Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * SLUB: A slab allocator that limits cache line use instead of queuing
4 : * objects in per cpu and per node lists.
5 : *
6 : * The allocator synchronizes using per slab locks or atomic operations
7 : * and only uses a centralized lock to manage a pool of partial slabs.
8 : *
9 : * (C) 2007 SGI, Christoph Lameter
10 : * (C) 2011 Linux Foundation, Christoph Lameter
11 : */
12 :
13 : #include <linux/mm.h>
14 : #include <linux/swap.h> /* mm_account_reclaimed_pages() */
15 : #include <linux/module.h>
16 : #include <linux/bit_spinlock.h>
17 : #include <linux/interrupt.h>
18 : #include <linux/swab.h>
19 : #include <linux/bitops.h>
20 : #include <linux/slab.h>
21 : #include "slab.h"
22 : #include <linux/proc_fs.h>
23 : #include <linux/seq_file.h>
24 : #include <linux/kasan.h>
25 : #include <linux/kmsan.h>
26 : #include <linux/cpu.h>
27 : #include <linux/cpuset.h>
28 : #include <linux/mempolicy.h>
29 : #include <linux/ctype.h>
30 : #include <linux/stackdepot.h>
31 : #include <linux/debugobjects.h>
32 : #include <linux/kallsyms.h>
33 : #include <linux/kfence.h>
34 : #include <linux/memory.h>
35 : #include <linux/math64.h>
36 : #include <linux/fault-inject.h>
37 : #include <linux/stacktrace.h>
38 : #include <linux/prefetch.h>
39 : #include <linux/memcontrol.h>
40 : #include <linux/random.h>
41 : #include <kunit/test.h>
42 : #include <kunit/test-bug.h>
43 : #include <linux/sort.h>
44 :
45 : #include <linux/debugfs.h>
46 : #include <trace/events/kmem.h>
47 :
48 : #include "internal.h"
49 :
50 : /*
51 : * Lock order:
52 : * 1. slab_mutex (Global Mutex)
53 : * 2. node->list_lock (Spinlock)
54 : * 3. kmem_cache->cpu_slab->lock (Local lock)
55 : * 4. slab_lock(slab) (Only on some arches)
56 : * 5. object_map_lock (Only for debugging)
57 : *
58 : * slab_mutex
59 : *
60 : * The role of the slab_mutex is to protect the list of all the slabs
61 : * and to synchronize major metadata changes to slab cache structures.
62 : * Also synchronizes memory hotplug callbacks.
63 : *
64 : * slab_lock
65 : *
66 : * The slab_lock is a wrapper around the page lock, thus it is a bit
67 : * spinlock.
68 : *
69 : * The slab_lock is only used on arches that do not have the ability
70 : * to do a cmpxchg_double. It only protects:
71 : *
72 : * A. slab->freelist -> List of free objects in a slab
73 : * B. slab->inuse -> Number of objects in use
74 : * C. slab->objects -> Number of objects in slab
75 : * D. slab->frozen -> frozen state
76 : *
77 : * Frozen slabs
78 : *
79 : * If a slab is frozen then it is exempt from list management. It is not
80 : * on any list except per cpu partial list. The processor that froze the
81 : * slab is the one who can perform list operations on the slab. Other
82 : * processors may put objects onto the freelist but the processor that
83 : * froze the slab is the only one that can retrieve the objects from the
84 : * slab's freelist.
85 : *
86 : * list_lock
87 : *
88 : * The list_lock protects the partial and full list on each node and
89 : * the partial slab counter. If taken then no new slabs may be added or
90 : * removed from the lists nor make the number of partial slabs be modified.
91 : * (Note that the total number of slabs is an atomic value that may be
92 : * modified without taking the list lock).
93 : *
94 : * The list_lock is a centralized lock and thus we avoid taking it as
95 : * much as possible. As long as SLUB does not have to handle partial
96 : * slabs, operations can continue without any centralized lock. F.e.
97 : * allocating a long series of objects that fill up slabs does not require
98 : * the list lock.
99 : *
100 : * For debug caches, all allocations are forced to go through a list_lock
101 : * protected region to serialize against concurrent validation.
102 : *
103 : * cpu_slab->lock local lock
104 : *
105 : * This locks protect slowpath manipulation of all kmem_cache_cpu fields
106 : * except the stat counters. This is a percpu structure manipulated only by
107 : * the local cpu, so the lock protects against being preempted or interrupted
108 : * by an irq. Fast path operations rely on lockless operations instead.
109 : *
110 : * On PREEMPT_RT, the local lock neither disables interrupts nor preemption
111 : * which means the lockless fastpath cannot be used as it might interfere with
112 : * an in-progress slow path operations. In this case the local lock is always
113 : * taken but it still utilizes the freelist for the common operations.
114 : *
115 : * lockless fastpaths
116 : *
117 : * The fast path allocation (slab_alloc_node()) and freeing (do_slab_free())
118 : * are fully lockless when satisfied from the percpu slab (and when
119 : * cmpxchg_double is possible to use, otherwise slab_lock is taken).
120 : * They also don't disable preemption or migration or irqs. They rely on
121 : * the transaction id (tid) field to detect being preempted or moved to
122 : * another cpu.
123 : *
124 : * irq, preemption, migration considerations
125 : *
126 : * Interrupts are disabled as part of list_lock or local_lock operations, or
127 : * around the slab_lock operation, in order to make the slab allocator safe
128 : * to use in the context of an irq.
129 : *
130 : * In addition, preemption (or migration on PREEMPT_RT) is disabled in the
131 : * allocation slowpath, bulk allocation, and put_cpu_partial(), so that the
132 : * local cpu doesn't change in the process and e.g. the kmem_cache_cpu pointer
133 : * doesn't have to be revalidated in each section protected by the local lock.
134 : *
135 : * SLUB assigns one slab for allocation to each processor.
136 : * Allocations only occur from these slabs called cpu slabs.
137 : *
138 : * Slabs with free elements are kept on a partial list and during regular
139 : * operations no list for full slabs is used. If an object in a full slab is
140 : * freed then the slab will show up again on the partial lists.
141 : * We track full slabs for debugging purposes though because otherwise we
142 : * cannot scan all objects.
143 : *
144 : * Slabs are freed when they become empty. Teardown and setup is
145 : * minimal so we rely on the page allocators per cpu caches for
146 : * fast frees and allocs.
147 : *
148 : * slab->frozen The slab is frozen and exempt from list processing.
149 : * This means that the slab is dedicated to a purpose
150 : * such as satisfying allocations for a specific
151 : * processor. Objects may be freed in the slab while
152 : * it is frozen but slab_free will then skip the usual
153 : * list operations. It is up to the processor holding
154 : * the slab to integrate the slab into the slab lists
155 : * when the slab is no longer needed.
156 : *
157 : * One use of this flag is to mark slabs that are
158 : * used for allocations. Then such a slab becomes a cpu
159 : * slab. The cpu slab may be equipped with an additional
160 : * freelist that allows lockless access to
161 : * free objects in addition to the regular freelist
162 : * that requires the slab lock.
163 : *
164 : * SLAB_DEBUG_FLAGS Slab requires special handling due to debug
165 : * options set. This moves slab handling out of
166 : * the fast path and disables lockless freelists.
167 : */
168 :
169 : /*
170 : * We could simply use migrate_disable()/enable() but as long as it's a
171 : * function call even on !PREEMPT_RT, use inline preempt_disable() there.
172 : */
173 : #ifndef CONFIG_PREEMPT_RT
174 : #define slub_get_cpu_ptr(var) get_cpu_ptr(var)
175 : #define slub_put_cpu_ptr(var) put_cpu_ptr(var)
176 : #define USE_LOCKLESS_FAST_PATH() (true)
177 : #else
178 : #define slub_get_cpu_ptr(var) \
179 : ({ \
180 : migrate_disable(); \
181 : this_cpu_ptr(var); \
182 : })
183 : #define slub_put_cpu_ptr(var) \
184 : do { \
185 : (void)(var); \
186 : migrate_enable(); \
187 : } while (0)
188 : #define USE_LOCKLESS_FAST_PATH() (false)
189 : #endif
190 :
191 : #ifndef CONFIG_SLUB_TINY
192 : #define __fastpath_inline __always_inline
193 : #else
194 : #define __fastpath_inline
195 : #endif
196 :
197 : #ifdef CONFIG_SLUB_DEBUG
198 : #ifdef CONFIG_SLUB_DEBUG_ON
199 : DEFINE_STATIC_KEY_TRUE(slub_debug_enabled);
200 : #else
201 : DEFINE_STATIC_KEY_FALSE(slub_debug_enabled);
202 : #endif
203 : #endif /* CONFIG_SLUB_DEBUG */
204 :
205 : /* Structure holding parameters for get_partial() call chain */
206 : struct partial_context {
207 : struct slab **slab;
208 : gfp_t flags;
209 : unsigned int orig_size;
210 : };
211 :
212 : static inline bool kmem_cache_debug(struct kmem_cache *s)
213 : {
214 88772 : return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS);
215 : }
216 :
217 : static inline bool slub_debug_orig_size(struct kmem_cache *s)
218 : {
219 342 : return (kmem_cache_debug_flags(s, SLAB_STORE_USER) &&
220 0 : (s->flags & SLAB_KMALLOC));
221 : }
222 :
223 0 : void *fixup_red_left(struct kmem_cache *s, void *p)
224 : {
225 4388 : if (kmem_cache_debug_flags(s, SLAB_RED_ZONE))
226 0 : p += s->red_left_pad;
227 :
228 0 : return p;
229 : }
230 :
231 : static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
232 : {
233 : #ifdef CONFIG_SLUB_CPU_PARTIAL
234 : return !kmem_cache_debug(s);
235 : #else
236 : return false;
237 : #endif
238 : }
239 :
240 : /*
241 : * Issues still to be resolved:
242 : *
243 : * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
244 : *
245 : * - Variable sizing of the per node arrays
246 : */
247 :
248 : /* Enable to log cmpxchg failures */
249 : #undef SLUB_DEBUG_CMPXCHG
250 :
251 : #ifndef CONFIG_SLUB_TINY
252 : /*
253 : * Minimum number of partial slabs. These will be left on the partial
254 : * lists even if they are empty. kmem_cache_shrink may reclaim them.
255 : */
256 : #define MIN_PARTIAL 5
257 :
258 : /*
259 : * Maximum number of desirable partial slabs.
260 : * The existence of more partial slabs makes kmem_cache_shrink
261 : * sort the partial list by the number of objects in use.
262 : */
263 : #define MAX_PARTIAL 10
264 : #else
265 : #define MIN_PARTIAL 0
266 : #define MAX_PARTIAL 0
267 : #endif
268 :
269 : #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
270 : SLAB_POISON | SLAB_STORE_USER)
271 :
272 : /*
273 : * These debug flags cannot use CMPXCHG because there might be consistency
274 : * issues when checking or reading debug information
275 : */
276 : #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
277 : SLAB_TRACE)
278 :
279 :
280 : /*
281 : * Debugging flags that require metadata to be stored in the slab. These get
282 : * disabled when slub_debug=O is used and a cache's min order increases with
283 : * metadata.
284 : */
285 : #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
286 :
287 : #define OO_SHIFT 16
288 : #define OO_MASK ((1 << OO_SHIFT) - 1)
289 : #define MAX_OBJS_PER_PAGE 32767 /* since slab.objects is u15 */
290 :
291 : /* Internal SLUB flags */
292 : /* Poison object */
293 : #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U)
294 : /* Use cmpxchg_double */
295 : #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U)
296 :
297 : /*
298 : * Tracking user of a slab.
299 : */
300 : #define TRACK_ADDRS_COUNT 16
301 : struct track {
302 : unsigned long addr; /* Called from address */
303 : #ifdef CONFIG_STACKDEPOT
304 : depot_stack_handle_t handle;
305 : #endif
306 : int cpu; /* Was running on cpu */
307 : int pid; /* Pid context */
308 : unsigned long when; /* When did the operation occur */
309 : };
310 :
311 : enum track_item { TRACK_ALLOC, TRACK_FREE };
312 :
313 : #ifdef SLAB_SUPPORTS_SYSFS
314 : static int sysfs_slab_add(struct kmem_cache *);
315 : static int sysfs_slab_alias(struct kmem_cache *, const char *);
316 : #else
317 : static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
318 : static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
319 : { return 0; }
320 : #endif
321 :
322 : #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
323 : static void debugfs_slab_add(struct kmem_cache *);
324 : #else
325 : static inline void debugfs_slab_add(struct kmem_cache *s) { }
326 : #endif
327 :
328 : static inline void stat(const struct kmem_cache *s, enum stat_item si)
329 : {
330 : #ifdef CONFIG_SLUB_STATS
331 : /*
332 : * The rmw is racy on a preemptible kernel but this is acceptable, so
333 : * avoid this_cpu_add()'s irq-disable overhead.
334 : */
335 : raw_cpu_inc(s->cpu_slab->stat[si]);
336 : #endif
337 : }
338 :
339 : /*
340 : * Tracks for which NUMA nodes we have kmem_cache_nodes allocated.
341 : * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily
342 : * differ during memory hotplug/hotremove operations.
343 : * Protected by slab_mutex.
344 : */
345 : static nodemask_t slab_nodes;
346 :
347 : #ifndef CONFIG_SLUB_TINY
348 : /*
349 : * Workqueue used for flush_cpu_slab().
350 : */
351 : static struct workqueue_struct *flushwq;
352 : #endif
353 :
354 : /********************************************************************
355 : * Core slab cache functions
356 : *******************************************************************/
357 :
358 : /*
359 : * Returns freelist pointer (ptr). With hardening, this is obfuscated
360 : * with an XOR of the address where the pointer is held and a per-cache
361 : * random number.
362 : */
363 : static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
364 : unsigned long ptr_addr)
365 : {
366 : #ifdef CONFIG_SLAB_FREELIST_HARDENED
367 : /*
368 : * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged.
369 : * Normally, this doesn't cause any issues, as both set_freepointer()
370 : * and get_freepointer() are called with a pointer with the same tag.
371 : * However, there are some issues with CONFIG_SLUB_DEBUG code. For
372 : * example, when __free_slub() iterates over objects in a cache, it
373 : * passes untagged pointers to check_object(). check_object() in turns
374 : * calls get_freepointer() with an untagged pointer, which causes the
375 : * freepointer to be restored incorrectly.
376 : */
377 : return (void *)((unsigned long)ptr ^ s->random ^
378 : swab((unsigned long)kasan_reset_tag((void *)ptr_addr)));
379 : #else
380 : return ptr;
381 : #endif
382 : }
383 :
384 : /* Returns the freelist pointer recorded at location ptr_addr. */
385 : static inline void *freelist_dereference(const struct kmem_cache *s,
386 : void *ptr_addr)
387 : {
388 110790 : return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr),
389 : (unsigned long)ptr_addr);
390 : }
391 :
392 : static inline void *get_freepointer(struct kmem_cache *s, void *object)
393 : {
394 110790 : object = kasan_reset_tag(object);
395 221580 : return freelist_dereference(s, object + s->offset);
396 : }
397 :
398 : #ifndef CONFIG_SLUB_TINY
399 : static void prefetch_freepointer(const struct kmem_cache *s, void *object)
400 : {
401 59373 : prefetchw(object + s->offset);
402 : }
403 : #endif
404 :
405 : /*
406 : * When running under KMSAN, get_freepointer_safe() may return an uninitialized
407 : * pointer value in the case the current thread loses the race for the next
408 : * memory chunk in the freelist. In that case this_cpu_cmpxchg_double() in
409 : * slab_alloc_node() will fail, so the uninitialized value won't be used, but
410 : * KMSAN will still check all arguments of cmpxchg because of imperfect
411 : * handling of inline assembly.
412 : * To work around this problem, we apply __no_kmsan_checks to ensure that
413 : * get_freepointer_safe() returns initialized memory.
414 : */
415 : __no_kmsan_checks
416 : static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
417 : {
418 : unsigned long freepointer_addr;
419 : void *p;
420 :
421 : if (!debug_pagealloc_enabled_static())
422 118746 : return get_freepointer(s, object);
423 :
424 : object = kasan_reset_tag(object);
425 : freepointer_addr = (unsigned long)object + s->offset;
426 : copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p));
427 : return freelist_ptr(s, p, freepointer_addr);
428 : }
429 :
430 : static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
431 : {
432 149485 : unsigned long freeptr_addr = (unsigned long)object + s->offset;
433 :
434 : #ifdef CONFIG_SLAB_FREELIST_HARDENED
435 : BUG_ON(object == fp); /* naive detection of double free or corruption */
436 : #endif
437 :
438 149485 : freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr);
439 149485 : *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
440 : }
441 :
442 : /* Loop over all objects in a slab */
443 : #define for_each_object(__p, __s, __addr, __objects) \
444 : for (__p = fixup_red_left(__s, __addr); \
445 : __p < (__addr) + (__objects) * (__s)->size; \
446 : __p += (__s)->size)
447 :
448 : static inline unsigned int order_objects(unsigned int order, unsigned int size)
449 : {
450 213 : return ((unsigned int)PAGE_SIZE << order) / size;
451 : }
452 :
453 : static inline struct kmem_cache_order_objects oo_make(unsigned int order,
454 : unsigned int size)
455 : {
456 106 : struct kmem_cache_order_objects x = {
457 212 : (order << OO_SHIFT) + order_objects(order, size)
458 : };
459 :
460 : return x;
461 : }
462 :
463 : static inline unsigned int oo_order(struct kmem_cache_order_objects x)
464 : {
465 6543 : return x.x >> OO_SHIFT;
466 : }
467 :
468 : static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
469 : {
470 53 : return x.x & OO_MASK;
471 : }
472 :
473 : #ifdef CONFIG_SLUB_CPU_PARTIAL
474 : static void slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects)
475 : {
476 : unsigned int nr_slabs;
477 :
478 : s->cpu_partial = nr_objects;
479 :
480 : /*
481 : * We take the number of objects but actually limit the number of
482 : * slabs on the per cpu partial list, in order to limit excessive
483 : * growth of the list. For simplicity we assume that the slabs will
484 : * be half-full.
485 : */
486 : nr_slabs = DIV_ROUND_UP(nr_objects * 2, oo_objects(s->oo));
487 : s->cpu_partial_slabs = nr_slabs;
488 : }
489 : #else
490 : static inline void
491 : slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects)
492 : {
493 : }
494 : #endif /* CONFIG_SLUB_CPU_PARTIAL */
495 :
496 : /*
497 : * Per slab locking using the pagelock
498 : */
499 : static __always_inline void slab_lock(struct slab *slab)
500 : {
501 42165 : struct page *page = slab_page(slab);
502 :
503 : VM_BUG_ON_PAGE(PageTail(page), page);
504 42165 : bit_spin_lock(PG_locked, &page->flags);
505 : }
506 :
507 : static __always_inline void slab_unlock(struct slab *slab)
508 : {
509 42165 : struct page *page = slab_page(slab);
510 :
511 : VM_BUG_ON_PAGE(PageTail(page), page);
512 42165 : __bit_spin_unlock(PG_locked, &page->flags);
513 : }
514 :
515 : /*
516 : * Interrupts must be disabled (for the fallback code to work right), typically
517 : * by an _irqsave() lock variant. On PREEMPT_RT the preempt_disable(), which is
518 : * part of bit_spin_lock(), is sufficient because the policy is not to allow any
519 : * allocation/ free operation in hardirq context. Therefore nothing can
520 : * interrupt the operation.
521 : */
522 : static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct slab *slab,
523 : void *freelist_old, unsigned long counters_old,
524 : void *freelist_new, unsigned long counters_new,
525 : const char *n)
526 : {
527 : if (USE_LOCKLESS_FAST_PATH())
528 : lockdep_assert_irqs_disabled();
529 : #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
530 : defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
531 : if (s->flags & __CMPXCHG_DOUBLE) {
532 : if (cmpxchg_double(&slab->freelist, &slab->counters,
533 : freelist_old, counters_old,
534 : freelist_new, counters_new))
535 : return true;
536 : } else
537 : #endif
538 : {
539 2327 : slab_lock(slab);
540 4654 : if (slab->freelist == freelist_old &&
541 2327 : slab->counters == counters_old) {
542 2327 : slab->freelist = freelist_new;
543 2327 : slab->counters = counters_new;
544 2327 : slab_unlock(slab);
545 : return true;
546 : }
547 0 : slab_unlock(slab);
548 : }
549 :
550 : cpu_relax();
551 0 : stat(s, CMPXCHG_DOUBLE_FAIL);
552 :
553 : #ifdef SLUB_DEBUG_CMPXCHG
554 : pr_info("%s %s: cmpxchg double redo ", n, s->name);
555 : #endif
556 :
557 : return false;
558 : }
559 :
560 39838 : static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct slab *slab,
561 : void *freelist_old, unsigned long counters_old,
562 : void *freelist_new, unsigned long counters_new,
563 : const char *n)
564 : {
565 : #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
566 : defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
567 : if (s->flags & __CMPXCHG_DOUBLE) {
568 : if (cmpxchg_double(&slab->freelist, &slab->counters,
569 : freelist_old, counters_old,
570 : freelist_new, counters_new))
571 : return true;
572 : } else
573 : #endif
574 : {
575 : unsigned long flags;
576 :
577 39838 : local_irq_save(flags);
578 39838 : slab_lock(slab);
579 79676 : if (slab->freelist == freelist_old &&
580 39838 : slab->counters == counters_old) {
581 39838 : slab->freelist = freelist_new;
582 39838 : slab->counters = counters_new;
583 39838 : slab_unlock(slab);
584 79676 : local_irq_restore(flags);
585 : return true;
586 : }
587 0 : slab_unlock(slab);
588 0 : local_irq_restore(flags);
589 : }
590 :
591 : cpu_relax();
592 0 : stat(s, CMPXCHG_DOUBLE_FAIL);
593 :
594 : #ifdef SLUB_DEBUG_CMPXCHG
595 : pr_info("%s %s: cmpxchg double redo ", n, s->name);
596 : #endif
597 :
598 : return false;
599 : }
600 :
601 : #ifdef CONFIG_SLUB_DEBUG
602 : static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)];
603 : static DEFINE_SPINLOCK(object_map_lock);
604 :
605 0 : static void __fill_map(unsigned long *obj_map, struct kmem_cache *s,
606 : struct slab *slab)
607 : {
608 0 : void *addr = slab_address(slab);
609 : void *p;
610 :
611 0 : bitmap_zero(obj_map, slab->objects);
612 :
613 0 : for (p = slab->freelist; p; p = get_freepointer(s, p))
614 0 : set_bit(__obj_to_index(s, addr, p), obj_map);
615 0 : }
616 :
617 : #if IS_ENABLED(CONFIG_KUNIT)
618 0 : static bool slab_add_kunit_errors(void)
619 : {
620 : struct kunit_resource *resource;
621 :
622 0 : if (!kunit_get_current_test())
623 : return false;
624 :
625 0 : resource = kunit_find_named_resource(current->kunit_test, "slab_errors");
626 0 : if (!resource)
627 : return false;
628 :
629 0 : (*(int *)resource->data)++;
630 0 : kunit_put_resource(resource);
631 0 : return true;
632 : }
633 : #else
634 : static inline bool slab_add_kunit_errors(void) { return false; }
635 : #endif
636 :
637 : static inline unsigned int size_from_object(struct kmem_cache *s)
638 : {
639 0 : if (s->flags & SLAB_RED_ZONE)
640 0 : return s->size - s->red_left_pad;
641 :
642 : return s->size;
643 : }
644 :
645 : static inline void *restore_red_left(struct kmem_cache *s, void *p)
646 : {
647 0 : if (s->flags & SLAB_RED_ZONE)
648 0 : p -= s->red_left_pad;
649 :
650 : return p;
651 : }
652 :
653 : /*
654 : * Debug settings:
655 : */
656 : #if defined(CONFIG_SLUB_DEBUG_ON)
657 : static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
658 : #else
659 : static slab_flags_t slub_debug;
660 : #endif
661 :
662 : static char *slub_debug_string;
663 : static int disable_higher_order_debug;
664 :
665 : /*
666 : * slub is about to manipulate internal object metadata. This memory lies
667 : * outside the range of the allocated object, so accessing it would normally
668 : * be reported by kasan as a bounds error. metadata_access_enable() is used
669 : * to tell kasan that these accesses are OK.
670 : */
671 : static inline void metadata_access_enable(void)
672 : {
673 : kasan_disable_current();
674 : }
675 :
676 : static inline void metadata_access_disable(void)
677 : {
678 : kasan_enable_current();
679 : }
680 :
681 : /*
682 : * Object debugging
683 : */
684 :
685 : /* Verify that a pointer has an address that is valid within a slab page */
686 0 : static inline int check_valid_pointer(struct kmem_cache *s,
687 : struct slab *slab, void *object)
688 : {
689 : void *base;
690 :
691 0 : if (!object)
692 : return 1;
693 :
694 0 : base = slab_address(slab);
695 0 : object = kasan_reset_tag(object);
696 0 : object = restore_red_left(s, object);
697 0 : if (object < base || object >= base + slab->objects * s->size ||
698 0 : (object - base) % s->size) {
699 : return 0;
700 : }
701 :
702 0 : return 1;
703 : }
704 :
705 : static void print_section(char *level, char *text, u8 *addr,
706 : unsigned int length)
707 : {
708 : metadata_access_enable();
709 0 : print_hex_dump(level, text, DUMP_PREFIX_ADDRESS,
710 0 : 16, 1, kasan_reset_tag((void *)addr), length, 1);
711 : metadata_access_disable();
712 : }
713 :
714 : /*
715 : * See comment in calculate_sizes().
716 : */
717 : static inline bool freeptr_outside_object(struct kmem_cache *s)
718 : {
719 : return s->offset >= s->inuse;
720 : }
721 :
722 : /*
723 : * Return offset of the end of info block which is inuse + free pointer if
724 : * not overlapping with object.
725 : */
726 : static inline unsigned int get_info_end(struct kmem_cache *s)
727 : {
728 0 : if (freeptr_outside_object(s))
729 0 : return s->inuse + sizeof(void *);
730 : else
731 : return s->inuse;
732 : }
733 :
734 : static struct track *get_track(struct kmem_cache *s, void *object,
735 : enum track_item alloc)
736 : {
737 : struct track *p;
738 :
739 0 : p = object + get_info_end(s);
740 :
741 0 : return kasan_reset_tag(p + alloc);
742 : }
743 :
744 : #ifdef CONFIG_STACKDEPOT
745 0 : static noinline depot_stack_handle_t set_track_prepare(void)
746 : {
747 : depot_stack_handle_t handle;
748 : unsigned long entries[TRACK_ADDRS_COUNT];
749 : unsigned int nr_entries;
750 :
751 0 : nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 3);
752 0 : handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT);
753 :
754 0 : return handle;
755 : }
756 : #else
757 : static inline depot_stack_handle_t set_track_prepare(void)
758 : {
759 : return 0;
760 : }
761 : #endif
762 :
763 : static void set_track_update(struct kmem_cache *s, void *object,
764 : enum track_item alloc, unsigned long addr,
765 : depot_stack_handle_t handle)
766 : {
767 0 : struct track *p = get_track(s, object, alloc);
768 :
769 : #ifdef CONFIG_STACKDEPOT
770 0 : p->handle = handle;
771 : #endif
772 0 : p->addr = addr;
773 0 : p->cpu = smp_processor_id();
774 0 : p->pid = current->pid;
775 0 : p->when = jiffies;
776 : }
777 :
778 : static __always_inline void set_track(struct kmem_cache *s, void *object,
779 : enum track_item alloc, unsigned long addr)
780 : {
781 0 : depot_stack_handle_t handle = set_track_prepare();
782 :
783 : set_track_update(s, object, alloc, addr, handle);
784 : }
785 :
786 1 : static void init_tracking(struct kmem_cache *s, void *object)
787 : {
788 : struct track *p;
789 :
790 1 : if (!(s->flags & SLAB_STORE_USER))
791 : return;
792 :
793 0 : p = get_track(s, object, TRACK_ALLOC);
794 0 : memset(p, 0, 2*sizeof(struct track));
795 : }
796 :
797 0 : static void print_track(const char *s, struct track *t, unsigned long pr_time)
798 : {
799 : depot_stack_handle_t handle __maybe_unused;
800 :
801 0 : if (!t->addr)
802 : return;
803 :
804 0 : pr_err("%s in %pS age=%lu cpu=%u pid=%d\n",
805 : s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
806 : #ifdef CONFIG_STACKDEPOT
807 0 : handle = READ_ONCE(t->handle);
808 0 : if (handle)
809 0 : stack_depot_print(handle);
810 : else
811 0 : pr_err("object allocation/free stack trace missing\n");
812 : #endif
813 : }
814 :
815 0 : void print_tracking(struct kmem_cache *s, void *object)
816 : {
817 0 : unsigned long pr_time = jiffies;
818 0 : if (!(s->flags & SLAB_STORE_USER))
819 : return;
820 :
821 0 : print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
822 0 : print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
823 : }
824 :
825 : static void print_slab_info(const struct slab *slab)
826 : {
827 0 : struct folio *folio = (struct folio *)slab_folio(slab);
828 :
829 0 : pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%pGp\n",
830 : slab, slab->objects, slab->inuse, slab->freelist,
831 : folio_flags(folio, 0));
832 : }
833 :
834 : /*
835 : * kmalloc caches has fixed sizes (mostly power of 2), and kmalloc() API
836 : * family will round up the real request size to these fixed ones, so
837 : * there could be an extra area than what is requested. Save the original
838 : * request size in the meta data area, for better debug and sanity check.
