~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/slub.c

Version: ~ [ linux-5.12-rc1 ] ~ [ linux-5.11.2 ] ~ [ linux-5.10.19 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.101 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.177 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.222 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.258 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.258 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 operatios
  7  * and only uses a centralized lock to manage a pool of partial slabs.
  8  *
  9  * (C) 2007 SGI, Christoph Lameter
 10  * (C) 2011 Linux Foundation, Christoph Lameter
 11  */
 12 
 13 #include <linux/mm.h>
 14 #include <linux/swap.h> /* struct reclaim_state */
 15 #include <linux/module.h>
 16 #include <linux/bit_spinlock.h>
 17 #include <linux/interrupt.h>
 18 #include <linux/bitops.h>
 19 #include <linux/slab.h>
 20 #include "slab.h"
 21 #include <linux/proc_fs.h>
 22 #include <linux/seq_file.h>
 23 #include <linux/kasan.h>
 24 #include <linux/cpu.h>
 25 #include <linux/cpuset.h>
 26 #include <linux/mempolicy.h>
 27 #include <linux/ctype.h>
 28 #include <linux/debugobjects.h>
 29 #include <linux/kallsyms.h>
 30 #include <linux/memory.h>
 31 #include <linux/math64.h>
 32 #include <linux/fault-inject.h>
 33 #include <linux/stacktrace.h>
 34 #include <linux/prefetch.h>
 35 #include <linux/memcontrol.h>
 36 #include <linux/random.h>
 37 
 38 #include <trace/events/kmem.h>
 39 
 40 #include "internal.h"
 41 
 42 /*
 43  * Lock order:
 44  *   1. slab_mutex (Global Mutex)
 45  *   2. node->list_lock
 46  *   3. slab_lock(page) (Only on some arches and for debugging)
 47  *
 48  *   slab_mutex
 49  *
 50  *   The role of the slab_mutex is to protect the list of all the slabs
 51  *   and to synchronize major metadata changes to slab cache structures.
 52  *
 53  *   The slab_lock is only used for debugging and on arches that do not
 54  *   have the ability to do a cmpxchg_double. It only protects:
 55  *      A. page->freelist       -> List of object free in a page
 56  *      B. page->inuse          -> Number of objects in use
 57  *      C. page->objects        -> Number of objects in page
 58  *      D. page->frozen         -> frozen state
 59  *
 60  *   If a slab is frozen then it is exempt from list management. It is not
 61  *   on any list. The processor that froze the slab is the one who can
 62  *   perform list operations on the page. Other processors may put objects
 63  *   onto the freelist but the processor that froze the slab is the only
 64  *   one that can retrieve the objects from the page's freelist.
 65  *
 66  *   The list_lock protects the partial and full list on each node and
 67  *   the partial slab counter. If taken then no new slabs may be added or
 68  *   removed from the lists nor make the number of partial slabs be modified.
 69  *   (Note that the total number of slabs is an atomic value that may be
 70  *   modified without taking the list lock).
 71  *
 72  *   The list_lock is a centralized lock and thus we avoid taking it as
 73  *   much as possible. As long as SLUB does not have to handle partial
 74  *   slabs, operations can continue without any centralized lock. F.e.
 75  *   allocating a long series of objects that fill up slabs does not require
 76  *   the list lock.
 77  *   Interrupts are disabled during allocation and deallocation in order to
 78  *   make the slab allocator safe to use in the context of an irq. In addition
 79  *   interrupts are disabled to ensure that the processor does not change
 80  *   while handling per_cpu slabs, due to kernel preemption.
 81  *
 82  * SLUB assigns one slab for allocation to each processor.
 83  * Allocations only occur from these slabs called cpu slabs.
 84  *
 85  * Slabs with free elements are kept on a partial list and during regular
 86  * operations no list for full slabs is used. If an object in a full slab is
 87  * freed then the slab will show up again on the partial lists.
 88  * We track full slabs for debugging purposes though because otherwise we
 89  * cannot scan all objects.
 90  *
 91  * Slabs are freed when they become empty. Teardown and setup is
 92  * minimal so we rely on the page allocators per cpu caches for
 93  * fast frees and allocs.
 94  *
 95  * Overloading of page flags that are otherwise used for LRU management.
 96  *
 97  * PageActive           The slab is frozen and exempt from list processing.
 98  *                      This means that the slab is dedicated to a purpose
 99  *                      such as satisfying allocations for a specific
100  *                      processor. Objects may be freed in the slab while
101  *                      it is frozen but slab_free will then skip the usual
102  *                      list operations. It is up to the processor holding
103  *                      the slab to integrate the slab into the slab lists
104  *                      when the slab is no longer needed.
105  *
106  *                      One use of this flag is to mark slabs that are
107  *                      used for allocations. Then such a slab becomes a cpu
108  *                      slab. The cpu slab may be equipped with an additional
109  *                      freelist that allows lockless access to
110  *                      free objects in addition to the regular freelist
111  *                      that requires the slab lock.
112  *
113  * PageError            Slab requires special handling due to debug
114  *                      options set. This moves slab handling out of
115  *                      the fast path and disables lockless freelists.
