		}
4919 		/* Derive kmem_cache from object */
4920 		df->slab = folio_slab(folio);
4921 		df->s = df->slab->slab_cache;
4922 	} else {
4923 		df->slab = folio_slab(folio);
4924 		df->s = cache_from_obj(s, object); /* Support for memcg */
4925 	}
4926 
4927 	/* Start new detached freelist */
4928 	df->tail = object;
4929 	df->freelist = object;
4930 	df->cnt = 1;
4931 
4932 	if (is_kfence_address(object))
4933 		return size;
4934 
4935 	set_freepointer(df->s, object, NULL);
4936 
4937 	same = size;
4938 	while (size) {
4939 		object = p[--size];
4940 		/* df->slab is always set at this point */
4941 		if (df->slab == virt_to_slab(object)) {
4942 			/* Opportunity build freelist */
4943 			set_freepointer(df->s, object, df->freelist);
4944 			df->freelist = object;
4945 			df->cnt++;
4946 			same--;
4947 			if (size != same)
4948 				swap(p[size], p[same]);
4949 			continue;
4950 		}
4951 
4952 		/* Limit look ahead search */
4953 		if (!--lookahead)
4954 			break;
4955 	}
4956 
4957 	return same;
4958 }
4959 
4960 /*
4961  * Internal bulk free of objects that were not initialised by the post alloc
4962  * hooks and thus should not be processed by the free hooks
4963  */
__kmem_cache_free_bulk(struct kmem_cache * s,size_t size,void ** p)4964 static void __kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
4965 {
4966 	if (!size)
4967 		return;
4968 
4969 	do {
4970 		struct detached_freelist df;
4971 
4972 		size = build_detached_freelist(s, size, p, &df);
4973 		if (!df.slab)
4974 			continue;
4975 
4976 		if (kfence_free(df.freelist))
4977 			continue;
4978 
4979 		do_slab_free(df.s, df.slab, df.freelist, df.tail, df.cnt,
4980 			     _RET_IP_);
4981 	} while (likely(size));
4982 }
4983 
4984 /* Note that interrupts must be enabled when calling this function. */
kmem_cache_free_bulk(struct kmem_cache * s,size_t size,void ** p)4985 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
4986 {
4987 	if (!size)
4988 		return;
4989 
4990 	do {
4991 		struct detached_freelist df;
4992 
4993 		size = build_detached_freelist(s, size, p, &df);
4994 		if (!df.slab)
4995 			continue;
4996 
4997 		slab_free_bulk(df.s, df.slab, df.freelist, df.tail, &p[size],
4998 			       df.cnt, _RET_IP_);
4999 	} while (likely(size));
5000 }
5001 EXPORT_SYMBOL(kmem_cache_free_bulk);
5002 
5003 #ifndef CONFIG_SLUB_TINY
5004 static inline
__kmem_cache_alloc_bulk(struct kmem_cache * s,gfp_t flags,size_t size,void ** p)5005 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
5006 			    void **p)
5007 {
5008 	struct kmem_cache_cpu *c;
5009 	unsigned long irqflags;
5010 	int i;
5011 
5012 	/*
5013 	 * Drain objects in the per cpu slab, while disabling local
5014 	 * IRQs, which protects against PREEMPT and interrupts
5015 	 * handlers invoking normal fastpath.
5016 	 */
5017 	c = slub_get_cpu_ptr(s->cpu_slab);
5018 	local_lock_irqsave(&s->cpu_slab->lock, irqflags);
5019 
5020 	for (i = 0; i < size; i++) {
5021 		void *object = kfence_alloc(s, s->object_size, flags);
5022 
5023 		if (unlikely(object)) {
5024 			p[i] = object;
5025 			continue;
5026 		}
5027 
5028 		object = c->freelist;
5029 		if (unlikely(!object)) {
5030 			/*
5031 			 * We may have removed an object from c->freelist using
5032 			 * the fastpath in the previous iteration; in that case,
5033 			 * c->tid has not been bumped yet.
5034 			 * Since ___slab_alloc() may reenable interrupts while
5035 			 * allocating memory, we should bump c->tid now.
5036 			 */
5037 			c->tid = next_tid(c->tid);
5038 
5039 			local_unlock_irqrestore(&s->cpu_slab->lock, irqflags);
5040 
5041 			/*
5042 			 * Invoking slow path likely have side-effect
5043 			 * of re-populating per CPU c->freelist
5044 			 */
5045 			p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
5046 					    _RET_IP_, c, s->object_size);
5047 			if (unlikely(!p[i]))
5048 				goto error;
5049 
5050 			c = this_cpu_ptr(s->cpu_slab);
5051 			maybe_wipe_obj_freeptr(s, p[i]);
5052 
5053 			local_lock_irqsave(&s->cpu_slab->lock, irqflags);
5054 
5055 			continue; /* goto for-loop */
5056 		}
5057 		c->freelist = get_freepointer(s, object);
5058 		p[i] = object;
5059 		maybe_wipe_obj_freeptr(s, p[i]);
5060 		stat(s, ALLOC_FASTPATH);
5061 	}
5062 	c->tid = next_tid(c->tid);
5063 	local_unlock_irqrestore(&s->cpu_slab->lock, irqflags);
5064 	slub_put_cpu_ptr(s->cpu_slab);
5065 
5066 	return i;
5067 
5068 error:
5069 	slub_put_cpu_ptr(s->cpu_slab);
5070 	__kmem_cache_free_bulk(s, i, p);
5071 	return 0;
5072 
5073 }
5074 #else /* CONFIG_SLUB_TINY */
__kmem_cache_alloc_bulk(struct kmem_cache * s,gfp_t flags,size_t size,void ** p)5075 static int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags,
5076 				   size_t size, void **p)
5077 {
5078 	int i;
5079 
5080 	for (i = 0; i < size; i++) {
5081 		void *object = kfence_alloc(s, s->object_size, flags);
5082 
5083 		if (unlikely(object)) {
5084 			p[i] = object;
5085 			continue;
5086 		}
5087 
5088 		p[i] = __slab_alloc_node(s, flags, NUMA_NO_NODE,
5089 					 _RET_IP_, s->object_size);
5090 		if (unlikely(!p[i]))
5091 			goto error;
5092 
5093 		maybe_wipe_obj_freeptr(s, p[i]);
5094 	}
5095 
5096 	return i;
5097 
5098 error:
5099 	__kmem_cache_free_bulk(s, i, p);
5100 	return 0;
5101 }
5102 #endif /* CONFIG_SLUB_TINY */
5103 
5104 /* Note that interrupts must be enabled when calling this function. */
kmem_cache_alloc_bulk_noprof(struct kmem_cache * s,gfp_t flags,size_t size,void ** p)5105 int kmem_cache_alloc_bulk_noprof(struct kmem_cache *s, gfp_t flags, size_t size,
5106 				 void **p)
5107 {
5108 	int i;
5109 
5110 	if (!size)
5111 		return 0;
5112 
5113 	s = slab_pre_alloc_hook(s, flags);
5114 	if (unlikely(!s))
5115 		return 0;
5116 
5117 	i = __kmem_cache_alloc_bulk(s, flags, size, p);
5118 	if (unlikely(i == 0))
5119 		return 0;
5120 
5121 	/*
5122 	 * memcg and kmem_cache debug support and memory initialization.
5123 	 * Done outside of the IRQ disabled fastpath loop.
5124 	 */
5125 	if (unlikely(!slab_post_alloc_hook(s, NULL, flags, size, p,
5126 		    slab_want_init_on_alloc(flags, s), s->object_size))) {
5127 		return 0;
5128 	}
5129 	return i;
5130 }
5131 EXPORT_SYMBOL(kmem_cache_alloc_bulk_noprof);
5132 
5133 
5134 /*
5135  * Object placement in a slab is made very easy because we always start at
5136  * offset 0. If we tune the size of the object to the alignment then we can
5137  * get the required alignment by putting one properly sized object after
5138  * another.
5139  *
5140  * Notice that the allocation order determines the sizes of the per cpu
5141  * caches. Each processor has always one slab available for allocations.
5142  * Increasing the allocation order reduces the number of times that slabs
5143  * must be moved on and off the partial lists and is therefore a factor in
5144  * locking overhead.
5145  */
5146 
5147 /*
5148  * Minimum / Maximum order of slab pages. This influences locking overhead
5149  * and slab fragmentation. A higher order reduces the number of partial slabs
5150  * and increases the number of allocations possible without having to
5151  * take the list_lock.
5152  */
5153 static unsigned int slub_min_order;
5154 static unsigned int slub_max_order =
5155 	IS_ENABLED(CONFIG_SLUB_TINY) ? 1 : PAGE_ALLOC_COSTLY_ORDER;
5156 static unsigned int slub_min_objects;
5157 
5158 /*
5159  * Calculate the order of allocation given an slab object size.
5160  *
5161  * The order of allocation has significant impact on performance and other
5162  * system components. Generally order 0 allocations should be preferred since
5163  * order 0 does not cause fragmentation in the page allocator. Larger objects
5164  * be problematic to put into order 0 slabs because there may be too much
5165  * unused space left. We go to a higher order if more than 1/16th of the slab
5166  * would be wasted.
5167  *
5168  * In order to reach satisfactory performance we must ensure that a minimum
5169  * number of objects is in one slab. Otherwise we may generate too much
5170  * activity on the partial lists which requires taking the list_lock. This is
5171  * less a concern for large slabs though which are rarely used.
5172  *
5173  * slab_max_order specifies the order where we begin to stop considering the
5174  * number of objects in a slab as critical. If we reach slab_max_order then
5175  * we try to keep the page order as low as possible. So we accept more waste
5176  * of space in favor of a small page order.
5177  *
5178  * Higher order allocations also allow the placement of more objects in a
5179  * slab and thereby reduce object handling overhead. If the user has
5180  * requested a higher minimum order then we start with that one instead of
5181  * the smallest order which will fit the object.
5182  */
calc_slab_order(unsigned int size,unsigned int min_order,unsigned int max_order,unsigned int fract_leftover)5183 static inline unsigned int calc_slab_order(unsigned int size,
5184 		unsigned int min_order, unsigned int max_order,
5185 		unsigned int fract_leftover)
5186 {
5187 	unsigned int order;
5188 
5189 	for (order = min_order; order <= max_order; order++) {
5190 
5191 		unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
5192 		unsigned int rem;
5193 
5194 		rem = slab_size % size;
5195 
5196 		if (rem <= slab_size / fract_leftover)
5197 			break;
5198 	}
5199 
5200 	return order;
5201 }
5202 
calculate_order(unsigned int size)5203 static inline int calculate_order(unsigned int size)
5204 {
5205 	unsigned int order;
5206 	unsigned int min_objects;
5207 	unsigned int max_objects;
5208 	unsigned int min_order;
5209 
5210 	min_objects = slub_min_objects;
5211 	if (!min_objects) {
5212 		/*
5213 		 * Some architectures will only update present cpus when
5214 		 * onlining them, so don't trust the number if it's just 1. But
5215 		 * we also don't want to use nr_cpu_ids always, as on some other
5216 		 * architectures, there can be many possible cpus, but never
5217 		 * onlined. Here we compromise between trying to avoid too high
5218 		 * order on systems that appear larger than they are, and too
5219 		 * low order on systems that appear smaller than they are.
5220 		 */
5221 		unsigned int nr_cpus = num_present_cpus();
5222 		if (nr_cpus <= 1)
5223 			nr_cpus = nr_cpu_ids;
5224 		min_objects = 4 * (fls(nr_cpus) + 1);
5225 	}
5226 	/* min_objects can't be 0 because get_order(0) is undefined */
5227 	max_objects = max(order_objects(slub_max_order, size), 1U);
5228 	min_objects = min(min_objects, max_objects);
5229 
5230 	min_order = max_t(unsigned int, slub_min_order,
5231 			  get_order(min_objects * size));
5232 	if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
5233 		return get_order(size * MAX_OBJS_PER_PAGE) - 1;
5234 
5235 	/*
5236 	 * Attempt to find best configuration for a slab. This works by first
5237 	 * attempting to generate a layout with the best possible configuration
5238 	 * and backing off gradually.
5239 	 *
5240 	 * We start with accepting at most 1/16 waste and try to find the
5241 	 * smallest order from min_objects-derived/slab_min_order up to
5242 	 * slab_max_order that will satisfy the constraint. Note that increasing
5243 	 * the order can only result in same or less fractional waste, not more.