839 : */
840 118 : static inline void set_orig_size(struct kmem_cache *s,
841 : void *object, unsigned int orig_size)
842 : {
843 118 : void *p = kasan_reset_tag(object);
844 :
845 118 : if (!slub_debug_orig_size(s))
846 : return;
847 :
848 : #ifdef CONFIG_KASAN_GENERIC
849 : /*
850 : * KASAN could save its free meta data in object's data area at
851 : * offset 0, if the size is larger than 'orig_size', it will
852 : * overlap the data redzone in [orig_size+1, object_size], and
853 : * the check should be skipped.
854 : */
855 : if (kasan_metadata_size(s, true) > orig_size)
856 : orig_size = s->object_size;
857 : #endif
858 :
859 0 : p += get_info_end(s);
860 0 : p += sizeof(struct track) * 2;
861 :
862 0 : *(unsigned int *)p = orig_size;
863 : }
864 :
865 0 : static inline unsigned int get_orig_size(struct kmem_cache *s, void *object)
866 : {
867 0 : void *p = kasan_reset_tag(object);
868 :
869 0 : if (!slub_debug_orig_size(s))
870 0 : return s->object_size;
871 :
872 0 : p += get_info_end(s);
873 0 : p += sizeof(struct track) * 2;
874 :
875 0 : return *(unsigned int *)p;
876 : }
877 :
878 118 : void skip_orig_size_check(struct kmem_cache *s, const void *object)
879 : {
880 118 : set_orig_size(s, (void *)object, s->object_size);
881 118 : }
882 :
883 0 : static void slab_bug(struct kmem_cache *s, char *fmt, ...)
884 : {
885 : struct va_format vaf;
886 : va_list args;
887 :
888 0 : va_start(args, fmt);
889 0 : vaf.fmt = fmt;
890 0 : vaf.va = &args;
891 0 : pr_err("=============================================================================\n");
892 0 : pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
893 0 : pr_err("-----------------------------------------------------------------------------\n\n");
894 0 : va_end(args);
895 0 : }
896 :
897 : __printf(2, 3)
898 0 : static void slab_fix(struct kmem_cache *s, char *fmt, ...)
899 : {
900 : struct va_format vaf;
901 : va_list args;
902 :
903 0 : if (slab_add_kunit_errors())
904 0 : return;
905 :
906 0 : va_start(args, fmt);
907 0 : vaf.fmt = fmt;
908 0 : vaf.va = &args;
909 0 : pr_err("FIX %s: %pV\n", s->name, &vaf);
910 0 : va_end(args);
911 : }
912 :
913 0 : static void print_trailer(struct kmem_cache *s, struct slab *slab, u8 *p)
914 : {
915 : unsigned int off; /* Offset of last byte */
916 0 : u8 *addr = slab_address(slab);
917 :
918 0 : print_tracking(s, p);
919 :
920 0 : print_slab_info(slab);
921 :
922 0 : pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n",
923 : p, p - addr, get_freepointer(s, p));
924 :
925 0 : if (s->flags & SLAB_RED_ZONE)
926 0 : print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
927 : s->red_left_pad);
928 0 : else if (p > addr + 16)
929 0 : print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
930 :
931 0 : print_section(KERN_ERR, "Object ", p,
932 0 : min_t(unsigned int, s->object_size, PAGE_SIZE));
933 0 : if (s->flags & SLAB_RED_ZONE)
934 0 : print_section(KERN_ERR, "Redzone ", p + s->object_size,
935 0 : s->inuse - s->object_size);
936 :
937 0 : off = get_info_end(s);
938 :
939 0 : if (s->flags & SLAB_STORE_USER)
940 0 : off += 2 * sizeof(struct track);
941 :
942 0 : if (slub_debug_orig_size(s))
943 0 : off += sizeof(unsigned int);
944 :
945 0 : off += kasan_metadata_size(s, false);
946 :
947 0 : if (off != size_from_object(s))
948 : /* Beginning of the filler is the free pointer */
949 0 : print_section(KERN_ERR, "Padding ", p + off,
950 0 : size_from_object(s) - off);
951 :
952 0 : dump_stack();
953 0 : }
954 :
955 0 : static void object_err(struct kmem_cache *s, struct slab *slab,
956 : u8 *object, char *reason)
957 : {
958 0 : if (slab_add_kunit_errors())
959 : return;
960 :
961 0 : slab_bug(s, "%s", reason);
962 0 : print_trailer(s, slab, object);
963 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
964 : }
965 :
966 82 : static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab,
967 : void **freelist, void *nextfree)
968 : {
969 82 : if ((s->flags & SLAB_CONSISTENCY_CHECKS) &&
970 0 : !check_valid_pointer(s, slab, nextfree) && freelist) {
971 0 : object_err(s, slab, *freelist, "Freechain corrupt");
972 0 : *freelist = NULL;
973 0 : slab_fix(s, "Isolate corrupted freechain");
974 0 : return true;
975 : }
976 :
977 : return false;
978 : }
979 :
980 0 : static __printf(3, 4) void slab_err(struct kmem_cache *s, struct slab *slab,
981 : const char *fmt, ...)
982 : {
983 : va_list args;
984 : char buf[100];
985 :
986 0 : if (slab_add_kunit_errors())
987 0 : return;
988 :
989 0 : va_start(args, fmt);
990 0 : vsnprintf(buf, sizeof(buf), fmt, args);
991 0 : va_end(args);
992 0 : slab_bug(s, "%s", buf);
993 0 : print_slab_info(slab);
994 0 : dump_stack();
995 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
996 : }
997 :
998 1 : static void init_object(struct kmem_cache *s, void *object, u8 val)
999 : {
1000 1 : u8 *p = kasan_reset_tag(object);
1001 1 : unsigned int poison_size = s->object_size;
1002 :
1003 1 : if (s->flags & SLAB_RED_ZONE) {
1004 0 : memset(p - s->red_left_pad, val, s->red_left_pad);
1005 :
1006 0 : if (slub_debug_orig_size(s) && val == SLUB_RED_ACTIVE) {
1007 : /*
1008 : * Redzone the extra allocated space by kmalloc than
1009 : * requested, and the poison size will be limited to
1010 : * the original request size accordingly.
1011 : */
1012 0 : poison_size = get_orig_size(s, object);
1013 : }
1014 : }
1015 :
1016 1 : if (s->flags & __OBJECT_POISON) {
1017 0 : memset(p, POISON_FREE, poison_size - 1);
1018 0 : p[poison_size - 1] = POISON_END;
1019 : }
1020 :
1021 1 : if (s->flags & SLAB_RED_ZONE)
1022 0 : memset(p + poison_size, val, s->inuse - poison_size);
1023 1 : }
1024 :
1025 0 : static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
1026 : void *from, void *to)
1027 : {
1028 0 : slab_fix(s, "Restoring %s 0x%p-0x%p=0x%x", message, from, to - 1, data);
1029 0 : memset(from, data, to - from);
1030 0 : }
1031 :
1032 0 : static int check_bytes_and_report(struct kmem_cache *s, struct slab *slab,
1033 : u8 *object, char *what,
1034 : u8 *start, unsigned int value, unsigned int bytes)
1035 : {
1036 : u8 *fault;
1037 : u8 *end;
1038 0 : u8 *addr = slab_address(slab);
1039 :
1040 : metadata_access_enable();
1041 0 : fault = memchr_inv(kasan_reset_tag(start), value, bytes);
1042 : metadata_access_disable();
1043 0 : if (!fault)
1044 : return 1;
1045 :
1046 0 : end = start + bytes;
1047 0 : while (end > fault && end[-1] == value)
1048 0 : end--;
1049 :
1050 0 : if (slab_add_kunit_errors())
1051 : goto skip_bug_print;
1052 :
1053 0 : slab_bug(s, "%s overwritten", what);
1054 0 : pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n",
1055 : fault, end - 1, fault - addr,
1056 : fault[0], value);
1057 0 : print_trailer(s, slab, object);
1058 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
1059 :
1060 : skip_bug_print:
1061 0 : restore_bytes(s, what, value, fault, end);
1062 0 : return 0;
1063 : }
1064 :
1065 : /*
1066 : * Object layout:
1067 : *
1068 : * object address
1069 : * Bytes of the object to be managed.
1070 : * If the freepointer may overlay the object then the free
1071 : * pointer is at the middle of the object.
1072 : *
1073 : * Poisoning uses 0x6b (POISON_FREE) and the last byte is
1074 : * 0xa5 (POISON_END)
1075 : *
1076 : * object + s->object_size
1077 : * Padding to reach word boundary. This is also used for Redzoning.
1078 : * Padding is extended by another word if Redzoning is enabled and
1079 : * object_size == inuse.
1080 : *
1081 : * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
1082 : * 0xcc (RED_ACTIVE) for objects in use.
1083 : *
1084 : * object + s->inuse
1085 : * Meta data starts here.
1086 : *
1087 : * A. Free pointer (if we cannot overwrite object on free)
1088 : * B. Tracking data for SLAB_STORE_USER
1089 : * C. Original request size for kmalloc object (SLAB_STORE_USER enabled)
1090 : * D. Padding to reach required alignment boundary or at minimum
1091 : * one word if debugging is on to be able to detect writes
1092 : * before the word boundary.
1093 : *
1094 : * Padding is done using 0x5a (POISON_INUSE)
1095 : *
1096 : * object + s->size
1097 : * Nothing is used beyond s->size.
1098 : *
1099 : * If slabcaches are merged then the object_size and inuse boundaries are mostly
1100 : * ignored. And therefore no slab options that rely on these boundaries
1101 : * may be used with merged slabcaches.
1102 : */
1103 :
1104 0 : static int check_pad_bytes(struct kmem_cache *s, struct slab *slab, u8 *p)
1105 : {
1106 0 : unsigned long off = get_info_end(s); /* The end of info */
1107 :
1108 0 : if (s->flags & SLAB_STORE_USER) {
1109 : /* We also have user information there */
1110 0 : off += 2 * sizeof(struct track);
1111 :
1112 0 : if (s->flags & SLAB_KMALLOC)
1113 0 : off += sizeof(unsigned int);
1114 : }
1115 :
1116 0 : off += kasan_metadata_size(s, false);
1117 :
1118 0 : if (size_from_object(s) == off)
1119 : return 1;
1120 :
1121 0 : return check_bytes_and_report(s, slab, p, "Object padding",
1122 0 : p + off, POISON_INUSE, size_from_object(s) - off);
1123 : }
1124 :
1125 : /* Check the pad bytes at the end of a slab page */
1126 0 : static void slab_pad_check(struct kmem_cache *s, struct slab *slab)
1127 : {
1128 : u8 *start;
1129 : u8 *fault;
1130 : u8 *end;
1131 : u8 *pad;
1132 : int length;
1133 : int remainder;
1134 :
1135 0 : if (!(s->flags & SLAB_POISON))
1136 : return;
1137 :
1138 0 : start = slab_address(slab);
1139 0 : length = slab_size(slab);
1140 0 : end = start + length;
1141 0 : remainder = length % s->size;
1142 0 : if (!remainder)
1143 : return;
1144 :
1145 0 : pad = end - remainder;
1146 : metadata_access_enable();
1147 0 : fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder);
1148 : metadata_access_disable();
1149 0 : if (!fault)
1150 : return;
1151 0 : while (end > fault && end[-1] == POISON_INUSE)
1152 0 : end--;
1153 :
1154 0 : slab_err(s, slab, "Padding overwritten. 0x%p-0x%p @offset=%tu",
1155 : fault, end - 1, fault - start);
1156 0 : print_section(KERN_ERR, "Padding ", pad, remainder);
1157 :
1158 0 : restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
1159 : }
1160 :
1161 0 : static int check_object(struct kmem_cache *s, struct slab *slab,
1162 : void *object, u8 val)
1163 : {
1164 0 : u8 *p = object;
1165 0 : u8 *endobject = object + s->object_size;
1166 : unsigned int orig_size;
1167 :
1168 0 : if (s->flags & SLAB_RED_ZONE) {
1169 0 : if (!check_bytes_and_report(s, slab, object, "Left Redzone",
1170 0 : object - s->red_left_pad, val, s->red_left_pad))
1171 : return 0;
1172 :
1173 0 : if (!check_bytes_and_report(s, slab, object, "Right Redzone",
1174 0 : endobject, val, s->inuse - s->object_size))
1175 : return 0;
1176 :
1177 0 : if (slub_debug_orig_size(s) && val == SLUB_RED_ACTIVE) {
1178 0 : orig_size = get_orig_size(s, object);
1179 :
1180 0 : if (s->object_size > orig_size &&
1181 0 : !check_bytes_and_report(s, slab, object,
1182 : "kmalloc Redzone", p + orig_size,
1183 : val, s->object_size - orig_size)) {
1184 : return 0;
1185 : }
1186 : }
1187 : } else {
1188 0 : if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
1189 0 : check_bytes_and_report(s, slab, p, "Alignment padding",
1190 : endobject, POISON_INUSE,
1191 : s->inuse - s->object_size);
1192 : }
1193 : }
1194 :
1195 0 : if (s->flags & SLAB_POISON) {
1196 0 : if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
1197 0 : (!check_bytes_and_report(s, slab, p, "Poison", p,
1198 0 : POISON_FREE, s->object_size - 1) ||
1199 0 : !check_bytes_and_report(s, slab, p, "End Poison",
1200 0 : p + s->object_size - 1, POISON_END, 1)))
1201 : return 0;
1202 : /*
1203 : * check_pad_bytes cleans up on its own.
1204 : */
1205 0 : check_pad_bytes(s, slab, p);
1206 : }
1207 :
1208 0 : if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE)
1209 : /*
1210 : * Object and freepointer overlap. Cannot check
1211 : * freepointer while object is allocated.
1212 : */
1213 : return 1;
1214 :
1215 : /* Check free pointer validity */
1216 0 : if (!check_valid_pointer(s, slab, get_freepointer(s, p))) {
1217 0 : object_err(s, slab, p, "Freepointer corrupt");
1218 : /*
1219 : * No choice but to zap it and thus lose the remainder
1220 : * of the free objects in this slab. May cause
1221 : * another error because the object count is now wrong.
1222 : */
1223 0 : set_freepointer(s, p, NULL);
1224 0 : return 0;
1225 : }
1226 : return 1;
1227 : }
1228 :
1229 0 : static int check_slab(struct kmem_cache *s, struct slab *slab)
1230 : {
1231 : int maxobj;
1232 :
1233 0 : if (!folio_test_slab(slab_folio(slab))) {
1234 0 : slab_err(s, slab, "Not a valid slab page");
1235 0 : return 0;
1236 : }
1237 :
1238 0 : maxobj = order_objects(slab_order(slab), s->size);
1239 0 : if (slab->objects > maxobj) {
1240 0 : slab_err(s, slab, "objects %u > max %u",
1241 : slab->objects, maxobj);
1242 0 : return 0;
1243 : }
1244 0 : if (slab->inuse > slab->objects) {
1245 0 : slab_err(s, slab, "inuse %u > max %u",
1246 : slab->inuse, slab->objects);
1247 0 : return 0;
1248 : }
1249 : /* Slab_pad_check fixes things up after itself */
1250 0 : slab_pad_check(s, slab);
1251 0 : return 1;
1252 : }
1253 :
1254 : /*
1255 : * Determine if a certain object in a slab is on the freelist. Must hold the
1256 : * slab lock to guarantee that the chains are in a consistent state.
1257 : */
1258 0 : static int on_freelist(struct kmem_cache *s, struct slab *slab, void *search)
1259 : {
1260 0 : int nr = 0;
1261 : void *fp;
1262 0 : void *object = NULL;
1263 : int max_objects;
1264 :
1265 0 : fp = slab->freelist;
1266 0 : while (fp && nr <= slab->objects) {
1267 0 : if (fp == search)
1268 : return 1;
1269 0 : if (!check_valid_pointer(s, slab, fp)) {
1270 0 : if (object) {
1271 0 : object_err(s, slab, object,
1272 : "Freechain corrupt");
1273 0 : set_freepointer(s, object, NULL);
1274 : } else {
1275 0 : slab_err(s, slab, "Freepointer corrupt");
1276 0 : slab->freelist = NULL;
1277 0 : slab->inuse = slab->objects;
1278 0 : slab_fix(s, "Freelist cleared");
1279 0 : return 0;
1280 : }
1281 : break;
1282 : }
1283 0 : object = fp;
1284 0 : fp = get_freepointer(s, object);
1285 0 : nr++;
1286 : }
1287 :
1288 0 : max_objects = order_objects(slab_order(slab), s->size);
1289 0 : if (max_objects > MAX_OBJS_PER_PAGE)
1290 0 : max_objects = MAX_OBJS_PER_PAGE;
1291 :
1292 0 : if (slab->objects != max_objects) {
1293 0 : slab_err(s, slab, "Wrong number of objects. Found %d but should be %d",
1294 : slab->objects, max_objects);
1295 0 : slab->objects = max_objects;
1296 0 : slab_fix(s, "Number of objects adjusted");
1297 : }
1298 0 : if (slab->inuse != slab->objects - nr) {
1299 0 : slab_err(s, slab, "Wrong object count. Counter is %d but counted were %d",
1300 : slab->inuse, slab->objects - nr);
1301 0 : slab->inuse = slab->objects - nr;
1302 0 : slab_fix(s, "Object count adjusted");
1303 : }
1304 0 : return search == NULL;
1305 : }
1306 :
1307 0 : static void trace(struct kmem_cache *s, struct slab *slab, void *object,
1308 : int alloc)
1309 : {
1310 0 : if (s->flags & SLAB_TRACE) {
1311 0 : pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
1312 : s->name,
1313 : alloc ? "alloc" : "free",
1314 : object, slab->inuse,
1315 : slab->freelist);
1316 :
1317 0 : if (!alloc)
1318 0 : print_section(KERN_INFO, "Object ", (void *)object,
1319 : s->object_size);
1320 :
1321 0 : dump_stack();
1322 : }
1323 0 : }
1324 :
1325 : /*
1326 : * Tracking of fully allocated slabs for debugging purposes.
1327 : */
1328 : static void add_full(struct kmem_cache *s,
1329 : struct kmem_cache_node *n, struct slab *slab)
1330 : {
1331 0 : if (!(s->flags & SLAB_STORE_USER))
1332 : return;
1333 :
1334 : lockdep_assert_held(&n->list_lock);
1335 0 : list_add(&slab->slab_list, &n->full);
1336 : }
1337 :
1338 : static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct slab *slab)
1339 : {
1340 1847 : if (!(s->flags & SLAB_STORE_USER))
1341 : return;
1342 :
1343 : lockdep_assert_held(&n->list_lock);
1344 0 : list_del(&slab->slab_list);
1345 : }
1346 :
1347 : /* Tracking of the number of slabs for debugging purposes */
1348 : static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1349 : {
1350 0 : struct kmem_cache_node *n = get_node(s, node);
1351 :
1352 0 : return atomic_long_read(&n->nr_slabs);
1353 : }
1354 :
1355 : static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1356 : {
1357 0 : return atomic_long_read(&n->nr_slabs);
1358 : }
1359 :
1360 : static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
1361 : {
1362 2195 : struct kmem_cache_node *n = get_node(s, node);
1363 :
1364 : /*
1365 : * May be called early in order to allocate a slab for the
1366 : * kmem_cache_node structure. Solve the chicken-egg
1367 : * dilemma by deferring the increment of the count during
1368 : * bootstrap (see early_kmem_cache_node_alloc).
1369 : */
1370 2195 : if (likely(n)) {
1371 4388 : atomic_long_inc(&n->nr_slabs);
1372 2194 : atomic_long_add(objects, &n->total_objects);
1373 : }
1374 : }
1375 : static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
1376 : {
1377 1753 : struct kmem_cache_node *n = get_node(s, node);
1378 :
1379 3506 : atomic_long_dec(&n->nr_slabs);
1380 3506 : atomic_long_sub(objects, &n->total_objects);
1381 : }
1382 :
1383 : /* Object debug checks for alloc/free paths */
1384 51365 : static void setup_object_debug(struct kmem_cache *s, void *object)
1385 : {
1386 102730 : if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))
1387 : return;
1388 :
1389 0 : init_object(s, object, SLUB_RED_INACTIVE);
1390 0 : init_tracking(s, object);
1391 : }
1392 :
1393 : static
1394 2194 : void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr)
1395 : {
1396 4388 : if (!kmem_cache_debug_flags(s, SLAB_POISON))
1397 : return;
1398 :
1399 0 : metadata_access_enable();
1400 0 : memset(kasan_reset_tag(addr), POISON_INUSE, slab_size(slab));
1401 : metadata_access_disable();
1402 : }
1403 :
1404 0 : static inline int alloc_consistency_checks(struct kmem_cache *s,
1405 : struct slab *slab, void *object)
1406 : {
1407 0 : if (!check_slab(s, slab))
1408 : return 0;
1409 :
1410 0 : if (!check_valid_pointer(s, slab, object)) {
1411 0 : object_err(s, slab, object, "Freelist Pointer check fails");
1412 0 : return 0;
1413 : }
1414 :
1415 0 : if (!check_object(s, slab, object, SLUB_RED_INACTIVE))
1416 : return 0;
1417 :
1418 0 : return 1;
1419 : }
1420 :
1421 0 : static noinline bool alloc_debug_processing(struct kmem_cache *s,
1422 : struct slab *slab, void *object, int orig_size)
1423 : {
1424 0 : if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1425 0 : if (!alloc_consistency_checks(s, slab, object))
1426 : goto bad;
1427 : }
1428 :
1429 : /* Success. Perform special debug activities for allocs */
1430 0 : trace(s, slab, object, 1);
1431 0 : set_orig_size(s, object, orig_size);
1432 0 : init_object(s, object, SLUB_RED_ACTIVE);
1433 0 : return true;
1434 :
1435 : bad:
1436 0 : if (folio_test_slab(slab_folio(slab))) {
1437 : /*
1438 : * If this is a slab page then lets do the best we can
1439 : * to avoid issues in the future. Marking all objects
1440 : * as used avoids touching the remaining objects.
1441 : */
1442 0 : slab_fix(s, "Marking all objects used");
1443 0 : slab->inuse = slab->objects;
1444 0 : slab->freelist = NULL;
1445 : }
1446 : return false;
1447 : }
1448 :
1449 0 : static inline int free_consistency_checks(struct kmem_cache *s,
1450 : struct slab *slab, void *object, unsigned long addr)
1451 : {
1452 0 : if (!check_valid_pointer(s, slab, object)) {
1453 0 : slab_err(s, slab, "Invalid object pointer 0x%p", object);
1454 : return 0;
1455 : }
1456 :
1457 0 : if (on_freelist(s, slab, object)) {
1458 0 : object_err(s, slab, object, "Object already free");
1459 : return 0;
1460 : }
1461 :
1462 0 : if (!check_object(s, slab, object, SLUB_RED_ACTIVE))
1463 : return 0;
1464 :
1465 0 : if (unlikely(s != slab->slab_cache)) {
1466 0 : if (!folio_test_slab(slab_folio(slab))) {
1467 0 : slab_err(s, slab, "Attempt to free object(0x%p) outside of slab",
1468 : object);
1469 0 : } else if (!slab->slab_cache) {
1470 0 : pr_err("SLUB <none>: no slab for object 0x%p.\n",
1471 : object);
1472 0 : dump_stack();
1473 : } else
1474 0 : object_err(s, slab, object,
1475 : "page slab pointer corrupt.");
1476 : return 0;
1477 : }
1478 : return 1;
1479 : }
1480 :
1481 : /*
1482 : * Parse a block of slub_debug options. Blocks are delimited by ';'
1483 : *
1484 : * @str: start of block
1485 : * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified
1486 : * @slabs: return start of list of slabs, or NULL when there's no list
1487 : * @init: assume this is initial parsing and not per-kmem-create parsing
1488 : *
1489 : * returns the start of next block if there's any, or NULL
1490 : */
1491 : static char *
1492 0 : parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init)
1493 : {
1494 0 : bool higher_order_disable = false;
1495 :
1496 : /* Skip any completely empty blocks */
1497 0 : while (*str && *str == ';')
1498 0 : str++;
1499 :
1500 0 : if (*str == ',') {
1501 : /*
1502 : * No options but restriction on slabs. This means full
1503 : * debugging for slabs matching a pattern.
1504 : */
1505 0 : *flags = DEBUG_DEFAULT_FLAGS;
1506 0 : goto check_slabs;
1507 : }
1508 0 : *flags = 0;
1509 :
1510 : /* Determine which debug features should be switched on */
1511 0 : for (; *str && *str != ',' && *str != ';'; str++) {
1512 0 : switch (tolower(*str)) {
1513 : case '-':
1514 0 : *flags = 0;
1515 0 : break;
1516 : case 'f':
1517 0 : *flags |= SLAB_CONSISTENCY_CHECKS;
1518 0 : break;
1519 : case 'z':
1520 0 : *flags |= SLAB_RED_ZONE;
1521 0 : break;
1522 : case 'p':
1523 0 : *flags |= SLAB_POISON;
1524 0 : break;
1525 : case 'u':
1526 0 : *flags |= SLAB_STORE_USER;
1527 0 : break;
1528 : case 't':
1529 0 : *flags |= SLAB_TRACE;
1530 0 : break;
1531 : case 'a':
1532 : *flags |= SLAB_FAILSLAB;
1533 0 : break;
1534 : case 'o':
1535 : /*
1536 : * Avoid enabling debugging on caches if its minimum
1537 : * order would increase as a result.
1538 : */
1539 : higher_order_disable = true;
1540 : break;
1541 : default:
1542 0 : if (init)
1543 0 : pr_err("slub_debug option '%c' unknown. skipped\n", *str);
1544 : }
1545 : }
1546 : check_slabs:
1547 0 : if (*str == ',')
1548 0 : *slabs = ++str;
1549 : else
1550 0 : *slabs = NULL;
1551 :
1552 : /* Skip over the slab list */
1553 0 : while (*str && *str != ';')
1554 0 : str++;
1555 :
1556 : /* Skip any completely empty blocks */
1557 0 : while (*str && *str == ';')
1558 0 : str++;
1559 :
1560 0 : if (init && higher_order_disable)
1561 0 : disable_higher_order_debug = 1;
1562 :
1563 0 : if (*str)
1564 : return str;
1565 : else
1566 0 : return NULL;
1567 : }
1568 :
1569 0 : static int __init setup_slub_debug(char *str)
1570 : {
1571 : slab_flags_t flags;
1572 : slab_flags_t global_flags;
1573 : char *saved_str;
1574 : char *slab_list;
1575 0 : bool global_slub_debug_changed = false;
1576 0 : bool slab_list_specified = false;
1577 :
1578 0 : global_flags = DEBUG_DEFAULT_FLAGS;
1579 0 : if (*str++ != '=' || !*str)
1580 : /*
1581 : * No options specified. Switch on full debugging.
1582 : */
1583 : goto out;
1584 :
1585 : saved_str = str;
1586 0 : while (str) {
1587 0 : str = parse_slub_debug_flags(str, &flags, &slab_list, true);
1588 :
1589 0 : if (!slab_list) {
1590 0 : global_flags = flags;
1591 0 : global_slub_debug_changed = true;
1592 : } else {
1593 0 : slab_list_specified = true;
1594 0 : if (flags & SLAB_STORE_USER)
1595 0 : stack_depot_request_early_init();
1596 : }
1597 : }
1598 :
1599 : /*
1600 : * For backwards compatibility, a single list of flags with list of
1601 : * slabs means debugging is only changed for those slabs, so the global
1602 : * slub_debug should be unchanged (0 or DEBUG_DEFAULT_FLAGS, depending
1603 : * on CONFIG_SLUB_DEBUG_ON). We can extended that to multiple lists as
1604 : * long as there is no option specifying flags without a slab list.
1605 : */
1606 0 : if (slab_list_specified) {
1607 0 : if (!global_slub_debug_changed)
1608 0 : global_flags = slub_debug;
1609 0 : slub_debug_string = saved_str;
1610 : }
1611 : out:
1612 0 : slub_debug = global_flags;
1613 0 : if (slub_debug & SLAB_STORE_USER)
1614 0 : stack_depot_request_early_init();
1615 0 : if (slub_debug != 0 || slub_debug_string)
1616 0 : static_branch_enable(&slub_debug_enabled);
1617 : else
1618 0 : static_branch_disable(&slub_debug_enabled);
1619 0 : if ((static_branch_unlikely(&init_on_alloc) ||
1620 0 : static_branch_unlikely(&init_on_free)) &&
1621 0 : (slub_debug & SLAB_POISON))
1622 0 : pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n");
1623 0 : return 1;
1624 : }
1625 :
1626 : __setup("slub_debug", setup_slub_debug);
1627 :
1628 : /*
1629 : * kmem_cache_flags - apply debugging options to the cache
1630 : * @object_size: the size of an object without meta data
1631 : * @flags: flags to set
1632 : * @name: name of the cache
1633 : *
1634 : * Debug option(s) are applied to @flags. In addition to the debug
1635 : * option(s), if a slab name (or multiple) is specified i.e.
1636 : * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ...