116  */
117 
118 static inline int kmem_cache_debug(struct kmem_cache *s)
119 {
120 #ifdef CONFIG_SLUB_DEBUG
121         return unlikely(s->flags & SLAB_DEBUG_FLAGS);
122 #else
123         return 0;
124 #endif
125 }
126 
127 void *fixup_red_left(struct kmem_cache *s, void *p)
128 {
129         if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE)
130                 p += s->red_left_pad;
131 
132         return p;
133 }
134 
135 static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
136 {
137 #ifdef CONFIG_SLUB_CPU_PARTIAL
138         return !kmem_cache_debug(s);
139 #else
140         return false;
141 #endif
142 }
143 
144 /*
145  * Issues still to be resolved:
146  *
147  * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
148  *
149  * - Variable sizing of the per node arrays
150  */
151 
152 /* Enable to test recovery from slab corruption on boot */
153 #undef SLUB_RESILIENCY_TEST
154 
155 /* Enable to log cmpxchg failures */
156 #undef SLUB_DEBUG_CMPXCHG
157 
158 /*
159  * Mininum number of partial slabs. These will be left on the partial
160  * lists even if they are empty. kmem_cache_shrink may reclaim them.
161  */
162 #define MIN_PARTIAL 5
163 
164 /*
165  * Maximum number of desirable partial slabs.
166  * The existence of more partial slabs makes kmem_cache_shrink
167  * sort the partial list by the number of objects in use.
168  */
169 #define MAX_PARTIAL 10
170 
171 #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
172                                 SLAB_POISON | SLAB_STORE_USER)
173 
174 /*
175  * These debug flags cannot use CMPXCHG because there might be consistency
176  * issues when checking or reading debug information
177  */
178 #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
179                                 SLAB_TRACE)
180 
181 
182 /*
183  * Debugging flags that require metadata to be stored in the slab.  These get
184  * disabled when slub_debug=O is used and a cache's min order increases with
185  * metadata.
186  */
187 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
188 
189 #define OO_SHIFT        16
190 #define OO_MASK         ((1 << OO_SHIFT) - 1)
191 #define MAX_OBJS_PER_PAGE       32767 /* since page.objects is u15 */
192 
193 /* Internal SLUB flags */
194 /* Poison object */
195 #define __OBJECT_POISON         ((slab_flags_t __force)0x80000000U)
196 /* Use cmpxchg_double */
197 #define __CMPXCHG_DOUBLE        ((slab_flags_t __force)0x40000000U)
198 
199 /*
200  * Tracking user of a slab.
201  */
202 #define TRACK_ADDRS_COUNT 16
203 struct track {
204         unsigned long addr;     /* Called from address */
205 #ifdef CONFIG_STACKTRACE
206         unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
207 #endif
208         int cpu;                /* Was running on cpu */
209         int pid;                /* Pid context */
210         unsigned long when;     /* When did the operation occur */
211 };
212 
213 enum track_item { TRACK_ALLOC, TRACK_FREE };
214 
215 #ifdef CONFIG_SYSFS
216 static int sysfs_slab_add(struct kmem_cache *);
217 static int sysfs_slab_alias(struct kmem_cache *, const char *);
218 static void memcg_propagate_slab_attrs(struct kmem_cache *s);
219 static void sysfs_slab_remove(struct kmem_cache *s);
220 #else
221 static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
222 static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
223                                                         { return 0; }
224 static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
225 static inline void sysfs_slab_remove(struct kmem_cache *s) { }
226 #endif
227 
228 static inline void stat(const struct kmem_cache *s, enum stat_item si)
229 {
230 #ifdef CONFIG_SLUB_STATS
231         /*
232          * The rmw is racy on a preemptible kernel but this is acceptable, so
233          * avoid this_cpu_add()'s irq-disable overhead.
234          */
235         raw_cpu_inc(s->cpu_slab->stat[si]);
236 #endif
237 }
238 
239 /********************************************************************
240  *                      Core slab cache functions
241  *******************************************************************/
242 
243 /*
244  * Returns freelist pointer (ptr). With hardening, this is obfuscated
245  * with an XOR of the address where the pointer is held and a per-cache
246  * random number.