5244 	 *
5245 	 * If that fails, we increase the acceptable fraction of waste and try
5246 	 * again. The last iteration with fraction of 1/2 would effectively
5247 	 * accept any waste and give us the order determined by min_objects, as
5248 	 * long as at least single object fits within slab_max_order.
5249 	 */
5250 	for (unsigned int fraction = 16; fraction > 1; fraction /= 2) {
5251 		order = calc_slab_order(size, min_order, slub_max_order,
5252 					fraction);
5253 		if (order <= slub_max_order)
5254 			return order;
5255 	}
5256 
5257 	/*
5258 	 * Doh this slab cannot be placed using slab_max_order.
5259 	 */
5260 	order = get_order(size);
5261 	if (order <= MAX_PAGE_ORDER)
5262 		return order;
5263 	return -ENOSYS;
5264 }
5265 
5266 static void
init_kmem_cache_node(struct kmem_cache_node * n)5267 init_kmem_cache_node(struct kmem_cache_node *n)
5268 {
5269 	n->nr_partial = 0;
5270 	spin_lock_init(&n->list_lock);
5271 	INIT_LIST_HEAD(&n->partial);
5272 #ifdef CONFIG_SLUB_DEBUG
5273 	atomic_long_set(&n->nr_slabs, 0);
5274 	atomic_long_set(&n->total_objects, 0);
5275 	INIT_LIST_HEAD(&n->full);
5276 #endif
5277 }
5278 
5279 #ifndef CONFIG_SLUB_TINY
alloc_kmem_cache_cpus(struct kmem_cache * s)5280 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
5281 {
5282 	BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
5283 			NR_KMALLOC_TYPES * KMALLOC_SHIFT_HIGH *
5284 			sizeof(struct kmem_cache_cpu));
5285 
5286 	/*
5287 	 * Must align to double word boundary for the double cmpxchg
5288 	 * instructions to work; see __pcpu_double_call_return_bool().
5289 	 */
5290 	s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
5291 				     2 * sizeof(void *));
5292 
5293 	if (!s->cpu_slab)
5294 		return 0;
5295 
5296 	init_kmem_cache_cpus(s);
5297 
5298 	return 1;
5299 }
5300 #else
alloc_kmem_cache_cpus(struct kmem_cache * s)5301 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
5302 {
5303 	return 1;
5304 }
5305 #endif /* CONFIG_SLUB_TINY */
5306 
5307 static struct kmem_cache *kmem_cache_node;
5308 
5309 /*
5310  * No kmalloc_node yet so do it by hand. We know that this is the first
5311  * slab on the node for this slabcache. There are no concurrent accesses
5312  * possible.
5313  *
5314  * Note that this function only works on the kmem_cache_node
5315  * when allocating for the kmem_cache_node. This is used for bootstrapping
5316  * memory on a fresh node that has no slab structures yet.
5317  */
early_kmem_cache_node_alloc(int node)5318 static void early_kmem_cache_node_alloc(int node)
5319 {
5320 	struct slab *slab;
5321 	struct kmem_cache_node *n;
5322 
5323 	BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
5324 
5325 	slab = new_slab(kmem_cache_node, GFP_NOWAIT, node);
5326 
5327 	BUG_ON(!slab);
5328 	if (slab_nid(slab) != node) {
5329 		pr_err("SLUB: Unable to allocate memory from node %d\n", node);
5330 		pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
5331 	}
5332 
5333 	n = slab->freelist;
5334 	BUG_ON(!n);
5335 #ifdef CONFIG_SLUB_DEBUG
5336 	init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
5337 #endif
5338 	n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false);
5339 	slab->freelist = get_freepointer(kmem_cache_node, n);
5340 	slab->inuse = 1;
5341 	kmem_cache_node->node[node] = n;
5342 	init_kmem_cache_node(n);
5343 	inc_slabs_node(kmem_cache_node, node, slab->objects);
5344 
5345 	/*
5346 	 * No locks need to be taken here as it has just been
5347 	 * initialized and there is no concurrent access.
5348 	 */
5349 	__add_partial(n, slab, DEACTIVATE_TO_HEAD);
5350 }
5351 
free_kmem_cache_nodes(struct kmem_cache * s)5352 static void free_kmem_cache_nodes(struct kmem_cache *s)
5353 {
5354 	int node;
5355 	struct kmem_cache_node *n;
5356 
5357 	for_each_kmem_cache_node(s, node, n) {
5358 		s->node[node] = NULL;
5359 		kmem_cache_free(kmem_cache_node, n);
5360 	}
5361 }
5362 
__kmem_cache_release(struct kmem_cache * s)5363 void __kmem_cache_release(struct kmem_cache *s)
5364 {
5365 	cache_random_seq_destroy(s);
5366 #ifndef CONFIG_SLUB_TINY
5367 	free_percpu(s->cpu_slab);
5368 #endif
5369 	free_kmem_cache_nodes(s);
5370 }
5371 
init_kmem_cache_nodes(struct kmem_cache * s)5372 static int init_kmem_cache_nodes(struct kmem_cache *s)
5373 {
5374 	int node;
5375 
5376 	for_each_node_mask(node, slab_nodes) {
5377 		struct kmem_cache_node *n;
5378 
5379 		if (slab_state == DOWN) {
5380 			early_kmem_cache_node_alloc(node);
5381 			continue;
5382 		}
5383 		n = kmem_cache_alloc_node(kmem_cache_node,
5384 						GFP_KERNEL, node);
5385 
5386 		if (!n) {
5387 			free_kmem_cache_nodes(s);
5388 			return 0;
5389 		}
5390 
5391 		init_kmem_cache_node(n);
5392 		s->node[node] = n;
5393 	}
5394 	return 1;
5395 }
5396 
set_cpu_partial(struct kmem_cache * s)5397 static void set_cpu_partial(struct kmem_cache *s)
5398 {
5399 #ifdef CONFIG_SLUB_CPU_PARTIAL
5400 	unsigned int nr_objects;
5401 
5402 	/*
5403 	 * cpu_partial determined the maximum number of objects kept in the
5404 	 * per cpu partial lists of a processor.
5405 	 *
5406 	 * Per cpu partial lists mainly contain slabs that just have one
5407 	 * object freed. If they are used for allocation then they can be
5408 	 * filled up again with minimal effort. The slab will never hit the
5409 	 * per node partial lists and therefore no locking will be required.
5410 	 *
5411 	 * For backwards compatibility reasons, this is determined as number
5412 	 * of objects, even though we now limit maximum number of pages, see
5413 	 * slub_set_cpu_partial()
5414 	 */
5415 	if (!kmem_cache_has_cpu_partial(s))
5416 		nr_objects = 0;
5417 	else if (s->size >= PAGE_SIZE)
5418 		nr_objects = 6;
5419 	else if (s->size >= 1024)
5420 		nr_objects = 24;
5421 	else if (s->size >= 256)
5422 		nr_objects = 52;
5423 	else
5424 		nr_objects = 120;
5425 
5426 	slub_set_cpu_partial(s, nr_objects);
5427 #endif
5428 }
5429 
5430 /*
5431  * calculate_sizes() determines the order and the distribution of data within
5432  * a slab object.
5433  */
calculate_sizes(struct kmem_cache_args * args,struct kmem_cache * s)5434 static int calculate_sizes(struct kmem_cache_args *args, struct kmem_cache *s)
5435 {
5436 	slab_flags_t flags = s->flags;
5437 	unsigned int size = s->object_size;
5438 	unsigned int order;
5439 
5440 	/*
5441 	 * Round up object size to the next word boundary. We can only
5442 	 * place the free pointer at word boundaries and this determines
5443 	 * the possible location of the free pointer.
5444 	 */
5445 	size = ALIGN(size, sizeof(void *));
5446 
5447 #ifdef CONFIG_SLUB_DEBUG
5448 	/*
5449 	 * Determine if we can poison the object itself. If the user of
5450 	 * the slab may touch the object after free or before allocation
5451 	 * then we should never poison the object itself.
5452 	 */
5453 	if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
5454 			!s->ctor)
5455 		s->flags |= __OBJECT_POISON;
5456 	else
5457 		s->flags &= ~__OBJECT_POISON;
5458 
5459 
5460 	/*
5461 	 * If we are Redzoning then check if there is some space between the
5462 	 * end of the object and the free pointer. If not then add an
5463 	 * additional word to have some bytes to store Redzone information.
5464 	 */
5465 	if ((flags & SLAB_RED_ZONE) && size == s->object_size)
5466 		size += sizeof(void *);
5467 #endif
5468 
5469 	/*
5470 	 * With that we have determined the number of bytes in actual use
5471 	 * by the object and redzoning.
5472 	 */
5473 	s->inuse = size;
5474 
5475 	if (((flags & SLAB_TYPESAFE_BY_RCU) && !args->use_freeptr_offset) ||
5476 	    (flags & SLAB_POISON) || s->ctor ||
5477 	    ((flags & SLAB_RED_ZONE) &&
5478 	     (s->object_size < sizeof(void *) || slub_debug_orig_size(s)))) {
5479 		/*
5480 		 * Relocate free pointer after the object if it is not
5481 		 * permitted to overwrite the first word of the object on
5482 		 * kmem_cache_free.
5483 		 *
5484 		 * This is the case if we do RCU, have a constructor or
5485 		 * destructor, are poisoning the objects, or are
5486 		 * redzoning an object smaller than sizeof(void *) or are
5487 		 * redzoning an object with slub_debug_orig_size() enabled,
5488 		 * in which case the right redzone may be extended.
5489 		 *
5490 		 * The assumption that s->offset >= s->inuse means free
5491 		 * pointer is outside of the object is used in the
5492 		 * freeptr_outside_object() function. If that is no
5493 		 * longer true, the function needs to be modified.
5494 		 */
5495 		s->offset = size;
5496 		size += sizeof(void *);
5497 	} else if ((flags & SLAB_TYPESAFE_BY_RCU) && args->use_freeptr_offset) {
5498 		s->offset = args->freeptr_offset;
5499 	} else {
5500 		/*
5501 		 * Store freelist pointer near middle of object to keep
5502 		 * it away from the edges of the object to avoid small
5503 		 * sized over/underflows from neighboring allocations.
5504 		 */
5505 		s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *));
5506 	}
5507 
5508 #ifdef CONFIG_SLUB_DEBUG
5509 	if (flags & SLAB_STORE_USER) {
5510 		/*
5511 		 * Need to store information about allocs and frees after
5512 		 * the object.
5513 		 */
5514 		size += 2 * sizeof(struct track);
5515 
5516 		/* Save the original kmalloc request size */
5517 		if (flags & SLAB_KMALLOC)
5518 			size += sizeof(unsigned int);
5519 	}
5520 #endif
5521 
5522 	kasan_cache_create(s, &size, &s->flags);
5523 #ifdef CONFIG_SLUB_DEBUG
5524 	if (flags & SLAB_RED_ZONE) {
5525 		/*
5526 		 * Add some empty padding so that we can catch
5527 		 * overwrites from earlier objects rather than let
5528 		 * tracking information or the free pointer be
5529 		 * corrupted if a user writes before the start
5530 		 * of the object.
5531 		 */
5532 		size += sizeof(void *);
5533 
5534 		s->red_left_pad = sizeof(void *);
5535 		s->red_left_pad = ALIGN(s->red_left_pad, s->align);
5536 		size += s->red_left_pad;
5537 	}
5538 #endif
5539 
5540 	/*
5541 	 * SLUB stores one object immediately after another beginning from
5542 	 * offset 0. In order to align the objects we have to simply size
5543 	 * each object to conform to the alignment.