1637 : * then only the select slabs will receive the debug option(s).
1638 : */
1639 103 : slab_flags_t kmem_cache_flags(unsigned int object_size,
1640 : slab_flags_t flags, const char *name)
1641 : {
1642 : char *iter;
1643 : size_t len;
1644 : char *next_block;
1645 : slab_flags_t block_flags;
1646 103 : slab_flags_t slub_debug_local = slub_debug;
1647 :
1648 103 : if (flags & SLAB_NO_USER_FLAGS)
1649 : return flags;
1650 :
1651 : /*
1652 : * If the slab cache is for debugging (e.g. kmemleak) then
1653 : * don't store user (stack trace) information by default,
1654 : * but let the user enable it via the command line below.
1655 : */
1656 103 : if (flags & SLAB_NOLEAKTRACE)
1657 0 : slub_debug_local &= ~SLAB_STORE_USER;
1658 :
1659 103 : len = strlen(name);
1660 103 : next_block = slub_debug_string;
1661 : /* Go through all blocks of debug options, see if any matches our slab's name */
1662 206 : while (next_block) {
1663 0 : next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false);
1664 0 : if (!iter)
1665 0 : continue;
1666 : /* Found a block that has a slab list, search it */
1667 0 : while (*iter) {
1668 : char *end, *glob;
1669 : size_t cmplen;
1670 :
1671 0 : end = strchrnul(iter, ',');
1672 0 : if (next_block && next_block < end)
1673 0 : end = next_block - 1;
1674 :
1675 0 : glob = strnchr(iter, end - iter, '*');
1676 0 : if (glob)
1677 0 : cmplen = glob - iter;
1678 : else
1679 0 : cmplen = max_t(size_t, len, (end - iter));
1680 :
1681 0 : if (!strncmp(name, iter, cmplen)) {
1682 0 : flags |= block_flags;
1683 0 : return flags;
1684 : }
1685 :
1686 0 : if (!*end || *end == ';')
1687 : break;
1688 0 : iter = end + 1;
1689 : }
1690 : }
1691 :
1692 103 : return flags | slub_debug_local;
1693 : }
1694 : #else /* !CONFIG_SLUB_DEBUG */
1695 : static inline void setup_object_debug(struct kmem_cache *s, void *object) {}
1696 : static inline
1697 : void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr) {}
1698 :
1699 : static inline bool alloc_debug_processing(struct kmem_cache *s,
1700 : struct slab *slab, void *object, int orig_size) { return true; }
1701 :
1702 : static inline bool free_debug_processing(struct kmem_cache *s,
1703 : struct slab *slab, void *head, void *tail, int *bulk_cnt,
1704 : unsigned long addr, depot_stack_handle_t handle) { return true; }
1705 :
1706 : static inline void slab_pad_check(struct kmem_cache *s, struct slab *slab) {}
1707 : static inline int check_object(struct kmem_cache *s, struct slab *slab,
1708 : void *object, u8 val) { return 1; }
1709 : static inline depot_stack_handle_t set_track_prepare(void) { return 0; }
1710 : static inline void set_track(struct kmem_cache *s, void *object,
1711 : enum track_item alloc, unsigned long addr) {}
1712 : static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1713 : struct slab *slab) {}
1714 : static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1715 : struct slab *slab) {}
1716 : slab_flags_t kmem_cache_flags(unsigned int object_size,
1717 : slab_flags_t flags, const char *name)
1718 : {
1719 : return flags;
1720 : }
1721 : #define slub_debug 0
1722 :
1723 : #define disable_higher_order_debug 0
1724 :
1725 : static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1726 : { return 0; }
1727 : static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1728 : { return 0; }
1729 : static inline void inc_slabs_node(struct kmem_cache *s, int node,
1730 : int objects) {}
1731 : static inline void dec_slabs_node(struct kmem_cache *s, int node,
1732 : int objects) {}
1733 :
1734 : #ifndef CONFIG_SLUB_TINY
1735 : static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab,
1736 : void **freelist, void *nextfree)
1737 : {
1738 : return false;
1739 : }
1740 : #endif
1741 : #endif /* CONFIG_SLUB_DEBUG */
1742 :
1743 : /*
1744 : * Hooks for other subsystems that check memory allocations. In a typical
1745 : * production configuration these hooks all should produce no code at all.
1746 : */
1747 : static __always_inline bool slab_free_hook(struct kmem_cache *s,
1748 : void *x, bool init)
1749 : {
1750 49059 : kmemleak_free_recursive(x, s->flags);
1751 49059 : kmsan_slab_free(s, x);
1752 :
1753 49059 : debug_check_no_locks_freed(x, s->object_size);
1754 :
1755 : if (!(s->flags & SLAB_DEBUG_OBJECTS))
1756 49059 : debug_check_no_obj_freed(x, s->object_size);
1757 :
1758 : /* Use KCSAN to help debug racy use-after-free. */
1759 : if (!(s->flags & SLAB_TYPESAFE_BY_RCU))
1760 : __kcsan_check_access(x, s->object_size,
1761 : KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT);
1762 :
1763 : /*
1764 : * As memory initialization might be integrated into KASAN,
1765 : * kasan_slab_free and initialization memset's must be
1766 : * kept together to avoid discrepancies in behavior.
1767 : *
1768 : * The initialization memset's clear the object and the metadata,
1769 : * but don't touch the SLAB redzone.
1770 : */
1771 49059 : if (init) {
1772 : int rsize;
1773 :
1774 : if (!kasan_has_integrated_init())
1775 0 : memset(kasan_reset_tag(x), 0, s->object_size);
1776 0 : rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0;
1777 0 : memset((char *)kasan_reset_tag(x) + s->inuse, 0,
1778 0 : s->size - s->inuse - rsize);
1779 : }
1780 : /* KASAN might put x into memory quarantine, delaying its reuse. */
1781 49059 : return kasan_slab_free(s, x, init);
1782 : }
1783 :
1784 49059 : static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1785 : void **head, void **tail,
1786 : int *cnt)
1787 : {
1788 :
1789 : void *object;
1790 49059 : void *next = *head;
1791 49059 : void *old_tail = *tail ? *tail : *head;
1792 :
1793 49059 : if (is_kfence_address(next)) {
1794 : slab_free_hook(s, next, false);
1795 : return true;
1796 : }
1797 :
1798 : /* Head and tail of the reconstructed freelist */
1799 49059 : *head = NULL;
1800 49059 : *tail = NULL;
1801 :
1802 : do {
1803 49059 : object = next;
1804 98118 : next = get_freepointer(s, object);
1805 :
1806 : /* If object's reuse doesn't have to be delayed */
1807 147177 : if (!slab_free_hook(s, object, slab_want_init_on_free(s))) {
1808 : /* Move object to the new freelist */
1809 98118 : set_freepointer(s, object, *head);
1810 49059 : *head = object;
1811 49059 : if (!*tail)
1812 49059 : *tail = object;
1813 : } else {
1814 : /*
1815 : * Adjust the reconstructed freelist depth
1816 : * accordingly if object's reuse is delayed.
1817 : */
1818 : --(*cnt);
1819 : }
1820 49059 : } while (object != old_tail);
1821 :
1822 49059 : if (*head == *tail)
1823 49059 : *tail = NULL;
1824 :
1825 49059 : return *head != NULL;
1826 : }
1827 :
1828 : static void *setup_object(struct kmem_cache *s, void *object)
1829 : {
1830 51365 : setup_object_debug(s, object);
1831 51365 : object = kasan_init_slab_obj(s, object);
1832 51365 : if (unlikely(s->ctor)) {
1833 260 : kasan_unpoison_object_data(s, object);
1834 260 : s->ctor(object);
1835 260 : kasan_poison_object_data(s, object);
1836 : }
1837 : return object;
1838 : }
1839 :
1840 : /*
1841 : * Slab allocation and freeing
1842 : */
1843 2194 : static inline struct slab *alloc_slab_page(gfp_t flags, int node,
1844 : struct kmem_cache_order_objects oo)
1845 : {
1846 : struct folio *folio;
1847 : struct slab *slab;
1848 2194 : unsigned int order = oo_order(oo);
1849 :
1850 2194 : if (node == NUMA_NO_NODE)
1851 2193 : folio = (struct folio *)alloc_pages(flags, order);
1852 : else
1853 1 : folio = (struct folio *)__alloc_pages_node(node, flags, order);
1854 :
1855 2194 : if (!folio)
1856 : return NULL;
1857 :
1858 2194 : slab = folio_slab(folio);
1859 2194 : __folio_set_slab(folio);
1860 : /* Make the flag visible before any changes to folio->mapping */
1861 2194 : smp_wmb();
1862 4388 : if (folio_is_pfmemalloc(folio))
1863 : slab_set_pfmemalloc(slab);
1864 :
1865 : return slab;
1866 : }
1867 :
1868 : #ifdef CONFIG_SLAB_FREELIST_RANDOM
1869 : /* Pre-initialize the random sequence cache */
1870 : static int init_cache_random_seq(struct kmem_cache *s)
1871 : {
1872 : unsigned int count = oo_objects(s->oo);
1873 : int err;
1874 :
1875 : /* Bailout if already initialised */
1876 : if (s->random_seq)
1877 : return 0;
1878 :
1879 : err = cache_random_seq_create(s, count, GFP_KERNEL);
1880 : if (err) {
1881 : pr_err("SLUB: Unable to initialize free list for %s\n",
1882 : s->name);
1883 : return err;
1884 : }
1885 :
1886 : /* Transform to an offset on the set of pages */
1887 : if (s->random_seq) {
1888 : unsigned int i;
1889 :
1890 : for (i = 0; i < count; i++)
1891 : s->random_seq[i] *= s->size;
1892 : }
1893 : return 0;
1894 : }
1895 :
1896 : /* Initialize each random sequence freelist per cache */
1897 : static void __init init_freelist_randomization(void)
1898 : {
1899 : struct kmem_cache *s;
1900 :
1901 : mutex_lock(&slab_mutex);
1902 :
1903 : list_for_each_entry(s, &slab_caches, list)
1904 : init_cache_random_seq(s);
1905 :
1906 : mutex_unlock(&slab_mutex);
1907 : }
1908 :
1909 : /* Get the next entry on the pre-computed freelist randomized */
1910 : static void *next_freelist_entry(struct kmem_cache *s, struct slab *slab,
1911 : unsigned long *pos, void *start,
1912 : unsigned long page_limit,
1913 : unsigned long freelist_count)
1914 : {
1915 : unsigned int idx;
1916 :
1917 : /*
1918 : * If the target page allocation failed, the number of objects on the
1919 : * page might be smaller than the usual size defined by the cache.
1920 : */
1921 : do {
1922 : idx = s->random_seq[*pos];
1923 : *pos += 1;
1924 : if (*pos >= freelist_count)
1925 : *pos = 0;
1926 : } while (unlikely(idx >= page_limit));
1927 :
1928 : return (char *)start + idx;
1929 : }
1930 :
1931 : /* Shuffle the single linked freelist based on a random pre-computed sequence */
1932 : static bool shuffle_freelist(struct kmem_cache *s, struct slab *slab)
1933 : {
1934 : void *start;
1935 : void *cur;
1936 : void *next;
1937 : unsigned long idx, pos, page_limit, freelist_count;
1938 :
1939 : if (slab->objects < 2 || !s->random_seq)
1940 : return false;
1941 :
1942 : freelist_count = oo_objects(s->oo);
1943 : pos = get_random_u32_below(freelist_count);
1944 :
1945 : page_limit = slab->objects * s->size;
1946 : start = fixup_red_left(s, slab_address(slab));
1947 :
1948 : /* First entry is used as the base of the freelist */
1949 : cur = next_freelist_entry(s, slab, &pos, start, page_limit,
1950 : freelist_count);
1951 : cur = setup_object(s, cur);
1952 : slab->freelist = cur;
1953 :
1954 : for (idx = 1; idx < slab->objects; idx++) {
1955 : next = next_freelist_entry(s, slab, &pos, start, page_limit,
1956 : freelist_count);
1957 : next = setup_object(s, next);
1958 : set_freepointer(s, cur, next);
1959 : cur = next;
1960 : }
1961 : set_freepointer(s, cur, NULL);
1962 :
1963 : return true;
1964 : }
1965 : #else
1966 : static inline int init_cache_random_seq(struct kmem_cache *s)
1967 : {
1968 : return 0;
1969 : }
1970 : static inline void init_freelist_randomization(void) { }
1971 : static inline bool shuffle_freelist(struct kmem_cache *s, struct slab *slab)
1972 : {
1973 : return false;
1974 : }
1975 : #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1976 :
1977 2194 : static struct slab *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1978 : {
1979 : struct slab *slab;
1980 2194 : struct kmem_cache_order_objects oo = s->oo;
1981 : gfp_t alloc_gfp;
1982 : void *start, *p, *next;
1983 : int idx;
1984 : bool shuffle;
1985 :
1986 2194 : flags &= gfp_allowed_mask;
1987 :
1988 2194 : flags |= s->allocflags;
1989 :
1990 : /*
1991 : * Let the initial higher-order allocation fail under memory pressure
1992 : * so we fall-back to the minimum order allocation.
1993 : */
1994 2194 : alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1995 4349 : if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
1996 74 : alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_RECLAIM;
1997 :
1998 2194 : slab = alloc_slab_page(alloc_gfp, node, oo);
1999 2194 : if (unlikely(!slab)) {
2000 0 : oo = s->min;
2001 0 : alloc_gfp = flags;
2002 : /*
2003 : * Allocation may have failed due to fragmentation.
2004 : * Try a lower order alloc if possible
2005 : */
2006 0 : slab = alloc_slab_page(alloc_gfp, node, oo);
2007 0 : if (unlikely(!slab))
2008 : return NULL;
2009 : stat(s, ORDER_FALLBACK);
2010 : }
2011 :
2012 2194 : slab->objects = oo_objects(oo);
2013 2194 : slab->inuse = 0;
2014 2194 : slab->frozen = 0;
2015 :
2016 4388 : account_slab(slab, oo_order(oo), s, flags);
2017 :
2018 2194 : slab->slab_cache = s;
2019 :
2020 2194 : kasan_poison_slab(slab);
2021 :
2022 2194 : start = slab_address(slab);
2023 :
2024 2194 : setup_slab_debug(s, slab, start);
2025 :
2026 2194 : shuffle = shuffle_freelist(s, slab);
2027 :
2028 : if (!shuffle) {
2029 2194 : start = fixup_red_left(s, start);
2030 2194 : start = setup_object(s, start);
2031 2194 : slab->freelist = start;
2032 51365 : for (idx = 0, p = start; idx < slab->objects - 1; idx++) {
2033 49171 : next = p + s->size;
2034 49171 : next = setup_object(s, next);
2035 98342 : set_freepointer(s, p, next);
2036 49171 : p = next;
2037 : }
2038 2194 : set_freepointer(s, p, NULL);
2039 : }
2040 :
2041 2194 : return slab;
2042 : }
2043 :
2044 2194 : static struct slab *new_slab(struct kmem_cache *s, gfp_t flags, int node)
2045 : {
2046 2194 : if (unlikely(flags & GFP_SLAB_BUG_MASK))
2047 0 : flags = kmalloc_fix_flags(flags);
2048 :
2049 2194 : WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2050 :
2051 2194 : return allocate_slab(s,
2052 : flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
2053 : }
2054 :
2055 1753 : static void __free_slab(struct kmem_cache *s, struct slab *slab)
2056 : {
2057 1753 : struct folio *folio = slab_folio(slab);
2058 1753 : int order = folio_order(folio);
2059 1753 : int pages = 1 << order;
2060 :
2061 1753 : __slab_clear_pfmemalloc(slab);
2062 1753 : folio->mapping = NULL;
2063 : /* Make the mapping reset visible before clearing the flag */
2064 1753 : smp_wmb();
2065 1753 : __folio_clear_slab(folio);
2066 3506 : mm_account_reclaimed_pages(pages);
2067 1753 : unaccount_slab(slab, order, s);
2068 1753 : __free_pages(&folio->page, order);
2069 1753 : }
2070 :
2071 0 : static void rcu_free_slab(struct rcu_head *h)
2072 : {
2073 0 : struct slab *slab = container_of(h, struct slab, rcu_head);
2074 :
2075 0 : __free_slab(slab->slab_cache, slab);
2076 0 : }
2077 :
2078 1753 : static void free_slab(struct kmem_cache *s, struct slab *slab)
2079 : {
2080 3506 : if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) {
2081 : void *p;
2082 :
2083 0 : slab_pad_check(s, slab);
2084 0 : for_each_object(p, s, slab_address(slab), slab->objects)
2085 0 : check_object(s, slab, p, SLUB_RED_INACTIVE);
2086 : }
2087 :
2088 1753 : if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU))
2089 0 : call_rcu(&slab->rcu_head, rcu_free_slab);
2090 : else
2091 1753 : __free_slab(s, slab);
2092 1753 : }
2093 :
2094 : static void discard_slab(struct kmem_cache *s, struct slab *slab)
2095 : {
2096 5259 : dec_slabs_node(s, slab_nid(slab), slab->objects);
2097 1753 : free_slab(s, slab);
2098 : }
2099 :
2100 : /*
2101 : * Management of partially allocated slabs.
2102 : */
2103 : static inline void
2104 : __add_partial(struct kmem_cache_node *n, struct slab *slab, int tail)
2105 : {
2106 1850 : n->nr_partial++;
2107 2 : if (tail == DEACTIVATE_TO_TAIL)
2108 1847 : list_add_tail(&slab->slab_list, &n->partial);
2109 : else
2110 3 : list_add(&slab->slab_list, &n->partial);
2111 : }
2112 :
2113 : static inline void add_partial(struct kmem_cache_node *n,
2114 : struct slab *slab, int tail)
2115 : {
2116 : lockdep_assert_held(&n->list_lock);
2117 2 : __add_partial(n, slab, tail);
2118 : }
2119 :
2120 : static inline void remove_partial(struct kmem_cache_node *n,
2121 : struct slab *slab)
2122 : {
2123 : lockdep_assert_held(&n->list_lock);
2124 3670 : list_del(&slab->slab_list);
2125 1835 : n->nr_partial--;
2126 : }
2127 :
2128 : /*
2129 : * Called only for kmem_cache_debug() caches instead of acquire_slab(), with a
2130 : * slab from the n->partial list. Remove only a single object from the slab, do
2131 : * the alloc_debug_processing() checks and leave the slab on the list, or move
2132 : * it to full list if it was the last free object.
2133 : */
2134 0 : static void *alloc_single_from_partial(struct kmem_cache *s,
2135 : struct kmem_cache_node *n, struct slab *slab, int orig_size)
2136 : {
2137 : void *object;
2138 :
2139 : lockdep_assert_held(&n->list_lock);
2140 :
2141 0 : object = slab->freelist;
2142 0 : slab->freelist = get_freepointer(s, object);
2143 0 : slab->inuse++;
2144 :
2145 0 : if (!alloc_debug_processing(s, slab, object, orig_size)) {
2146 0 : remove_partial(n, slab);
2147 0 : return NULL;
2148 : }
2149 :
2150 0 : if (slab->inuse == slab->objects) {
2151 0 : remove_partial(n, slab);
2152 0 : add_full(s, n, slab);
2153 : }
2154 :
2155 : return object;
2156 : }
2157 :
2158 : /*
2159 : * Called only for kmem_cache_debug() caches to allocate from a freshly
2160 : * allocated slab. Allocate a single object instead of whole freelist
2161 : * and put the slab to the partial (or full) list.
2162 : */
2163 0 : static void *alloc_single_from_new_slab(struct kmem_cache *s,
2164 : struct slab *slab, int orig_size)
2165 : {
2166 0 : int nid = slab_nid(slab);
2167 0 : struct kmem_cache_node *n = get_node(s, nid);
2168 : unsigned long flags;
2169 : void *object;
2170 :
2171 :
2172 0 : object = slab->freelist;
2173 0 : slab->freelist = get_freepointer(s, object);
2174 0 : slab->inuse = 1;
2175 :
2176 0 : if (!alloc_debug_processing(s, slab, object, orig_size))
2177 : /*
2178 : * It's not really expected that this would fail on a
2179 : * freshly allocated slab, but a concurrent memory
2180 : * corruption in theory could cause that.
2181 : */
2182 : return NULL;
2183 :
2184 0 : spin_lock_irqsave(&n->list_lock, flags);
2185 :
2186 0 : if (slab->inuse == slab->objects)
2187 0 : add_full(s, n, slab);
2188 : else
2189 : add_partial(n, slab, DEACTIVATE_TO_HEAD);
2190 :
2191 0 : inc_slabs_node(s, nid, slab->objects);
2192 0 : spin_unlock_irqrestore(&n->list_lock, flags);
2193 :
2194 0 : return object;
2195 : }
2196 :
2197 : /*
2198 : * Remove slab from the partial list, freeze it and
2199 : * return the pointer to the freelist.
2200 : *
2201 : * Returns a list of objects or NULL if it fails.
2202 : */
2203 82 : static inline void *acquire_slab(struct kmem_cache *s,
2204 : struct kmem_cache_node *n, struct slab *slab,
2205 : int mode)
2206 : {
2207 : void *freelist;
2208 : unsigned long counters;
2209 : struct slab new;
2210 :
2211 : lockdep_assert_held(&n->list_lock);
2212 :
2213 : /*
2214 : * Zap the freelist and set the frozen bit.
2215 : * The old freelist is the list of objects for the
2216 : * per cpu allocation list.
2217 : */
2218 82 : freelist = slab->freelist;
2219 82 : counters = slab->counters;
2220 82 : new.counters = counters;
2221 82 : if (mode) {
2222 82 : new.inuse = slab->objects;
2223 82 : new.freelist = NULL;
2224 : } else {
2225 : new.freelist = freelist;
2226 : }
2227 :
2228 : VM_BUG_ON(new.frozen);
2229 82 : new.frozen = 1;
2230 :
2231 164 : if (!__cmpxchg_double_slab(s, slab,
2232 : freelist, counters,
2233 : new.freelist, new.counters,
2234 : "acquire_slab"))
2235 : return NULL;
2236 :
2237 164 : remove_partial(n, slab);
2238 82 : WARN_ON(!freelist);
2239 : return freelist;
2240 : }
2241 :
2242 : #ifdef CONFIG_SLUB_CPU_PARTIAL
2243 : static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain);
2244 : #else
2245 : static inline void put_cpu_partial(struct kmem_cache *s, struct slab *slab,
2246 : int drain) { }
2247 : #endif
2248 : static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags);
2249 :
2250 : /*
2251 : * Try to allocate a partial slab from a specific node.
2252 : */
2253 2275 : static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
2254 : struct partial_context *pc)
2255 : {
2256 : struct slab *slab, *slab2;
2257 2275 : void *object = NULL;
2258 : unsigned long flags;
2259 2275 : unsigned int partial_slabs = 0;
2260 :
2261 : /*
2262 : * Racy check. If we mistakenly see no partial slabs then we
2263 : * just allocate an empty slab. If we mistakenly try to get a
2264 : * partial slab and there is none available then get_partial()
2265 : * will return NULL.
2266 : */
2267 2275 : if (!n || !n->nr_partial)
2268 : return NULL;
2269 :
2270 82 : spin_lock_irqsave(&n->list_lock, flags);
2271 82 : list_for_each_entry_safe(slab, slab2, &n->partial, slab_list) {
2272 : void *t;
2273 :
2274 164 : if (!pfmemalloc_match(slab, pc->flags))
2275 0 : continue;
2276 :
2277 82 : if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) {
2278 0 : object = alloc_single_from_partial(s, n, slab,
2279 0 : pc->orig_size);
2280 0 : if (object)
2281 : break;
2282 0 : continue;
2283 : }
2284 :
2285 82 : t = acquire_slab(s, n, slab, object == NULL);
2286 82 : if (!t)
2287 : break;
2288 :
2289 82 : if (!object) {
2290 82 : *pc->slab = slab;
2291 82 : stat(s, ALLOC_FROM_PARTIAL);
2292 82 : object = t;
2293 : } else {
2294 : put_cpu_partial(s, slab, 0);
2295 : stat(s, CPU_PARTIAL_NODE);
2296 : partial_slabs++;
2297 : }
2298 : #ifdef CONFIG_SLUB_CPU_PARTIAL
2299 : if (!kmem_cache_has_cpu_partial(s)
2300 : || partial_slabs > s->cpu_partial_slabs / 2)
2301 : break;
2302 : #else
2303 : break;
2304 : #endif
2305 :
2306 : }
2307 164 : spin_unlock_irqrestore(&n->list_lock, flags);
2308 82 : return object;
2309 : }
2310 :
2311 : /*
2312 : * Get a slab from somewhere. Search in increasing NUMA distances.
2313 : */
2314 : static void *get_any_partial(struct kmem_cache *s, struct partial_context *pc)
2315 : {
2316 : #ifdef CONFIG_NUMA
2317 : struct zonelist *zonelist;
2318 : struct zoneref *z;
2319 : struct zone *zone;
2320 : enum zone_type highest_zoneidx = gfp_zone(pc->flags);
2321 : void *object;
2322 : unsigned int cpuset_mems_cookie;
2323 :
2324 : /*
2325 : * The defrag ratio allows a configuration of the tradeoffs between
2326 : * inter node defragmentation and node local allocations. A lower
2327 : * defrag_ratio increases the tendency to do local allocations
2328 : * instead of attempting to obtain partial slabs from other nodes.
2329 : *
2330 : * If the defrag_ratio is set to 0 then kmalloc() always
2331 : * returns node local objects. If the ratio is higher then kmalloc()
2332 : * may return off node objects because partial slabs are obtained
2333 : * from other nodes and filled up.
2334 : *
2335 : * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
2336 : * (which makes defrag_ratio = 1000) then every (well almost)
2337 : * allocation will first attempt to defrag slab caches on other nodes.
2338 : * This means scanning over all nodes to look for partial slabs which
2339 : * may be expensive if we do it every time we are trying to find a slab
2340 : * with available objects.
2341 : */
2342 : if (!s->remote_node_defrag_ratio ||
2343 : get_cycles() % 1024 > s->remote_node_defrag_ratio)
2344 : return NULL;
2345 :
2346 : do {
2347 : cpuset_mems_cookie = read_mems_allowed_begin();
2348 : zonelist = node_zonelist(mempolicy_slab_node(), pc->flags);
2349 : for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) {
2350 : struct kmem_cache_node *n;
2351 :
2352 : n = get_node(s, zone_to_nid(zone));
2353 :
2354 : if (n && cpuset_zone_allowed(zone, pc->flags) &&
2355 : n->nr_partial > s->min_partial) {
2356 : object = get_partial_node(s, n, pc);
2357 : if (object) {
2358 : /*
2359 : * Don't check read_mems_allowed_retry()
2360 : * here - if mems_allowed was updated in
2361 : * parallel, that was a harmless race
2362 : * between allocation and the cpuset
2363 : * update
2364 : */
2365 : return object;
2366 : }
2367 : }
2368 : }
2369 : } while (read_mems_allowed_retry(cpuset_mems_cookie));
2370 : #endif /* CONFIG_NUMA */
2371 : return NULL;
2372 : }
2373 :
2374 : /*
2375 : * Get a partial slab, lock it and return it.
2376 : */
2377 2275 : static void *get_partial(struct kmem_cache *s, int node, struct partial_context *pc)
2378 : {
2379 : void *object;
2380 2275 : int searchnode = node;
2381 :
2382 2275 : if (node == NUMA_NO_NODE)
2383 2273 : searchnode = numa_mem_id();
2384 :
2385 2275 : object = get_partial_node(s, get_node(s, searchnode), pc);
2386 2275 : if (object || node != NUMA_NO_NODE)
2387 : return object;
2388 :
2389 2193 : return get_any_partial(s, pc);
2390 : }
2391 :
2392 : #ifndef CONFIG_SLUB_TINY
2393 :
2394 : #ifdef CONFIG_PREEMPTION
2395 : /*
2396 : * Calculate the next globally unique transaction for disambiguation
2397 : * during cmpxchg. The transactions start with the cpu number and are then
2398 : * incremented by CONFIG_NR_CPUS.
2399 : */
2400 : #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
2401 : #else
2402 : /*
2403 : * No preemption supported therefore also no need to check for
2404 : * different cpus.