247  */
248 static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
249                                  unsigned long ptr_addr)
250 {
251 #ifdef CONFIG_SLAB_FREELIST_HARDENED
252         return (void *)((unsigned long)ptr ^ s->random ^ ptr_addr);
253 #else
254         return ptr;
255 #endif
256 }
257 
258 /* Returns the freelist pointer recorded at location ptr_addr. */
259 static inline void *freelist_dereference(const struct kmem_cache *s,
260                                          void *ptr_addr)
261 {
262         return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr),
263                             (unsigned long)ptr_addr);
264 }
265 
266 static inline void *get_freepointer(struct kmem_cache *s, void *object)
267 {
268         return freelist_dereference(s, object + s->offset);
269 }
270 
271 static void prefetch_freepointer(const struct kmem_cache *s, void *object)
272 {
273         prefetch(object + s->offset);
274 }
275 
276 static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
277 {
278         unsigned long freepointer_addr;
279         void *p;
280 
281         if (!debug_pagealloc_enabled())
282                 return get_freepointer(s, object);
283 
284         freepointer_addr = (unsigned long)object + s->offset;
285         probe_kernel_read(&p, (void **)freepointer_addr, sizeof(p));
286         return freelist_ptr(s, p, freepointer_addr);
287 }
288 
289 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
290 {
291         unsigned long freeptr_addr = (unsigned long)object + s->offset;
292 
293 #ifdef CONFIG_SLAB_FREELIST_HARDENED
294         BUG_ON(object == fp); /* naive detection of double free or corruption */
295 #endif
296 
297         *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
298 }
299 
300 /* Loop over all objects in a slab */
301 #define for_each_object(__p, __s, __addr, __objects) \
302         for (__p = fixup_red_left(__s, __addr); \
303                 __p < (__addr) + (__objects) * (__s)->size; \
304                 __p += (__s)->size)
305 
306 #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \
307         for (__p = fixup_red_left(__s, __addr), __idx = 1; \
308                 __idx <= __objects; \
309                 __p += (__s)->size, __idx++)
310 
311 /* Determine object index from a given position */
312 static inline unsigned int slab_index(void *p, struct kmem_cache *s, void *addr)
313 {
314         return (p - addr) / s->size;
315 }
316 
317 static inline unsigned int order_objects(unsigned int order, unsigned int size)
318 {
319         return ((unsigned int)PAGE_SIZE << order) / size;
320 }
321 
322 static inline struct kmem_cache_order_objects oo_make(unsigned int order,
323                 unsigned int size)
324 {
325         struct kmem_cache_order_objects x = {
326                 (order << OO_SHIFT) + order_objects(order, size)
327         };
328 
329         return x;
330 }
331 
332 static inline unsigned int oo_order(struct kmem_cache_order_objects x)
333 {
334         return x.x >> OO_SHIFT;
335 }
336 
337 static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
338 {
339         return x.x & OO_MASK;
340 }
341 
342 /*
343  * Per slab locking using the pagelock
344  */
345 static __always_inline void slab_lock(struct page *page)
346 {
347         VM_BUG_ON_PAGE(PageTail(page), page);
348         bit_spin_lock(PG_locked, &page->flags);
349 }
350 
351 static __always_inline void slab_unlock(struct page *page)
352 {
353         VM_BUG_ON_PAGE(PageTail(page), page);
354         __bit_spin_unlock(PG_locked, &page->flags);
355 }
356 
357 /* Interrupts must be disabled (for the fallback code to work right) */
358 static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
359                 void *freelist_old, unsigned long counters_old,
360                 void *freelist_new, unsigned long counters_new,
361                 const char *n)
362 {
363         VM_BUG_ON(!irqs_disabled());
364 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
365     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
366         if (s->flags & __CMPXCHG_DOUBLE) {
367                 if (cmpxchg_double(&page->freelist, &page->counters,
368                                    freelist_old, counters_old,
369                                    freelist_new, counters_new))
370                         return true;
371         } else
372 #endif
373         {
374                 slab_lock(page);
375                 if (page->freelist == freelist_old &&
376                                         page->counters == counters_old) {
377                         page->freelist = freelist_new;
378                         page->counters = counters_new;
379                         slab_unlock(page);
380                         return true;
381                 }
382                 slab_unlock(page);
383         }
384 
385         cpu_relax();
386         stat(s, CMPXCHG_DOUBLE_FAIL);
387 
388 #ifdef SLUB_DEBUG_CMPXCHG
389         pr_info("%s %s: cmpxchg double redo ", n, s->name);
390 #endif
391 
392         return false;
393 }
394 
395 static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
396                 void *freelist_old, unsigned long counters_old,
397                 void *freelist_new, unsigned long counters_new,
398                 const char *n)
399 {
400 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
401     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
402         if (s->flags & __CMPXCHG_DOUBLE) {
403                 if (cmpxchg_double(&page->freelist, &page->counters,
404                                    freelist_old, counters_old,
405                                    freelist_new, counters_new))
406                         return true;
407         } else
408 #endif
409         {
410                 unsigned long flags;
411 
412                 local_irq_save(flags);
413                 slab_lock(page);
414                 if (page->freelist == freelist_old &&
415                                         page->counters == counters_old) {
416                         page->freelist = freelist_new;
417                         page->counters = counters_new;
418                         slab_unlock(page);
419                         local_irq_restore(flags);
420                         return true;
421                 }
422                 slab_unlock(page);
423                 local_irq_restore(flags);
424         }
425 
426         cpu_relax();
427         stat(s, CMPXCHG_DOUBLE_FAIL);
428 
429 #ifdef SLUB_DEBUG_CMPXCHG
430         pr_info("%s %s: cmpxchg double redo ", n, s->name);
431 #endif
432 
433         return false;
434 }
435 
436 #ifdef CONFIG_SLUB_DEBUG
437 /*
438  * Determine a map of object in use on a page.
439  *
440  * Node listlock must be held to guarantee that the page does
441  * not vanish from under us.