5544 	 */
5545 	size = ALIGN(size, s->align);
5546 	s->size = size;
5547 	s->reciprocal_size = reciprocal_value(size);
5548 	order = calculate_order(size);
5549 
5550 	if ((int)order < 0)
5551 		return 0;
5552 
5553 	s->allocflags = __GFP_COMP;
5554 
5555 	if (s->flags & SLAB_CACHE_DMA)
5556 		s->allocflags |= GFP_DMA;
5557 
5558 	if (s->flags & SLAB_CACHE_DMA32)
5559 		s->allocflags |= GFP_DMA32;
5560 
5561 	if (s->flags & SLAB_RECLAIM_ACCOUNT)
5562 		s->allocflags |= __GFP_RECLAIMABLE;
5563 
5564 	/*
5565 	 * Determine the number of objects per slab
5566 	 */
5567 	s->oo = oo_make(order, size);
5568 	s->min = oo_make(get_order(size), size);
5569 
5570 	return !!oo_objects(s->oo);
5571 }
5572 
list_slab_objects(struct kmem_cache * s,struct slab * slab,const char * text)5573 static void list_slab_objects(struct kmem_cache *s, struct slab *slab,
5574 			      const char *text)
5575 {
5576 #ifdef CONFIG_SLUB_DEBUG
5577 	void *addr = slab_address(slab);
5578 	void *p;
5579 
5580 	slab_err(s, slab, text, s->name);
5581 
5582 	spin_lock(&object_map_lock);
5583 	__fill_map(object_map, s, slab);
5584 
5585 	for_each_object(p, s, addr, slab->objects) {
5586 
5587 		if (!test_bit(__obj_to_index(s, addr, p), object_map)) {
5588 			if (slab_add_kunit_errors())
5589 				continue;
5590 			pr_err("Object 0x%p @offset=%tu\n", p, p - addr);
5591 			print_tracking(s, p);
5592 		}
5593 	}
5594 	spin_unlock(&object_map_lock);
5595 #endif
5596 }
5597 
5598 /*
5599  * Attempt to free all partial slabs on a node.
5600  * This is called from __kmem_cache_shutdown(). We must take list_lock
5601  * because sysfs file might still access partial list after the shutdowning.
5602  */
free_partial(struct kmem_cache * s,struct kmem_cache_node * n)5603 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
5604 {
5605 	LIST_HEAD(discard);
5606 	struct slab *slab, *h;
5607 
5608 	BUG_ON(irqs_disabled());
5609 	spin_lock_irq(&n->list_lock);
5610 	list_for_each_entry_safe(slab, h, &n->partial, slab_list) {
5611 		if (!slab->inuse) {
5612 			remove_partial(n, slab);
5613 			list_add(&slab->slab_list, &discard);
5614 		} else {
5615 			list_slab_objects(s, slab,
5616 			  "Objects remaining in %s on __kmem_cache_shutdown()");
5617 		}
5618 	}
5619 	spin_unlock_irq(&n->list_lock);
5620 
5621 	list_for_each_entry_safe(slab, h, &discard, slab_list)
5622 		discard_slab(s, slab);
5623 }
5624 
__kmem_cache_empty(struct kmem_cache * s)5625 bool __kmem_cache_empty(struct kmem_cache *s)
5626 {
5627 	int node;
5628 	struct kmem_cache_node *n;
5629 
5630 	for_each_kmem_cache_node(s, node, n)
5631 		if (n->nr_partial || node_nr_slabs(n))
5632 			return false;
5633 	return true;
5634 }
5635 
5636 /*
5637  * Release all resources used by a slab cache.
5638  */
__kmem_cache_shutdown(struct kmem_cache * s)5639 int __kmem_cache_shutdown(struct kmem_cache *s)
5640 {
5641 	int node;
5642 	struct kmem_cache_node *n;
5643 
5644 	flush_all_cpus_locked(s);
5645 	/* Attempt to free all objects */
5646 	for_each_kmem_cache_node(s, node, n) {
5647 		free_partial(s, n);
5648 		if (n->nr_partial || node_nr_slabs(n))
5649 			return 1;
5650 	}
5651 	return 0;
5652 }
5653 
5654 #ifdef CONFIG_PRINTK
__kmem_obj_info(struct kmem_obj_info * kpp,void * object,struct slab * slab)5655 void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
5656 {
5657 	void *base;
5658 	int __maybe_unused i;
5659 	unsigned int objnr;
5660 	void *objp;
5661 	void *objp0;
5662 	struct kmem_cache *s = slab->slab_cache;
5663 	struct track __maybe_unused *trackp;
5664 
5665 	kpp->kp_ptr = object;
5666 	kpp->kp_slab = slab;
5667 	kpp->kp_slab_cache = s;
5668 	base = slab_address(slab);
5669 	objp0 = kasan_reset_tag(object);
5670 #ifdef CONFIG_SLUB_DEBUG
5671 	objp = restore_red_left(s, objp0);
5672 #else
5673 	objp = objp0;
5674 #endif
5675 	objnr = obj_to_index(s, slab, objp);
5676 	kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp);
5677 	objp = base + s->size * objnr;
5678 	kpp->kp_objp = objp;
5679 	if (WARN_ON_ONCE(objp < base || objp >= base + slab->objects * s->size
5680 			 || (objp - base) % s->size) ||
5681 	    !(s->flags & SLAB_STORE_USER))
5682 		return;
5683 #ifdef CONFIG_SLUB_DEBUG
5684 	objp = fixup_red_left(s, objp);
5685 	trackp = get_track(s, objp, TRACK_ALLOC);
5686 	kpp->kp_ret = (void *)trackp->addr;
5687 #ifdef CONFIG_STACKDEPOT
5688 	{
5689 		depot_stack_handle_t handle;
5690 		unsigned long *entries;
5691 		unsigned int nr_entries;
5692 
5693 		handle = READ_ONCE(trackp->handle);
5694 		if (handle) {
5695 			nr_entries = stack_depot_fetch(handle, &entries);
5696 			for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
5697 				kpp->kp_stack[i] = (void *)entries[i];
5698 		}
5699 
5700 		trackp = get_track(s, objp, TRACK_FREE);
5701 		handle = READ_ONCE(trackp->handle);
5702 		if (handle) {
5703 			nr_entries = stack_depot_fetch(handle, &entries);
5704 			for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
5705 				kpp->kp_free_stack[i] = (void *)entries[i];
5706 		}
5707 	}
5708 #endif
5709 #endif
5710 }
5711 #endif
5712 
5713 /********************************************************************
5714  *		Kmalloc subsystem
5715  *******************************************************************/
5716 
setup_slub_min_order(char * str)5717 static int __init setup_slub_min_order(char *str)
5718 {
5719 	get_option(&str, (int *)&slub_min_order);
5720 
5721 	if (slub_min_order > slub_max_order)
5722 		slub_max_order = slub_min_order;
5723 
5724 	return 1;
5725 }
5726 
5727 __setup("slab_min_order=", setup_slub_min_order);
5728 __setup_param("slub_min_order=", slub_min_order, setup_slub_min_order, 0);
5729 
5730 
setup_slub_max_order(char * str)5731 static int __init setup_slub_max_order(char *str)
5732 {
5733 	get_option(&str, (int *)&slub_max_order);
5734 	slub_max_order = min_t(unsigned int, slub_max_order, MAX_PAGE_ORDER);
5735 
5736 	if (slub_min_order > slub_max_order)
5737 		slub_min_order = slub_max_order;
5738 
5739 	return 1;
5740 }
5741 
5742 __setup("slab_max_order=", setup_slub_max_order);
5743 __setup_param("slub_max_order=", slub_max_order, setup_slub_max_order, 0);
5744 
setup_slub_min_objects(char * str)5745 static int __init setup_slub_min_objects(char *str)
5746 {
5747 	get_option(&str, (int *)&slub_min_objects);
5748 
5749 	return 1;
5750 }
5751 
5752 __setup("slab_min_objects=", setup_slub_min_objects);
5753 __setup_param("slub_min_objects=", slub_min_objects, setup_slub_min_objects, 0);
5754 
5755 #ifdef CONFIG_NUMA
setup_slab_strict_numa(char * str)5756 static int __init setup_slab_strict_numa(char *str)
5757 {
5758 	if (nr_node_ids > 1) {
5759 		static_branch_enable(&strict_numa);
5760 		pr_info("SLUB: Strict NUMA enabled.\n");
5761 	} else {
5762 		pr_warn("slab_strict_numa parameter set on non NUMA system.\n");
5763 	}
5764 
5765 	return 1;
5766 }
5767 
5768 __setup("slab_strict_numa", setup_slab_strict_numa);
5769 #endif
5770 
5771 
5772 #ifdef CONFIG_HARDENED_USERCOPY
5773 /*
5774  * Rejects incorrectly sized objects and objects that are to be copied
5775  * to/from userspace but do not fall entirely within the containing slab
5776  * cache's usercopy region.
5777  *
5778  * Returns NULL if check passes, otherwise const char * to name of cache
5779  * to indicate an error.
5780  */
__check_heap_object(const void * ptr,unsigned long n,const struct slab * slab,bool to_user)5781 void __check_heap_object(const void *ptr, unsigned long n,
5782 			 const struct slab *slab, bool to_user)
5783 {
5784 	struct kmem_cache *s;
5785 	unsigned int offset;
5786 	bool is_kfence = is_kfence_address(ptr);
5787 
5788 	ptr = kasan_reset_tag(ptr);
5789 
5790 	/* Find object and usable object size. */
5791 	s = slab->slab_cache;
5792 
5793 	/* Reject impossible pointers. */
5794 	if (ptr < slab_address(slab))
5795 		usercopy_abort("SLUB object not in SLUB page?!", NULL,
5796 			       to_user, 0, n);
5797 
5798 	/* Find offset within object. */
5799 	if (is_kfence)
5800 		offset = ptr - kfence_object_start(ptr);
5801 	else
5802 		offset = (ptr - slab_address(slab)) % s->size;
5803 
5804 	/* Adjust for redzone and reject if within the redzone. */
5805 	if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) {
5806 		if (offset < s->red_left_pad)
5807 			usercopy_abort("SLUB object in left red zone",
5808 				       s->name, to_user, offset, n);
5809 		offset -= s->red_left_pad;
5810 	}
5811 
5812 	/* Allow address range falling entirely within usercopy region. */
5813 	if (offset >= s->useroffset &&
5814 	    offset - s->useroffset <= s->usersize &&
5815 	    n <= s->useroffset - offset + s->usersize)
5816 		return;
5817 
5818 	usercopy_abort("SLUB object", s->name, to_user, offset, n);
5819 }
5820 #endif /* CONFIG_HARDENED_USERCOPY */
5821 
5822 #define SHRINK_PROMOTE_MAX 32
5823 
5824 /*
5825  * kmem_cache_shrink discards empty slabs and promotes the slabs filled
5826  * up most to the head of the partial lists. New allocations will then
5827  * fill those up and thus they can be removed from the partial lists.
5828  *
5829  * The slabs with the least items are placed last. This results in them
5830  * being allocated from last increasing the chance that the last objects
5831  * are freed in them.
5832  */
__kmem_cache_do_shrink(struct kmem_cache * s)5833 static int __kmem_cache_do_shrink(struct kmem_cache *s)
5834 {
5835 	int node;
5836 	int i;
5837 	struct kmem_cache_node *n;
5838 	struct slab *slab;
5839 	struct slab *t;
5840 	struct list_head discard;
5841 	struct list_head promote[SHRINK_PROMOTE_MAX];
5842 	unsigned long flags;
5843 	int ret = 0;
5844 
5845 	for_each_kmem_cache_node(s, node, n) {
5846 		INIT_LIST_HEAD(&discard);
5847 		for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
5848 			INIT_LIST_HEAD(promote + i);
5849 
5850 		spin_lock_irqsave(&n->list_lock, flags);
5851 
5852 		/*
5853 		 * Build lists of slabs to discard or promote.
5854 		 *
5855 		 * Note that concurrent frees may occur while we hold the
5856 		 * list_lock. slab->inuse here is the upper limit.
5857 		 */
5858 		list_for_each_entry_safe(slab, t, &n->partial, slab_list) {
5859 			int free = slab->objects - slab->inuse;
5860 
5861 			/* Do not reread slab->inuse */
5862 			barrier();
5863 
5864 			/* We do not keep full slabs on the list */
5865 			BUG_ON(free <= 0);
5866 
5867 			if (free == slab->objects) {
5868 				list_move(&slab->slab_list, &discard);
5869 				slab_clear_node_partial(slab);
5870 				n->nr_partial--;
5871 				dec_slabs_node(s, node, slab->objects);
5872 			} else if (free <= SHRINK_PROMOTE_MAX)
5873 				list_move(&slab->slab_list, promote + free - 1);
5874 		}
5875 
5876 		/*
5877 		 * Promote the slabs filled up most to the head of the
5878 		 * partial list.