2405 : */
2406 : #define TID_STEP 1
2407 : #endif /* CONFIG_PREEMPTION */
2408 :
2409 : static inline unsigned long next_tid(unsigned long tid)
2410 : {
2411 73118 : return tid + TID_STEP;
2412 : }
2413 :
2414 : #ifdef SLUB_DEBUG_CMPXCHG
2415 : static inline unsigned int tid_to_cpu(unsigned long tid)
2416 : {
2417 : return tid % TID_STEP;
2418 : }
2419 :
2420 : static inline unsigned long tid_to_event(unsigned long tid)
2421 : {
2422 : return tid / TID_STEP;
2423 : }
2424 : #endif
2425 :
2426 : static inline unsigned int init_tid(int cpu)
2427 : {
2428 53 : return cpu;
2429 : }
2430 :
2431 : static inline void note_cmpxchg_failure(const char *n,
2432 : const struct kmem_cache *s, unsigned long tid)
2433 : {
2434 : #ifdef SLUB_DEBUG_CMPXCHG
2435 : unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
2436 :
2437 : pr_info("%s %s: cmpxchg redo ", n, s->name);
2438 :
2439 : #ifdef CONFIG_PREEMPTION
2440 : if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
2441 : pr_warn("due to cpu change %d -> %d\n",
2442 : tid_to_cpu(tid), tid_to_cpu(actual_tid));
2443 : else
2444 : #endif
2445 : if (tid_to_event(tid) != tid_to_event(actual_tid))
2446 : pr_warn("due to cpu running other code. Event %ld->%ld\n",
2447 : tid_to_event(tid), tid_to_event(actual_tid));
2448 : else
2449 : pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
2450 : actual_tid, tid, next_tid(tid));
2451 : #endif
2452 : stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
2453 : }
2454 :
2455 : static void init_kmem_cache_cpus(struct kmem_cache *s)
2456 : {
2457 : int cpu;
2458 : struct kmem_cache_cpu *c;
2459 :
2460 53 : for_each_possible_cpu(cpu) {
2461 53 : c = per_cpu_ptr(s->cpu_slab, cpu);
2462 53 : local_lock_init(&c->lock);
2463 53 : c->tid = init_tid(cpu);
2464 : }
2465 : }
2466 :
2467 : /*
2468 : * Finishes removing the cpu slab. Merges cpu's freelist with slab's freelist,
2469 : * unfreezes the slabs and puts it on the proper list.
2470 : * Assumes the slab has been already safely taken away from kmem_cache_cpu
2471 : * by the caller.
2472 : */
2473 2 : static void deactivate_slab(struct kmem_cache *s, struct slab *slab,
2474 : void *freelist)
2475 : {
2476 : enum slab_modes { M_NONE, M_PARTIAL, M_FREE, M_FULL_NOLIST };
2477 6 : struct kmem_cache_node *n = get_node(s, slab_nid(slab));
2478 2 : int free_delta = 0;
2479 2 : enum slab_modes mode = M_NONE;
2480 : void *nextfree, *freelist_iter, *freelist_tail;
2481 2 : int tail = DEACTIVATE_TO_HEAD;
2482 2 : unsigned long flags = 0;
2483 : struct slab new;
2484 : struct slab old;
2485 :
2486 2 : if (slab->freelist) {
2487 0 : stat(s, DEACTIVATE_REMOTE_FREES);
2488 0 : tail = DEACTIVATE_TO_TAIL;
2489 : }
2490 :
2491 : /*
2492 : * Stage one: Count the objects on cpu's freelist as free_delta and
2493 : * remember the last object in freelist_tail for later splicing.
2494 : */
2495 2 : freelist_tail = NULL;
2496 2 : freelist_iter = freelist;
2497 86 : while (freelist_iter) {
2498 164 : nextfree = get_freepointer(s, freelist_iter);
2499 :
2500 : /*
2501 : * If 'nextfree' is invalid, it is possible that the object at
2502 : * 'freelist_iter' is already corrupted. So isolate all objects
2503 : * starting at 'freelist_iter' by skipping them.
2504 : */
2505 82 : if (freelist_corrupted(s, slab, &freelist_iter, nextfree))
2506 : break;
2507 :
2508 82 : freelist_tail = freelist_iter;
2509 82 : free_delta++;
2510 :
2511 82 : freelist_iter = nextfree;
2512 : }
2513 :
2514 : /*
2515 : * Stage two: Unfreeze the slab while splicing the per-cpu
2516 : * freelist to the head of slab's freelist.
2517 : *
2518 : * Ensure that the slab is unfrozen while the list presence
2519 : * reflects the actual number of objects during unfreeze.
2520 : *
2521 : * We first perform cmpxchg holding lock and insert to list
2522 : * when it succeed. If there is mismatch then the slab is not
2523 : * unfrozen and number of objects in the slab may have changed.
2524 : * Then release lock and retry cmpxchg again.
2525 : */
2526 : redo:
2527 :
2528 2 : old.freelist = READ_ONCE(slab->freelist);
2529 2 : old.counters = READ_ONCE(slab->counters);
2530 : VM_BUG_ON(!old.frozen);
2531 :
2532 : /* Determine target state of the slab */
2533 2 : new.counters = old.counters;
2534 2 : if (freelist_tail) {
2535 2 : new.inuse -= free_delta;
2536 4 : set_freepointer(s, freelist_tail, old.freelist);
2537 2 : new.freelist = freelist;
2538 : } else
2539 : new.freelist = old.freelist;
2540 :
2541 2 : new.frozen = 0;
2542 :
2543 2 : if (!new.inuse && n->nr_partial >= s->min_partial) {
2544 : mode = M_FREE;
2545 2 : } else if (new.freelist) {
2546 2 : mode = M_PARTIAL;
2547 : /*
2548 : * Taking the spinlock removes the possibility that
2549 : * acquire_slab() will see a slab that is frozen
2550 : */
2551 2 : spin_lock_irqsave(&n->list_lock, flags);
2552 : } else {
2553 : mode = M_FULL_NOLIST;
2554 : }
2555 :
2556 :
2557 2 : if (!cmpxchg_double_slab(s, slab,
2558 : old.freelist, old.counters,
2559 : new.freelist, new.counters,
2560 : "unfreezing slab")) {
2561 0 : if (mode == M_PARTIAL)
2562 0 : spin_unlock_irqrestore(&n->list_lock, flags);
2563 : goto redo;
2564 : }
2565 :
2566 :
2567 2 : if (mode == M_PARTIAL) {
2568 2 : add_partial(n, slab, tail);
2569 4 : spin_unlock_irqrestore(&n->list_lock, flags);
2570 2 : stat(s, tail);
2571 0 : } else if (mode == M_FREE) {
2572 0 : stat(s, DEACTIVATE_EMPTY);
2573 : discard_slab(s, slab);
2574 : stat(s, FREE_SLAB);
2575 : } else if (mode == M_FULL_NOLIST) {
2576 : stat(s, DEACTIVATE_FULL);
2577 : }
2578 2 : }
2579 :
2580 : #ifdef CONFIG_SLUB_CPU_PARTIAL
2581 : static void __unfreeze_partials(struct kmem_cache *s, struct slab *partial_slab)
2582 : {
2583 : struct kmem_cache_node *n = NULL, *n2 = NULL;
2584 : struct slab *slab, *slab_to_discard = NULL;
2585 : unsigned long flags = 0;
2586 :
2587 : while (partial_slab) {
2588 : struct slab new;
2589 : struct slab old;
2590 :
2591 : slab = partial_slab;
2592 : partial_slab = slab->next;
2593 :
2594 : n2 = get_node(s, slab_nid(slab));
2595 : if (n != n2) {
2596 : if (n)
2597 : spin_unlock_irqrestore(&n->list_lock, flags);
2598 :
2599 : n = n2;
2600 : spin_lock_irqsave(&n->list_lock, flags);
2601 : }
2602 :
2603 : do {
2604 :
2605 : old.freelist = slab->freelist;
2606 : old.counters = slab->counters;
2607 : VM_BUG_ON(!old.frozen);
2608 :
2609 : new.counters = old.counters;
2610 : new.freelist = old.freelist;
2611 :
2612 : new.frozen = 0;
2613 :
2614 : } while (!__cmpxchg_double_slab(s, slab,
2615 : old.freelist, old.counters,
2616 : new.freelist, new.counters,
2617 : "unfreezing slab"));
2618 :
2619 : if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
2620 : slab->next = slab_to_discard;
2621 : slab_to_discard = slab;
2622 : } else {
2623 : add_partial(n, slab, DEACTIVATE_TO_TAIL);
2624 : stat(s, FREE_ADD_PARTIAL);
2625 : }
2626 : }
2627 :
2628 : if (n)
2629 : spin_unlock_irqrestore(&n->list_lock, flags);
2630 :
2631 : while (slab_to_discard) {
2632 : slab = slab_to_discard;
2633 : slab_to_discard = slab_to_discard->next;
2634 :
2635 : stat(s, DEACTIVATE_EMPTY);
2636 : discard_slab(s, slab);
2637 : stat(s, FREE_SLAB);
2638 : }
2639 : }
2640 :
2641 : /*
2642 : * Unfreeze all the cpu partial slabs.
2643 : */
2644 : static void unfreeze_partials(struct kmem_cache *s)
2645 : {
2646 : struct slab *partial_slab;
2647 : unsigned long flags;
2648 :
2649 : local_lock_irqsave(&s->cpu_slab->lock, flags);
2650 : partial_slab = this_cpu_read(s->cpu_slab->partial);
2651 : this_cpu_write(s->cpu_slab->partial, NULL);
2652 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2653 :
2654 : if (partial_slab)
2655 : __unfreeze_partials(s, partial_slab);
2656 : }
2657 :
2658 : static void unfreeze_partials_cpu(struct kmem_cache *s,
2659 : struct kmem_cache_cpu *c)
2660 : {
2661 : struct slab *partial_slab;
2662 :
2663 : partial_slab = slub_percpu_partial(c);
2664 : c->partial = NULL;
2665 :
2666 : if (partial_slab)
2667 : __unfreeze_partials(s, partial_slab);
2668 : }
2669 :
2670 : /*
2671 : * Put a slab that was just frozen (in __slab_free|get_partial_node) into a
2672 : * partial slab slot if available.
2673 : *
2674 : * If we did not find a slot then simply move all the partials to the
2675 : * per node partial list.
2676 : */
2677 : static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain)
2678 : {
2679 : struct slab *oldslab;
2680 : struct slab *slab_to_unfreeze = NULL;
2681 : unsigned long flags;
2682 : int slabs = 0;
2683 :
2684 : local_lock_irqsave(&s->cpu_slab->lock, flags);
2685 :
2686 : oldslab = this_cpu_read(s->cpu_slab->partial);
2687 :
2688 : if (oldslab) {
2689 : if (drain && oldslab->slabs >= s->cpu_partial_slabs) {
2690 : /*
2691 : * Partial array is full. Move the existing set to the
2692 : * per node partial list. Postpone the actual unfreezing
2693 : * outside of the critical section.
2694 : */
2695 : slab_to_unfreeze = oldslab;
2696 : oldslab = NULL;
2697 : } else {
2698 : slabs = oldslab->slabs;
2699 : }
2700 : }
2701 :
2702 : slabs++;
2703 :
2704 : slab->slabs = slabs;
2705 : slab->next = oldslab;
2706 :
2707 : this_cpu_write(s->cpu_slab->partial, slab);
2708 :
2709 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2710 :
2711 : if (slab_to_unfreeze) {
2712 : __unfreeze_partials(s, slab_to_unfreeze);
2713 : stat(s, CPU_PARTIAL_DRAIN);
2714 : }
2715 : }
2716 :
2717 : #else /* CONFIG_SLUB_CPU_PARTIAL */
2718 :
2719 : static inline void unfreeze_partials(struct kmem_cache *s) { }
2720 : static inline void unfreeze_partials_cpu(struct kmem_cache *s,
2721 : struct kmem_cache_cpu *c) { }
2722 :
2723 : #endif /* CONFIG_SLUB_CPU_PARTIAL */
2724 :
2725 0 : static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
2726 : {
2727 : unsigned long flags;
2728 : struct slab *slab;
2729 : void *freelist;
2730 :
2731 0 : local_lock_irqsave(&s->cpu_slab->lock, flags);
2732 :
2733 0 : slab = c->slab;
2734 0 : freelist = c->freelist;
2735 :
2736 0 : c->slab = NULL;
2737 0 : c->freelist = NULL;
2738 0 : c->tid = next_tid(c->tid);
2739 :
2740 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
2741 :
2742 0 : if (slab) {
2743 0 : deactivate_slab(s, slab, freelist);
2744 0 : stat(s, CPUSLAB_FLUSH);
2745 : }
2746 0 : }
2747 :
2748 2 : static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
2749 : {
2750 2 : struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2751 2 : void *freelist = c->freelist;
2752 2 : struct slab *slab = c->slab;
2753 :
2754 2 : c->slab = NULL;
2755 2 : c->freelist = NULL;
2756 4 : c->tid = next_tid(c->tid);
2757 :
2758 2 : if (slab) {
2759 2 : deactivate_slab(s, slab, freelist);
2760 2 : stat(s, CPUSLAB_FLUSH);
2761 : }
2762 :
2763 2 : unfreeze_partials_cpu(s, c);
2764 2 : }
2765 :
2766 : struct slub_flush_work {
2767 : struct work_struct work;
2768 : struct kmem_cache *s;
2769 : bool skip;
2770 : };
2771 :
2772 : /*
2773 : * Flush cpu slab.
2774 : *
2775 : * Called from CPU work handler with migration disabled.
2776 : */
2777 0 : static void flush_cpu_slab(struct work_struct *w)
2778 : {
2779 : struct kmem_cache *s;
2780 : struct kmem_cache_cpu *c;
2781 : struct slub_flush_work *sfw;
2782 :
2783 0 : sfw = container_of(w, struct slub_flush_work, work);
2784 :
2785 0 : s = sfw->s;
2786 0 : c = this_cpu_ptr(s->cpu_slab);
2787 :
2788 0 : if (c->slab)
2789 0 : flush_slab(s, c);
2790 :
2791 0 : unfreeze_partials(s);
2792 0 : }
2793 :
2794 : static bool has_cpu_slab(int cpu, struct kmem_cache *s)
2795 : {
2796 0 : struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2797 :
2798 0 : return c->slab || slub_percpu_partial(c);
2799 : }
2800 :
2801 : static DEFINE_MUTEX(flush_lock);
2802 : static DEFINE_PER_CPU(struct slub_flush_work, slub_flush);
2803 :
2804 0 : static void flush_all_cpus_locked(struct kmem_cache *s)
2805 : {
2806 : struct slub_flush_work *sfw;
2807 : unsigned int cpu;
2808 :
2809 : lockdep_assert_cpus_held();
2810 0 : mutex_lock(&flush_lock);
2811 :
2812 0 : for_each_online_cpu(cpu) {
2813 0 : sfw = &per_cpu(slub_flush, cpu);
2814 0 : if (!has_cpu_slab(cpu, s)) {
2815 0 : sfw->skip = true;
2816 0 : continue;
2817 : }
2818 0 : INIT_WORK(&sfw->work, flush_cpu_slab);
2819 0 : sfw->skip = false;
2820 0 : sfw->s = s;
2821 0 : queue_work_on(cpu, flushwq, &sfw->work);
2822 : }
2823 :
2824 0 : for_each_online_cpu(cpu) {
2825 0 : sfw = &per_cpu(slub_flush, cpu);
2826 0 : if (sfw->skip)
2827 0 : continue;
2828 0 : flush_work(&sfw->work);
2829 : }
2830 :
2831 0 : mutex_unlock(&flush_lock);
2832 0 : }
2833 :
2834 : static void flush_all(struct kmem_cache *s)
2835 : {
2836 : cpus_read_lock();
2837 0 : flush_all_cpus_locked(s);
2838 : cpus_read_unlock();
2839 : }
2840 :
2841 : /*
2842 : * Use the cpu notifier to insure that the cpu slabs are flushed when
2843 : * necessary.
2844 : */
2845 0 : static int slub_cpu_dead(unsigned int cpu)
2846 : {
2847 : struct kmem_cache *s;
2848 :
2849 0 : mutex_lock(&slab_mutex);
2850 0 : list_for_each_entry(s, &slab_caches, list)
2851 0 : __flush_cpu_slab(s, cpu);
2852 0 : mutex_unlock(&slab_mutex);
2853 0 : return 0;
2854 : }
2855 :
2856 : #else /* CONFIG_SLUB_TINY */
2857 : static inline void flush_all_cpus_locked(struct kmem_cache *s) { }
2858 : static inline void flush_all(struct kmem_cache *s) { }
2859 : static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) { }
2860 : static inline int slub_cpu_dead(unsigned int cpu) { return 0; }
2861 : #endif /* CONFIG_SLUB_TINY */
2862 :
2863 : /*
2864 : * Check if the objects in a per cpu structure fit numa
2865 : * locality expectations.
2866 : */
2867 : static inline int node_match(struct slab *slab, int node)
2868 : {
2869 : #ifdef CONFIG_NUMA
2870 : if (node != NUMA_NO_NODE && slab_nid(slab) != node)
2871 : return 0;
2872 : #endif
2873 : return 1;
2874 : }
2875 :
2876 : #ifdef CONFIG_SLUB_DEBUG
2877 0 : static int count_free(struct slab *slab)
2878 : {
2879 0 : return slab->objects - slab->inuse;
2880 : }
2881 :
2882 : static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2883 : {
2884 0 : return atomic_long_read(&n->total_objects);
2885 : }
2886 :
2887 : /* Supports checking bulk free of a constructed freelist */
2888 0 : static inline bool free_debug_processing(struct kmem_cache *s,
2889 : struct slab *slab, void *head, void *tail, int *bulk_cnt,
2890 : unsigned long addr, depot_stack_handle_t handle)
2891 : {
2892 0 : bool checks_ok = false;
2893 0 : void *object = head;
2894 0 : int cnt = 0;
2895 :
2896 0 : if (s->flags & SLAB_CONSISTENCY_CHECKS) {
2897 0 : if (!check_slab(s, slab))
2898 : goto out;
2899 : }
2900 :
2901 0 : if (slab->inuse < *bulk_cnt) {
2902 0 : slab_err(s, slab, "Slab has %d allocated objects but %d are to be freed\n",
2903 : slab->inuse, *bulk_cnt);
2904 0 : goto out;
2905 : }
2906 :
2907 : next_object:
2908 :
2909 0 : if (++cnt > *bulk_cnt)
2910 : goto out_cnt;
2911 :
2912 0 : if (s->flags & SLAB_CONSISTENCY_CHECKS) {
2913 0 : if (!free_consistency_checks(s, slab, object, addr))
2914 : goto out;
2915 : }
2916 :
2917 0 : if (s->flags & SLAB_STORE_USER)
2918 : set_track_update(s, object, TRACK_FREE, addr, handle);
2919 0 : trace(s, slab, object, 0);
2920 : /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
2921 0 : init_object(s, object, SLUB_RED_INACTIVE);
2922 :
2923 : /* Reached end of constructed freelist yet? */
2924 0 : if (object != tail) {
2925 0 : object = get_freepointer(s, object);
2926 0 : goto next_object;
2927 : }
2928 : checks_ok = true;
2929 :
2930 : out_cnt:
2931 0 : if (cnt != *bulk_cnt) {
2932 0 : slab_err(s, slab, "Bulk free expected %d objects but found %d\n",
2933 : *bulk_cnt, cnt);
2934 0 : *bulk_cnt = cnt;
2935 : }
2936 :
2937 : out:
2938 :
2939 0 : if (!checks_ok)
2940 0 : slab_fix(s, "Object at 0x%p not freed", object);
2941 :
2942 0 : return checks_ok;
2943 : }
2944 : #endif /* CONFIG_SLUB_DEBUG */
2945 :
2946 : #if defined(CONFIG_SLUB_DEBUG) || defined(SLAB_SUPPORTS_SYSFS)
2947 0 : static unsigned long count_partial(struct kmem_cache_node *n,
2948 : int (*get_count)(struct slab *))
2949 : {
2950 : unsigned long flags;
2951 0 : unsigned long x = 0;
2952 : struct slab *slab;
2953 :
2954 0 : spin_lock_irqsave(&n->list_lock, flags);
2955 0 : list_for_each_entry(slab, &n->partial, slab_list)
2956 0 : x += get_count(slab);
2957 0 : spin_unlock_irqrestore(&n->list_lock, flags);
2958 0 : return x;
2959 : }
2960 : #endif /* CONFIG_SLUB_DEBUG || SLAB_SUPPORTS_SYSFS */
2961 :
2962 : #ifdef CONFIG_SLUB_DEBUG
2963 : static noinline void
2964 0 : slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2965 : {
2966 : static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2967 : DEFAULT_RATELIMIT_BURST);
2968 : int node;
2969 : struct kmem_cache_node *n;
2970 :
2971 0 : if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2972 : return;
2973 :
2974 0 : pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2975 : nid, gfpflags, &gfpflags);
2976 0 : pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
2977 : s->name, s->object_size, s->size, oo_order(s->oo),
2978 : oo_order(s->min));
2979 :
2980 0 : if (oo_order(s->min) > get_order(s->object_size))
2981 0 : pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2982 : s->name);
2983 :
2984 0 : for_each_kmem_cache_node(s, node, n) {
2985 : unsigned long nr_slabs;
2986 : unsigned long nr_objs;
2987 : unsigned long nr_free;
2988 :
2989 0 : nr_free = count_partial(n, count_free);
2990 0 : nr_slabs = node_nr_slabs(n);
2991 0 : nr_objs = node_nr_objs(n);
2992 :
2993 0 : pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
2994 : node, nr_slabs, nr_objs, nr_free);
2995 : }
2996 : }
2997 : #else /* CONFIG_SLUB_DEBUG */
2998 : static inline void
2999 : slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) { }
3000 : #endif
3001 :
3002 : static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags)
3003 : {
3004 4602 : if (unlikely(slab_test_pfmemalloc(slab)))
3005 0 : return gfp_pfmemalloc_allowed(gfpflags);
3006 :
3007 : return true;
3008 : }
3009 :
3010 : #ifndef CONFIG_SLUB_TINY
3011 : /*
3012 : * Check the slab->freelist and either transfer the freelist to the
3013 : * per cpu freelist or deactivate the slab.
3014 : *
3015 : * The slab is still frozen if the return value is not NULL.
3016 : *
3017 : * If this function returns NULL then the slab has been unfrozen.
3018 : */
3019 : static inline void *get_freelist(struct kmem_cache *s, struct slab *slab)
3020 : {
3021 : struct slab new;
3022 : unsigned long counters;
3023 : void *freelist;
3024 :
3025 : lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock));
3026 :
3027 : do {
3028 2245 : freelist = slab->freelist;
3029 2245 : counters = slab->counters;
3030 :
3031 2245 : new.counters = counters;
3032 : VM_BUG_ON(!new.frozen);
3033 :
3034 2245 : new.inuse = slab->objects;
3035 2245 : new.frozen = freelist != NULL;
3036 :
3037 4490 : } while (!__cmpxchg_double_slab(s, slab,
3038 : freelist, counters,
3039 : NULL, new.counters,
3040 2245 : "get_freelist"));
3041 :
3042 : return freelist;
3043 : }
3044 :
3045 : /*
3046 : * Slow path. The lockless freelist is empty or we need to perform
3047 : * debugging duties.
3048 : *
3049 : * Processing is still very fast if new objects have been freed to the
3050 : * regular freelist. In that case we simply take over the regular freelist
3051 : * as the lockless freelist and zap the regular freelist.
3052 : *
3053 : * If that is not working then we fall back to the partial lists. We take the
3054 : * first element of the freelist as the object to allocate now and move the
3055 : * rest of the freelist to the lockless freelist.
3056 : *
3057 : * And if we were unable to get a new slab from the partial slab lists then
3058 : * we need to allocate a new slab. This is the slowest path since it involves
3059 : * a call to the page allocator and the setup of a new slab.
3060 : *
3061 : * Version of __slab_alloc to use when we know that preemption is
3062 : * already disabled (which is the case for bulk allocation).
3063 : */
3064 2275 : static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
3065 : unsigned long addr, struct kmem_cache_cpu *c, unsigned int orig_size)
3066 : {
3067 : void *freelist;
3068 : struct slab *slab;
3069 : unsigned long flags;
3070 : struct partial_context pc;
3071 :
3072 2275 : stat(s, ALLOC_SLOWPATH);
3073 :
3074 : reread_slab:
3075 :
3076 2275 : slab = READ_ONCE(c->slab);
3077 2275 : if (!slab) {
3078 : /*
3079 : * if the node is not online or has no normal memory, just
3080 : * ignore the node constraint
3081 : */
3082 32 : if (unlikely(node != NUMA_NO_NODE &&
3083 : !node_isset(node, slab_nodes)))
3084 0 : node = NUMA_NO_NODE;
3085 : goto new_slab;
3086 : }
3087 : redo:
3088 :
3089 2245 : if (unlikely(!node_match(slab, node))) {
3090 : /*
3091 : * same as above but node_match() being false already
3092 : * implies node != NUMA_NO_NODE
3093 : */
3094 : if (!node_isset(node, slab_nodes)) {
3095 : node = NUMA_NO_NODE;
3096 : } else {
3097 : stat(s, ALLOC_NODE_MISMATCH);
3098 : goto deactivate_slab;
3099 : }
3100 : }
3101 :
3102 : /*
3103 : * By rights, we should be searching for a slab page that was
3104 : * PFMEMALLOC but right now, we are losing the pfmemalloc
3105 : * information when the page leaves the per-cpu allocator
3106 : */
3107 4490 : if (unlikely(!pfmemalloc_match(slab, gfpflags)))
3108 : goto deactivate_slab;
3109 :
3110 : /* must check again c->slab in case we got preempted and it changed */
3111 2245 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3112 2245 : if (unlikely(slab != c->slab)) {
3113 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3114 : goto reread_slab;
3115 : }
3116 2245 : freelist = c->freelist;
3117 2245 : if (freelist)
3118 : goto load_freelist;
3119 :
3120 2245 : freelist = get_freelist(s, slab);
3121 :
3122 2245 : if (!freelist) {
3123 2245 : c->slab = NULL;
3124 4490 : c->tid = next_tid(c->tid);
3125 2245 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3126 : stat(s, DEACTIVATE_BYPASS);
3127 : goto new_slab;
3128 : }
3129 :
3130 : stat(s, ALLOC_REFILL);
3131 :
3132 : load_freelist:
3133 :
3134 2275 : lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock));
3135 :
3136 : /*
3137 : * freelist is pointing to the list of objects to be used.
3138 : * slab is pointing to the slab from which the objects are obtained.
3139 : * That slab must be frozen for per cpu allocations to work.
3140 : */
3141 : VM_BUG_ON(!c->slab->frozen);
3142 4550 : c->freelist = get_freepointer(s, freelist);
3143 4550 : c->tid = next_tid(c->tid);
3144 4550 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3145 2275 : return freelist;
3146 :
3147 : deactivate_slab:
3148 :
3149 0 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3150 0 : if (slab != c->slab) {
3151 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3152 : goto reread_slab;
3153 : }
3154 0 : freelist = c->freelist;
3155 0 : c->slab = NULL;
3156 0 : c->freelist = NULL;
3157 0 : c->tid = next_tid(c->tid);
3158 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3159 0 : deactivate_slab(s, slab, freelist);
3160 :
3161 : new_slab:
3162 :
3163 : if (slub_percpu_partial(c)) {
3164 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3165 : if (unlikely(c->slab)) {
3166 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3167 : goto reread_slab;
3168 : }
3169 : if (unlikely(!slub_percpu_partial(c))) {
3170 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3171 : /* we were preempted and partial list got empty */
3172 : goto new_objects;
3173 : }
3174 :
3175 : slab = c->slab = slub_percpu_partial(c);
3176 : slub_set_percpu_partial(c, slab);
3177 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3178 : stat(s, CPU_PARTIAL_ALLOC);
3179 : goto redo;
3180 : }
3181 :
3182 : new_objects:
3183 :
3184 2275 : pc.flags = gfpflags;
3185 2275 : pc.slab = &slab;
3186 2275 : pc.orig_size = orig_size;
3187 2275 : freelist = get_partial(s, node, &pc);
3188 2275 : if (freelist)
3189 : goto check_new_slab;
3190 :
3191 2193 : slub_put_cpu_ptr(s->cpu_slab);
3192 2193 : slab = new_slab(s, gfpflags, node);
3193 2193 : c = slub_get_cpu_ptr(s->cpu_slab);
3194 :
3195 2193 : if (unlikely(!slab)) {
3196 0 : slab_out_of_memory(s, gfpflags, node);
3197 0 : return NULL;
3198 : }
3199 :
3200 2193 : stat(s, ALLOC_SLAB);
3201 :
3202 2193 : if (kmem_cache_debug(s)) {
3203 0 : freelist = alloc_single_from_new_slab(s, slab, orig_size);
3204 :
3205 0 : if (unlikely(!freelist))
3206 : goto new_objects;
3207 :
3208 0 : if (s->flags & SLAB_STORE_USER)
3209 : set_track(s, freelist, TRACK_ALLOC, addr);
3210 :
3211 : return freelist;
3212 : }
3213 :
3214 : /*
3215 : * No other reference to the slab yet so we can
3216 : * muck around with it freely without cmpxchg
3217 : */
3218 2193 : freelist = slab->freelist;
3219 2193 : slab->freelist = NULL;
3220 2193 : slab->inuse = slab->objects;
3221 2193 : slab->frozen = 1;
3222 :
3223 4386 : inc_slabs_node(s, slab_nid(slab), slab->objects);
3224 :
3225 : check_new_slab:
3226 :
3227 2275 : if (kmem_cache_debug(s)) {
3228 : /*
3229 : * For debug caches here we had to go through
3230 : * alloc_single_from_partial() so just store the tracking info
3231 : * and return the object
3232 : */
3233 0 : if (s->flags & SLAB_STORE_USER)
3234 : set_track(s, freelist, TRACK_ALLOC, addr);
3235 :
3236 : return freelist;
3237 : }
3238 :
3239 4550 : if (unlikely(!pfmemalloc_match(slab, gfpflags))) {
3240 : /*
3241 : * For !pfmemalloc_match() case we don't load freelist so that
3242 : * we don't make further mismatched allocations easier.