442  */
443 static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
444 {
445         void *p;
446         void *addr = page_address(page);
447 
448         for (p = page->freelist; p; p = get_freepointer(s, p))
449                 set_bit(slab_index(p, s, addr), map);
450 }
451 
452 static inline unsigned int size_from_object(struct kmem_cache *s)
453 {
454         if (s->flags & SLAB_RED_ZONE)
455                 return s->size - s->red_left_pad;
456 
457         return s->size;
458 }
459 
460 static inline void *restore_red_left(struct kmem_cache *s, void *p)
461 {
462         if (s->flags & SLAB_RED_ZONE)
463                 p -= s->red_left_pad;
464 
465         return p;
466 }
467 
468 /*
469  * Debug settings:
470  */
471 #if defined(CONFIG_SLUB_DEBUG_ON)
472 static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
473 #else
474 static slab_flags_t slub_debug;
475 #endif
476 
477 static char *slub_debug_slabs;
478 static int disable_higher_order_debug;
479 
480 /*
481  * slub is about to manipulate internal object metadata.  This memory lies
482  * outside the range of the allocated object, so accessing it would normally
483  * be reported by kasan as a bounds error.  metadata_access_enable() is used
484  * to tell kasan that these accesses are OK.
485  */
486 static inline void metadata_access_enable(void)
487 {
488         kasan_disable_current();
489 }
490 
491 static inline void metadata_access_disable(void)
492 {
493         kasan_enable_current();
494 }
495 
496 /*
497  * Object debugging
498  */
499 
500 /* Verify that a pointer has an address that is valid within a slab page */
501 static inline int check_valid_pointer(struct kmem_cache *s,
502                                 struct page *page, void *object)
503 {
504         void *base;
505 
506         if (!object)
507                 return 1;
508 
509         base = page_address(page);
510         object = restore_red_left(s, object);
511         if (object < base || object >= base + page->objects * s->size ||
512                 (object - base) % s->size) {
513                 return 0;
514         }
515 
516         return 1;
517 }
518 
519 static void print_section(char *level, char *text, u8 *addr,
520                           unsigned int length)
521 {
522         metadata_access_enable();
523         print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
524                         length, 1);
525         metadata_access_disable();
526 }
527 
528 static struct track *get_track(struct kmem_cache *s, void *object,
529         enum track_item alloc)
530 {
531         struct track *p;
532 
533         if (s->offset)
534                 p = object + s->offset + sizeof(void *);
535         else
536                 p = object + s->inuse;
537 
538         return p + alloc;
539 }
540 
541 static void set_track(struct kmem_cache *s, void *object,
542                         enum track_item alloc, unsigned long addr)
543 {
544         struct track *p = get_track(s, object, alloc);
545 
546         if (addr) {
547 #ifdef CONFIG_STACKTRACE
548                 struct stack_trace trace;
549                 int i;
550 
551                 trace.nr_entries = 0;
552                 trace.max_entries = TRACK_ADDRS_COUNT;
553                 trace.entries = p->addrs;
554                 trace.skip = 3;
555                 metadata_access_enable();
556                 save_stack_trace(&trace);
557                 metadata_access_disable();
558 
559                 /* See rant in lockdep.c */
560                 if (trace.nr_entries != 0 &&
561                     trace.entries[trace.nr_entries - 1] == ULONG_MAX)
562                         trace.nr_entries--;
563 
564                 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
565                         p->addrs[i] = 0;
566 #endif
567                 p->addr = addr;
568                 p->cpu = smp_processor_id();
569                 p->pid = current->pid;
570                 p->when = jiffies;
571         } else
572                 memset(p, 0, sizeof(struct track));
573 }
574 
575 static void init_tracking(struct kmem_cache *s, void *object)
576 {
577         if (!(s->flags & SLAB_STORE_USER))
578                 return;
579 
580         set_track(s, object, TRACK_FREE, 0UL);
581         set_track(s, object, TRACK_ALLOC, 0UL);
582 }
583 
584 static void print_track(const char *s, struct track *t, unsigned long pr_time)
585 {
586         if (!t->addr)
587                 return;
588 
589         pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
590                s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
591 #ifdef CONFIG_STACKTRACE
592         {
593                 int i;
594                 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
595                         if (t->addrs[i])
596                                 pr_err("\t%pS\n", (void *)t->addrs[i]);
597                         else
598                                 break;
599         }
600 #endif
601 }
602 
603 static void print_tracking(struct kmem_cache *s, void *object)
604 {
605         unsigned long pr_time = jiffies;
606         if (!(s->flags & SLAB_STORE_USER))
607                 return;
608 
609         print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
610         print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
611 }
612 
613 static void print_page_info(struct page *page)
614 {
615         pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
616                page, page->objects, page->inuse, page->freelist, page->flags);
617 
618 }
619 
620 static void slab_bug(struct kmem_cache *s, char *fmt, ...)