5879 		 */
5880 		for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
5881 			list_splice(promote + i, &n->partial);
5882 
5883 		spin_unlock_irqrestore(&n->list_lock, flags);
5884 
5885 		/* Release empty slabs */
5886 		list_for_each_entry_safe(slab, t, &discard, slab_list)
5887 			free_slab(s, slab);
5888 
5889 		if (node_nr_slabs(n))
5890 			ret = 1;
5891 	}
5892 
5893 	return ret;
5894 }
5895 
__kmem_cache_shrink(struct kmem_cache * s)5896 int __kmem_cache_shrink(struct kmem_cache *s)
5897 {
5898 	flush_all(s);
5899 	return __kmem_cache_do_shrink(s);
5900 }
5901 
slab_mem_going_offline_callback(void * arg)5902 static int slab_mem_going_offline_callback(void *arg)
5903 {
5904 	struct kmem_cache *s;
5905 
5906 	mutex_lock(&slab_mutex);
5907 	list_for_each_entry(s, &slab_caches, list) {
5908 		flush_all_cpus_locked(s);
5909 		__kmem_cache_do_shrink(s);
5910 	}
5911 	mutex_unlock(&slab_mutex);
5912 
5913 	return 0;
5914 }
5915 
slab_mem_offline_callback(void * arg)5916 static void slab_mem_offline_callback(void *arg)
5917 {
5918 	struct memory_notify *marg = arg;
5919 	int offline_node;
5920 
5921 	offline_node = marg->status_change_nid_normal;
5922 
5923 	/*
5924 	 * If the node still has available memory. we need kmem_cache_node
5925 	 * for it yet.
5926 	 */
5927 	if (offline_node < 0)
5928 		return;
5929 
5930 	mutex_lock(&slab_mutex);
5931 	node_clear(offline_node, slab_nodes);
5932 	/*
5933 	 * We no longer free kmem_cache_node structures here, as it would be
5934 	 * racy with all get_node() users, and infeasible to protect them with
5935 	 * slab_mutex.
5936 	 */
5937 	mutex_unlock(&slab_mutex);
5938 }
5939 
slab_mem_going_online_callback(void * arg)5940 static int slab_mem_going_online_callback(void *arg)
5941 {
5942 	struct kmem_cache_node *n;
5943 	struct kmem_cache *s;
5944 	struct memory_notify *marg = arg;
5945 	int nid = marg->status_change_nid_normal;
5946 	int ret = 0;
5947 
5948 	/*
5949 	 * If the node's memory is already available, then kmem_cache_node is
5950 	 * already created. Nothing to do.
5951 	 */
5952 	if (nid < 0)
5953 		return 0;
5954 
5955 	/*
5956 	 * We are bringing a node online. No memory is available yet. We must
5957 	 * allocate a kmem_cache_node structure in order to bring the node
5958 	 * online.
5959 	 */
5960 	mutex_lock(&slab_mutex);
5961 	list_for_each_entry(s, &slab_caches, list) {
5962 		/*
5963 		 * The structure may already exist if the node was previously
5964 		 * onlined and offlined.
5965 		 */
5966 		if (get_node(s, nid))
5967 			continue;
5968 		/*
5969 		 * XXX: kmem_cache_alloc_node will fallback to other nodes
5970 		 *      since memory is not yet available from the node that
5971 		 *      is brought up.
5972 		 */
5973 		n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
5974 		if (!n) {
5975 			ret = -ENOMEM;
5976 			goto out;
5977 		}
5978 		init_kmem_cache_node(n);
5979 		s->node[nid] = n;
5980 	}
5981 	/*
5982 	 * Any cache created after this point will also have kmem_cache_node
5983 	 * initialized for the new node.
5984 	 */
5985 	node_set(nid, slab_nodes);
5986 out:
5987 	mutex_unlock(&slab_mutex);
5988 	return ret;
5989 }
5990 
slab_memory_callback(struct notifier_block * self,unsigned long action,void * arg)5991 static int slab_memory_callback(struct notifier_block *self,
5992 				unsigned long action, void *arg)
5993 {
5994 	int ret = 0;
5995 
5996 	switch (action) {
5997 	case MEM_GOING_ONLINE:
5998 		ret = slab_mem_going_online_callback(arg);
5999 		break;
6000 	case MEM_GOING_OFFLINE:
6001 		ret = slab_mem_going_offline_callback(arg);
6002 		break;
6003 	case MEM_OFFLINE:
6004 	case MEM_CANCEL_ONLINE:
6005 		slab_mem_offline_callback(arg);
6006 		break;
6007 	case MEM_ONLINE:
6008 	case MEM_CANCEL_OFFLINE:
6009 		break;
6010 	}
6011 	if (ret)
6012 		ret = notifier_from_errno(ret);
6013 	else
6014 		ret = NOTIFY_OK;
6015 	return ret;
6016 }
6017 
6018 /********************************************************************
6019  *			Basic setup of slabs
6020  *******************************************************************/
6021 
6022 /*
6023  * Used for early kmem_cache structures that were allocated using
6024  * the page allocator. Allocate them properly then fix up the pointers
6025  * that may be pointing to the wrong kmem_cache structure.
6026  */
6027 
bootstrap(struct kmem_cache * static_cache)6028 static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
6029 {
6030 	int node;
6031 	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
6032 	struct kmem_cache_node *n;
6033 
6034 	memcpy(s, static_cache, kmem_cache->object_size);
6035 
6036 	/*
6037 	 * This runs very early, and only the boot processor is supposed to be
6038 	 * up.  Even if it weren't true, IRQs are not up so we couldn't fire
6039 	 * IPIs around.
6040 	 */
6041 	__flush_cpu_slab(s, smp_processor_id());
6042 	for_each_kmem_cache_node(s, node, n) {
6043 		struct slab *p;
6044 
6045 		list_for_each_entry(p, &n->partial, slab_list)
6046 			p->slab_cache = s;
6047 
6048 #ifdef CONFIG_SLUB_DEBUG
6049 		list_for_each_entry(p, &n->full, slab_list)
6050 			p->slab_cache = s;
6051 #endif
6052 	}
6053 	list_add(&s->list, &slab_caches);
6054 	return s;
6055 }
6056 
kmem_cache_init(void)6057 void __init kmem_cache_init(void)
6058 {
6059 	static __initdata struct kmem_cache boot_kmem_cache,
6060 		boot_kmem_cache_node;
6061 	int node;
6062 
6063 	if (debug_guardpage_minorder())
6064 		slub_max_order = 0;
6065 
6066 	/* Print slub debugging pointers without hashing */
6067 	if (__slub_debug_enabled())
6068 		no_hash_pointers_enable(NULL);
6069 
6070 	kmem_cache_node = &boot_kmem_cache_node;
6071 	kmem_cache = &boot_kmem_cache;
6072 
6073 	/*
6074 	 * Initialize the nodemask for which we will allocate per node
6075 	 * structures. Here we don't need taking slab_mutex yet.
6076 	 */
6077 	for_each_node_state(node, N_NORMAL_MEMORY)
6078 		node_set(node, slab_nodes);
6079 
6080 	create_boot_cache(kmem_cache_node, "kmem_cache_node",
6081 			sizeof(struct kmem_cache_node),
6082 			SLAB_HWCACHE_ALIGN | SLAB_NO_OBJ_EXT, 0, 0);
6083 
6084 	hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
6085 
6086 	/* Able to allocate the per node structures */
6087 	slab_state = PARTIAL;
6088 
6089 	create_boot_cache(kmem_cache, "kmem_cache",
6090 			offsetof(struct kmem_cache, node) +
6091 				nr_node_ids * sizeof(struct kmem_cache_node *),
6092 			SLAB_HWCACHE_ALIGN | SLAB_NO_OBJ_EXT, 0, 0);
6093 
6094 	kmem_cache = bootstrap(&boot_kmem_cache);
6095 	kmem_cache_node = bootstrap(&boot_kmem_cache_node);
6096 
6097 	/* Now we can use the kmem_cache to allocate kmalloc slabs */
6098 	setup_kmalloc_cache_index_table();
6099 	create_kmalloc_caches();
6100 
6101 	/* Setup random freelists for each cache */
6102 	init_freelist_randomization();
6103 
6104 	cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
6105 				  slub_cpu_dead);
6106 
6107 	pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n",
6108 		cache_line_size(),
6109 		slub_min_order, slub_max_order, slub_min_objects,
6110 		nr_cpu_ids, nr_node_ids);
6111 }
6112 
kmem_cache_init_late(void)6113 void __init kmem_cache_init_late(void)
6114 {
6115 #ifndef CONFIG_SLUB_TINY
6116 	flushwq = alloc_workqueue("slub_flushwq", WQ_MEM_RECLAIM, 0);
6117 	WARN_ON(!flushwq);
6118 #endif
6119 }
6120 
6121 struct kmem_cache *
__kmem_cache_alias(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,void (* ctor)(void *))6122 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
6123 		   slab_flags_t flags, void (*ctor)(void *))
6124 {
6125 	struct kmem_cache *s;
6126 
6127 	s = find_mergeable(size, align, flags, name, ctor);
6128 	if (s) {
6129 		if (sysfs_slab_alias(s, name))
6130 			pr_err("SLUB: Unable to add cache alias %s to sysfs\n",
6131 			       name);
6132 
6133 		s->refcount++;
6134 
6135 		/*
6136 		 * Adjust the object sizes so that we clear
6137 		 * the complete object on kzalloc.
6138 		 */
6139 		s->object_size = max(s->object_size, size);
6140 		s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
6141 	}
6142 
6143 	return s;
6144 }
6145 
do_kmem_cache_create(struct kmem_cache * s,const char * name,unsigned int size,struct kmem_cache_args * args,slab_flags_t flags)6146 int do_kmem_cache_create(struct kmem_cache *s, const char *name,
6147 			 unsigned int size, struct kmem_cache_args *args,
6148 			 slab_flags_t flags)
6149 {
6150 	int err = -EINVAL;
6151 
6152 	s->name = name;
6153 	s->size = s->object_size = size;
6154 
6155 	s->flags = kmem_cache_flags(flags, s->name);
6156 #ifdef CONFIG_SLAB_FREELIST_HARDENED
6157 	s->random = get_random_long();
6158 #endif
6159 	s->align = args->align;
6160 	s->ctor = args->ctor;
6161 #ifdef CONFIG_HARDENED_USERCOPY
6162 	s->useroffset = args->useroffset;
6163 	s->usersize = args->usersize;
6164 #endif
6165 
6166 	if (!calculate_sizes(args, s))
6167 		goto out;
6168 	if (disable_higher_order_debug) {
6169 		/*
6170 		 * Disable debugging flags that store metadata if the min slab
6171 		 * order increased.
6172 		 */
6173 		if (get_order(s->size) > get_order(s->object_size)) {
6174 			s->flags &= ~DEBUG_METADATA_FLAGS;
6175 			s->offset = 0;
6176 			if (!calculate_sizes(args, s))
6177 				goto out;
6178 		}
6179 	}
6180 
6181 #ifdef system_has_freelist_aba
6182 	if (system_has_freelist_aba() && !(s->flags & SLAB_NO_CMPXCHG)) {
6183 		/* Enable fast mode */
6184 		s->flags |= __CMPXCHG_DOUBLE;
6185 	}
6186 #endif
6187 
6188 	/*
6189 	 * The larger the object size is, the more slabs we want on the partial
6190 	 * list to avoid pounding the page allocator excessively.
6191 	 */
6192 	s->min_partial = min_t(unsigned long, MAX_PARTIAL, ilog2(s->size) / 2);
6193 	s->min_partial = max_t(unsigned long, MIN_PARTIAL, s->min_partial);
6194 
6195 	set_cpu_partial(s);
6196 
6197 #ifdef CONFIG_NUMA
6198 	s->remote_node_defrag_ratio = 1000;
6199 #endif
6200 
6201 	/* Initialize the pre-computed randomized freelist if slab is up */
6202 	if (slab_state >= UP) {
6203 		if (init_cache_random_seq(s))
6204 			goto out;
6205 	}
6206 
6207 	if (!init_kmem_cache_nodes(s))
6208 		goto out;
6209 
6210 	if (!alloc_kmem_cache_cpus(s))
6211 		goto out;
6212 
6213 	err = 0;
6214 
6215 	/* Mutex is not taken during early boot */
6216 	if (slab_state <= UP)
6217 		goto out;
6218 
6219 	/*
6220 	 * Failing to create sysfs files is not critical to SLUB functionality.
6221 	 * If it fails, proceed with cache creation without these files.