3243 : */
3244 0 : deactivate_slab(s, slab, get_freepointer(s, freelist));
3245 0 : return freelist;
3246 : }
3247 :
3248 : retry_load_slab:
3249 :
3250 2275 : local_lock_irqsave(&s->cpu_slab->lock, flags);
3251 2275 : if (unlikely(c->slab)) {
3252 0 : void *flush_freelist = c->freelist;
3253 0 : struct slab *flush_slab = c->slab;
3254 :
3255 0 : c->slab = NULL;
3256 0 : c->freelist = NULL;
3257 0 : c->tid = next_tid(c->tid);
3258 :
3259 0 : local_unlock_irqrestore(&s->cpu_slab->lock, flags);
3260 :
3261 0 : deactivate_slab(s, flush_slab, flush_freelist);
3262 :
3263 0 : stat(s, CPUSLAB_FLUSH);
3264 :
3265 : goto retry_load_slab;
3266 : }
3267 2275 : c->slab = slab;
3268 :
3269 2275 : goto load_freelist;
3270 : }
3271 :
3272 : /*
3273 : * A wrapper for ___slab_alloc() for contexts where preemption is not yet
3274 : * disabled. Compensates for possible cpu changes by refetching the per cpu area
3275 : * pointer.
3276 : */
3277 : static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
3278 : unsigned long addr, struct kmem_cache_cpu *c, unsigned int orig_size)
3279 : {
3280 : void *p;
3281 :
3282 : #ifdef CONFIG_PREEMPT_COUNT
3283 : /*
3284 : * We may have been preempted and rescheduled on a different
3285 : * cpu before disabling preemption. Need to reload cpu area
3286 : * pointer.
3287 : */
3288 : c = slub_get_cpu_ptr(s->cpu_slab);
3289 : #endif
3290 :
3291 2275 : p = ___slab_alloc(s, gfpflags, node, addr, c, orig_size);
3292 : #ifdef CONFIG_PREEMPT_COUNT
3293 : slub_put_cpu_ptr(s->cpu_slab);
3294 : #endif
3295 : return p;
3296 : }
3297 :
3298 : static __always_inline void *__slab_alloc_node(struct kmem_cache *s,
3299 : gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
3300 : {
3301 : struct kmem_cache_cpu *c;
3302 : struct slab *slab;
3303 : unsigned long tid;
3304 : void *object;
3305 :
3306 : redo:
3307 : /*
3308 : * Must read kmem_cache cpu data via this cpu ptr. Preemption is
3309 : * enabled. We may switch back and forth between cpus while
3310 : * reading from one cpu area. That does not matter as long
3311 : * as we end up on the original cpu again when doing the cmpxchg.
3312 : *
3313 : * We must guarantee that tid and kmem_cache_cpu are retrieved on the
3314 : * same cpu. We read first the kmem_cache_cpu pointer and use it to read
3315 : * the tid. If we are preempted and switched to another cpu between the
3316 : * two reads, it's OK as the two are still associated with the same cpu
3317 : * and cmpxchg later will validate the cpu.
3318 : */
3319 61648 : c = raw_cpu_ptr(s->cpu_slab);
3320 61648 : tid = READ_ONCE(c->tid);
3321 :
3322 : /*
3323 : * Irqless object alloc/free algorithm used here depends on sequence
3324 : * of fetching cpu_slab's data. tid should be fetched before anything
3325 : * on c to guarantee that object and slab associated with previous tid
3326 : * won't be used with current tid. If we fetch tid first, object and
3327 : * slab could be one associated with next tid and our alloc/free
3328 : * request will be failed. In this case, we will retry. So, no problem.
3329 : */
3330 61648 : barrier();
3331 :
3332 : /*
3333 : * The transaction ids are globally unique per cpu and per operation on
3334 : * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
3335 : * occurs on the right processor and that there was no operation on the
3336 : * linked list in between.
3337 : */
3338 :
3339 61648 : object = c->freelist;
3340 61648 : slab = c->slab;
3341 :
3342 61648 : if (!USE_LOCKLESS_FAST_PATH() ||
3343 121021 : unlikely(!object || !slab || !node_match(slab, node))) {
3344 4206 : object = __slab_alloc(s, gfpflags, node, addr, c, orig_size);
3345 : } else {
3346 59373 : void *next_object = get_freepointer_safe(s, object);
3347 :
3348 : /*
3349 : * The cmpxchg will only match if there was no additional
3350 : * operation and if we are on the right processor.
3351 : *
3352 : * The cmpxchg does the following atomically (without lock
3353 : * semantics!)
3354 : * 1. Relocate first pointer to the current per cpu area.
3355 : * 2. Verify that tid and freelist have not been changed
3356 : * 3. If they were not changed replace tid and freelist
3357 : *
3358 : * Since this is without lock semantics the protection is only
3359 : * against code executing on this cpu *not* from access by
3360 : * other cpus.
3361 : */
3362 237492 : if (unlikely(!this_cpu_cmpxchg_double(
3363 : s->cpu_slab->freelist, s->cpu_slab->tid,
3364 : object, tid,
3365 : next_object, next_tid(tid)))) {
3366 :
3367 : note_cmpxchg_failure("slab_alloc", s, tid);
3368 : goto redo;
3369 : }
3370 59373 : prefetch_freepointer(s, next_object);
3371 : stat(s, ALLOC_FASTPATH);
3372 : }
3373 :
3374 : return object;
3375 : }
3376 : #else /* CONFIG_SLUB_TINY */
3377 : static void *__slab_alloc_node(struct kmem_cache *s,
3378 : gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
3379 : {
3380 : struct partial_context pc;
3381 : struct slab *slab;
3382 : void *object;
3383 :
3384 : pc.flags = gfpflags;
3385 : pc.slab = &slab;
3386 : pc.orig_size = orig_size;
3387 : object = get_partial(s, node, &pc);
3388 :
3389 : if (object)
3390 : return object;
3391 :
3392 : slab = new_slab(s, gfpflags, node);
3393 : if (unlikely(!slab)) {
3394 : slab_out_of_memory(s, gfpflags, node);
3395 : return NULL;
3396 : }
3397 :
3398 : object = alloc_single_from_new_slab(s, slab, orig_size);
3399 :
3400 : return object;
3401 : }
3402 : #endif /* CONFIG_SLUB_TINY */
3403 :
3404 : /*
3405 : * If the object has been wiped upon free, make sure it's fully initialized by
3406 : * zeroing out freelist pointer.
3407 : */
3408 : static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s,
3409 : void *obj)
3410 : {
3411 61648 : if (unlikely(slab_want_init_on_free(s)) && obj)
3412 0 : memset((void *)((char *)kasan_reset_tag(obj) + s->offset),
3413 : 0, sizeof(void *));
3414 : }
3415 :
3416 : /*
3417 : * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
3418 : * have the fastpath folded into their functions. So no function call
3419 : * overhead for requests that can be satisfied on the fastpath.
3420 : *
3421 : * The fastpath works by first checking if the lockless freelist can be used.
3422 : * If not then __slab_alloc is called for slow processing.
3423 : *
3424 : * Otherwise we can simply pick the next object from the lockless free list.
3425 : */
3426 : static __fastpath_inline void *slab_alloc_node(struct kmem_cache *s, struct list_lru *lru,
3427 : gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
3428 : {
3429 : void *object;
3430 61648 : struct obj_cgroup *objcg = NULL;
3431 61648 : bool init = false;
3432 :
3433 123296 : s = slab_pre_alloc_hook(s, lru, &objcg, 1, gfpflags);
3434 61648 : if (!s)
3435 : return NULL;
3436 :
3437 61648 : object = kfence_alloc(s, orig_size, gfpflags);
3438 : if (unlikely(object))
3439 : goto out;
3440 :
3441 61648 : object = __slab_alloc_node(s, gfpflags, node, addr, orig_size);
3442 :
3443 123296 : maybe_wipe_obj_freeptr(s, object);
3444 123296 : init = slab_want_init_on_alloc(gfpflags, s);
3445 :
3446 : out:
3447 : /*
3448 : * When init equals 'true', like for kzalloc() family, only
3449 : * @orig_size bytes might be zeroed instead of s->object_size
3450 : */
3451 61648 : slab_post_alloc_hook(s, objcg, gfpflags, 1, &object, init, orig_size);
3452 :
3453 61648 : return object;
3454 : }
3455 :
3456 : static __fastpath_inline void *slab_alloc(struct kmem_cache *s, struct list_lru *lru,
3457 : gfp_t gfpflags, unsigned long addr, size_t orig_size)
3458 : {
3459 13669 : return slab_alloc_node(s, lru, gfpflags, NUMA_NO_NODE, addr, orig_size);
3460 : }
3461 :
3462 : static __fastpath_inline
3463 : void *__kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
3464 : gfp_t gfpflags)
3465 : {
3466 27338 : void *ret = slab_alloc(s, lru, gfpflags, _RET_IP_, s->object_size);
3467 :
3468 13669 : trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, NUMA_NO_NODE);
3469 :
3470 : return ret;
3471 : }
3472 :
3473 13569 : void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
3474 : {
3475 13569 : return __kmem_cache_alloc_lru(s, NULL, gfpflags);
3476 : }
3477 : EXPORT_SYMBOL(kmem_cache_alloc);
3478 :
3479 100 : void *kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
3480 : gfp_t gfpflags)
3481 : {
3482 100 : return __kmem_cache_alloc_lru(s, lru, gfpflags);
3483 : }
3484 : EXPORT_SYMBOL(kmem_cache_alloc_lru);
3485 :
3486 47251 : void *__kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags,
3487 : int node, size_t orig_size,
3488 : unsigned long caller)
3489 : {
3490 47251 : return slab_alloc_node(s, NULL, gfpflags, node,
3491 : caller, orig_size);
3492 : }
3493 :
3494 728 : void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
3495 : {
3496 1456 : void *ret = slab_alloc_node(s, NULL, gfpflags, node, _RET_IP_, s->object_size);
3497 :
3498 728 : trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, node);
3499 :
3500 728 : return ret;
3501 : }
3502 : EXPORT_SYMBOL(kmem_cache_alloc_node);
3503 :
3504 0 : static noinline void free_to_partial_list(
3505 : struct kmem_cache *s, struct slab *slab,
3506 : void *head, void *tail, int bulk_cnt,
3507 : unsigned long addr)
3508 : {
3509 0 : struct kmem_cache_node *n = get_node(s, slab_nid(slab));
3510 0 : struct slab *slab_free = NULL;
3511 0 : int cnt = bulk_cnt;
3512 : unsigned long flags;
3513 0 : depot_stack_handle_t handle = 0;
3514 :
3515 0 : if (s->flags & SLAB_STORE_USER)
3516 0 : handle = set_track_prepare();
3517 :
3518 0 : spin_lock_irqsave(&n->list_lock, flags);
3519 :
3520 0 : if (free_debug_processing(s, slab, head, tail, &cnt, addr, handle)) {
3521 0 : void *prior = slab->freelist;
3522 :
3523 : /* Perform the actual freeing while we still hold the locks */
3524 0 : slab->inuse -= cnt;
3525 0 : set_freepointer(s, tail, prior);
3526 0 : slab->freelist = head;
3527 :
3528 : /*
3529 : * If the slab is empty, and node's partial list is full,
3530 : * it should be discarded anyway no matter it's on full or
3531 : * partial list.
3532 : */
3533 0 : if (slab->inuse == 0 && n->nr_partial >= s->min_partial)
3534 0 : slab_free = slab;
3535 :
3536 0 : if (!prior) {
3537 : /* was on full list */
3538 0 : remove_full(s, n, slab);
3539 0 : if (!slab_free) {
3540 : add_partial(n, slab, DEACTIVATE_TO_TAIL);
3541 : stat(s, FREE_ADD_PARTIAL);
3542 : }
3543 0 : } else if (slab_free) {
3544 0 : remove_partial(n, slab);
3545 : stat(s, FREE_REMOVE_PARTIAL);
3546 : }
3547 : }
3548 :
3549 0 : if (slab_free) {
3550 : /*
3551 : * Update the counters while still holding n->list_lock to
3552 : * prevent spurious validation warnings
3553 : */
3554 0 : dec_slabs_node(s, slab_nid(slab_free), slab_free->objects);
3555 : }
3556 :
3557 0 : spin_unlock_irqrestore(&n->list_lock, flags);
3558 :
3559 0 : if (slab_free) {
3560 0 : stat(s, FREE_SLAB);
3561 0 : free_slab(s, slab_free);
3562 : }
3563 0 : }
3564 :
3565 : /*
3566 : * Slow path handling. This may still be called frequently since objects
3567 : * have a longer lifetime than the cpu slabs in most processing loads.
3568 : *
3569 : * So we still attempt to reduce cache line usage. Just take the slab
3570 : * lock and free the item. If there is no additional partial slab
3571 : * handling required then we can return immediately.
3572 : */
3573 39836 : static void __slab_free(struct kmem_cache *s, struct slab *slab,
3574 : void *head, void *tail, int cnt,
3575 : unsigned long addr)
3576 :
3577 : {
3578 : void *prior;
3579 : int was_frozen;
3580 : struct slab new;
3581 : unsigned long counters;
3582 39836 : struct kmem_cache_node *n = NULL;
3583 : unsigned long flags;
3584 :
3585 39836 : stat(s, FREE_SLOWPATH);
3586 :
3587 39836 : if (kfence_free(head))
3588 38083 : return;
3589 :
3590 39836 : if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) {
3591 0 : free_to_partial_list(s, slab, head, tail, cnt, addr);
3592 0 : return;
3593 : }
3594 :
3595 : do {
3596 39836 : if (unlikely(n)) {
3597 0 : spin_unlock_irqrestore(&n->list_lock, flags);
3598 0 : n = NULL;
3599 : }
3600 39836 : prior = slab->freelist;
3601 39836 : counters = slab->counters;
3602 79672 : set_freepointer(s, tail, prior);
3603 39836 : new.counters = counters;
3604 39836 : was_frozen = new.frozen;
3605 39836 : new.inuse -= cnt;
3606 39836 : if ((!new.inuse || !prior) && !was_frozen) {
3607 :
3608 3634 : if (kmem_cache_has_cpu_partial(s) && !prior) {
3609 :
3610 : /*
3611 : * Slab was on no list before and will be
3612 : * partially empty
3613 : * We can defer the list move and instead
3614 : * freeze it.
3615 : */
3616 : new.frozen = 1;
3617 :
3618 : } else { /* Needs to be taken off a list */
3619 :
3620 10902 : n = get_node(s, slab_nid(slab));
3621 : /*
3622 : * Speculatively acquire the list_lock.
3623 : * If the cmpxchg does not succeed then we may
3624 : * drop the list_lock without any processing.
3625 : *
3626 : * Otherwise the list_lock will synchronize with
3627 : * other processors updating the list of slabs.
3628 : */
3629 3634 : spin_lock_irqsave(&n->list_lock, flags);
3630 :
3631 : }
3632 : }
3633 :
3634 39836 : } while (!cmpxchg_double_slab(s, slab,
3635 : prior, counters,
3636 : head, new.counters,
3637 39836 : "__slab_free"));
3638 :
3639 39836 : if (likely(!n)) {
3640 :
3641 : if (likely(was_frozen)) {
3642 : /*
3643 : * The list lock was not taken therefore no list
3644 : * activity can be necessary.
3645 : */
3646 : stat(s, FREE_FROZEN);
3647 : } else if (new.frozen) {
3648 : /*
3649 : * If we just froze the slab then put it onto the
3650 : * per cpu partial list.
3651 : */
3652 : put_cpu_partial(s, slab, 1);
3653 : stat(s, CPU_PARTIAL_FREE);
3654 : }
3655 :
3656 : return;
3657 : }
3658 :
3659 3634 : if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
3660 : goto slab_empty;
3661 :
3662 : /*
3663 : * Objects left in the slab. If it was not on the partial list before
3664 : * then add it.
3665 : */
3666 1881 : if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
3667 3694 : remove_full(s, n, slab);
3668 : add_partial(n, slab, DEACTIVATE_TO_TAIL);
3669 : stat(s, FREE_ADD_PARTIAL);
3670 : }
3671 1881 : spin_unlock_irqrestore(&n->list_lock, flags);
3672 : return;
3673 :
3674 : slab_empty:
3675 1753 : if (prior) {
3676 : /*
3677 : * Slab on the partial list.
3678 : */
3679 1753 : remove_partial(n, slab);
3680 : stat(s, FREE_REMOVE_PARTIAL);
3681 : } else {
3682 : /* Slab must be on the full list */
3683 0 : remove_full(s, n, slab);
3684 : }
3685 :
3686 3506 : spin_unlock_irqrestore(&n->list_lock, flags);
3687 1753 : stat(s, FREE_SLAB);
3688 1753 : discard_slab(s, slab);
3689 : }
3690 :
3691 : #ifndef CONFIG_SLUB_TINY
3692 : /*
3693 : * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
3694 : * can perform fastpath freeing without additional function calls.
3695 : *
3696 : * The fastpath is only possible if we are freeing to the current cpu slab
3697 : * of this processor. This typically the case if we have just allocated
3698 : * the item before.
3699 : *
3700 : * If fastpath is not possible then fall back to __slab_free where we deal
3701 : * with all sorts of special processing.
3702 : *
3703 : * Bulk free of a freelist with several objects (all pointing to the
3704 : * same slab) possible by specifying head and tail ptr, plus objects
3705 : * count (cnt). Bulk free indicated by tail pointer being set.
3706 : */
3707 : static __always_inline void do_slab_free(struct kmem_cache *s,
3708 : struct slab *slab, void *head, void *tail,
3709 : int cnt, unsigned long addr)
3710 : {
3711 49059 : void *tail_obj = tail ? : head;
3712 : struct kmem_cache_cpu *c;
3713 : unsigned long tid;
3714 : void **freelist;
3715 :
3716 : redo:
3717 : /*
3718 : * Determine the currently cpus per cpu slab.
3719 : * The cpu may change afterward. However that does not matter since
3720 : * data is retrieved via this pointer. If we are on the same cpu
3721 : * during the cmpxchg then the free will succeed.
3722 : */
3723 49059 : c = raw_cpu_ptr(s->cpu_slab);
3724 49059 : tid = READ_ONCE(c->tid);
3725 :
3726 : /* Same with comment on barrier() in slab_alloc_node() */
3727 49059 : barrier();
3728 :
3729 49059 : if (unlikely(slab != c->slab)) {
3730 39836 : __slab_free(s, slab, head, tail_obj, cnt, addr);
3731 : return;
3732 : }
3733 :
3734 : if (USE_LOCKLESS_FAST_PATH()) {
3735 9223 : freelist = READ_ONCE(c->freelist);
3736 :
3737 18446 : set_freepointer(s, tail_obj, freelist);
3738 :
3739 36892 : if (unlikely(!this_cpu_cmpxchg_double(
3740 : s->cpu_slab->freelist, s->cpu_slab->tid,
3741 : freelist, tid,
3742 : head, next_tid(tid)))) {
3743 :
3744 : note_cmpxchg_failure("slab_free", s, tid);
3745 : goto redo;
3746 : }
3747 : } else {
3748 : /* Update the free list under the local lock */
3749 : local_lock(&s->cpu_slab->lock);
3750 : c = this_cpu_ptr(s->cpu_slab);
3751 : if (unlikely(slab != c->slab)) {
3752 : local_unlock(&s->cpu_slab->lock);
3753 : goto redo;
3754 : }
3755 : tid = c->tid;
3756 : freelist = c->freelist;
3757 :
3758 : set_freepointer(s, tail_obj, freelist);
3759 : c->freelist = head;
3760 : c->tid = next_tid(tid);
3761 :
3762 : local_unlock(&s->cpu_slab->lock);
3763 : }
3764 : stat(s, FREE_FASTPATH);
3765 : }
3766 : #else /* CONFIG_SLUB_TINY */
3767 : static void do_slab_free(struct kmem_cache *s,
3768 : struct slab *slab, void *head, void *tail,
3769 : int cnt, unsigned long addr)
3770 : {
3771 : void *tail_obj = tail ? : head;
3772 :
3773 : __slab_free(s, slab, head, tail_obj, cnt, addr);
3774 : }
3775 : #endif /* CONFIG_SLUB_TINY */
3776 :
3777 : static __fastpath_inline void slab_free(struct kmem_cache *s, struct slab *slab,
3778 : void *head, void *tail, void **p, int cnt,
3779 : unsigned long addr)
3780 : {
3781 49059 : memcg_slab_free_hook(s, slab, p, cnt);
3782 : /*
3783 : * With KASAN enabled slab_free_freelist_hook modifies the freelist
3784 : * to remove objects, whose reuse must be delayed.
3785 : */
3786 49059 : if (slab_free_freelist_hook(s, &head, &tail, &cnt))
3787 49059 : do_slab_free(s, slab, head, tail, cnt, addr);
3788 : }
3789 :
3790 : #ifdef CONFIG_KASAN_GENERIC
3791 : void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
3792 : {
3793 : do_slab_free(cache, virt_to_slab(x), x, NULL, 1, addr);
3794 : }
3795 : #endif
3796 :
3797 43506 : void __kmem_cache_free(struct kmem_cache *s, void *x, unsigned long caller)
3798 : {
3799 130518 : slab_free(s, virt_to_slab(x), x, NULL, &x, 1, caller);
3800 43506 : }
3801 :
3802 5553 : void kmem_cache_free(struct kmem_cache *s, void *x)
3803 : {
3804 5553 : s = cache_from_obj(s, x);
3805 5553 : if (!s)
3806 : return;
3807 5553 : trace_kmem_cache_free(_RET_IP_, x, s);
3808 11106 : slab_free(s, virt_to_slab(x), x, NULL, &x, 1, _RET_IP_);
3809 : }
3810 : EXPORT_SYMBOL(kmem_cache_free);
3811 :
3812 : struct detached_freelist {
3813 : struct slab *slab;
3814 : void *tail;
3815 : void *freelist;
3816 : int cnt;
3817 : struct kmem_cache *s;
3818 : };
3819 :
3820 : /*
3821 : * This function progressively scans the array with free objects (with
3822 : * a limited look ahead) and extract objects belonging to the same
3823 : * slab. It builds a detached freelist directly within the given
3824 : * slab/objects. This can happen without any need for
3825 : * synchronization, because the objects are owned by running process.
3826 : * The freelist is build up as a single linked list in the objects.
3827 : * The idea is, that this detached freelist can then be bulk
3828 : * transferred to the real freelist(s), but only requiring a single
3829 : * synchronization primitive. Look ahead in the array is limited due
3830 : * to performance reasons.
3831 : */
3832 : static inline
3833 0 : int build_detached_freelist(struct kmem_cache *s, size_t size,
3834 : void **p, struct detached_freelist *df)
3835 : {
3836 0 : int lookahead = 3;
3837 : void *object;
3838 : struct folio *folio;
3839 : size_t same;
3840 :
3841 0 : object = p[--size];
3842 0 : folio = virt_to_folio(object);
3843 0 : if (!s) {
3844 : /* Handle kalloc'ed objects */
3845 0 : if (unlikely(!folio_test_slab(folio))) {
3846 0 : free_large_kmalloc(folio, object);
3847 0 : df->slab = NULL;
3848 0 : return size;
3849 : }
3850 : /* Derive kmem_cache from object */
3851 0 : df->slab = folio_slab(folio);
3852 0 : df->s = df->slab->slab_cache;
3853 : } else {
3854 0 : df->slab = folio_slab(folio);
3855 0 : df->s = cache_from_obj(s, object); /* Support for memcg */
3856 : }
3857 :
3858 : /* Start new detached freelist */
3859 0 : df->tail = object;
3860 0 : df->freelist = object;
3861 0 : df->cnt = 1;
3862 :
3863 0 : if (is_kfence_address(object))
3864 : return size;
3865 :
3866 0 : set_freepointer(df->s, object, NULL);
3867 :
3868 0 : same = size;
3869 0 : while (size) {
3870 0 : object = p[--size];
3871 : /* df->slab is always set at this point */
3872 0 : if (df->slab == virt_to_slab(object)) {
3873 : /* Opportunity build freelist */
3874 0 : set_freepointer(df->s, object, df->freelist);
3875 0 : df->freelist = object;
3876 0 : df->cnt++;
3877 0 : same--;
3878 0 : if (size != same)
3879 0 : swap(p[size], p[same]);
3880 0 : continue;
3881 : }
3882 :
3883 : /* Limit look ahead search */
3884 0 : if (!--lookahead)
3885 : break;
3886 : }
3887 :
3888 0 : return same;
3889 : }
3890 :
3891 : /* Note that interrupts must be enabled when calling this function. */
3892 0 : void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
3893 : {
3894 0 : if (!size)
3895 : return;
3896 :
3897 : do {
3898 : struct detached_freelist df;
3899 :
3900 0 : size = build_detached_freelist(s, size, p, &df);
3901 0 : if (!df.slab)
3902 0 : continue;
3903 :
3904 0 : slab_free(df.s, df.slab, df.freelist, df.tail, &p[size], df.cnt,
3905 0 : _RET_IP_);
3906 0 : } while (likely(size));
3907 : }
3908 : EXPORT_SYMBOL(kmem_cache_free_bulk);
3909 :
3910 : #ifndef CONFIG_SLUB_TINY
3911 0 : static inline int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags,
3912 : size_t size, void **p, struct obj_cgroup *objcg)
3913 : {
3914 : struct kmem_cache_cpu *c;
3915 : unsigned long irqflags;
3916 : int i;
3917 :
3918 : /*
3919 : * Drain objects in the per cpu slab, while disabling local
3920 : * IRQs, which protects against PREEMPT and interrupts
3921 : * handlers invoking normal fastpath.
3922 : */
3923 0 : c = slub_get_cpu_ptr(s->cpu_slab);
3924 0 : local_lock_irqsave(&s->cpu_slab->lock, irqflags);
3925 :
3926 0 : for (i = 0; i < size; i++) {
3927 0 : void *object = kfence_alloc(s, s->object_size, flags);
3928 :
3929 : if (unlikely(object)) {
3930 : p[i] = object;
3931 : continue;
3932 : }
3933 :
3934 0 : object = c->freelist;
3935 0 : if (unlikely(!object)) {
3936 : /*
3937 : * We may have removed an object from c->freelist using
3938 : * the fastpath in the previous iteration; in that case,
3939 : * c->tid has not been bumped yet.
3940 : * Since ___slab_alloc() may reenable interrupts while
3941 : * allocating memory, we should bump c->tid now.
3942 : */
3943 0 : c->tid = next_tid(c->tid);
3944 :
3945 0 : local_unlock_irqrestore(&s->cpu_slab->lock, irqflags);
3946 :
3947 : /*
3948 : * Invoking slow path likely have side-effect
3949 : * of re-populating per CPU c->freelist
3950 : */
3951 0 : p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
3952 0 : _RET_IP_, c, s->object_size);
3953 0 : if (unlikely(!p[i]))
3954 : goto error;
3955 :
3956 0 : c = this_cpu_ptr(s->cpu_slab);
3957 0 : maybe_wipe_obj_freeptr(s, p[i]);
3958 :
3959 0 : local_lock_irqsave(&s->cpu_slab->lock, irqflags);
3960 :
3961 0 : continue; /* goto for-loop */
3962 : }
3963 0 : c->freelist = get_freepointer(s, object);
3964 0 : p[i] = object;
3965 0 : maybe_wipe_obj_freeptr(s, p[i]);
3966 : }
3967 0 : c->tid = next_tid(c->tid);
3968 0 : local_unlock_irqrestore(&s->cpu_slab->lock, irqflags);
3969 0 : slub_put_cpu_ptr(s->cpu_slab);
3970 :
3971 0 : return i;
3972 :
3973 : error:
3974 0 : slub_put_cpu_ptr(s->cpu_slab);
3975 0 : slab_post_alloc_hook(s, objcg, flags, i, p, false, s->object_size);
3976 0 : kmem_cache_free_bulk(s, i, p);
3977 0 : return 0;
3978 :
3979 : }
3980 : #else /* CONFIG_SLUB_TINY */
3981 : static int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags,
3982 : size_t size, void **p, struct obj_cgroup *objcg)
3983 : {
3984 : int i;
3985 :
3986 : for (i = 0; i < size; i++) {
3987 : void *object = kfence_alloc(s, s->object_size, flags);
3988 :
3989 : if (unlikely(object)) {
3990 : p[i] = object;
3991 : continue;
3992 : }
3993 :
3994 : p[i] = __slab_alloc_node(s, flags, NUMA_NO_NODE,
3995 : _RET_IP_, s->object_size);
3996 : if (unlikely(!p[i]))
3997 : goto error;
3998 :
3999 : maybe_wipe_obj_freeptr(s, p[i]);
4000 : }
4001 :
4002 : return i;
4003 :
4004 : error:
4005 : slab_post_alloc_hook(s, objcg, flags, i, p, false, s->object_size);
4006 : kmem_cache_free_bulk(s, i, p);
4007 : return 0;
4008 : }
4009 : #endif /* CONFIG_SLUB_TINY */
4010 :
4011 : /* Note that interrupts must be enabled when calling this function. */
4012 0 : int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
4013 : void **p)
4014 : {
4015 : int i;
4016 0 : struct obj_cgroup *objcg = NULL;
4017 :
4018 0 : if (!size)
4019 : return 0;
4020 :
4021 : /* memcg and kmem_cache debug support */
4022 0 : s = slab_pre_alloc_hook(s, NULL, &objcg, size, flags);
4023 0 : if (unlikely(!s))
4024 : return 0;
4025 :
4026 0 : i = __kmem_cache_alloc_bulk(s, flags, size, p, objcg);
4027 :
4028 : /*
4029 : * memcg and kmem_cache debug support and memory initialization.