621 {
622         struct va_format vaf;
623         va_list args;
624 
625         va_start(args, fmt);
626         vaf.fmt = fmt;
627         vaf.va = &args;
628         pr_err("=============================================================================\n");
629         pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
630         pr_err("-----------------------------------------------------------------------------\n\n");
631 
632         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
633         va_end(args);
634 }
635 
636 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
637 {
638         struct va_format vaf;
639         va_list args;
640 
641         va_start(args, fmt);
642         vaf.fmt = fmt;
643         vaf.va = &args;
644         pr_err("FIX %s: %pV\n", s->name, &vaf);
645         va_end(args);
646 }
647 
648 static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
649 {
650         unsigned int off;       /* Offset of last byte */
651         u8 *addr = page_address(page);
652 
653         print_tracking(s, p);
654 
655         print_page_info(page);
656 
657         pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
658                p, p - addr, get_freepointer(s, p));
659 
660         if (s->flags & SLAB_RED_ZONE)
661                 print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
662                               s->red_left_pad);
663         else if (p > addr + 16)
664                 print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
665 
666         print_section(KERN_ERR, "Object ", p,
667                       min_t(unsigned int, s->object_size, PAGE_SIZE));
668         if (s->flags & SLAB_RED_ZONE)
669                 print_section(KERN_ERR, "Redzone ", p + s->object_size,
670                         s->inuse - s->object_size);
671 
672         if (s->offset)
673                 off = s->offset + sizeof(void *);
674         else
675                 off = s->inuse;
676 
677         if (s->flags & SLAB_STORE_USER)
678                 off += 2 * sizeof(struct track);
679 
680         off += kasan_metadata_size(s);
681 
682         if (off != size_from_object(s))
683                 /* Beginning of the filler is the free pointer */
684                 print_section(KERN_ERR, "Padding ", p + off,
685                               size_from_object(s) - off);
686 
687         dump_stack();
688 }
689 
690 void object_err(struct kmem_cache *s, struct page *page,
691                         u8 *object, char *reason)
692 {
693         slab_bug(s, "%s", reason);
694         print_trailer(s, page, object);
695 }
696 
697 static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page,
698                         const char *fmt, ...)
699 {
700         va_list args;
701         char buf[100];
702 
703         va_start(args, fmt);
704         vsnprintf(buf, sizeof(buf), fmt, args);
705         va_end(args);
706         slab_bug(s, "%s", buf);
707         print_page_info(page);
708         dump_stack();
709 }
710 
711 static void init_object(struct kmem_cache *s, void *object, u8 val)
712 {
713         u8 *p = object;
714 
715         if (s->flags & SLAB_RED_ZONE)
716                 memset(p - s->red_left_pad, val, s->red_left_pad);
717 
718         if (s->flags & __OBJECT_POISON) {
719                 memset(p, POISON_FREE, s->object_size - 1);
720                 p[s->object_size - 1] = POISON_END;
721         }
722 
723         if (s->flags & SLAB_RED_ZONE)
724                 memset(p + s->object_size, val, s->inuse - s->object_size);
725 }
726 
727 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
728                                                 void *from, void *to)
729 {
730         slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
731         memset(from, data, to - from);
732 }
733 
734 static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
735                         u8 *object, char *what,
736                         u8 *start, unsigned int value, unsigned int bytes)
737 {
738         u8 *fault;
739         u8 *end;
740 
741         metadata_access_enable();
742         fault = memchr_inv(start, value, bytes);
743         metadata_access_disable();
744         if (!fault)
745                 return 1;
746 
747         end = start + bytes;
748         while (end > fault && end[-1] == value)
749                 end--;
750 
751         slab_bug(s, "%s overwritten", what);
752         pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
753                                         fault, end - 1, fault[0], value);
754         print_trailer(s, page, object);
755 
756         restore_bytes(s, what, value, fault, end);
757         return 0;
758 }
759 
760 /*
761  * Object layout:
762  *
763  * object address
764  *      Bytes of the object to be managed.
765  *      If the freepointer may overlay the object then the free
766  *      pointer is the first word of the object.
767  *
768  *      Poisoning uses 0x6b (POISON_FREE) and the last byte is
769  *      0xa5 (POISON_END)
770  *
771  * object + s->object_size
772  *      Padding to reach word boundary. This is also used for Redzoning.
773  *      Padding is extended by another word if Redzoning is enabled and
774  *      object_size == inuse.
775  *
776  *      We fill with 0xbb (RED_INACTIVE) for inactive objects and with
777  *      0xcc (RED_ACTIVE) for objects in use.
778  *
779  * object + s->inuse
780  *      Meta data starts here.
781  *
782  *      A. Free pointer (if we cannot overwrite object on free)
783  *      B. Tracking data for SLAB_STORE_USER
784  *      C. Padding to reach required alignment boundary or at mininum
785  *              one word if debugging is on to be able to detect writes
786  *              before the word boundary.
787  *
788  *      Padding is done using 0x5a (POISON_INUSE)
789  *
790  * object + s->size
791  *      Nothing is used beyond s->size.
792  *
793  * If slabcaches are merged then the object_size and inuse boundaries are mostly
794  * ignored. And therefore no slab options that rely on these boundaries
795  * may be used with merged slabcaches.