6222 	 */
6223 	if (sysfs_slab_add(s))
6224 		pr_err("SLUB: Unable to add cache %s to sysfs\n", s->name);
6225 
6226 	if (s->flags & SLAB_STORE_USER)
6227 		debugfs_slab_add(s);
6228 
6229 out:
6230 	if (err)
6231 		__kmem_cache_release(s);
6232 	return err;
6233 }
6234 
6235 #ifdef SLAB_SUPPORTS_SYSFS
count_inuse(struct slab * slab)6236 static int count_inuse(struct slab *slab)
6237 {
6238 	return slab->inuse;
6239 }
6240 
count_total(struct slab * slab)6241 static int count_total(struct slab *slab)
6242 {
6243 	return slab->objects;
6244 }
6245 #endif
6246 
6247 #ifdef CONFIG_SLUB_DEBUG
validate_slab(struct kmem_cache * s,struct slab * slab,unsigned long * obj_map)6248 static void validate_slab(struct kmem_cache *s, struct slab *slab,
6249 			  unsigned long *obj_map)
6250 {
6251 	void *p;
6252 	void *addr = slab_address(slab);
6253 
6254 	if (!check_slab(s, slab) || !on_freelist(s, slab, NULL))
6255 		return;
6256 
6257 	/* Now we know that a valid freelist exists */
6258 	__fill_map(obj_map, s, slab);
6259 	for_each_object(p, s, addr, slab->objects) {
6260 		u8 val = test_bit(__obj_to_index(s, addr, p), obj_map) ?
6261 			 SLUB_RED_INACTIVE : SLUB_RED_ACTIVE;
6262 
6263 		if (!check_object(s, slab, p, val))
6264 			break;
6265 	}
6266 }
6267 
validate_slab_node(struct kmem_cache * s,struct kmem_cache_node * n,unsigned long * obj_map)6268 static int validate_slab_node(struct kmem_cache *s,
6269 		struct kmem_cache_node *n, unsigned long *obj_map)
6270 {
6271 	unsigned long count = 0;
6272 	struct slab *slab;
6273 	unsigned long flags;
6274 
6275 	spin_lock_irqsave(&n->list_lock, flags);
6276 
6277 	list_for_each_entry(slab, &n->partial, slab_list) {
6278 		validate_slab(s, slab, obj_map);
6279 		count++;
6280 	}
6281 	if (count != n->nr_partial) {
6282 		pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
6283 		       s->name, count, n->nr_partial);
6284 		slab_add_kunit_errors();
6285 	}
6286 
6287 	if (!(s->flags & SLAB_STORE_USER))
6288 		goto out;
6289 
6290 	list_for_each_entry(slab, &n->full, slab_list) {
6291 		validate_slab(s, slab, obj_map);
6292 		count++;
6293 	}
6294 	if (count != node_nr_slabs(n)) {
6295 		pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
6296 		       s->name, count, node_nr_slabs(n));
6297 		slab_add_kunit_errors();
6298 	}
6299 
6300 out:
6301 	spin_unlock_irqrestore(&n->list_lock, flags);
6302 	return count;
6303 }
6304 
validate_slab_cache(struct kmem_cache * s)6305 long validate_slab_cache(struct kmem_cache *s)
6306 {
6307 	int node;
6308 	unsigned long count = 0;
6309 	struct kmem_cache_node *n;
6310 	unsigned long *obj_map;
6311 
6312 	obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL);
6313 	if (!obj_map)
6314 		return -ENOMEM;
6315 
6316 	flush_all(s);
6317 	for_each_kmem_cache_node(s, node, n)
6318 		count += validate_slab_node(s, n, obj_map);
6319 
6320 	bitmap_free(obj_map);
6321 
6322 	return count;
6323 }
6324 EXPORT_SYMBOL(validate_slab_cache);
6325 
6326 #ifdef CONFIG_DEBUG_FS
6327 /*
6328  * Generate lists of code addresses where slabcache objects are allocated
6329  * and freed.
6330  */
6331 
6332 struct location {
6333 	depot_stack_handle_t handle;
6334 	unsigned long count;
6335 	unsigned long addr;
6336 	unsigned long waste;
6337 	long long sum_time;
6338 	long min_time;
6339 	long max_time;
6340 	long min_pid;
6341 	long max_pid;
6342 	DECLARE_BITMAP(cpus, NR_CPUS);
6343 	nodemask_t nodes;
6344 };
6345 
6346 struct loc_track {
6347 	unsigned long max;
6348 	unsigned long count;
6349 	struct location *loc;
6350 	loff_t idx;
6351 };
6352 
6353 static struct dentry *slab_debugfs_root;
6354 
free_loc_track(struct loc_track * t)6355 static void free_loc_track(struct loc_track *t)
6356 {
6357 	if (t->max)
6358 		free_pages((unsigned long)t->loc,
6359 			get_order(sizeof(struct location) * t->max));
6360 }
6361 
alloc_loc_track(struct loc_track * t,unsigned long max,gfp_t flags)6362 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
6363 {
6364 	struct location *l;
6365 	int order;
6366 
6367 	order = get_order(sizeof(struct location) * max);
6368 
6369 	l = (void *)__get_free_pages(flags, order);
6370 	if (!l)
6371 		return 0;
6372 
6373 	if (t->count) {
6374 		memcpy(l, t->loc, sizeof(struct location) * t->count);
6375 		free_loc_track(t);
6376 	}
6377 	t->max = max;
6378 	t->loc = l;
6379 	return 1;
6380 }
6381 
add_location(struct loc_track * t,struct kmem_cache * s,const struct track * track,unsigned int orig_size)6382 static int add_location(struct loc_track *t, struct kmem_cache *s,
6383 				const struct track *track,
6384 				unsigned int orig_size)
6385 {
6386 	long start, end, pos;
6387 	struct location *l;
6388 	unsigned long caddr, chandle, cwaste;
6389 	unsigned long age = jiffies - track->when;
6390 	depot_stack_handle_t handle = 0;
6391 	unsigned int waste = s->object_size - orig_size;
6392 
6393 #ifdef CONFIG_STACKDEPOT
6394 	handle = READ_ONCE(track->handle);
6395 #endif
6396 	start = -1;
6397 	end = t->count;
6398 
6399 	for ( ; ; ) {
6400 		pos = start + (end - start + 1) / 2;
6401 
6402 		/*
6403 		 * There is nothing at "end". If we end up there
6404 		 * we need to add something to before end.
6405 		 */
6406 		if (pos == end)
6407 			break;
6408 
6409 		l = &t->loc[pos];
6410 		caddr = l->addr;
6411 		chandle = l->handle;
6412 		cwaste = l->waste;
6413 		if ((track->addr == caddr) && (handle == chandle) &&
6414 			(waste == cwaste)) {
6415 
6416 			l->count++;
6417 			if (track->when) {
6418 				l->sum_time += age;
6419 				if (age < l->min_time)
6420 					l->min_time = age;
6421 				if (age > l->max_time)
6422 					l->max_time = age;
6423 
6424 				if (track->pid < l->min_pid)
6425 					l->min_pid = track->pid;
6426 				if (track->pid > l->max_pid)
6427 					l->max_pid = track->pid;
6428 
6429 				cpumask_set_cpu(track->cpu,
6430 						to_cpumask(l->cpus));
6431 			}
6432 			node_set(page_to_nid(virt_to_page(track)), l->nodes);
6433 			return 1;
6434 		}
6435 
6436 		if (track->addr < caddr)
6437 			end = pos;
6438 		else if (track->addr == caddr && handle < chandle)
6439 			end = pos;
6440 		else if (track->addr == caddr && handle == chandle &&
6441 				waste < cwaste)
6442 			end = pos;
6443 		else
6444 			start = pos;
6445 	}
6446 
6447 	/*
6448 	 * Not found. Insert new tracking element.
6449 	 */
6450 	if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
6451 		return 0;
6452 
6453 	l = t->loc + pos;
6454 	if (pos < t->count)
6455 		memmove(l + 1, l,
6456 			(t->count - pos) * sizeof(struct location));
6457 	t->count++;
6458 	l->count = 1;
6459 	l->addr = track->addr;
6460 	l->sum_time = age;
6461 	l->min_time = age;
6462 	l->max_time = age;
6463 	l->min_pid = track->pid;
6464 	l->max_pid = track->pid;
6465 	l->handle = handle;
6466 	l->waste = waste;
6467 	cpumask_clear(to_cpumask(l->cpus));
6468 	cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
6469 	nodes_clear(l->nodes);
6470 	node_set(page_to_nid(virt_to_page(track)), l->nodes);
6471 	return 1;
6472 }
6473 
process_slab(struct loc_track * t,struct kmem_cache * s,struct slab * slab,enum track_item alloc,unsigned long * obj_map)6474 static void process_slab(struct loc_track *t, struct kmem_cache *s,
6475 		struct slab *slab, enum track_item alloc,
6476 		unsigned long *obj_map)
6477 {
6478 	void *addr = slab_address(slab);
6479 	bool is_alloc = (alloc == TRACK_ALLOC);
6480 	void *p;
6481 
6482 	__fill_map(obj_map, s, slab);
6483 
6484 	for_each_object(p, s, addr, slab->objects)
6485 		if (!test_bit(__obj_to_index(s, addr, p), obj_map))
6486 			add_location(t, s, get_track(s, p, alloc),
6487 				     is_alloc ? get_orig_size(s, p) :
6488 						s->object_size);
6489 }
6490 #endif  /* CONFIG_DEBUG_FS   */
6491 #endif	/* CONFIG_SLUB_DEBUG */
6492 
6493 #ifdef SLAB_SUPPORTS_SYSFS
6494 enum slab_stat_type {
6495 	SL_ALL,			/* All slabs */
6496 	SL_PARTIAL,		/* Only partially allocated slabs */
6497 	SL_CPU,			/* Only slabs used for cpu caches */
6498 	SL_OBJECTS,		/* Determine allocated objects not slabs */
6499 	SL_TOTAL		/* Determine object capacity not slabs */
6500 };
6501 
6502 #define SO_ALL		(1 << SL_ALL)
6503 #define SO_PARTIAL	(1 << SL_PARTIAL)
6504 #define SO_CPU		(1 << SL_CPU)
6505 #define SO_OBJECTS	(1 << SL_OBJECTS)
6506 #define SO_TOTAL	(1 << SL_TOTAL)
6507 
show_slab_objects(struct kmem_cache * s,char * buf,unsigned long flags)6508 static ssize_t show_slab_objects(struct kmem_cache *s,
6509 				 char *buf, unsigned long flags)
6510 {
6511 	unsigned long total = 0;
6512 	int node;
6513 	int x;
6514 	unsigned long *nodes;
6515 	int len = 0;
6516 
6517 	nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
6518 	if (!nodes)
6519 		return -ENOMEM;
6520 
6521 	if (flags & SO_CPU) {
6522 		int cpu;
6523 
6524 		for_each_possible_cpu(cpu) {
6525 			struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
6526 							       cpu);
6527 			int node;
6528 			struct slab *slab;
6529 
6530 			slab = READ_ONCE(c->slab);
6531 			if (!slab)
6532 				continue;
6533 
6534 			node = slab_nid(slab);
6535 			if (flags & SO_TOTAL)
6536 				x = slab->objects;
6537 			else if (flags & SO_OBJECTS)
6538 				x = slab->inuse;
6539 			else
6540 				x = 1;
6541 
6542 			total += x;
6543 			nodes[node] += x;
6544 
6545 #ifdef CONFIG_SLUB_CPU_PARTIAL
6546 			slab = slub_percpu_partial_read_once(c);
6547 			if (slab) {
6548 				node = slab_nid(slab);
6549 				if (flags & SO_TOTAL)
6550 					WARN_ON_ONCE(1);
6551 				else if (flags & SO_OBJECTS)
6552 					WARN_ON_ONCE(1);
6553 				else
6554 					x = data_race(slab->slabs);
6555 				total += x;
6556 				nodes[node] += x;
6557 			}
6558 #endif
6559 		}
6560 	}
6561 
6562 	/*
6563 	 * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex"
6564 	 * already held which will conflict with an existing lock order:
6565 	 *
6566 	 * mem_hotplug_lock->slab_mutex->kernfs_mutex
6567 	 *
6568 	 * We don't really need mem_hotplug_lock (to hold off
6569 	 * slab_mem_going_offline_callback) here because slab's memory hot
6570 	 * unplug code doesn't destroy the kmem_cache->node[] data.