4030 : * Done outside of the IRQ disabled fastpath loop.
4031 : */
4032 0 : if (i != 0)
4033 0 : slab_post_alloc_hook(s, objcg, flags, size, p,
4034 0 : slab_want_init_on_alloc(flags, s), s->object_size);
4035 : return i;
4036 : }
4037 : EXPORT_SYMBOL(kmem_cache_alloc_bulk);
4038 :
4039 :
4040 : /*
4041 : * Object placement in a slab is made very easy because we always start at
4042 : * offset 0. If we tune the size of the object to the alignment then we can
4043 : * get the required alignment by putting one properly sized object after
4044 : * another.
4045 : *
4046 : * Notice that the allocation order determines the sizes of the per cpu
4047 : * caches. Each processor has always one slab available for allocations.
4048 : * Increasing the allocation order reduces the number of times that slabs
4049 : * must be moved on and off the partial lists and is therefore a factor in
4050 : * locking overhead.
4051 : */
4052 :
4053 : /*
4054 : * Minimum / Maximum order of slab pages. This influences locking overhead
4055 : * and slab fragmentation. A higher order reduces the number of partial slabs
4056 : * and increases the number of allocations possible without having to
4057 : * take the list_lock.
4058 : */
4059 : static unsigned int slub_min_order;
4060 : static unsigned int slub_max_order =
4061 : IS_ENABLED(CONFIG_SLUB_TINY) ? 1 : PAGE_ALLOC_COSTLY_ORDER;
4062 : static unsigned int slub_min_objects;
4063 :
4064 : /*
4065 : * Calculate the order of allocation given an slab object size.
4066 : *
4067 : * The order of allocation has significant impact on performance and other
4068 : * system components. Generally order 0 allocations should be preferred since
4069 : * order 0 does not cause fragmentation in the page allocator. Larger objects
4070 : * be problematic to put into order 0 slabs because there may be too much
4071 : * unused space left. We go to a higher order if more than 1/16th of the slab
4072 : * would be wasted.
4073 : *
4074 : * In order to reach satisfactory performance we must ensure that a minimum
4075 : * number of objects is in one slab. Otherwise we may generate too much
4076 : * activity on the partial lists which requires taking the list_lock. This is
4077 : * less a concern for large slabs though which are rarely used.
4078 : *
4079 : * slub_max_order specifies the order where we begin to stop considering the
4080 : * number of objects in a slab as critical. If we reach slub_max_order then
4081 : * we try to keep the page order as low as possible. So we accept more waste
4082 : * of space in favor of a small page order.
4083 : *
4084 : * Higher order allocations also allow the placement of more objects in a
4085 : * slab and thereby reduce object handling overhead. If the user has
4086 : * requested a higher minimum order then we start with that one instead of
4087 : * the smallest order which will fit the object.
4088 : */
4089 54 : static inline unsigned int calc_slab_order(unsigned int size,
4090 : unsigned int min_objects, unsigned int max_order,
4091 : unsigned int fract_leftover)
4092 : {
4093 54 : unsigned int min_order = slub_min_order;
4094 : unsigned int order;
4095 :
4096 54 : if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
4097 0 : return get_order(size * MAX_OBJS_PER_PAGE) - 1;
4098 :
4099 164 : for (order = max(min_order, (unsigned int)get_order(min_objects * size));
4100 2 : order <= max_order; order++) {
4101 :
4102 55 : unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
4103 : unsigned int rem;
4104 :
4105 55 : rem = slab_size % size;
4106 :
4107 55 : if (rem <= slab_size / fract_leftover)
4108 : break;
4109 : }
4110 :
4111 : return order;
4112 : }
4113 :
4114 53 : static inline int calculate_order(unsigned int size)
4115 : {
4116 : unsigned int order;
4117 : unsigned int min_objects;
4118 : unsigned int max_objects;
4119 : unsigned int nr_cpus;
4120 :
4121 : /*
4122 : * Attempt to find best configuration for a slab. This
4123 : * works by first attempting to generate a layout with
4124 : * the best configuration and backing off gradually.
4125 : *
4126 : * First we increase the acceptable waste in a slab. Then
4127 : * we reduce the minimum objects required in a slab.
4128 : */
4129 53 : min_objects = slub_min_objects;
4130 53 : if (!min_objects) {
4131 : /*
4132 : * Some architectures will only update present cpus when
4133 : * onlining them, so don't trust the number if it's just 1. But
4134 : * we also don't want to use nr_cpu_ids always, as on some other
4135 : * architectures, there can be many possible cpus, but never
4136 : * onlined. Here we compromise between trying to avoid too high
4137 : * order on systems that appear larger than they are, and too
4138 : * low order on systems that appear smaller than they are.
4139 : */
4140 53 : nr_cpus = num_present_cpus();
4141 : if (nr_cpus <= 1)
4142 53 : nr_cpus = nr_cpu_ids;
4143 53 : min_objects = 4 * (fls(nr_cpus) + 1);
4144 : }
4145 106 : max_objects = order_objects(slub_max_order, size);
4146 53 : min_objects = min(min_objects, max_objects);
4147 :
4148 106 : while (min_objects > 1) {
4149 : unsigned int fraction;
4150 :
4151 : fraction = 16;
4152 54 : while (fraction >= 4) {
4153 54 : order = calc_slab_order(size, min_objects,
4154 : slub_max_order, fraction);
4155 54 : if (order <= slub_max_order)
4156 53 : return order;
4157 1 : fraction /= 2;
4158 : }
4159 0 : min_objects--;
4160 : }
4161 :
4162 : /*
4163 : * We were unable to place multiple objects in a slab. Now
4164 : * lets see if we can place a single object there.
4165 : */
4166 0 : order = calc_slab_order(size, 1, slub_max_order, 1);
4167 0 : if (order <= slub_max_order)
4168 0 : return order;
4169 :
4170 : /*
4171 : * Doh this slab cannot be placed using slub_max_order.
4172 : */
4173 0 : order = calc_slab_order(size, 1, MAX_ORDER, 1);
4174 0 : if (order <= MAX_ORDER)
4175 0 : return order;
4176 : return -ENOSYS;
4177 : }
4178 :
4179 : static void
4180 : init_kmem_cache_node(struct kmem_cache_node *n)
4181 : {
4182 53 : n->nr_partial = 0;
4183 53 : spin_lock_init(&n->list_lock);
4184 106 : INIT_LIST_HEAD(&n->partial);
4185 : #ifdef CONFIG_SLUB_DEBUG
4186 106 : atomic_long_set(&n->nr_slabs, 0);
4187 106 : atomic_long_set(&n->total_objects, 0);
4188 106 : INIT_LIST_HEAD(&n->full);
4189 : #endif
4190 : }
4191 :
4192 : #ifndef CONFIG_SLUB_TINY
4193 53 : static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
4194 : {
4195 : BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
4196 : NR_KMALLOC_TYPES * KMALLOC_SHIFT_HIGH *
4197 : sizeof(struct kmem_cache_cpu));
4198 :
4199 : /*
4200 : * Must align to double word boundary for the double cmpxchg
4201 : * instructions to work; see __pcpu_double_call_return_bool().
4202 : */
4203 53 : s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
4204 : 2 * sizeof(void *));
4205 :
4206 53 : if (!s->cpu_slab)
4207 : return 0;
4208 :
4209 : init_kmem_cache_cpus(s);
4210 :
4211 : return 1;
4212 : }
4213 : #else
4214 : static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
4215 : {
4216 : return 1;
4217 : }
4218 : #endif /* CONFIG_SLUB_TINY */
4219 :
4220 : static struct kmem_cache *kmem_cache_node;
4221 :
4222 : /*
4223 : * No kmalloc_node yet so do it by hand. We know that this is the first
4224 : * slab on the node for this slabcache. There are no concurrent accesses
4225 : * possible.
4226 : *
4227 : * Note that this function only works on the kmem_cache_node
4228 : * when allocating for the kmem_cache_node. This is used for bootstrapping
4229 : * memory on a fresh node that has no slab structures yet.
4230 : */
4231 1 : static void early_kmem_cache_node_alloc(int node)
4232 : {
4233 : struct slab *slab;
4234 : struct kmem_cache_node *n;
4235 :
4236 1 : BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
4237 :
4238 1 : slab = new_slab(kmem_cache_node, GFP_NOWAIT, node);
4239 :
4240 1 : BUG_ON(!slab);
4241 3 : inc_slabs_node(kmem_cache_node, slab_nid(slab), slab->objects);
4242 2 : if (slab_nid(slab) != node) {
4243 0 : pr_err("SLUB: Unable to allocate memory from node %d\n", node);
4244 0 : pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
4245 : }
4246 :
4247 1 : n = slab->freelist;
4248 1 : BUG_ON(!n);
4249 : #ifdef CONFIG_SLUB_DEBUG
4250 1 : init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
4251 1 : init_tracking(kmem_cache_node, n);
4252 : #endif
4253 1 : n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false);
4254 2 : slab->freelist = get_freepointer(kmem_cache_node, n);
4255 1 : slab->inuse = 1;
4256 1 : kmem_cache_node->node[node] = n;
4257 1 : init_kmem_cache_node(n);
4258 2 : inc_slabs_node(kmem_cache_node, node, slab->objects);
4259 :
4260 : /*
4261 : * No locks need to be taken here as it has just been
4262 : * initialized and there is no concurrent access.
4263 : */
4264 1 : __add_partial(n, slab, DEACTIVATE_TO_HEAD);
4265 1 : }
4266 :
4267 0 : static void free_kmem_cache_nodes(struct kmem_cache *s)
4268 : {
4269 : int node;
4270 : struct kmem_cache_node *n;
4271 :
4272 0 : for_each_kmem_cache_node(s, node, n) {
4273 0 : s->node[node] = NULL;
4274 0 : kmem_cache_free(kmem_cache_node, n);
4275 : }
4276 0 : }
4277 :
4278 0 : void __kmem_cache_release(struct kmem_cache *s)
4279 : {
4280 0 : cache_random_seq_destroy(s);
4281 : #ifndef CONFIG_SLUB_TINY
4282 0 : free_percpu(s->cpu_slab);
4283 : #endif
4284 0 : free_kmem_cache_nodes(s);
4285 0 : }
4286 :
4287 53 : static int init_kmem_cache_nodes(struct kmem_cache *s)
4288 : {
4289 : int node;
4290 :
4291 159 : for_each_node_mask(node, slab_nodes) {
4292 : struct kmem_cache_node *n;
4293 :
4294 53 : if (slab_state == DOWN) {
4295 1 : early_kmem_cache_node_alloc(node);
4296 1 : continue;
4297 : }
4298 52 : n = kmem_cache_alloc_node(kmem_cache_node,
4299 : GFP_KERNEL, node);
4300 :
4301 52 : if (!n) {
4302 0 : free_kmem_cache_nodes(s);
4303 0 : return 0;
4304 : }
4305 :
4306 52 : init_kmem_cache_node(n);
4307 52 : s->node[node] = n;
4308 : }
4309 : return 1;
4310 : }
4311 :
4312 : static void set_cpu_partial(struct kmem_cache *s)
4313 : {
4314 : #ifdef CONFIG_SLUB_CPU_PARTIAL
4315 : unsigned int nr_objects;
4316 :
4317 : /*
4318 : * cpu_partial determined the maximum number of objects kept in the
4319 : * per cpu partial lists of a processor.
4320 : *
4321 : * Per cpu partial lists mainly contain slabs that just have one
4322 : * object freed. If they are used for allocation then they can be
4323 : * filled up again with minimal effort. The slab will never hit the
4324 : * per node partial lists and therefore no locking will be required.
4325 : *
4326 : * For backwards compatibility reasons, this is determined as number
4327 : * of objects, even though we now limit maximum number of pages, see
4328 : * slub_set_cpu_partial()
4329 : */
4330 : if (!kmem_cache_has_cpu_partial(s))
4331 : nr_objects = 0;
4332 : else if (s->size >= PAGE_SIZE)
4333 : nr_objects = 6;
4334 : else if (s->size >= 1024)
4335 : nr_objects = 24;
4336 : else if (s->size >= 256)
4337 : nr_objects = 52;
4338 : else
4339 : nr_objects = 120;
4340 :
4341 : slub_set_cpu_partial(s, nr_objects);
4342 : #endif
4343 : }
4344 :
4345 : /*
4346 : * calculate_sizes() determines the order and the distribution of data within
4347 : * a slab object.
4348 : */
4349 53 : static int calculate_sizes(struct kmem_cache *s)
4350 : {
4351 53 : slab_flags_t flags = s->flags;
4352 53 : unsigned int size = s->object_size;
4353 : unsigned int order;
4354 :
4355 : /*
4356 : * Round up object size to the next word boundary. We can only
4357 : * place the free pointer at word boundaries and this determines
4358 : * the possible location of the free pointer.
4359 : */
4360 53 : size = ALIGN(size, sizeof(void *));
4361 :
4362 : #ifdef CONFIG_SLUB_DEBUG
4363 : /*
4364 : * Determine if we can poison the object itself. If the user of
4365 : * the slab may touch the object after free or before allocation
4366 : * then we should never poison the object itself.
4367 : */
4368 53 : if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
4369 0 : !s->ctor)
4370 0 : s->flags |= __OBJECT_POISON;
4371 : else
4372 53 : s->flags &= ~__OBJECT_POISON;
4373 :
4374 :
4375 : /*
4376 : * If we are Redzoning then check if there is some space between the
4377 : * end of the object and the free pointer. If not then add an
4378 : * additional word to have some bytes to store Redzone information.
4379 : */
4380 53 : if ((flags & SLAB_RED_ZONE) && size == s->object_size)
4381 0 : size += sizeof(void *);
4382 : #endif
4383 :
4384 : /*
4385 : * With that we have determined the number of bytes in actual use
4386 : * by the object and redzoning.
4387 : */
4388 53 : s->inuse = size;
4389 :
4390 53 : if (slub_debug_orig_size(s) ||
4391 48 : (flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
4392 48 : ((flags & SLAB_RED_ZONE) && s->object_size < sizeof(void *)) ||
4393 48 : s->ctor) {
4394 : /*
4395 : * Relocate free pointer after the object if it is not
4396 : * permitted to overwrite the first word of the object on
4397 : * kmem_cache_free.
4398 : *
4399 : * This is the case if we do RCU, have a constructor or
4400 : * destructor, are poisoning the objects, or are
4401 : * redzoning an object smaller than sizeof(void *).
4402 : *
4403 : * The assumption that s->offset >= s->inuse means free
4404 : * pointer is outside of the object is used in the
4405 : * freeptr_outside_object() function. If that is no
4406 : * longer true, the function needs to be modified.
4407 : */
4408 10 : s->offset = size;
4409 10 : size += sizeof(void *);
4410 : } else {
4411 : /*
4412 : * Store freelist pointer near middle of object to keep
4413 : * it away from the edges of the object to avoid small
4414 : * sized over/underflows from neighboring allocations.
4415 : */
4416 43 : s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *));
4417 : }
4418 :
4419 : #ifdef CONFIG_SLUB_DEBUG
4420 53 : if (flags & SLAB_STORE_USER) {
4421 : /*
4422 : * Need to store information about allocs and frees after
4423 : * the object.
4424 : */
4425 0 : size += 2 * sizeof(struct track);
4426 :
4427 : /* Save the original kmalloc request size */
4428 0 : if (flags & SLAB_KMALLOC)
4429 0 : size += sizeof(unsigned int);
4430 : }
4431 : #endif
4432 :
4433 53 : kasan_cache_create(s, &size, &s->flags);
4434 : #ifdef CONFIG_SLUB_DEBUG
4435 53 : if (flags & SLAB_RED_ZONE) {
4436 : /*
4437 : * Add some empty padding so that we can catch
4438 : * overwrites from earlier objects rather than let
4439 : * tracking information or the free pointer be
4440 : * corrupted if a user writes before the start
4441 : * of the object.
4442 : */
4443 0 : size += sizeof(void *);
4444 :
4445 : s->red_left_pad = sizeof(void *);
4446 0 : s->red_left_pad = ALIGN(s->red_left_pad, s->align);
4447 0 : size += s->red_left_pad;
4448 : }
4449 : #endif
4450 :
4451 : /*
4452 : * SLUB stores one object immediately after another beginning from
4453 : * offset 0. In order to align the objects we have to simply size
4454 : * each object to conform to the alignment.
4455 : */
4456 53 : size = ALIGN(size, s->align);
4457 53 : s->size = size;
4458 53 : s->reciprocal_size = reciprocal_value(size);
4459 53 : order = calculate_order(size);
4460 :
4461 53 : if ((int)order < 0)
4462 : return 0;
4463 :
4464 53 : s->allocflags = 0;
4465 53 : if (order)
4466 18 : s->allocflags |= __GFP_COMP;
4467 :
4468 53 : if (s->flags & SLAB_CACHE_DMA)
4469 0 : s->allocflags |= GFP_DMA;
4470 :
4471 53 : if (s->flags & SLAB_CACHE_DMA32)
4472 0 : s->allocflags |= GFP_DMA32;
4473 :
4474 53 : if (s->flags & SLAB_RECLAIM_ACCOUNT)
4475 18 : s->allocflags |= __GFP_RECLAIMABLE;
4476 :
4477 : /*
4478 : * Determine the number of objects per slab
4479 : */
4480 106 : s->oo = oo_make(order, size);
4481 159 : s->min = oo_make(get_order(size), size);
4482 :
4483 53 : return !!oo_objects(s->oo);
4484 : }
4485 :
4486 53 : static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
4487 : {
4488 53 : s->flags = kmem_cache_flags(s->size, flags, s->name);
4489 : #ifdef CONFIG_SLAB_FREELIST_HARDENED
4490 : s->random = get_random_long();
4491 : #endif
4492 :
4493 53 : if (!calculate_sizes(s))
4494 : goto error;
4495 53 : if (disable_higher_order_debug) {
4496 : /*
4497 : * Disable debugging flags that store metadata if the min slab
4498 : * order increased.
4499 : */
4500 0 : if (get_order(s->size) > get_order(s->object_size)) {
4501 0 : s->flags &= ~DEBUG_METADATA_FLAGS;
4502 0 : s->offset = 0;
4503 0 : if (!calculate_sizes(s))
4504 : goto error;
4505 : }
4506 : }
4507 :
4508 : #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
4509 : defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
4510 : if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
4511 : /* Enable fast mode */
4512 : s->flags |= __CMPXCHG_DOUBLE;
4513 : #endif
4514 :
4515 : /*
4516 : * The larger the object size is, the more slabs we want on the partial
4517 : * list to avoid pounding the page allocator excessively.
4518 : */
4519 106 : s->min_partial = min_t(unsigned long, MAX_PARTIAL, ilog2(s->size) / 2);
4520 53 : s->min_partial = max_t(unsigned long, MIN_PARTIAL, s->min_partial);
4521 :
4522 53 : set_cpu_partial(s);
4523 :
4524 : #ifdef CONFIG_NUMA
4525 : s->remote_node_defrag_ratio = 1000;
4526 : #endif
4527 :
4528 : /* Initialize the pre-computed randomized freelist if slab is up */
4529 : if (slab_state >= UP) {
4530 : if (init_cache_random_seq(s))
4531 : goto error;
4532 : }
4533 :
4534 53 : if (!init_kmem_cache_nodes(s))
4535 : goto error;
4536 :
4537 53 : if (alloc_kmem_cache_cpus(s))
4538 : return 0;
4539 :
4540 : error:
4541 0 : __kmem_cache_release(s);
4542 0 : return -EINVAL;
4543 : }
4544 :
4545 0 : static void list_slab_objects(struct kmem_cache *s, struct slab *slab,
4546 : const char *text)
4547 : {
4548 : #ifdef CONFIG_SLUB_DEBUG
4549 0 : void *addr = slab_address(slab);
4550 : void *p;
4551 :
4552 0 : slab_err(s, slab, text, s->name);
4553 :
4554 0 : spin_lock(&object_map_lock);
4555 0 : __fill_map(object_map, s, slab);
4556 :
4557 0 : for_each_object(p, s, addr, slab->objects) {
4558 :
4559 0 : if (!test_bit(__obj_to_index(s, addr, p), object_map)) {
4560 0 : pr_err("Object 0x%p @offset=%tu\n", p, p - addr);
4561 0 : print_tracking(s, p);
4562 : }
4563 : }
4564 0 : spin_unlock(&object_map_lock);
4565 : #endif
4566 0 : }
4567 :
4568 : /*
4569 : * Attempt to free all partial slabs on a node.
4570 : * This is called from __kmem_cache_shutdown(). We must take list_lock
4571 : * because sysfs file might still access partial list after the shutdowning.
4572 : */
4573 0 : static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
4574 : {
4575 0 : LIST_HEAD(discard);
4576 : struct slab *slab, *h;
4577 :
4578 0 : BUG_ON(irqs_disabled());
4579 0 : spin_lock_irq(&n->list_lock);
4580 0 : list_for_each_entry_safe(slab, h, &n->partial, slab_list) {
4581 0 : if (!slab->inuse) {
4582 0 : remove_partial(n, slab);
4583 0 : list_add(&slab->slab_list, &discard);
4584 : } else {
4585 0 : list_slab_objects(s, slab,
4586 : "Objects remaining in %s on __kmem_cache_shutdown()");
4587 : }
4588 : }
4589 0 : spin_unlock_irq(&n->list_lock);
4590 :
4591 0 : list_for_each_entry_safe(slab, h, &discard, slab_list)
4592 0 : discard_slab(s, slab);
4593 0 : }
4594 :
4595 0 : bool __kmem_cache_empty(struct kmem_cache *s)
4596 : {
4597 : int node;
4598 : struct kmem_cache_node *n;
4599 :
4600 0 : for_each_kmem_cache_node(s, node, n)
4601 0 : if (n->nr_partial || slabs_node(s, node))
4602 : return false;
4603 : return true;
4604 : }
4605 :
4606 : /*
4607 : * Release all resources used by a slab cache.
4608 : */
4609 0 : int __kmem_cache_shutdown(struct kmem_cache *s)
4610 : {
4611 : int node;
4612 : struct kmem_cache_node *n;
4613 :
4614 0 : flush_all_cpus_locked(s);
4615 : /* Attempt to free all objects */
4616 0 : for_each_kmem_cache_node(s, node, n) {
4617 0 : free_partial(s, n);
4618 0 : if (n->nr_partial || slabs_node(s, node))
4619 : return 1;
4620 : }
4621 : return 0;
4622 : }
4623 :
4624 : #ifdef CONFIG_PRINTK
4625 0 : void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
4626 : {
4627 : void *base;
4628 : int __maybe_unused i;
4629 : unsigned int objnr;
4630 : void *objp;
4631 : void *objp0;
4632 0 : struct kmem_cache *s = slab->slab_cache;
4633 : struct track __maybe_unused *trackp;
4634 :
4635 0 : kpp->kp_ptr = object;
4636 0 : kpp->kp_slab = slab;
4637 0 : kpp->kp_slab_cache = s;
4638 0 : base = slab_address(slab);
4639 0 : objp0 = kasan_reset_tag(object);
4640 : #ifdef CONFIG_SLUB_DEBUG
4641 0 : objp = restore_red_left(s, objp0);
4642 : #else
4643 : objp = objp0;
4644 : #endif
4645 0 : objnr = obj_to_index(s, slab, objp);
4646 0 : kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp);
4647 0 : objp = base + s->size * objnr;
4648 0 : kpp->kp_objp = objp;
4649 0 : if (WARN_ON_ONCE(objp < base || objp >= base + slab->objects * s->size
4650 0 : || (objp - base) % s->size) ||
4651 0 : !(s->flags & SLAB_STORE_USER))
4652 : return;
4653 : #ifdef CONFIG_SLUB_DEBUG
4654 0 : objp = fixup_red_left(s, objp);
4655 0 : trackp = get_track(s, objp, TRACK_ALLOC);
4656 0 : kpp->kp_ret = (void *)trackp->addr;
4657 : #ifdef CONFIG_STACKDEPOT
4658 : {
4659 : depot_stack_handle_t handle;
4660 : unsigned long *entries;
4661 : unsigned int nr_entries;
4662 :
4663 0 : handle = READ_ONCE(trackp->handle);
4664 0 : if (handle) {
4665 0 : nr_entries = stack_depot_fetch(handle, &entries);
4666 0 : for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
4667 0 : kpp->kp_stack[i] = (void *)entries[i];
4668 : }
4669 :
4670 0 : trackp = get_track(s, objp, TRACK_FREE);
4671 0 : handle = READ_ONCE(trackp->handle);
4672 0 : if (handle) {
4673 0 : nr_entries = stack_depot_fetch(handle, &entries);
4674 0 : for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++)
4675 0 : kpp->kp_free_stack[i] = (void *)entries[i];
4676 : }
4677 : }
4678 : #endif
4679 : #endif
4680 : }
4681 : #endif
4682 :
4683 : /********************************************************************
4684 : * Kmalloc subsystem
4685 : *******************************************************************/
4686 :
4687 0 : static int __init setup_slub_min_order(char *str)
4688 : {
4689 0 : get_option(&str, (int *)&slub_min_order);
4690 :
4691 0 : return 1;
4692 : }
4693 :
4694 : __setup("slub_min_order=", setup_slub_min_order);
4695 :
4696 0 : static int __init setup_slub_max_order(char *str)
4697 : {
4698 0 : get_option(&str, (int *)&slub_max_order);
4699 0 : slub_max_order = min_t(unsigned int, slub_max_order, MAX_ORDER);
4700 :
4701 0 : return 1;
4702 : }
4703 :
4704 : __setup("slub_max_order=", setup_slub_max_order);
4705 :
4706 0 : static int __init setup_slub_min_objects(char *str)
4707 : {
4708 0 : get_option(&str, (int *)&slub_min_objects);
4709 :
4710 0 : return 1;
4711 : }
4712 :
4713 : __setup("slub_min_objects=", setup_slub_min_objects);
4714 :
4715 : #ifdef CONFIG_HARDENED_USERCOPY
4716 : /*
4717 : * Rejects incorrectly sized objects and objects that are to be copied
4718 : * to/from userspace but do not fall entirely within the containing slab
4719 : * cache's usercopy region.
4720 : *
4721 : * Returns NULL if check passes, otherwise const char * to name of cache
4722 : * to indicate an error.
4723 : */
4724 : void __check_heap_object(const void *ptr, unsigned long n,
4725 : const struct slab *slab, bool to_user)
4726 : {
4727 : struct kmem_cache *s;
4728 : unsigned int offset;
4729 : bool is_kfence = is_kfence_address(ptr);
4730 :
4731 : ptr = kasan_reset_tag(ptr);
4732 :
4733 : /* Find object and usable object size. */
4734 : s = slab->slab_cache;
4735 :
4736 : /* Reject impossible pointers. */
4737 : if (ptr < slab_address(slab))
4738 : usercopy_abort("SLUB object not in SLUB page?!", NULL,
4739 : to_user, 0, n);
4740 :
4741 : /* Find offset within object. */
4742 : if (is_kfence)
4743 : offset = ptr - kfence_object_start(ptr);
4744 : else
4745 : offset = (ptr - slab_address(slab)) % s->size;
4746 :
4747 : /* Adjust for redzone and reject if within the redzone. */
4748 : if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) {
4749 : if (offset < s->red_left_pad)
4750 : usercopy_abort("SLUB object in left red zone",
4751 : s->name, to_user, offset, n);
4752 : offset -= s->red_left_pad;
4753 : }
4754 :
4755 : /* Allow address range falling entirely within usercopy region. */
4756 : if (offset >= s->useroffset &&
4757 : offset - s->useroffset <= s->usersize &&
4758 : n <= s->useroffset - offset + s->usersize)
4759 : return;
4760 :
4761 : usercopy_abort("SLUB object", s->name, to_user, offset, n);
4762 : }
4763 : #endif /* CONFIG_HARDENED_USERCOPY */
4764 :
4765 : #define SHRINK_PROMOTE_MAX 32
4766 :
4767 : /*
4768 : * kmem_cache_shrink discards empty slabs and promotes the slabs filled
4769 : * up most to the head of the partial lists. New allocations will then
4770 : * fill those up and thus they can be removed from the partial lists.
4771 : *
4772 : * The slabs with the least items are placed last. This results in them
4773 : * being allocated from last increasing the chance that the last objects
4774 : * are freed in them.
4775 : */
4776 0 : static int __kmem_cache_do_shrink(struct kmem_cache *s)
4777 : {
4778 : int node;
4779 : int i;
4780 : struct kmem_cache_node *n;
4781 : struct slab *slab;
4782 : struct slab *t;
4783 : struct list_head discard;
4784 : struct list_head promote[SHRINK_PROMOTE_MAX];
4785 : unsigned long flags;
4786 0 : int ret = 0;
4787 :
4788 0 : for_each_kmem_cache_node(s, node, n) {
4789 0 : INIT_LIST_HEAD(&discard);
4790 0 : for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
4791 0 : INIT_LIST_HEAD(promote + i);
4792 :
4793 0 : spin_lock_irqsave(&n->list_lock, flags);
4794 :
4795 : /*
4796 : * Build lists of slabs to discard or promote.