796  */
797 
798 static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
799 {
800         unsigned long off = s->inuse;   /* The end of info */
801 
802         if (s->offset)
803                 /* Freepointer is placed after the object. */
804                 off += sizeof(void *);
805 
806         if (s->flags & SLAB_STORE_USER)
807                 /* We also have user information there */
808                 off += 2 * sizeof(struct track);
809 
810         off += kasan_metadata_size(s);
811 
812         if (size_from_object(s) == off)
813                 return 1;
814 
815         return check_bytes_and_report(s, page, p, "Object padding",
816                         p + off, POISON_INUSE, size_from_object(s) - off);
817 }
818 
819 /* Check the pad bytes at the end of a slab page */
820 static int slab_pad_check(struct kmem_cache *s, struct page *page)
821 {
822         u8 *start;
823         u8 *fault;
824         u8 *end;
825         u8 *pad;
826         int length;
827         int remainder;
828 
829         if (!(s->flags & SLAB_POISON))
830                 return 1;
831 
832         start = page_address(page);
833         length = PAGE_SIZE << compound_order(page);
834         end = start + length;
835         remainder = length % s->size;
836         if (!remainder)
837                 return 1;
838 
839         pad = end - remainder;
840         metadata_access_enable();
841         fault = memchr_inv(pad, POISON_INUSE, remainder);
842         metadata_access_disable();
843         if (!fault)
844                 return 1;
845         while (end > fault && end[-1] == POISON_INUSE)
846                 end--;
847 
848         slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
849         print_section(KERN_ERR, "Padding ", pad, remainder);
850 
851         restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
852         return 0;
853 }
854 
855 static int check_object(struct kmem_cache *s, struct page *page,
856                                         void *object, u8 val)
857 {
858         u8 *p = object;
859         u8 *endobject = object + s->object_size;
860 
861         if (s->flags & SLAB_RED_ZONE) {
862                 if (!check_bytes_and_report(s, page, object, "Redzone",
863                         object - s->red_left_pad, val, s->red_left_pad))
864                         return 0;
865 
866                 if (!check_bytes_and_report(s, page, object, "Redzone",
867                         endobject, val, s->inuse - s->object_size))
868                         return 0;
869         } else {
870                 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
871                         check_bytes_and_report(s, page, p, "Alignment padding",
872                                 endobject, POISON_INUSE,
873                                 s->inuse - s->object_size);
874                 }
875         }
876 
877         if (s->flags & SLAB_POISON) {
878                 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
879                         (!check_bytes_and_report(s, page, p, "Poison", p,
880                                         POISON_FREE, s->object_size - 1) ||
881                          !check_bytes_and_report(s, page, p, "Poison",
882                                 p + s->object_size - 1, POISON_END, 1)))
883                         return 0;
884                 /*
885                  * check_pad_bytes cleans up on its own.
886                  */
887                 check_pad_bytes(s, page, p);
888         }
889 
890         if (!s->offset && val == SLUB_RED_ACTIVE)
891                 /*
892                  * Object and freepointer overlap. Cannot check
893                  * freepointer while object is allocated.
894                  */
895                 return 1;
896 
897         /* Check free pointer validity */
898         if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
899                 object_err(s, page, p, "Freepointer corrupt");
900                 /*
901                  * No choice but to zap it and thus lose the remainder
902                  * of the free objects in this slab. May cause
903                  * another error because the object count is now wrong.
904                  */
905                 set_freepointer(s, p, NULL);
906                 return 0;
907         }
908         return 1;
909 }
910 
911 static int check_slab(struct kmem_cache *s, struct page *page)
912 {
913         int maxobj;
914 
915         VM_BUG_ON(!irqs_disabled());
916 
917         if (!PageSlab(page)) {
918                 slab_err(s, page, "Not a valid slab page");
919                 return 0;
920         }
921 
922         maxobj = order_objects(compound_order(page), s->size);
923         if (page->objects > maxobj) {
924                 slab_err(s, page, "objects %u > max %u",
925                         page->objects, maxobj);
926                 return 0;
927         }
928         if (page->inuse > page->objects) {
929                 slab_err(s, page, "inuse %u > max %u",
930                         page->inuse, page->objects);
931                 return 0;
932         }
933         /* Slab_pad_check fixes things up after itself */
934         slab_pad_check(s, page);
935         return 1;
936 }
937 
938 /*
939  * Determine if a certain object on a page is on the freelist. Must hold the
940  * slab lock to guarantee that the chains are in a consistent state.
941  */
942 static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
943 {
944         int nr = 0;
945         void *fp;
946         void *object = NULL;
947         int max_objects;
948 
949         fp = page->freelist;
950         while (fp && nr <= page->objects) {
951                 if (fp == search)
952                         return 1;
953                 if (!check_valid_pointer(s, page, fp)) {
954                         if (object) {
955                                 object_err(s, page, object,
956                                         "Freechain corrupt");
957                                 set_freepointer(s, object, NULL);
958                         } else {
959                                 slab_err(s, page, "Freepointer corrupt");
960                                 page->freelist = NULL;
961                                 page->inuse = page->objects;
962                                 slab_fix(s, "Freelist cleared");
963                                 return 0;
964                         }
965                         break;
966                 }
967                 object = fp;
968                 fp = get_freepointer(s, object);
969                 nr++;
970         }
971 
972         max_objects = order_objects(compound_order(page), s->size);
973         if (max_objects > MAX_OBJS_PER_PAGE)
974                 max_objects = MAX_OBJS_PER_PAGE;
975 
976         if (page->objects != max_objects) {
977                 slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
978                          page->objects, max_objects);
979                 page->objects = max_objects;
980                 slab_fix(s, "Number of objects adjusted.");
981         }
982         if (page->inuse != page->objects - nr) {
983                 slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
984                          page->inuse, page->objects - nr);
985                 page->inuse = page->objects - nr;
986                 slab_fix(s, "Object count adjusted.");
987         }
988         return search == NULL;
989 }
990 
991 static void trace(struct kmem_cache *s, struct page *page, void *object,
992                                                                 int alloc)
993 {
994         if (s->flags & SLAB_TRACE) {
995                 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
996                         s->name,
997                         alloc ? "alloc" : "free",
998                         object, page->inuse,
999                         page->freelist);
1000 
1001                 if (!alloc)
1002                         print_section(KERN_INFO, "Object ", (void *)object,
1003                                         s->object_size);
1004 
1005                 dump_stack();
1006         }
1007 }
1008 
1009 /*
1010  * Tracking of fully allocated slabs for debugging purposes.