6571 	 */
6572 
6573 #ifdef CONFIG_SLUB_DEBUG
6574 	if (flags & SO_ALL) {
6575 		struct kmem_cache_node *n;
6576 
6577 		for_each_kmem_cache_node(s, node, n) {
6578 
6579 			if (flags & SO_TOTAL)
6580 				x = node_nr_objs(n);
6581 			else if (flags & SO_OBJECTS)
6582 				x = node_nr_objs(n) - count_partial(n, count_free);
6583 			else
6584 				x = node_nr_slabs(n);
6585 			total += x;
6586 			nodes[node] += x;
6587 		}
6588 
6589 	} else
6590 #endif
6591 	if (flags & SO_PARTIAL) {
6592 		struct kmem_cache_node *n;
6593 
6594 		for_each_kmem_cache_node(s, node, n) {
6595 			if (flags & SO_TOTAL)
6596 				x = count_partial(n, count_total);
6597 			else if (flags & SO_OBJECTS)
6598 				x = count_partial(n, count_inuse);
6599 			else
6600 				x = n->nr_partial;
6601 			total += x;
6602 			nodes[node] += x;
6603 		}
6604 	}
6605 
6606 	len += sysfs_emit_at(buf, len, "%lu", total);
6607 #ifdef CONFIG_NUMA
6608 	for (node = 0; node < nr_node_ids; node++) {
6609 		if (nodes[node])
6610 			len += sysfs_emit_at(buf, len, " N%d=%lu",
6611 					     node, nodes[node]);
6612 	}
6613 #endif
6614 	len += sysfs_emit_at(buf, len, "\n");
6615 	kfree(nodes);
6616 
6617 	return len;
6618 }
6619 
6620 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
6621 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
6622 
6623 struct slab_attribute {
6624 	struct attribute attr;
6625 	ssize_t (*show)(struct kmem_cache *s, char *buf);
6626 	ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
6627 };
6628 
6629 #define SLAB_ATTR_RO(_name) \
6630 	static struct slab_attribute _name##_attr = __ATTR_RO_MODE(_name, 0400)
6631 
6632 #define SLAB_ATTR(_name) \
6633 	static struct slab_attribute _name##_attr = __ATTR_RW_MODE(_name, 0600)
6634 
slab_size_show(struct kmem_cache * s,char * buf)6635 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
6636 {
6637 	return sysfs_emit(buf, "%u\n", s->size);
6638 }
6639 SLAB_ATTR_RO(slab_size);
6640 
align_show(struct kmem_cache * s,char * buf)6641 static ssize_t align_show(struct kmem_cache *s, char *buf)
6642 {
6643 	return sysfs_emit(buf, "%u\n", s->align);
6644 }
6645 SLAB_ATTR_RO(align);
6646 
object_size_show(struct kmem_cache * s,char * buf)6647 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
6648 {
6649 	return sysfs_emit(buf, "%u\n", s->object_size);
6650 }
6651 SLAB_ATTR_RO(object_size);
6652 
objs_per_slab_show(struct kmem_cache * s,char * buf)6653 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
6654 {
6655 	return sysfs_emit(buf, "%u\n", oo_objects(s->oo));
6656 }
6657 SLAB_ATTR_RO(objs_per_slab);
6658 
order_show(struct kmem_cache * s,char * buf)6659 static ssize_t order_show(struct kmem_cache *s, char *buf)
6660 {
6661 	return sysfs_emit(buf, "%u\n", oo_order(s->oo));
6662 }
6663 SLAB_ATTR_RO(order);
6664 
min_partial_show(struct kmem_cache * s,char * buf)6665 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
6666 {
6667 	return sysfs_emit(buf, "%lu\n", s->min_partial);
6668 }
6669 
min_partial_store(struct kmem_cache * s,const char * buf,size_t length)6670 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
6671 				 size_t length)
6672 {
6673 	unsigned long min;
6674 	int err;
6675 
6676 	err = kstrtoul(buf, 10, &min);
6677 	if (err)
6678 		return err;
6679 
6680 	s->min_partial = min;
6681 	return length;
6682 }
6683 SLAB_ATTR(min_partial);
6684 
cpu_partial_show(struct kmem_cache * s,char * buf)6685 static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
6686 {
6687 	unsigned int nr_partial = 0;
6688 #ifdef CONFIG_SLUB_CPU_PARTIAL
6689 	nr_partial = s->cpu_partial;
6690 #endif
6691 
6692 	return sysfs_emit(buf, "%u\n", nr_partial);
6693 }
6694 
cpu_partial_store(struct kmem_cache * s,const char * buf,size_t length)6695 static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
6696 				 size_t length)
6697 {
6698 	unsigned int objects;
6699 	int err;
6700 
6701 	err = kstrtouint(buf, 10, &objects);
6702 	if (err)
6703 		return err;
6704 	if (objects && !kmem_cache_has_cpu_partial(s))
6705 		return -EINVAL;
6706 
6707 	slub_set_cpu_partial(s, objects);
6708 	flush_all(s);
6709 	return length;
6710 }
6711 SLAB_ATTR(cpu_partial);
6712 
ctor_show(struct kmem_cache * s,char * buf)6713 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
6714 {
6715 	if (!s->ctor)
6716 		return 0;
6717 	return sysfs_emit(buf, "%pS\n", s->ctor);
6718 }
6719 SLAB_ATTR_RO(ctor);
6720 
aliases_show(struct kmem_cache * s,char * buf)6721 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
6722 {
6723 	return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
6724 }
6725 SLAB_ATTR_RO(aliases);
6726 
partial_show(struct kmem_cache * s,char * buf)6727 static ssize_t partial_show(struct kmem_cache *s, char *buf)
6728 {
6729 	return show_slab_objects(s, buf, SO_PARTIAL);
6730 }
6731 SLAB_ATTR_RO(partial);
6732 
cpu_slabs_show(struct kmem_cache * s,char * buf)6733 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
6734 {
6735 	return show_slab_objects(s, buf, SO_CPU);
6736 }
6737 SLAB_ATTR_RO(cpu_slabs);
6738 
objects_partial_show(struct kmem_cache * s,char * buf)6739 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
6740 {
6741 	return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
6742 }
6743 SLAB_ATTR_RO(objects_partial);
6744 
slabs_cpu_partial_show(struct kmem_cache * s,char * buf)6745 static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
6746 {
6747 	int objects = 0;
6748 	int slabs = 0;
6749 	int cpu __maybe_unused;
6750 	int len = 0;
6751 
6752 #ifdef CONFIG_SLUB_CPU_PARTIAL
6753 	for_each_online_cpu(cpu) {
6754 		struct slab *slab;
6755 
6756 		slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
6757 
6758 		if (slab)
6759 			slabs += data_race(slab->slabs);
6760 	}
6761 #endif
6762 
6763 	/* Approximate half-full slabs, see slub_set_cpu_partial() */
6764 	objects = (slabs * oo_objects(s->oo)) / 2;
6765 	len += sysfs_emit_at(buf, len, "%d(%d)", objects, slabs);
6766 
6767 #ifdef CONFIG_SLUB_CPU_PARTIAL
6768 	for_each_online_cpu(cpu) {
6769 		struct slab *slab;
6770 
6771 		slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
6772 		if (slab) {
6773 			slabs = data_race(slab->slabs);
6774 			objects = (slabs * oo_objects(s->oo)) / 2;
6775 			len += sysfs_emit_at(buf, len, " C%d=%d(%d)",
6776 					     cpu, objects, slabs);
6777 		}
6778 	}
6779 #endif
6780 	len += sysfs_emit_at(buf, len, "\n");
6781 
6782 	return len;
6783 }
6784 SLAB_ATTR_RO(slabs_cpu_partial);
6785 
reclaim_account_show(struct kmem_cache * s,char * buf)6786 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
6787 {
6788 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
6789 }
6790 SLAB_ATTR_RO(reclaim_account);
6791 
hwcache_align_show(struct kmem_cache * s,char * buf)6792 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
6793 {
6794 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
6795 }
6796 SLAB_ATTR_RO(hwcache_align);
6797 
6798 #ifdef CONFIG_ZONE_DMA
cache_dma_show(struct kmem_cache * s,char * buf)6799 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
6800 {
6801 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
6802 }
6803 SLAB_ATTR_RO(cache_dma);
6804 #endif
6805 
6806 #ifdef CONFIG_HARDENED_USERCOPY
usersize_show(struct kmem_cache * s,char * buf)6807 static ssize_t usersize_show(struct kmem_cache *s, char *buf)
6808 {
6809 	return sysfs_emit(buf, "%u\n", s->usersize);
6810 }
6811 SLAB_ATTR_RO(usersize);
6812 #endif
6813 
destroy_by_rcu_show(struct kmem_cache * s,char * buf)6814 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
6815 {
6816 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
6817 }
6818 SLAB_ATTR_RO(destroy_by_rcu);
6819 
6820 #ifdef CONFIG_SLUB_DEBUG
slabs_show(struct kmem_cache * s,char * buf)6821 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
6822 {
6823 	return show_slab_objects(s, buf, SO_ALL);
6824 }
6825 SLAB_ATTR_RO(slabs);
6826 
total_objects_show(struct kmem_cache * s,char * buf)6827 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
6828 {
6829 	return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
6830 }
6831 SLAB_ATTR_RO(total_objects);
6832 
objects_show(struct kmem_cache * s,char * buf)6833 static ssize_t objects_show(struct kmem_cache *s, char *buf)
6834 {
6835 	return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
6836 }
6837 SLAB_ATTR_RO(objects);
6838 
sanity_checks_show(struct kmem_cache * s,char * buf)6839 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
6840 {
6841 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
6842 }
6843 SLAB_ATTR_RO(sanity_checks);
6844 
trace_show(struct kmem_cache * s,char * buf)6845 static ssize_t trace_show(struct kmem_cache *s, char *buf)
6846 {
6847 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE));
6848 }
6849 SLAB_ATTR_RO(trace);
6850 
red_zone_show(struct kmem_cache * s,char * buf)6851 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
6852 {
6853 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
6854 }
6855 
6856 SLAB_ATTR_RO(red_zone);
6857 
poison_show(struct kmem_cache * s,char * buf)6858 static ssize_t poison_show(struct kmem_cache *s, char *buf)
6859 {
6860 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON));
6861 }
6862 
6863 SLAB_ATTR_RO(poison);
6864 
store_user_show(struct kmem_cache * s,char * buf)6865 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
6866 {
6867 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
6868 }
6869 
6870 SLAB_ATTR_RO(store_user);
6871 
validate_show(struct kmem_cache * s,char * buf)6872 static ssize_t validate_show(struct kmem_cache *s, char *buf)
6873 {
6874 	return 0;
6875 }
6876 
validate_store(struct kmem_cache * s,const char * buf,size_t length)6877 static ssize_t validate_store(struct kmem_cache *s,
6878 			const char *buf, size_t length)
6879 {
6880 	int ret = -EINVAL;
6881 
6882 	if (buf[0] == '1' && kmem_cache_debug(s)) {
6883 		ret = validate_slab_cache(s);
6884 		if (ret >= 0)
6885 			ret = length;
6886 	}
6887 	return ret;
6888 }
6889 SLAB_ATTR(validate);
6890 
6891 #endif /* CONFIG_SLUB_DEBUG */
6892 
6893 #ifdef CONFIG_FAILSLAB
failslab_show(struct kmem_cache * s,char * buf)6894 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
6895 {
6896 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
6897 }
6898 
failslab_store(struct kmem_cache * s,const char * buf,size_t length)6899 static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
6900 				size_t length)
6901 {
6902 	if (s->refcount > 1)
6903 		return -EINVAL;
6904 
6905 	if (buf[0] == '1')
6906 		WRITE_ONCE(s->flags, s->flags | SLAB_FAILSLAB);
6907 	else
6908 		WRITE_ONCE(s->flags, s->flags & ~SLAB_FAILSLAB);
6909 
6910 	return length;
6911 }
6912 SLAB_ATTR(failslab);
6913 #endif
6914 
shrink_show(struct kmem_cache * s,char * buf)6915 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