4797 : *
4798 : * Note that concurrent frees may occur while we hold the
4799 : * list_lock. slab->inuse here is the upper limit.
4800 : */
4801 0 : list_for_each_entry_safe(slab, t, &n->partial, slab_list) {
4802 0 : int free = slab->objects - slab->inuse;
4803 :
4804 : /* Do not reread slab->inuse */
4805 0 : barrier();
4806 :
4807 : /* We do not keep full slabs on the list */
4808 0 : BUG_ON(free <= 0);
4809 :
4810 0 : if (free == slab->objects) {
4811 0 : list_move(&slab->slab_list, &discard);
4812 0 : n->nr_partial--;
4813 0 : dec_slabs_node(s, node, slab->objects);
4814 0 : } else if (free <= SHRINK_PROMOTE_MAX)
4815 0 : list_move(&slab->slab_list, promote + free - 1);
4816 : }
4817 :
4818 : /*
4819 : * Promote the slabs filled up most to the head of the
4820 : * partial list.
4821 : */
4822 0 : for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
4823 0 : list_splice(promote + i, &n->partial);
4824 :
4825 0 : spin_unlock_irqrestore(&n->list_lock, flags);
4826 :
4827 : /* Release empty slabs */
4828 0 : list_for_each_entry_safe(slab, t, &discard, slab_list)
4829 0 : free_slab(s, slab);
4830 :
4831 0 : if (slabs_node(s, node))
4832 0 : ret = 1;
4833 : }
4834 :
4835 0 : return ret;
4836 : }
4837 :
4838 0 : int __kmem_cache_shrink(struct kmem_cache *s)
4839 : {
4840 0 : flush_all(s);
4841 0 : return __kmem_cache_do_shrink(s);
4842 : }
4843 :
4844 : static int slab_mem_going_offline_callback(void *arg)
4845 : {
4846 : struct kmem_cache *s;
4847 :
4848 : mutex_lock(&slab_mutex);
4849 : list_for_each_entry(s, &slab_caches, list) {
4850 : flush_all_cpus_locked(s);
4851 : __kmem_cache_do_shrink(s);
4852 : }
4853 : mutex_unlock(&slab_mutex);
4854 :
4855 : return 0;
4856 : }
4857 :
4858 : static void slab_mem_offline_callback(void *arg)
4859 : {
4860 : struct memory_notify *marg = arg;
4861 : int offline_node;
4862 :
4863 : offline_node = marg->status_change_nid_normal;
4864 :
4865 : /*
4866 : * If the node still has available memory. we need kmem_cache_node
4867 : * for it yet.
4868 : */
4869 : if (offline_node < 0)
4870 : return;
4871 :
4872 : mutex_lock(&slab_mutex);
4873 : node_clear(offline_node, slab_nodes);
4874 : /*
4875 : * We no longer free kmem_cache_node structures here, as it would be
4876 : * racy with all get_node() users, and infeasible to protect them with
4877 : * slab_mutex.
4878 : */
4879 : mutex_unlock(&slab_mutex);
4880 : }
4881 :
4882 : static int slab_mem_going_online_callback(void *arg)
4883 : {
4884 : struct kmem_cache_node *n;
4885 : struct kmem_cache *s;
4886 : struct memory_notify *marg = arg;
4887 : int nid = marg->status_change_nid_normal;
4888 : int ret = 0;
4889 :
4890 : /*
4891 : * If the node's memory is already available, then kmem_cache_node is
4892 : * already created. Nothing to do.
4893 : */
4894 : if (nid < 0)
4895 : return 0;
4896 :
4897 : /*
4898 : * We are bringing a node online. No memory is available yet. We must
4899 : * allocate a kmem_cache_node structure in order to bring the node
4900 : * online.
4901 : */
4902 : mutex_lock(&slab_mutex);
4903 : list_for_each_entry(s, &slab_caches, list) {
4904 : /*
4905 : * The structure may already exist if the node was previously
4906 : * onlined and offlined.
4907 : */
4908 : if (get_node(s, nid))
4909 : continue;
4910 : /*
4911 : * XXX: kmem_cache_alloc_node will fallback to other nodes
4912 : * since memory is not yet available from the node that
4913 : * is brought up.
4914 : */
4915 : n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
4916 : if (!n) {
4917 : ret = -ENOMEM;
4918 : goto out;
4919 : }
4920 : init_kmem_cache_node(n);
4921 : s->node[nid] = n;
4922 : }
4923 : /*
4924 : * Any cache created after this point will also have kmem_cache_node
4925 : * initialized for the new node.
4926 : */
4927 : node_set(nid, slab_nodes);
4928 : out:
4929 : mutex_unlock(&slab_mutex);
4930 : return ret;
4931 : }
4932 :
4933 : static int slab_memory_callback(struct notifier_block *self,
4934 : unsigned long action, void *arg)
4935 : {
4936 : int ret = 0;
4937 :
4938 : switch (action) {
4939 : case MEM_GOING_ONLINE:
4940 : ret = slab_mem_going_online_callback(arg);
4941 : break;
4942 : case MEM_GOING_OFFLINE:
4943 : ret = slab_mem_going_offline_callback(arg);
4944 : break;
4945 : case MEM_OFFLINE:
4946 : case MEM_CANCEL_ONLINE:
4947 : slab_mem_offline_callback(arg);
4948 : break;
4949 : case MEM_ONLINE:
4950 : case MEM_CANCEL_OFFLINE:
4951 : break;
4952 : }
4953 : if (ret)
4954 : ret = notifier_from_errno(ret);
4955 : else
4956 : ret = NOTIFY_OK;
4957 : return ret;
4958 : }
4959 :
4960 : /********************************************************************
4961 : * Basic setup of slabs
4962 : *******************************************************************/
4963 :
4964 : /*
4965 : * Used for early kmem_cache structures that were allocated using
4966 : * the page allocator. Allocate them properly then fix up the pointers
4967 : * that may be pointing to the wrong kmem_cache structure.
4968 : */
4969 :
4970 2 : static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
4971 : {
4972 : int node;
4973 4 : struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
4974 : struct kmem_cache_node *n;
4975 :
4976 2 : memcpy(s, static_cache, kmem_cache->object_size);
4977 :
4978 : /*
4979 : * This runs very early, and only the boot processor is supposed to be
4980 : * up. Even if it weren't true, IRQs are not up so we couldn't fire
4981 : * IPIs around.
4982 : */
4983 2 : __flush_cpu_slab(s, smp_processor_id());
4984 6 : for_each_kmem_cache_node(s, node, n) {
4985 : struct slab *p;
4986 :
4987 4 : list_for_each_entry(p, &n->partial, slab_list)
4988 2 : p->slab_cache = s;
4989 :
4990 : #ifdef CONFIG_SLUB_DEBUG
4991 2 : list_for_each_entry(p, &n->full, slab_list)
4992 0 : p->slab_cache = s;
4993 : #endif
4994 : }
4995 4 : list_add(&s->list, &slab_caches);
4996 2 : return s;
4997 : }
4998 :
4999 1 : void __init kmem_cache_init(void)
5000 : {
5001 : static __initdata struct kmem_cache boot_kmem_cache,
5002 : boot_kmem_cache_node;
5003 : int node;
5004 :
5005 : if (debug_guardpage_minorder())
5006 : slub_max_order = 0;
5007 :
5008 : /* Print slub debugging pointers without hashing */
5009 1 : if (__slub_debug_enabled())
5010 0 : no_hash_pointers_enable(NULL);
5011 :
5012 1 : kmem_cache_node = &boot_kmem_cache_node;
5013 1 : kmem_cache = &boot_kmem_cache;
5014 :
5015 : /*
5016 : * Initialize the nodemask for which we will allocate per node
5017 : * structures. Here we don't need taking slab_mutex yet.
5018 : */
5019 3 : for_each_node_state(node, N_NORMAL_MEMORY)
5020 : node_set(node, slab_nodes);
5021 :
5022 1 : create_boot_cache(kmem_cache_node, "kmem_cache_node",
5023 : sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
5024 :
5025 1 : hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
5026 :
5027 : /* Able to allocate the per node structures */
5028 1 : slab_state = PARTIAL;
5029 :
5030 1 : create_boot_cache(kmem_cache, "kmem_cache",
5031 : offsetof(struct kmem_cache, node) +
5032 : nr_node_ids * sizeof(struct kmem_cache_node *),
5033 : SLAB_HWCACHE_ALIGN, 0, 0);
5034 :
5035 1 : kmem_cache = bootstrap(&boot_kmem_cache);
5036 1 : kmem_cache_node = bootstrap(&boot_kmem_cache_node);
5037 :
5038 : /* Now we can use the kmem_cache to allocate kmalloc slabs */
5039 1 : setup_kmalloc_cache_index_table();
5040 1 : create_kmalloc_caches(0);
5041 :
5042 : /* Setup random freelists for each cache */
5043 1 : init_freelist_randomization();
5044 :
5045 1 : cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
5046 : slub_cpu_dead);
5047 :
5048 1 : pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n",
5049 : cache_line_size(),
5050 : slub_min_order, slub_max_order, slub_min_objects,
5051 : nr_cpu_ids, nr_node_ids);
5052 1 : }
5053 :
5054 1 : void __init kmem_cache_init_late(void)
5055 : {
5056 : #ifndef CONFIG_SLUB_TINY
5057 1 : flushwq = alloc_workqueue("slub_flushwq", WQ_MEM_RECLAIM, 0);
5058 1 : WARN_ON(!flushwq);
5059 : #endif
5060 1 : }
5061 :
5062 : struct kmem_cache *
5063 57 : __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
5064 : slab_flags_t flags, void (*ctor)(void *))
5065 : {
5066 : struct kmem_cache *s;
5067 :
5068 57 : s = find_mergeable(size, align, flags, name, ctor);
5069 57 : if (s) {
5070 32 : if (sysfs_slab_alias(s, name))
5071 : return NULL;
5072 :
5073 32 : s->refcount++;
5074 :
5075 : /*
5076 : * Adjust the object sizes so that we clear
5077 : * the complete object on kzalloc.
5078 : */
5079 32 : s->object_size = max(s->object_size, size);
5080 32 : s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
5081 : }
5082 :
5083 : return s;
5084 : }
5085 :
5086 53 : int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
5087 : {
5088 : int err;
5089 :
5090 53 : err = kmem_cache_open(s, flags);
5091 53 : if (err)
5092 : return err;
5093 :
5094 : /* Mutex is not taken during early boot */
5095 53 : if (slab_state <= UP)
5096 : return 0;
5097 :
5098 0 : err = sysfs_slab_add(s);
5099 0 : if (err) {
5100 0 : __kmem_cache_release(s);
5101 0 : return err;
5102 : }
5103 :
5104 : if (s->flags & SLAB_STORE_USER)
5105 : debugfs_slab_add(s);
5106 :
5107 : return 0;
5108 : }
5109 :
5110 : #ifdef SLAB_SUPPORTS_SYSFS
5111 0 : static int count_inuse(struct slab *slab)
5112 : {
5113 0 : return slab->inuse;
5114 : }
5115 :
5116 0 : static int count_total(struct slab *slab)
5117 : {
5118 0 : return slab->objects;
5119 : }
5120 : #endif
5121 :
5122 : #ifdef CONFIG_SLUB_DEBUG
5123 0 : static void validate_slab(struct kmem_cache *s, struct slab *slab,
5124 : unsigned long *obj_map)
5125 : {
5126 : void *p;
5127 0 : void *addr = slab_address(slab);
5128 :
5129 0 : if (!check_slab(s, slab) || !on_freelist(s, slab, NULL))
5130 : return;
5131 :
5132 : /* Now we know that a valid freelist exists */
5133 0 : __fill_map(obj_map, s, slab);
5134 0 : for_each_object(p, s, addr, slab->objects) {
5135 0 : u8 val = test_bit(__obj_to_index(s, addr, p), obj_map) ?
5136 : SLUB_RED_INACTIVE : SLUB_RED_ACTIVE;
5137 :
5138 0 : if (!check_object(s, slab, p, val))
5139 : break;
5140 : }
5141 : }
5142 :
5143 0 : static int validate_slab_node(struct kmem_cache *s,
5144 : struct kmem_cache_node *n, unsigned long *obj_map)
5145 : {
5146 0 : unsigned long count = 0;
5147 : struct slab *slab;
5148 : unsigned long flags;
5149 :
5150 0 : spin_lock_irqsave(&n->list_lock, flags);
5151 :
5152 0 : list_for_each_entry(slab, &n->partial, slab_list) {
5153 0 : validate_slab(s, slab, obj_map);
5154 0 : count++;
5155 : }
5156 0 : if (count != n->nr_partial) {
5157 0 : pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
5158 : s->name, count, n->nr_partial);
5159 0 : slab_add_kunit_errors();
5160 : }
5161 :
5162 0 : if (!(s->flags & SLAB_STORE_USER))
5163 : goto out;
5164 :
5165 0 : list_for_each_entry(slab, &n->full, slab_list) {
5166 0 : validate_slab(s, slab, obj_map);
5167 0 : count++;
5168 : }
5169 0 : if (count != atomic_long_read(&n->nr_slabs)) {
5170 0 : pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
5171 : s->name, count, atomic_long_read(&n->nr_slabs));
5172 0 : slab_add_kunit_errors();
5173 : }
5174 :
5175 : out:
5176 0 : spin_unlock_irqrestore(&n->list_lock, flags);
5177 0 : return count;
5178 : }
5179 :
5180 0 : long validate_slab_cache(struct kmem_cache *s)
5181 : {
5182 : int node;
5183 0 : unsigned long count = 0;
5184 : struct kmem_cache_node *n;
5185 : unsigned long *obj_map;
5186 :
5187 0 : obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL);
5188 0 : if (!obj_map)
5189 : return -ENOMEM;
5190 :
5191 0 : flush_all(s);
5192 0 : for_each_kmem_cache_node(s, node, n)
5193 0 : count += validate_slab_node(s, n, obj_map);
5194 :
5195 0 : bitmap_free(obj_map);
5196 :
5197 0 : return count;
5198 : }
5199 : EXPORT_SYMBOL(validate_slab_cache);
5200 :
5201 : #ifdef CONFIG_DEBUG_FS
5202 : /*
5203 : * Generate lists of code addresses where slabcache objects are allocated
5204 : * and freed.
5205 : */
5206 :
5207 : struct location {
5208 : depot_stack_handle_t handle;
5209 : unsigned long count;
5210 : unsigned long addr;
5211 : unsigned long waste;
5212 : long long sum_time;
5213 : long min_time;
5214 : long max_time;
5215 : long min_pid;
5216 : long max_pid;
5217 : DECLARE_BITMAP(cpus, NR_CPUS);
5218 : nodemask_t nodes;
5219 : };
5220 :
5221 : struct loc_track {
5222 : unsigned long max;
5223 : unsigned long count;
5224 : struct location *loc;
5225 : loff_t idx;
5226 : };
5227 :
5228 : static struct dentry *slab_debugfs_root;
5229 :
5230 : static void free_loc_track(struct loc_track *t)
5231 : {
5232 : if (t->max)
5233 : free_pages((unsigned long)t->loc,
5234 : get_order(sizeof(struct location) * t->max));
5235 : }
5236 :
5237 : static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
5238 : {
5239 : struct location *l;
5240 : int order;
5241 :
5242 : order = get_order(sizeof(struct location) * max);
5243 :
5244 : l = (void *)__get_free_pages(flags, order);
5245 : if (!l)
5246 : return 0;
5247 :
5248 : if (t->count) {
5249 : memcpy(l, t->loc, sizeof(struct location) * t->count);
5250 : free_loc_track(t);
5251 : }
5252 : t->max = max;
5253 : t->loc = l;
5254 : return 1;
5255 : }
5256 :
5257 : static int add_location(struct loc_track *t, struct kmem_cache *s,
5258 : const struct track *track,
5259 : unsigned int orig_size)
5260 : {
5261 : long start, end, pos;
5262 : struct location *l;
5263 : unsigned long caddr, chandle, cwaste;
5264 : unsigned long age = jiffies - track->when;
5265 : depot_stack_handle_t handle = 0;
5266 : unsigned int waste = s->object_size - orig_size;
5267 :
5268 : #ifdef CONFIG_STACKDEPOT
5269 : handle = READ_ONCE(track->handle);
5270 : #endif
5271 : start = -1;
5272 : end = t->count;
5273 :
5274 : for ( ; ; ) {
5275 : pos = start + (end - start + 1) / 2;
5276 :
5277 : /*
5278 : * There is nothing at "end". If we end up there
5279 : * we need to add something to before end.
5280 : */
5281 : if (pos == end)
5282 : break;
5283 :
5284 : l = &t->loc[pos];
5285 : caddr = l->addr;
5286 : chandle = l->handle;
5287 : cwaste = l->waste;
5288 : if ((track->addr == caddr) && (handle == chandle) &&
5289 : (waste == cwaste)) {
5290 :
5291 : l->count++;
5292 : if (track->when) {
5293 : l->sum_time += age;
5294 : if (age < l->min_time)
5295 : l->min_time = age;
5296 : if (age > l->max_time)
5297 : l->max_time = age;
5298 :
5299 : if (track->pid < l->min_pid)
5300 : l->min_pid = track->pid;
5301 : if (track->pid > l->max_pid)
5302 : l->max_pid = track->pid;
5303 :
5304 : cpumask_set_cpu(track->cpu,
5305 : to_cpumask(l->cpus));
5306 : }
5307 : node_set(page_to_nid(virt_to_page(track)), l->nodes);
5308 : return 1;
5309 : }
5310 :
5311 : if (track->addr < caddr)
5312 : end = pos;
5313 : else if (track->addr == caddr && handle < chandle)
5314 : end = pos;
5315 : else if (track->addr == caddr && handle == chandle &&
5316 : waste < cwaste)
5317 : end = pos;
5318 : else
5319 : start = pos;
5320 : }
5321 :
5322 : /*
5323 : * Not found. Insert new tracking element.
5324 : */
5325 : if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
5326 : return 0;
5327 :
5328 : l = t->loc + pos;
5329 : if (pos < t->count)
5330 : memmove(l + 1, l,
5331 : (t->count - pos) * sizeof(struct location));
5332 : t->count++;
5333 : l->count = 1;
5334 : l->addr = track->addr;
5335 : l->sum_time = age;
5336 : l->min_time = age;
5337 : l->max_time = age;
5338 : l->min_pid = track->pid;
5339 : l->max_pid = track->pid;
5340 : l->handle = handle;
5341 : l->waste = waste;
5342 : cpumask_clear(to_cpumask(l->cpus));
5343 : cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
5344 : nodes_clear(l->nodes);
5345 : node_set(page_to_nid(virt_to_page(track)), l->nodes);
5346 : return 1;
5347 : }
5348 :
5349 : static void process_slab(struct loc_track *t, struct kmem_cache *s,
5350 : struct slab *slab, enum track_item alloc,
5351 : unsigned long *obj_map)
5352 : {
5353 : void *addr = slab_address(slab);
5354 : bool is_alloc = (alloc == TRACK_ALLOC);
5355 : void *p;
5356 :
5357 : __fill_map(obj_map, s, slab);
5358 :
5359 : for_each_object(p, s, addr, slab->objects)
5360 : if (!test_bit(__obj_to_index(s, addr, p), obj_map))
5361 : add_location(t, s, get_track(s, p, alloc),
5362 : is_alloc ? get_orig_size(s, p) :
5363 : s->object_size);
5364 : }
5365 : #endif /* CONFIG_DEBUG_FS */
5366 : #endif /* CONFIG_SLUB_DEBUG */
5367 :
5368 : #ifdef SLAB_SUPPORTS_SYSFS
5369 : enum slab_stat_type {
5370 : SL_ALL, /* All slabs */
5371 : SL_PARTIAL, /* Only partially allocated slabs */
5372 : SL_CPU, /* Only slabs used for cpu caches */
5373 : SL_OBJECTS, /* Determine allocated objects not slabs */
5374 : SL_TOTAL /* Determine object capacity not slabs */
5375 : };
5376 :
5377 : #define SO_ALL (1 << SL_ALL)
5378 : #define SO_PARTIAL (1 << SL_PARTIAL)
5379 : #define SO_CPU (1 << SL_CPU)
5380 : #define SO_OBJECTS (1 << SL_OBJECTS)
5381 : #define SO_TOTAL (1 << SL_TOTAL)
5382 :
5383 0 : static ssize_t show_slab_objects(struct kmem_cache *s,
5384 : char *buf, unsigned long flags)
5385 : {
5386 0 : unsigned long total = 0;
5387 : int node;
5388 : int x;
5389 : unsigned long *nodes;
5390 0 : int len = 0;
5391 :
5392 0 : nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
5393 0 : if (!nodes)
5394 : return -ENOMEM;
5395 :
5396 0 : if (flags & SO_CPU) {
5397 : int cpu;
5398 :
5399 0 : for_each_possible_cpu(cpu) {
5400 0 : struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
5401 : cpu);
5402 : int node;
5403 : struct slab *slab;
5404 :
5405 0 : slab = READ_ONCE(c->slab);
5406 0 : if (!slab)
5407 0 : continue;
5408 :
5409 0 : node = slab_nid(slab);
5410 0 : if (flags & SO_TOTAL)
5411 0 : x = slab->objects;
5412 0 : else if (flags & SO_OBJECTS)
5413 0 : x = slab->inuse;
5414 : else
5415 : x = 1;
5416 :
5417 0 : total += x;
5418 0 : nodes[node] += x;
5419 :
5420 : #ifdef CONFIG_SLUB_CPU_PARTIAL
5421 : slab = slub_percpu_partial_read_once(c);
5422 : if (slab) {
5423 : node = slab_nid(slab);
5424 : if (flags & SO_TOTAL)
5425 : WARN_ON_ONCE(1);
5426 : else if (flags & SO_OBJECTS)
5427 : WARN_ON_ONCE(1);
5428 : else
5429 : x = slab->slabs;
5430 : total += x;
5431 : nodes[node] += x;
5432 : }
5433 : #endif
5434 : }
5435 : }
5436 :
5437 : /*
5438 : * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex"
5439 : * already held which will conflict with an existing lock order:
5440 : *
5441 : * mem_hotplug_lock->slab_mutex->kernfs_mutex
5442 : *
5443 : * We don't really need mem_hotplug_lock (to hold off
5444 : * slab_mem_going_offline_callback) here because slab's memory hot
5445 : * unplug code doesn't destroy the kmem_cache->node[] data.