1011  */
1012 static void add_full(struct kmem_cache *s,
1013         struct kmem_cache_node *n, struct page *page)
1014 {
1015         if (!(s->flags & SLAB_STORE_USER))
1016                 return;
1017 
1018         lockdep_assert_held(&n->list_lock);
1019         list_add(&page->lru, &n->full);
1020 }
1021 
1022 static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
1023 {
1024         if (!(s->flags & SLAB_STORE_USER))
1025                 return;
1026 
1027         lockdep_assert_held(&n->list_lock);
1028         list_del(&page->lru);
1029 }
1030 
1031 /* Tracking of the number of slabs for debugging purposes */
1032 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1033 {
1034         struct kmem_cache_node *n = get_node(s, node);
1035 
1036         return atomic_long_read(&n->nr_slabs);
1037 }
1038 
1039 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1040 {
1041         return atomic_long_read(&n->nr_slabs);
1042 }
1043 
1044 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
1045 {
1046         struct kmem_cache_node *n = get_node(s, node);
1047 
1048         /*
1049          * May be called early in order to allocate a slab for the
1050          * kmem_cache_node structure. Solve the chicken-egg
1051          * dilemma by deferring the increment of the count during
1052          * bootstrap (see early_kmem_cache_node_alloc).
1053          */
1054         if (likely(n)) {
1055                 atomic_long_inc(&n->nr_slabs);
1056                 atomic_long_add(objects, &n->total_objects);
1057         }
1058 }
1059 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
1060 {
1061         struct kmem_cache_node *n = get_node(s, node);
1062 
1063         atomic_long_dec(&n->nr_slabs);
1064         atomic_long_sub(objects, &n->total_objects);
1065 }
1066 
1067 /* Object debug checks for alloc/free paths */
1068 static void setup_object_debug(struct kmem_cache *s, struct page *page,
1069                                                                 void *object)
1070 {
1071         if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1072                 return;
1073 
1074         init_object(s, object, SLUB_RED_INACTIVE);
1075         init_tracking(s, object);
1076 }
1077 
1078 static inline int alloc_consistency_checks(struct kmem_cache *s,
1079                                         struct page *page,
1080                                         void *object, unsigned long addr)
1081 {
1082         if (!check_slab(s, page))
1083                 return 0;
1084 
1085         if (!check_valid_pointer(s, page, object)) {
1086                 object_err(s, page, object, "Freelist Pointer check fails");
1087                 return 0;
1088         }
1089 
1090         if (!check_object(s, page, object, SLUB_RED_INACTIVE))
1091                 return 0;
1092 
1093         return 1;
1094 }
1095 
1096 static noinline int alloc_debug_processing(struct kmem_cache *s,
1097                                         struct page *page,
1098                                         void *object, unsigned long addr)
1099 {
1100         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1101                 if (!alloc_consistency_checks(s, page, object, addr))
1102                         goto bad;
1103         }
1104 
1105         /* Success perform special debug activities for allocs */
1106         if (s->flags & SLAB_STORE_USER)
1107                 set_track(s, object, TRACK_ALLOC, addr);
1108         trace(s, page, object, 1);
1109         init_object(s, object, SLUB_RED_ACTIVE);
1110         return 1;
1111 
1112 bad:
1113         if (PageSlab(page)) {
1114                 /*
1115                  * If this is a slab page then lets do the best we can
1116                  * to avoid issues in the future. Marking all objects
1117                  * as used avoids touching the remaining objects.