6916 {
6917 	return 0;
6918 }
6919 
shrink_store(struct kmem_cache * s,const char * buf,size_t length)6920 static ssize_t shrink_store(struct kmem_cache *s,
6921 			const char *buf, size_t length)
6922 {
6923 	if (buf[0] == '1')
6924 		kmem_cache_shrink(s);
6925 	else
6926 		return -EINVAL;
6927 	return length;
6928 }
6929 SLAB_ATTR(shrink);
6930 
6931 #ifdef CONFIG_NUMA
remote_node_defrag_ratio_show(struct kmem_cache * s,char * buf)6932 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
6933 {
6934 	return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10);
6935 }
6936 
remote_node_defrag_ratio_store(struct kmem_cache * s,const char * buf,size_t length)6937 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
6938 				const char *buf, size_t length)
6939 {
6940 	unsigned int ratio;
6941 	int err;
6942 
6943 	err = kstrtouint(buf, 10, &ratio);
6944 	if (err)
6945 		return err;
6946 	if (ratio > 100)
6947 		return -ERANGE;
6948 
6949 	s->remote_node_defrag_ratio = ratio * 10;
6950 
6951 	return length;
6952 }
6953 SLAB_ATTR(remote_node_defrag_ratio);
6954 #endif
6955 
6956 #ifdef CONFIG_SLUB_STATS
show_stat(struct kmem_cache * s,char * buf,enum stat_item si)6957 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
6958 {
6959 	unsigned long sum  = 0;
6960 	int cpu;
6961 	int len = 0;
6962 	int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
6963 
6964 	if (!data)
6965 		return -ENOMEM;
6966 
6967 	for_each_online_cpu(cpu) {
6968 		unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
6969 
6970 		data[cpu] = x;
6971 		sum += x;
6972 	}
6973 
6974 	len += sysfs_emit_at(buf, len, "%lu", sum);
6975 
6976 #ifdef CONFIG_SMP
6977 	for_each_online_cpu(cpu) {
6978 		if (data[cpu])
6979 			len += sysfs_emit_at(buf, len, " C%d=%u",
6980 					     cpu, data[cpu]);
6981 	}
6982 #endif
6983 	kfree(data);
6984 	len += sysfs_emit_at(buf, len, "\n");
6985 
6986 	return len;
6987 }
6988 
clear_stat(struct kmem_cache * s,enum stat_item si)6989 static void clear_stat(struct kmem_cache *s, enum stat_item si)
6990 {
6991 	int cpu;
6992 
6993 	for_each_online_cpu(cpu)
6994 		per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
6995 }
6996 
6997 #define STAT_ATTR(si, text) 					\
6998 static ssize_t text##_show(struct kmem_cache *s, char *buf)	\
6999 {								\
7000 	return show_stat(s, buf, si);				\
7001 }								\
7002 static ssize_t text##_store(struct kmem_cache *s,		\
7003 				const char *buf, size_t length)	\
7004 {								\
7005 	if (buf[0] != '0')					\
7006 		return -EINVAL;					\
7007 	clear_stat(s, si);					\
7008 	return length;						\
7009 }								\
7010 SLAB_ATTR(text);						\
7011 
7012 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
7013 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
7014 STAT_ATTR(FREE_FASTPATH, free_fastpath);
7015 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
7016 STAT_ATTR(FREE_FROZEN, free_frozen);
7017 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
7018 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
7019 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
7020 STAT_ATTR(ALLOC_SLAB, alloc_slab);
7021 STAT_ATTR(ALLOC_REFILL, alloc_refill);
7022 STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
7023 STAT_ATTR(FREE_SLAB, free_slab);
7024 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
7025 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
7026 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
7027 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
7028 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
7029 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
7030 STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
7031 STAT_ATTR(ORDER_FALLBACK, order_fallback);
7032 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
7033 STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
7034 STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
7035 STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
7036 STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
7037 STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
7038 #endif	/* CONFIG_SLUB_STATS */
7039 
7040 #ifdef CONFIG_KFENCE
skip_kfence_show(struct kmem_cache * s,char * buf)7041 static ssize_t skip_kfence_show(struct kmem_cache *s, char *buf)
7042 {
7043 	return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_SKIP_KFENCE));
7044 }
7045 
skip_kfence_store(struct kmem_cache * s,const char * buf,size_t length)7046 static ssize_t skip_kfence_store(struct kmem_cache *s,
7047 			const char *buf, size_t length)
7048 {
7049 	int ret = length;
7050 
7051 	if (buf[0] == '0')
7052 		s->flags &= ~SLAB_SKIP_KFENCE;
7053 	else if (buf[0] == '1')
7054 		s->flags |= SLAB_SKIP_KFENCE;
7055 	else
7056 		ret = -EINVAL;
7057 
7058 	return ret;
7059 }
7060 SLAB_ATTR(skip_kfence);
7061 #endif
7062 
7063 static struct attribute *slab_attrs[] = {
7064 	&slab_size_attr.attr,
7065 	&object_size_attr.attr,
7066 	&objs_per_slab_attr.attr,
7067 	&order_attr.attr,
7068 	&min_partial_attr.attr,
7069 	&cpu_partial_attr.attr,
7070 	&objects_partial_attr.attr,
7071 	&partial_attr.attr,
7072 	&cpu_slabs_attr.attr,
7073 	&ctor_attr.attr,
7074 	&aliases_attr.attr,
7075 	&align_attr.attr,
7076 	&hwcache_align_attr.attr,
7077 	&reclaim_account_attr.attr,
7078 	&destroy_by_rcu_attr.attr,
7079 	&shrink_attr.attr,
7080 	&slabs_cpu_partial_attr.attr,
7081 #ifdef CONFIG_SLUB_DEBUG
7082 	&total_objects_attr.attr,
7083 	&objects_attr.attr,
7084 	&slabs_attr.attr,
7085 	&sanity_checks_attr.attr,
7086 	&trace_attr.attr,
7087 	&red_zone_attr.attr,
7088 	&poison_attr.attr,
7089 	&store_user_attr.attr,
7090 	&validate_attr.attr,
7091 #endif
7092 #ifdef CONFIG_ZONE_DMA
7093 	&cache_dma_attr.attr,
7094 #endif
7095 #ifdef CONFIG_NUMA
7096 	&remote_node_defrag_ratio_attr.attr,
7097 #endif
7098 #ifdef CONFIG_SLUB_STATS
7099 	&alloc_fastpath_attr.attr,
7100 	&alloc_slowpath_attr.attr,
7101 	&free_fastpath_attr.attr,
7102 	&free_slowpath_attr.attr,
7103 	&free_frozen_attr.attr,
7104 	&free_add_partial_attr.attr,
7105 	&free_remove_partial_attr.attr,
7106 	&alloc_from_partial_attr.attr,
7107 	&alloc_slab_attr.attr,
7108 	&alloc_refill_attr.attr,
7109 	&alloc_node_mismatch_attr.attr,
7110 	&free_slab_attr.attr,
7111 	&cpuslab_flush_attr.attr,
7112 	&deactivate_full_attr.attr,
7113 	&deactivate_empty_attr.attr,
7114 	&deactivate_to_head_attr.attr,
7115 	&deactivate_to_tail_attr.attr,
7116 	&deactivate_remote_frees_attr.attr,
7117 	&deactivate_bypass_attr.attr,
7118 	&order_fallback_attr.attr,
7119 	&cmpxchg_double_fail_attr.attr,
7120 	&cmpxchg_double_cpu_fail_attr.attr,
7121 	&cpu_partial_alloc_attr.attr,
7122 	&cpu_partial_free_attr.attr,
7123 	&cpu_partial_node_attr.attr,
7124 	&cpu_partial_drain_attr.attr,
7125 #endif
7126 #ifdef CONFIG_FAILSLAB
7127 	&failslab_attr.attr,
7128 #endif
7129 #ifdef CONFIG_HARDENED_USERCOPY
7130 	&usersize_attr.attr,
7131 #endif
7132 #ifdef CONFIG_KFENCE
7133 	&skip_kfence_attr.attr,
7134 #endif
7135 
7136 	NULL
7137 };
7138 
7139 static const struct attribute_group slab_attr_group = {
7140 	.attrs = slab_attrs,
7141 };
7142 
slab_attr_show(struct kobject * kobj,struct attribute * attr,char * buf)7143 static ssize_t slab_attr_show(struct kobject *kobj,
7144 				struct attribute *attr,
7145 				char *buf)
7146 {
7147 	struct slab_attribute *attribute;
7148 	struct kmem_cache *s;
7149 
7150 	attribute = to_slab_attr(attr);
7151 	s = to_slab(kobj);
7152 
7153 	if (!attribute->show)
7154 		return -EIO;
7155 
7156 	return attribute->show(s, buf);
7157 }
7158 
slab_attr_store(struct kobject * kobj,struct attribute * attr,const char * buf,size_t len)7159 static ssize_t slab_attr_store(struct kobject *kobj,
7160 				struct attribute *attr,
7161 				const char *buf, size_t len)
7162 {
7163 	struct slab_attribute *attribute;
7164 	struct kmem_cache *s;
7165 
7166 	attribute = to_slab_attr(attr);
7167 	s = to_slab(kobj);
7168 
7169 	if (!attribute->store)
7170 		return -EIO;
7171 
7172 	return attribute->store(s, buf, len);
7173 }
7174 
kmem_cache_release(struct kobject * k)7175 static void kmem_cache_release(struct kobject *k)
7176 {
7177 	slab_kmem_cache_release(to_slab(k));
7178 }
7179 
7180 static const struct sysfs_ops slab_sysfs_ops = {
7181 	.show = slab_attr_show,
7182 	.store = slab_attr_store,
7183 };
7184 
7185 static const struct kobj_type slab_ktype = {
7186 	.sysfs_ops = &slab_sysfs_ops,
7187 	.release = kmem_cache_release,
7188 };
7189 
7190 static struct kset *slab_kset;
7191 
cache_kset(struct kmem_cache * s)7192 static inline struct kset *cache_kset(struct kmem_cache *s)
7193 {
7194 	return slab_kset;
7195 }
7196 
7197 #define ID_STR_LENGTH 32
7198 
7199 /* Create a unique string id for a slab cache:
7200  *
7201  * Format	:[flags-]size
7202  */
create_unique_id(struct kmem_cache * s)7203 static char *create_unique_id(struct kmem_cache *s)
7204 {
7205 	char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
7206 	char *p = name;
7207 
7208 	if (!name)
7209 		return ERR_PTR(-ENOMEM);
7210 
7211 	*p++ = ':';
7212 	/*
7213 	 * First flags affecting slabcache operations. We will only
7214 	 * get here for aliasable slabs so we do not need to support
7215 	 * too many flags. The flags here must cover all flags that
7216 	 * are matched during merging to guarantee that the id is
7217 	 * unique.
7218 	 */
7219 	if (s->flags & SLAB_CACHE_DMA)
7220 		*p++ = 'd';
7221 	if (s->flags & SLAB_CACHE_DMA32)
7222 		*p++ = 'D';
7223 	if (s->flags & SLAB_RECLAIM_ACCOUNT)
7224 		*p++ = 'a';
7225 	if (s->flags & SLAB_CONSISTENCY_CHECKS)
7226 		*p++ = 'F';
7227 	if (s->flags & SLAB_ACCOUNT)
7228 		*p++ = 'A';
7229 	if (p != name + 1)
7230 		*p++ = '-';
7231 	p += snprintf(p, ID_STR_LENGTH - (p - name), "%07u", s->size);
7232 
7233 	if (WARN_ON(p > name + ID_STR_LENGTH - 1)) {
7234 		kfree(name);
7235 		return ERR_PTR(-EINVAL);
7236 	}
7237 	kmsan_unpoison_memory(name, p - name);
7238 	return name;
7239 }
7240 
sysfs_slab_add(struct kmem_cache * s)7241 static int sysfs_slab_add(struct kmem_cache *s)
7242 {
7243 	int err;
7244 	const char *name;
7245 	struct kset *kset = cache_kset(s);
7246 	int unmergeable = slab_unmergeable(s);
7247 
7248 	if (!unmergeable && disable_higher_order_debug &&
7249 			(slub_debug & DEBUG_METADATA_FLAGS))
7250 		unmergeable = 1;
7251 
7252 	if (unmergeable) {
7253 		/*
7254 		 * Slabcache can never be merged so we can use the name proper.
7255 		 * This is typically the case for debug situations. In that
7256 		 * case we can catch duplicate names easily.
7257 		 */
7258 		sysfs_remove_link(&slab_kset->kobj, s->name);
7259 		name = s->name;
7260 	} else {
7261 		/*
7262 		 * Create a unique name for the slab as a target
7263 		 * for the symlinks.