5446 : */
5447 :
5448 : #ifdef CONFIG_SLUB_DEBUG
5449 0 : if (flags & SO_ALL) {
5450 : struct kmem_cache_node *n;
5451 :
5452 0 : for_each_kmem_cache_node(s, node, n) {
5453 :
5454 0 : if (flags & SO_TOTAL)
5455 0 : x = atomic_long_read(&n->total_objects);
5456 0 : else if (flags & SO_OBJECTS)
5457 0 : x = atomic_long_read(&n->total_objects) -
5458 0 : count_partial(n, count_free);
5459 : else
5460 0 : x = atomic_long_read(&n->nr_slabs);
5461 0 : total += x;
5462 0 : nodes[node] += x;
5463 : }
5464 :
5465 : } else
5466 : #endif
5467 0 : if (flags & SO_PARTIAL) {
5468 : struct kmem_cache_node *n;
5469 :
5470 0 : for_each_kmem_cache_node(s, node, n) {
5471 0 : if (flags & SO_TOTAL)
5472 0 : x = count_partial(n, count_total);
5473 0 : else if (flags & SO_OBJECTS)
5474 0 : x = count_partial(n, count_inuse);
5475 : else
5476 0 : x = n->nr_partial;
5477 0 : total += x;
5478 0 : nodes[node] += x;
5479 : }
5480 : }
5481 :
5482 0 : len += sysfs_emit_at(buf, len, "%lu", total);
5483 : #ifdef CONFIG_NUMA
5484 : for (node = 0; node < nr_node_ids; node++) {
5485 : if (nodes[node])
5486 : len += sysfs_emit_at(buf, len, " N%d=%lu",
5487 : node, nodes[node]);
5488 : }
5489 : #endif
5490 0 : len += sysfs_emit_at(buf, len, "\n");
5491 0 : kfree(nodes);
5492 :
5493 0 : return len;
5494 : }
5495 :
5496 : #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
5497 : #define to_slab(n) container_of(n, struct kmem_cache, kobj)
5498 :
5499 : struct slab_attribute {
5500 : struct attribute attr;
5501 : ssize_t (*show)(struct kmem_cache *s, char *buf);
5502 : ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
5503 : };
5504 :
5505 : #define SLAB_ATTR_RO(_name) \
5506 : static struct slab_attribute _name##_attr = __ATTR_RO_MODE(_name, 0400)
5507 :
5508 : #define SLAB_ATTR(_name) \
5509 : static struct slab_attribute _name##_attr = __ATTR_RW_MODE(_name, 0600)
5510 :
5511 0 : static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
5512 : {
5513 0 : return sysfs_emit(buf, "%u\n", s->size);
5514 : }
5515 : SLAB_ATTR_RO(slab_size);
5516 :
5517 0 : static ssize_t align_show(struct kmem_cache *s, char *buf)
5518 : {
5519 0 : return sysfs_emit(buf, "%u\n", s->align);
5520 : }
5521 : SLAB_ATTR_RO(align);
5522 :
5523 0 : static ssize_t object_size_show(struct kmem_cache *s, char *buf)
5524 : {
5525 0 : return sysfs_emit(buf, "%u\n", s->object_size);
5526 : }
5527 : SLAB_ATTR_RO(object_size);
5528 :
5529 0 : static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
5530 : {
5531 0 : return sysfs_emit(buf, "%u\n", oo_objects(s->oo));
5532 : }
5533 : SLAB_ATTR_RO(objs_per_slab);
5534 :
5535 0 : static ssize_t order_show(struct kmem_cache *s, char *buf)
5536 : {
5537 0 : return sysfs_emit(buf, "%u\n", oo_order(s->oo));
5538 : }
5539 : SLAB_ATTR_RO(order);
5540 :
5541 0 : static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
5542 : {
5543 0 : return sysfs_emit(buf, "%lu\n", s->min_partial);
5544 : }
5545 :
5546 0 : static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
5547 : size_t length)
5548 : {
5549 : unsigned long min;
5550 : int err;
5551 :
5552 0 : err = kstrtoul(buf, 10, &min);
5553 0 : if (err)
5554 0 : return err;
5555 :
5556 0 : s->min_partial = min;
5557 0 : return length;
5558 : }
5559 : SLAB_ATTR(min_partial);
5560 :
5561 0 : static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
5562 : {
5563 0 : unsigned int nr_partial = 0;
5564 : #ifdef CONFIG_SLUB_CPU_PARTIAL
5565 : nr_partial = s->cpu_partial;
5566 : #endif
5567 :
5568 0 : return sysfs_emit(buf, "%u\n", nr_partial);
5569 : }
5570 :
5571 0 : static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
5572 : size_t length)
5573 : {
5574 : unsigned int objects;
5575 : int err;
5576 :
5577 0 : err = kstrtouint(buf, 10, &objects);
5578 0 : if (err)
5579 0 : return err;
5580 0 : if (objects && !kmem_cache_has_cpu_partial(s))
5581 : return -EINVAL;
5582 :
5583 0 : slub_set_cpu_partial(s, objects);
5584 0 : flush_all(s);
5585 0 : return length;
5586 : }
5587 : SLAB_ATTR(cpu_partial);
5588 :
5589 0 : static ssize_t ctor_show(struct kmem_cache *s, char *buf)
5590 : {
5591 0 : if (!s->ctor)
5592 : return 0;
5593 0 : return sysfs_emit(buf, "%pS\n", s->ctor);
5594 : }
5595 : SLAB_ATTR_RO(ctor);
5596 :
5597 0 : static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5598 : {
5599 0 : return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
5600 : }
5601 : SLAB_ATTR_RO(aliases);
5602 :
5603 0 : static ssize_t partial_show(struct kmem_cache *s, char *buf)
5604 : {
5605 0 : return show_slab_objects(s, buf, SO_PARTIAL);
5606 : }
5607 : SLAB_ATTR_RO(partial);
5608 :
5609 0 : static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5610 : {
5611 0 : return show_slab_objects(s, buf, SO_CPU);
5612 : }
5613 : SLAB_ATTR_RO(cpu_slabs);
5614 :
5615 0 : static ssize_t objects_show(struct kmem_cache *s, char *buf)
5616 : {
5617 0 : return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
5618 : }
5619 : SLAB_ATTR_RO(objects);
5620 :
5621 0 : static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5622 : {
5623 0 : return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5624 : }
5625 : SLAB_ATTR_RO(objects_partial);
5626 :
5627 0 : static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5628 : {
5629 0 : int objects = 0;
5630 0 : int slabs = 0;
5631 : int cpu __maybe_unused;
5632 0 : int len = 0;
5633 :
5634 : #ifdef CONFIG_SLUB_CPU_PARTIAL
5635 : for_each_online_cpu(cpu) {
5636 : struct slab *slab;
5637 :
5638 : slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5639 :
5640 : if (slab)
5641 : slabs += slab->slabs;
5642 : }
5643 : #endif
5644 :
5645 : /* Approximate half-full slabs, see slub_set_cpu_partial() */
5646 0 : objects = (slabs * oo_objects(s->oo)) / 2;
5647 0 : len += sysfs_emit_at(buf, len, "%d(%d)", objects, slabs);
5648 :
5649 : #if defined(CONFIG_SLUB_CPU_PARTIAL) && defined(CONFIG_SMP)
5650 : for_each_online_cpu(cpu) {
5651 : struct slab *slab;
5652 :
5653 : slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5654 : if (slab) {
5655 : slabs = READ_ONCE(slab->slabs);
5656 : objects = (slabs * oo_objects(s->oo)) / 2;
5657 : len += sysfs_emit_at(buf, len, " C%d=%d(%d)",
5658 : cpu, objects, slabs);
5659 : }
5660 : }
5661 : #endif
5662 0 : len += sysfs_emit_at(buf, len, "\n");
5663 :
5664 0 : return len;
5665 : }
5666 : SLAB_ATTR_RO(slabs_cpu_partial);
5667 :
5668 0 : static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5669 : {
5670 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5671 : }
5672 : SLAB_ATTR_RO(reclaim_account);
5673 :
5674 0 : static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5675 : {
5676 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
5677 : }
5678 : SLAB_ATTR_RO(hwcache_align);
5679 :
5680 : #ifdef CONFIG_ZONE_DMA
5681 : static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5682 : {
5683 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5684 : }
5685 : SLAB_ATTR_RO(cache_dma);
5686 : #endif
5687 :
5688 : #ifdef CONFIG_HARDENED_USERCOPY
5689 : static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5690 : {
5691 : return sysfs_emit(buf, "%u\n", s->usersize);
5692 : }
5693 : SLAB_ATTR_RO(usersize);
5694 : #endif
5695 :
5696 0 : static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5697 : {
5698 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
5699 : }
5700 : SLAB_ATTR_RO(destroy_by_rcu);
5701 :
5702 : #ifdef CONFIG_SLUB_DEBUG
5703 0 : static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5704 : {
5705 0 : return show_slab_objects(s, buf, SO_ALL);
5706 : }
5707 : SLAB_ATTR_RO(slabs);
5708 :
5709 0 : static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5710 : {
5711 0 : return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5712 : }
5713 : SLAB_ATTR_RO(total_objects);
5714 :
5715 0 : static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5716 : {
5717 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
5718 : }
5719 : SLAB_ATTR_RO(sanity_checks);
5720 :
5721 0 : static ssize_t trace_show(struct kmem_cache *s, char *buf)
5722 : {
5723 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5724 : }
5725 : SLAB_ATTR_RO(trace);
5726 :
5727 0 : static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5728 : {
5729 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5730 : }
5731 :
5732 : SLAB_ATTR_RO(red_zone);
5733 :
5734 0 : static ssize_t poison_show(struct kmem_cache *s, char *buf)
5735 : {
5736 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON));
5737 : }
5738 :
5739 : SLAB_ATTR_RO(poison);
5740 :
5741 0 : static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5742 : {
5743 0 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5744 : }
5745 :
5746 : SLAB_ATTR_RO(store_user);
5747 :
5748 0 : static ssize_t validate_show(struct kmem_cache *s, char *buf)
5749 : {
5750 0 : return 0;
5751 : }
5752 :
5753 0 : static ssize_t validate_store(struct kmem_cache *s,
5754 : const char *buf, size_t length)
5755 : {
5756 0 : int ret = -EINVAL;
5757 :
5758 0 : if (buf[0] == '1' && kmem_cache_debug(s)) {
5759 0 : ret = validate_slab_cache(s);
5760 0 : if (ret >= 0)
5761 0 : ret = length;
5762 : }
5763 0 : return ret;
5764 : }
5765 : SLAB_ATTR(validate);
5766 :
5767 : #endif /* CONFIG_SLUB_DEBUG */
5768 :
5769 : #ifdef CONFIG_FAILSLAB
5770 : static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5771 : {
5772 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5773 : }
5774 :
5775 : static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5776 : size_t length)
5777 : {
5778 : if (s->refcount > 1)
5779 : return -EINVAL;
5780 :
5781 : if (buf[0] == '1')
5782 : WRITE_ONCE(s->flags, s->flags | SLAB_FAILSLAB);
5783 : else
5784 : WRITE_ONCE(s->flags, s->flags & ~SLAB_FAILSLAB);
5785 :
5786 : return length;
5787 : }
5788 : SLAB_ATTR(failslab);
5789 : #endif
5790 :
5791 0 : static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5792 : {
5793 0 : return 0;
5794 : }
5795 :
5796 0 : static ssize_t shrink_store(struct kmem_cache *s,
5797 : const char *buf, size_t length)
5798 : {
5799 0 : if (buf[0] == '1')
5800 0 : kmem_cache_shrink(s);
5801 : else
5802 : return -EINVAL;
5803 0 : return length;
5804 : }
5805 : SLAB_ATTR(shrink);
5806 :
5807 : #ifdef CONFIG_NUMA
5808 : static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
5809 : {
5810 : return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10);
5811 : }
5812 :
5813 : static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
5814 : const char *buf, size_t length)
5815 : {
5816 : unsigned int ratio;
5817 : int err;
5818 :
5819 : err = kstrtouint(buf, 10, &ratio);
5820 : if (err)
5821 : return err;
5822 : if (ratio > 100)
5823 : return -ERANGE;
5824 :
5825 : s->remote_node_defrag_ratio = ratio * 10;
5826 :
5827 : return length;
5828 : }
5829 : SLAB_ATTR(remote_node_defrag_ratio);
5830 : #endif
5831 :
5832 : #ifdef CONFIG_SLUB_STATS
5833 : static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5834 : {
5835 : unsigned long sum = 0;
5836 : int cpu;
5837 : int len = 0;
5838 : int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
5839 :
5840 : if (!data)
5841 : return -ENOMEM;
5842 :
5843 : for_each_online_cpu(cpu) {
5844 : unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
5845 :
5846 : data[cpu] = x;
5847 : sum += x;
5848 : }
5849 :
5850 : len += sysfs_emit_at(buf, len, "%lu", sum);
5851 :
5852 : #ifdef CONFIG_SMP
5853 : for_each_online_cpu(cpu) {
5854 : if (data[cpu])
5855 : len += sysfs_emit_at(buf, len, " C%d=%u",
5856 : cpu, data[cpu]);
5857 : }
5858 : #endif
5859 : kfree(data);
5860 : len += sysfs_emit_at(buf, len, "\n");
5861 :
5862 : return len;
5863 : }
5864 :
5865 : static void clear_stat(struct kmem_cache *s, enum stat_item si)
5866 : {
5867 : int cpu;
5868 :
5869 : for_each_online_cpu(cpu)
5870 : per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
5871 : }
5872 :
5873 : #define STAT_ATTR(si, text) \
5874 : static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5875 : { \
5876 : return show_stat(s, buf, si); \
5877 : } \
5878 : static ssize_t text##_store(struct kmem_cache *s, \
5879 : const char *buf, size_t length) \
5880 : { \
5881 : if (buf[0] != '0') \
5882 : return -EINVAL; \
5883 : clear_stat(s, si); \
5884 : return length; \
5885 : } \
5886 : SLAB_ATTR(text); \
5887 :
5888 : STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5889 : STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5890 : STAT_ATTR(FREE_FASTPATH, free_fastpath);
5891 : STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5892 : STAT_ATTR(FREE_FROZEN, free_frozen);
5893 : STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5894 : STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5895 : STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5896 : STAT_ATTR(ALLOC_SLAB, alloc_slab);
5897 : STAT_ATTR(ALLOC_REFILL, alloc_refill);
5898 : STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
5899 : STAT_ATTR(FREE_SLAB, free_slab);
5900 : STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5901 : STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5902 : STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5903 : STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5904 : STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5905 : STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
5906 : STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
5907 : STAT_ATTR(ORDER_FALLBACK, order_fallback);
5908 : STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5909 : STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
5910 : STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5911 : STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
5912 : STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5913 : STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
5914 : #endif /* CONFIG_SLUB_STATS */
5915 :
5916 : #ifdef CONFIG_KFENCE
5917 : static ssize_t skip_kfence_show(struct kmem_cache *s, char *buf)
5918 : {
5919 : return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_SKIP_KFENCE));
5920 : }
5921 :
5922 : static ssize_t skip_kfence_store(struct kmem_cache *s,
5923 : const char *buf, size_t length)
5924 : {
5925 : int ret = length;
5926 :
5927 : if (buf[0] == '0')
5928 : s->flags &= ~SLAB_SKIP_KFENCE;
5929 : else if (buf[0] == '1')
5930 : s->flags |= SLAB_SKIP_KFENCE;
5931 : else
5932 : ret = -EINVAL;
5933 :
5934 : return ret;
5935 : }
5936 : SLAB_ATTR(skip_kfence);
5937 : #endif
5938 :
5939 : static struct attribute *slab_attrs[] = {
5940 : &slab_size_attr.attr,
5941 : &object_size_attr.attr,
5942 : &objs_per_slab_attr.attr,
5943 : &order_attr.attr,
5944 : &min_partial_attr.attr,
5945 : &cpu_partial_attr.attr,
5946 : &objects_attr.attr,
5947 : &objects_partial_attr.attr,
5948 : &partial_attr.attr,
5949 : &cpu_slabs_attr.attr,
5950 : &ctor_attr.attr,
5951 : &aliases_attr.attr,
5952 : &align_attr.attr,
5953 : &hwcache_align_attr.attr,
5954 : &reclaim_account_attr.attr,
5955 : &destroy_by_rcu_attr.attr,
5956 : &shrink_attr.attr,
5957 : &slabs_cpu_partial_attr.attr,
5958 : #ifdef CONFIG_SLUB_DEBUG
5959 : &total_objects_attr.attr,
5960 : &slabs_attr.attr,
5961 : &sanity_checks_attr.attr,
5962 : &trace_attr.attr,
5963 : &red_zone_attr.attr,
5964 : &poison_attr.attr,
5965 : &store_user_attr.attr,
5966 : &validate_attr.attr,
5967 : #endif
5968 : #ifdef CONFIG_ZONE_DMA
5969 : &cache_dma_attr.attr,
5970 : #endif
5971 : #ifdef CONFIG_NUMA
5972 : &remote_node_defrag_ratio_attr.attr,
5973 : #endif
5974 : #ifdef CONFIG_SLUB_STATS
5975 : &alloc_fastpath_attr.attr,
5976 : &alloc_slowpath_attr.attr,
5977 : &free_fastpath_attr.attr,
5978 : &free_slowpath_attr.attr,
5979 : &free_frozen_attr.attr,
5980 : &free_add_partial_attr.attr,
5981 : &free_remove_partial_attr.attr,
5982 : &alloc_from_partial_attr.attr,
5983 : &alloc_slab_attr.attr,
5984 : &alloc_refill_attr.attr,
5985 : &alloc_node_mismatch_attr.attr,
5986 : &free_slab_attr.attr,
5987 : &cpuslab_flush_attr.attr,
5988 : &deactivate_full_attr.attr,
5989 : &deactivate_empty_attr.attr,
5990 : &deactivate_to_head_attr.attr,
5991 : &deactivate_to_tail_attr.attr,
5992 : &deactivate_remote_frees_attr.attr,
5993 : &deactivate_bypass_attr.attr,
5994 : &order_fallback_attr.attr,
5995 : &cmpxchg_double_fail_attr.attr,
5996 : &cmpxchg_double_cpu_fail_attr.attr,
5997 : &cpu_partial_alloc_attr.attr,
5998 : &cpu_partial_free_attr.attr,
5999 : &cpu_partial_node_attr.attr,
6000 : &cpu_partial_drain_attr.attr,
6001 : #endif
6002 : #ifdef CONFIG_FAILSLAB
6003 : &failslab_attr.attr,
6004 : #endif
6005 : #ifdef CONFIG_HARDENED_USERCOPY
6006 : &usersize_attr.attr,
6007 : #endif
6008 : #ifdef CONFIG_KFENCE
6009 : &skip_kfence_attr.attr,
6010 : #endif
6011 :
6012 : NULL
6013 : };
6014 :
6015 : static const struct attribute_group slab_attr_group = {
6016 : .attrs = slab_attrs,
6017 : };
6018 :
6019 0 : static ssize_t slab_attr_show(struct kobject *kobj,
6020 : struct attribute *attr,
6021 : char *buf)
6022 : {
6023 : struct slab_attribute *attribute;
6024 : struct kmem_cache *s;
6025 :
6026 0 : attribute = to_slab_attr(attr);
6027 0 : s = to_slab(kobj);
6028 :
6029 0 : if (!attribute->show)
6030 : return -EIO;
6031 :
6032 0 : return attribute->show(s, buf);
6033 : }
6034 :
6035 0 : static ssize_t slab_attr_store(struct kobject *kobj,
6036 : struct attribute *attr,
6037 : const char *buf, size_t len)
6038 : {
6039 : struct slab_attribute *attribute;
6040 : struct kmem_cache *s;
6041 :
6042 0 : attribute = to_slab_attr(attr);
6043 0 : s = to_slab(kobj);
6044 :
6045 0 : if (!attribute->store)
6046 : return -EIO;
6047 :
6048 0 : return attribute->store(s, buf, len);
6049 : }
6050 :
6051 0 : static void kmem_cache_release(struct kobject *k)
6052 : {
6053 0 : slab_kmem_cache_release(to_slab(k));
6054 0 : }
6055 :
6056 : static const struct sysfs_ops slab_sysfs_ops = {
6057 : .show = slab_attr_show,
6058 : .store = slab_attr_store,
6059 : };
6060 :
6061 : static const struct kobj_type slab_ktype = {
6062 : .sysfs_ops = &slab_sysfs_ops,
6063 : .release = kmem_cache_release,
6064 : };
6065 :
6066 : static struct kset *slab_kset;
6067 :
6068 : static inline struct kset *cache_kset(struct kmem_cache *s)
6069 : {
6070 53 : return slab_kset;
6071 : }
6072 :
6073 : #define ID_STR_LENGTH 32
6074 :
6075 : /* Create a unique string id for a slab cache:
6076 : *
6077 : * Format :[flags-]size
6078 : */
6079 41 : static char *create_unique_id(struct kmem_cache *s)
6080 : {
6081 41 : char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
6082 41 : char *p = name;
6083 :
6084 41 : if (!name)
6085 : return ERR_PTR(-ENOMEM);
6086 :
6087 41 : *p++ = ':';
6088 : /*
6089 : * First flags affecting slabcache operations. We will only
6090 : * get here for aliasable slabs so we do not need to support
6091 : * too many flags. The flags here must cover all flags that
6092 : * are matched during merging to guarantee that the id is
6093 : * unique.
6094 : */
6095 41 : if (s->flags & SLAB_CACHE_DMA)
6096 0 : *p++ = 'd';
6097 41 : if (s->flags & SLAB_CACHE_DMA32)
6098 0 : *p++ = 'D';
6099 41 : if (s->flags & SLAB_RECLAIM_ACCOUNT)
6100 14 : *p++ = 'a';
6101 41 : if (s->flags & SLAB_CONSISTENCY_CHECKS)
6102 0 : *p++ = 'F';
6103 : if (s->flags & SLAB_ACCOUNT)
6104 : *p++ = 'A';
6105 41 : if (p != name + 1)
6106 14 : *p++ = '-';
6107 41 : p += snprintf(p, ID_STR_LENGTH - (p - name), "%07u", s->size);
6108 :
6109 41 : if (WARN_ON(p > name + ID_STR_LENGTH - 1)) {
6110 0 : kfree(name);
6111 0 : return ERR_PTR(-EINVAL);
6112 : }
6113 : kmsan_unpoison_memory(name, p - name);
6114 : return name;
6115 : }
6116 :
6117 53 : static int sysfs_slab_add(struct kmem_cache *s)
6118 : {
6119 : int err;
6120 : const char *name;
6121 106 : struct kset *kset = cache_kset(s);
6122 53 : int unmergeable = slab_unmergeable(s);
6123 :
6124 53 : if (!unmergeable && disable_higher_order_debug &&
6125 0 : (slub_debug & DEBUG_METADATA_FLAGS))
6126 0 : unmergeable = 1;
6127 :
6128 53 : if (unmergeable) {
6129 : /*
6130 : * Slabcache can never be merged so we can use the name proper.
6131 : * This is typically the case for debug situations. In that
6132 : * case we can catch duplicate names easily.
6133 : */
6134 12 : sysfs_remove_link(&slab_kset->kobj, s->name);
6135 12 : name = s->name;
6136 : } else {
6137 : /*
6138 : * Create a unique name for the slab as a target
6139 : * for the symlinks.
6140 : */
6141 41 : name = create_unique_id(s);
6142 41 : if (IS_ERR(name))
6143 0 : return PTR_ERR(name);
6144 : }
6145 :
6146 53 : s->kobj.kset = kset;
6147 53 : err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
6148 53 : if (err)
6149 : goto out;
6150 :
6151 53 : err = sysfs_create_group(&s->kobj, &slab_attr_group);
6152 53 : if (err)
6153 : goto out_del_kobj;
6154 :
6155 53 : if (!unmergeable) {
6156 : /* Setup first alias */
6157 41 : sysfs_slab_alias(s, s->name);
6158 : }
6159 : out:
6160 53 : if (!unmergeable)
6161 41 : kfree(name);
6162 : return err;
6163 : out_del_kobj:
6164 0 : kobject_del(&s->kobj);
6165 0 : goto out;
6166 : }
6167 :
6168 0 : void sysfs_slab_unlink(struct kmem_cache *s)
6169 : {
6170 0 : if (slab_state >= FULL)
6171 0 : kobject_del(&s->kobj);
6172 0 : }
6173 :
6174 0 : void sysfs_slab_release(struct kmem_cache *s)
6175 : {
6176 0 : if (slab_state >= FULL)
6177 0 : kobject_put(&s->kobj);
6178 0 : }
6179 :
6180 : /*
6181 : * Need to buffer aliases during bootup until sysfs becomes
6182 : * available lest we lose that information.
6183 : */
6184 : struct saved_alias {
6185 : struct kmem_cache *s;
6186 : const char *name;
6187 : struct saved_alias *next;
6188 : };
6189 :
6190 : static struct saved_alias *alias_list;
6191 :
6192 105 : static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
6193 : {
6194 : struct saved_alias *al;
6195 :
6196 105 : if (slab_state == FULL) {
6197 : /*
6198 : * If we have a leftover link then remove it.
6199 : */
6200 73 : sysfs_remove_link(&slab_kset->kobj, name);
6201 73 : return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
6202 : }
6203 :
6204 32 : al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
6205 32 : if (!al)
6206 : return -ENOMEM;
6207 :
6208 32 : al->s = s;
6209 32 : al->name = name;
6210 32 : al->next = alias_list;
6211 32 : alias_list = al;
6212 32 : kmsan_unpoison_memory(al, sizeof(*al));
6213 32 : return 0;
6214 : }
6215 :
6216 1 : static int __init slab_sysfs_init(void)
6217 : {
6218 : struct kmem_cache *s;
6219 : int err;
6220 :
6221 1 : mutex_lock(&slab_mutex);
6222 :
6223 1 : slab_kset = kset_create_and_add("slab", NULL, kernel_kobj);
6224 1 : if (!slab_kset) {
6225 0 : mutex_unlock(&slab_mutex);
6226 0 : pr_err("Cannot register slab subsystem.\n");
6227 0 : return -ENOSYS;
6228 : }
6229 :
6230 1 : slab_state = FULL;
6231 :
6232 54 : list_for_each_entry(s, &slab_caches, list) {
6233 53 : err = sysfs_slab_add(s);
6234 53 : if (err)
6235 0 : pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
6236 : s->name);
6237 : }
6238 :
6239 33 : while (alias_list) {
6240 32 : struct saved_alias *al = alias_list;
6241 :
6242 32 : alias_list = alias_list->next;
6243 32 : err = sysfs_slab_alias(al->s, al->name);
6244 32 : if (err)
6245 0 : pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
6246 : al->name);
6247 32 : kfree(al);
6248 : }
6249 :
6250 1 : mutex_unlock(&slab_mutex);
6251 1 : return 0;
6252 : }
6253 : late_initcall(slab_sysfs_init);
6254 : #endif /* SLAB_SUPPORTS_SYSFS */
6255 :
6256 : #if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS)
6257 : static int slab_debugfs_show(struct seq_file *seq, void *v)
6258 : {
6259 : struct loc_track *t = seq->private;
6260 : struct location *l;
6261 : unsigned long idx;
6262 :
6263 : idx = (unsigned long) t->idx;
6264 : if (idx < t->count) {
6265 : l = &t->loc[idx];
6266 :
6267 : seq_printf(seq, "%7ld ", l->count);
6268 :
6269 : if (l->addr)
6270 : seq_printf(seq, "%pS", (void *)l->addr);
6271 : else
6272 : seq_puts(seq, "<not-available>");
6273 :
6274 : if (l->waste)
6275 : seq_printf(seq, " waste=%lu/%lu",
6276 : l->count * l->waste, l->waste);
6277 :
6278 : if (l->sum_time != l->min_time) {
6279 : seq_printf(seq, " age=%ld/%llu/%ld",
6280 : l->min_time, div_u64(l->sum_time, l->count),
6281 : l->max_time);
6282 : } else
6283 : seq_printf(seq, " age=%ld", l->min_time);
6284 :
6285 : if (l->min_pid != l->max_pid)
6286 : seq_printf(seq, " pid=%ld-%ld", l->min_pid, l->max_pid);
6287 : else
6288 : seq_printf(seq, " pid=%ld",
6289 : l->min_pid);
6290 :
6291 : if (num_online_cpus() > 1 && !cpumask_empty(to_cpumask(l->cpus)))
6292 : seq_printf(seq, " cpus=%*pbl",
6293 : cpumask_pr_args(to_cpumask(l->cpus)));
6294 :
6295 : if (nr_online_nodes > 1 && !nodes_empty(l->nodes))
6296 : seq_printf(seq, " nodes=%*pbl",
6297 : nodemask_pr_args(&l->nodes));
6298 :
6299 : #ifdef CONFIG_STACKDEPOT
6300 : {
6301 : depot_stack_handle_t handle;
6302 : unsigned long *entries;
6303 : unsigned int nr_entries, j;
6304 :
6305 : handle = READ_ONCE(l->handle);
6306 : if (handle) {
6307 : nr_entries = stack_depot_fetch(handle, &entries);
6308 : seq_puts(seq, "\n");
6309 : for (j = 0; j < nr_entries; j++)
6310 : seq_printf(seq, " %pS\n", (void *)entries[j]);
6311 : }
6312 : }
6313 : #endif
6314 : seq_puts(seq, "\n");
6315 : }
6316 :
6317 : if (!idx && !t->count)
6318 : seq_puts(seq, "No data\n");
6319 :
6320 : return 0;
6321 : }
6322 :
6323 : static void slab_debugfs_stop(struct seq_file *seq, void *v)
6324 : {
6325 : }
6326 :
6327 : static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos)
6328 : {
6329 : struct loc_track *t = seq->private;
6330 :
6331 : t->idx = ++(*ppos);
6332 : if (*ppos <= t->count)
6333 : return ppos;
6334 :
6335 : return NULL;
6336 : }
6337 :
6338 : static int cmp_loc_by_count(const void *a, const void *b, const void *data)
6339 : {
6340 : struct location *loc1 = (struct location *)a;
6341 : struct location *loc2 = (struct location *)b;
6342 :
6343 : if (loc1->count > loc2->count)
6344 : return -1;
6345 : else
6346 : return 1;
6347 : }
6348 :
6349 : static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos)
6350 : {
6351 : struct loc_track *t = seq->private;
6352 :
6353 : t->idx = *ppos;
6354 : return ppos;
6355 : }
6356 :
6357 : static const struct seq_operations slab_debugfs_sops = {
6358 : .start = slab_debugfs_start,
6359 : .next = slab_debugfs_next,
6360 : .stop = slab_debugfs_stop,
6361 : .show = slab_debugfs_show,
6362 : };
6363 :
6364 : static int slab_debug_trace_open(struct inode *inode, struct file *filep)
6365 : {
6366 :
6367 : struct kmem_cache_node *n;
6368 : enum track_item alloc;
6369 : int node;
6370 : struct loc_track *t = __seq_open_private(filep, &slab_debugfs_sops,
6371 : sizeof(struct loc_track));
6372 : struct kmem_cache *s = file_inode(filep)->i_private;
6373 : unsigned long *obj_map;
6374 :
6375 : if (!t)
6376 : return -ENOMEM;
6377 :
6378 : obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL);
6379 : if (!obj_map) {
6380 : seq_release_private(inode, filep);
6381 : return -ENOMEM;
6382 : }
6383 :
6384 : if (strcmp(filep->f_path.dentry->d_name.name, "alloc_traces") == 0)
6385 : alloc = TRACK_ALLOC;
6386 : else
6387 : alloc = TRACK_FREE;
6388 :
6389 : if (!alloc_loc_track(t, PAGE_SIZE / sizeof(struct location), GFP_KERNEL)) {
6390 : bitmap_free(obj_map);
6391 : seq_release_private(inode, filep);
6392 : return -ENOMEM;
6393 : }
6394 :
6395 : for_each_kmem_cache_node(s, node, n) {
6396 : unsigned long flags;
6397 : struct slab *slab;
6398 :
6399 : if (!atomic_long_read(&n->nr_slabs))
6400 : continue;
6401 :
6402 : spin_lock_irqsave(&n->list_lock, flags);
6403 : list_for_each_entry(slab, &n->partial, slab_list)
6404 : process_slab(t, s, slab, alloc, obj_map);
6405 : list_for_each_entry(slab, &n->full, slab_list)
6406 : process_slab(t, s, slab, alloc, obj_map);
6407 : spin_unlock_irqrestore(&n->list_lock, flags);
6408 : }
6409 :
6410 : /* Sort locations by count */
6411 : sort_r(t->loc, t->count, sizeof(struct location),
6412 : cmp_loc_by_count, NULL, NULL);
6413 :
6414 : bitmap_free(obj_map);
6415 : return 0;
6416 : }
6417 :
6418 : static int slab_debug_trace_release(struct inode *inode, struct file *file)
6419 : {
6420 : struct seq_file *seq = file->private_data;
6421 : struct loc_track *t = seq->private;
6422 :
6423 : free_loc_track(t);
6424 : return seq_release_private(inode, file);
6425 : }
6426 :
6427 : static const struct file_operations slab_debugfs_fops = {
6428 : .open = slab_debug_trace_open,
6429 : .read = seq_read,
6430 : .llseek = seq_lseek,
6431 : .release = slab_debug_trace_release,
6432 : };
6433 :
6434 : static void debugfs_slab_add(struct kmem_cache *s)
6435 : {
6436 : struct dentry *slab_cache_dir;
6437 :
6438 : if (unlikely(!slab_debugfs_root))
6439 : return;
6440 :
6441 : slab_cache_dir = debugfs_create_dir(s->name, slab_debugfs_root);
6442 :
6443 : debugfs_create_file("alloc_traces", 0400,
6444 : slab_cache_dir, s, &slab_debugfs_fops);
6445 :
6446 : debugfs_create_file("free_traces", 0400,
6447 : slab_cache_dir, s, &slab_debugfs_fops);
6448 : }
6449 :
6450 : void debugfs_slab_release(struct kmem_cache *s)
6451 : {
6452 : debugfs_lookup_and_remove(s->name, slab_debugfs_root);
6453 : }
6454 :
6455 : static int __init slab_debugfs_init(void)
6456 : {
6457 : struct kmem_cache *s;
6458 :
6459 : slab_debugfs_root = debugfs_create_dir("slab", NULL);
6460 :
6461 : list_for_each_entry(s, &slab_caches, list)
6462 : if (s->flags & SLAB_STORE_USER)
6463 : debugfs_slab_add(s);
6464 :
6465 : return 0;
6466 :
6467 : }
6468 : __initcall(slab_debugfs_init);
6469 : #endif
6470 : /*
6471 : * The /proc/slabinfo ABI
6472 : */
6473 : #ifdef CONFIG_SLUB_DEBUG
6474 0 : void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
6475 : {
6476 0 : unsigned long nr_slabs = 0;
6477 0 : unsigned long nr_objs = 0;
6478 0 : unsigned long nr_free = 0;
6479 : int node;
6480 : struct kmem_cache_node *n;
6481 :
6482 0 : for_each_kmem_cache_node(s, node, n) {
6483 0 : nr_slabs += node_nr_slabs(n);
6484 0 : nr_objs += node_nr_objs(n);
6485 0 : nr_free += count_partial(n, count_free);
6486 : }
6487 :
6488 0 : sinfo->active_objs = nr_objs - nr_free;
6489 0 : sinfo->num_objs = nr_objs;
6490 0 : sinfo->active_slabs = nr_slabs;
6491 0 : sinfo->num_slabs = nr_slabs;
6492 0 : sinfo->objects_per_slab = oo_objects(s->oo);
6493 0 : sinfo->cache_order = oo_order(s->oo);
6494 0 : }
6495 :
6496 0 : void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
6497 : {
6498 0 : }
6499 :
6500 0 : ssize_t slabinfo_write(struct file *file, const char __user *buffer,
6501 : size_t count, loff_t *ppos)
6502 : {
6503 0 : return -EIO;
6504 : }
6505 : #endif /* CONFIG_SLUB_DEBUG */
|