7264 		 */
7265 		name = create_unique_id(s);
7266 		if (IS_ERR(name))
7267 			return PTR_ERR(name);
7268 	}
7269 
7270 	s->kobj.kset = kset;
7271 	err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
7272 	if (err)
7273 		goto out;
7274 
7275 	err = sysfs_create_group(&s->kobj, &slab_attr_group);
7276 	if (err)
7277 		goto out_del_kobj;
7278 
7279 	if (!unmergeable) {
7280 		/* Setup first alias */
7281 		sysfs_slab_alias(s, s->name);
7282 	}
7283 out:
7284 	if (!unmergeable)
7285 		kfree(name);
7286 	return err;
7287 out_del_kobj:
7288 	kobject_del(&s->kobj);
7289 	goto out;
7290 }
7291 
sysfs_slab_unlink(struct kmem_cache * s)7292 void sysfs_slab_unlink(struct kmem_cache *s)
7293 {
7294 	if (s->kobj.state_in_sysfs)
7295 		kobject_del(&s->kobj);
7296 }
7297 
sysfs_slab_release(struct kmem_cache * s)7298 void sysfs_slab_release(struct kmem_cache *s)
7299 {
7300 	kobject_put(&s->kobj);
7301 }
7302 
7303 /*
7304  * Need to buffer aliases during bootup until sysfs becomes
7305  * available lest we lose that information.
7306  */
7307 struct saved_alias {
7308 	struct kmem_cache *s;
7309 	const char *name;
7310 	struct saved_alias *next;
7311 };
7312 
7313 static struct saved_alias *alias_list;
7314 
sysfs_slab_alias(struct kmem_cache * s,const char * name)7315 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
7316 {
7317 	struct saved_alias *al;
7318 
7319 	if (slab_state == FULL) {
7320 		/*
7321 		 * If we have a leftover link then remove it.
7322 		 */
7323 		sysfs_remove_link(&slab_kset->kobj, name);
7324 		/*
7325 		 * The original cache may have failed to generate sysfs file.
7326 		 * In that case, sysfs_create_link() returns -ENOENT and
7327 		 * symbolic link creation is skipped.
7328 		 */
7329 		return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
7330 	}
7331 
7332 	al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
7333 	if (!al)
7334 		return -ENOMEM;
7335 
7336 	al->s = s;
7337 	al->name = name;
7338 	al->next = alias_list;
7339 	alias_list = al;
7340 	kmsan_unpoison_memory(al, sizeof(*al));
7341 	return 0;
7342 }
7343 
slab_sysfs_init(void)7344 static int __init slab_sysfs_init(void)
7345 {
7346 	struct kmem_cache *s;
7347 	int err;
7348 
7349 	mutex_lock(&slab_mutex);
7350 
7351 	slab_kset = kset_create_and_add("slab", NULL, kernel_kobj);
7352 	if (!slab_kset) {
7353 		mutex_unlock(&slab_mutex);
7354 		pr_err("Cannot register slab subsystem.\n");
7355 		return -ENOMEM;
7356 	}
7357 
7358 	slab_state = FULL;
7359 
7360 	list_for_each_entry(s, &slab_caches, list) {
7361 		err = sysfs_slab_add(s);
7362 		if (err)
7363 			pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
7364 			       s->name);
7365 	}
7366 
7367 	while (alias_list) {
7368 		struct saved_alias *al = alias_list;
7369 
7370 		alias_list = alias_list->next;
7371 		err = sysfs_slab_alias(al->s, al->name);
7372 		if (err)
7373 			pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
7374 			       al->name);
7375 		kfree(al);
7376 	}
7377 
7378 	mutex_unlock(&slab_mutex);
7379 	return 0;
7380 }
7381 late_initcall(slab_sysfs_init);
7382 #endif /* SLAB_SUPPORTS_SYSFS */
7383 
7384 #if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS)
slab_debugfs_show(struct seq_file * seq,void * v)7385 static int slab_debugfs_show(struct seq_file *seq, void *v)
7386 {
7387 	struct loc_track *t = seq->private;
7388 	struct location *l;
7389 	unsigned long idx;
7390 
7391 	idx = (unsigned long) t->idx;
7392 	if (idx < t->count) {
7393 		l = &t->loc[idx];
7394 
7395 		seq_printf(seq, "%7ld ", l->count);
7396 
7397 		if (l->addr)
7398 			seq_printf(seq, "%pS", (void *)l->addr);
7399 		else
7400 			seq_puts(seq, "<not-available>");
7401 
7402 		if (l->waste)
7403 			seq_printf(seq, " waste=%lu/%lu",
7404 				l->count * l->waste, l->waste);
7405 
7406 		if (l->sum_time != l->min_time) {
7407 			seq_printf(seq, " age=%ld/%llu/%ld",
7408 				l->min_time, div_u64(l->sum_time, l->count),
7409 				l->max_time);
7410 		} else
7411 			seq_printf(seq, " age=%ld", l->min_time);
7412 
7413 		if (l->min_pid != l->max_pid)
7414 			seq_printf(seq, " pid=%ld-%ld", l->min_pid, l->max_pid);
7415 		else
7416 			seq_printf(seq, " pid=%ld",
7417 				l->min_pid);
7418 
7419 		if (num_online_cpus() > 1 && !cpumask_empty(to_cpumask(l->cpus)))
7420 			seq_printf(seq, " cpus=%*pbl",
7421 				 cpumask_pr_args(to_cpumask(l->cpus)));
7422 
7423 		if (nr_online_nodes > 1 && !nodes_empty(l->nodes))
7424 			seq_printf(seq, " nodes=%*pbl",
7425 				 nodemask_pr_args(&l->nodes));
7426 
7427 #ifdef CONFIG_STACKDEPOT
7428 		{
7429 			depot_stack_handle_t handle;
7430 			unsigned long *entries;
7431 			unsigned int nr_entries, j;
7432 
7433 			handle = READ_ONCE(l->handle);
7434 			if (handle) {
7435 				nr_entries = stack_depot_fetch(handle, &entries);
7436 				seq_puts(seq, "\n");
7437 				for (j = 0; j < nr_entries; j++)
7438 					seq_printf(seq, "        %pS\n", (void *)entries[j]);
7439 			}
7440 		}
7441 #endif
7442 		seq_puts(seq, "\n");
7443 	}
7444 
7445 	if (!idx && !t->count)
7446 		seq_puts(seq, "No data\n");
7447 
7448 	return 0;
7449 }
7450 
slab_debugfs_stop(struct seq_file * seq,void * v)7451 static void slab_debugfs_stop(struct seq_file *seq, void *v)
7452 {
7453 }
7454 
slab_debugfs_next(struct seq_file * seq,void * v,loff_t * ppos)7455 static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos)
7456 {
7457 	struct loc_track *t = seq->private;
7458 
7459 	t->idx = ++(*ppos);
7460 	if (*ppos <= t->count)
7461 		return ppos;
7462 
7463 	return NULL;
7464 }
7465 
cmp_loc_by_count(const void * a,const void * b,const void * data)7466 static int cmp_loc_by_count(const void *a, const void *b, const void *data)
7467 {
7468 	struct location *loc1 = (struct location *)a;
7469 	struct location *loc2 = (struct location *)b;
7470 
7471 	if (loc1->count > loc2->count)
7472 		return -1;
7473 	else
7474 		return 1;
7475 }
7476 
slab_debugfs_start(struct seq_file * seq,loff_t * ppos)7477 static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos)
7478 {
7479 	struct loc_track *t = seq->private;
7480 
7481 	t->idx = *ppos;
7482 	return ppos;
7483 }
7484 
7485 static const struct seq_operations slab_debugfs_sops = {
7486 	.start  = slab_debugfs_start,
7487 	.next   = slab_debugfs_next,
7488 	.stop   = slab_debugfs_stop,
7489 	.show   = slab_debugfs_show,
7490 };
7491 
slab_debug_trace_open(struct inode * inode,struct file * filep)7492 static int slab_debug_trace_open(struct inode *inode, struct file *filep)
7493 {
7494 
7495 	struct kmem_cache_node *n;
7496 	enum track_item alloc;
7497 	int node;
7498 	struct loc_track *t = __seq_open_private(filep, &slab_debugfs_sops,
7499 						sizeof(struct loc_track));
7500 	struct kmem_cache *s = file_inode(filep)->i_private;
7501 	unsigned long *obj_map;
7502 
7503 	if (!t)
7504 		return -ENOMEM;
7505 
7506 	obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL);
7507 	if (!obj_map) {
7508 		seq_release_private(inode, filep);
7509 		return -ENOMEM;
7510 	}
7511 
7512 	alloc = debugfs_get_aux_num(filep);
7513 
7514 	if (!alloc_loc_track(t, PAGE_SIZE / sizeof(struct location), GFP_KERNEL)) {
7515 		bitmap_free(obj_map);
7516 		seq_release_private(inode, filep);
7517 		return -ENOMEM;
7518 	}
7519 
7520 	for_each_kmem_cache_node(s, node, n) {
7521 		unsigned long flags;
7522 		struct slab *slab;
7523 
7524 		if (!node_nr_slabs(n))
7525 			continue;
7526 
7527 		spin_lock_irqsave(&n->list_lock, flags);
7528 		list_for_each_entry(slab, &n->partial, slab_list)
7529 			process_slab(t, s, slab, alloc, obj_map);
7530 		list_for_each_entry(slab, &n->full, slab_list)
7531 			process_slab(t, s, slab, alloc, obj_map);
7532 		spin_unlock_irqrestore(&n->list_lock, flags);
7533 	}
7534 
7535 	/* Sort locations by count */
7536 	sort_r(t->loc, t->count, sizeof(struct location),
7537 		cmp_loc_by_count, NULL, NULL);
7538 
7539 	bitmap_free(obj_map);
7540 	return 0;
7541 }
7542 
slab_debug_trace_release(struct inode * inode,struct file * file)7543 static int slab_debug_trace_release(struct inode *inode, struct file *file)
7544 {
7545 	struct seq_file *seq = file->private_data;
7546 	struct loc_track *t = seq->private;
7547 
7548 	free_loc_track(t);
7549 	return seq_release_private(inode, file);
7550 }
7551 
7552 static const struct file_operations slab_debugfs_fops = {
7553 	.open    = slab_debug_trace_open,
7554 	.read    = seq_read,
7555 	.llseek  = seq_lseek,
7556 	.release = slab_debug_trace_release,
7557 };
7558 
debugfs_slab_add(struct kmem_cache * s)7559 static void debugfs_slab_add(struct kmem_cache *s)
7560 {
7561 	struct dentry *slab_cache_dir;
7562 
7563 	if (unlikely(!slab_debugfs_root))
7564 		return;
7565 
7566 	slab_cache_dir = debugfs_create_dir(s->name, slab_debugfs_root);
7567 
7568 	debugfs_create_file_aux_num("alloc_traces", 0400, slab_cache_dir, s,
7569 					TRACK_ALLOC, &slab_debugfs_fops);
7570 
7571 	debugfs_create_file_aux_num("free_traces", 0400, slab_cache_dir, s,
7572 					TRACK_FREE, &slab_debugfs_fops);
7573 }
7574 
debugfs_slab_release(struct kmem_cache * s)7575 void debugfs_slab_release(struct kmem_cache *s)
7576 {
7577 	debugfs_lookup_and_remove(s->name, slab_debugfs_root);
7578 }
7579 
slab_debugfs_init(void)7580 static int __init slab_debugfs_init(void)
7581 {
7582 	struct kmem_cache *s;
7583 
7584 	slab_debugfs_root = debugfs_create_dir("slab", NULL);
7585 
7586 	list_for_each_entry(s, &slab_caches, list)
7587 		if (s->flags & SLAB_STORE_USER)
7588 			debugfs_slab_add(s);
7589 
7590 	return 0;
7591 
7592 }
7593 __initcall(slab_debugfs_init);
7594 #endif
7595 /*
7596  * The /proc/slabinfo ABI
7597  */
7598 #ifdef CONFIG_SLUB_DEBUG
get_slabinfo(struct kmem_cache * s,struct slabinfo * sinfo)7599 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
7600 {
7601 	unsigned long nr_slabs = 0;
7602 	unsigned long nr_objs = 0;
7603 	unsigned long nr_free = 0;
7604 	int node;
7605 	struct kmem_cache_node *n;
7606 
7607 	for_each_kmem_cache_node(s, node, n) {
7608 		nr_slabs += node_nr_slabs(n);
7609 		nr_objs += node_nr_objs(n);
7610 		nr_free += count_partial_free_approx(n);
7611 	}
7612 
7613 	sinfo->active_objs = nr_objs - nr_free;
7614 	sinfo->num_objs = nr_objs;
7615 	sinfo->active_slabs = nr_slabs;
7616 	sinfo->num_slabs = nr_slabs;
7617 	sinfo->objects_per_slab = oo_objects(s->oo);
7618 	sinfo->cache_order = oo_order(s->oo);
7619 }
7620 #endif /* CONFIG_SLUB_DEBUG */
7621