1118                  */
1119                 slab_fix(s, "Marking all objects used");
1120                 page->inuse = page->objects;
1121                 page->freelist = NULL;
1122         }
1123         return 0;
1124 }
1125 
1126 static inline int free_consistency_checks(struct kmem_cache *s,
1127                 struct page *page, void *object, unsigned long addr)
1128 {
1129         if (!check_valid_pointer(s, page, object)) {
1130                 slab_err(s, page, "Invalid object pointer 0x%p", object);
1131                 return 0;
1132         }
1133 
1134         if (on_freelist(s, page, object)) {
1135                 object_err(s, page, object, "Object already free");
1136                 return 0;
1137         }
1138 
1139         if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1140                 return 0;
1141 
1142         if (unlikely(s != page->slab_cache)) {
1143                 if (!PageSlab(page)) {
1144                         slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1145                                  object);
1146                 } else if (!page->slab_cache) {
1147                         pr_err("SLUB <none>: no slab for object 0x%p.\n",
1148                                object);
1149                         dump_stack();
1150                 } else
1151                         object_err(s, page, object,
1152                                         "page slab pointer corrupt.");
1153                 return 0;
1154         }
1155         return 1;
1156 }
1157 
1158 /* Supports checking bulk free of a constructed freelist */
1159 static noinline int free_debug_processing(
1160         struct kmem_cache *s, struct page *page,
1161         void *head, void *tail, int bulk_cnt,
1162         unsigned long addr)
1163 {
1164         struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1165         void *object = head;
1166         int cnt = 0;
1167         unsigned long uninitialized_var(flags);
1168         int ret = 0;
1169 
1170         spin_lock_irqsave(&n->list_lock, flags);
1171         slab_lock(page);
1172 
1173         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1174                 if (!check_slab(s, page))
1175                         goto out;
1176         }
1177 
1178 next_object:
1179         cnt++;
1180 
1181         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1182                 if (!free_consistency_checks(s, page, object, addr))
1183                         goto out;
1184         }
1185 
1186         if (s->flags & SLAB_STORE_USER)
1187                 set_track(s, object, TRACK_FREE, addr);
1188         trace(s, page, object, 0);
1189         /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
1190         init_object(s, object, SLUB_RED_INACTIVE);
1191 
1192         /* Reached end of constructed freelist yet? */
1193         if (object != tail) {
1194                 object = get_freepointer(s, object);
1195                 goto next_object;
1196         }
1197         ret = 1;
1198 
1199 out:
1200         if (cnt != bulk_cnt)
1201                 slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1202                          bulk_cnt, cnt);
1203 
1204         slab_unlock(page);
1205         spin_unlock_irqrestore(&n->list_lock, flags);
1206         if (!ret)
1207                 slab_fix(s, "Object at 0x%p not freed", object);
1208         return ret;
1209 }
1210 
1211 static int __init setup_slub_debug(char *str)
1212 {
1213         slub_debug = DEBUG_DEFAULT_FLAGS;
1214         if (*str++ != '=' || !*str)
1215                 /*
1216                  * No options specified. Switch on full debugging.
1217                  */
1218                 goto out;
1219 
1220         if (*str == ',')
1221                 /*
1222                  * No options but restriction on slabs. This means full
1223                  * debugging for slabs matching a pattern.
1224                  */
1225                 goto check_slabs;
1226 
1227         slub_debug = 0;
1228         if (*str == '-')
1229                 /*
1230                  * Switch off all debugging measures.
1231                  */
1232                 goto out;
1233 
1234         /*
1235          * Determine which debug features should be switched on
1236          */
1237         for (; *str && *str != ','; str++) {
1238                 switch (tolower(*str)) {
1239                 case 'f':
1240                         slub_debug |= SLAB_CONSISTENCY_CHECKS;
1241                         break;
1242                 case 'z':
1243                         slub_debug |= SLAB_RED_ZONE;
1244                         break;
1245                 case 'p':
1246                         slub_debug |= SLAB_POISON;
1247                         break;
1248                 case 'u':
1249                         slub_debug |= SLAB_STORE_USER;
1250                         break;
1251                 case 't':
1252                         slub_debug |= SLAB_TRACE;
1253                         break;
1254                 case 'a':
1255                         slub_debug |= SLAB_FAILSLAB;
1256                         break;
1257                 case 'o':
1258                         /*
1259                          * Avoid enabling debugging on caches if its minimum
1260                          * order would increase as a result.
1261                          */
1262                         disable_higher_order_debug = 1;
1263                         break;
1264                 default:
1265                         pr_err("slub_debug option '%c' unknown. skipped\n",
1266                                *str);
1267                 }
1268         }
1269 
1270 check_slabs:
1271         if (*str == ',')
1272                 slub_debug_slabs = str + 1;
1273 out:
1274         return 1;
1275 }
1276 
1277 __setup("slub_debug", setup_slub_debug);
1278 
1279 /*
1280  * kmem_cache_flags - apply debugging options to the cache
1281  * @object_size:        the size of an object without meta data
1282  * @flags:              flags to set
1283  * @name:               name of the cache
1284  * @ctor:               constructor function
1285  *
1286  * Debug option(s) are applied to @flags. In addition to the debug
1287  * option(s), if a slab name (or multiple) is specified i.e.
1288  * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ...
1289  * then only the select slabs will receive the debug option(s).
1290  */
1291 slab_flags_t kmem_cache_flags(unsigned int object_size,
1292         slab_flags_t flags, const char *name,
1293         void (*ctor)(void *))
1294 {
1295         char *iter;
1296         size_t len;
1297 
1298         /* If slub_debug = 0, it folds into the if conditional. */
1299         if (!slub_debug_slabs)
1300                 return flags | slub_debug;
1301 
1302         len = strlen(name);
1303         iter = slub_debug_slabs;
1304         while (*iter) {
1305                 char *end, *glob;
1306                 size_t cmplen;
1307 
1308                 end = strchr(iter, ',');
1309                 if (!end)
1310                         end = iter + strlen(iter);
1311 
1312                 glob = strnchr(iter, end - iter, '*');
1313                 if (glob)
1314                         cmplen = glob - iter;
1315                 else
1316                         cmplen = max_t(size_t, len, (end - iter));
1317 
1318                 if (!strncmp(name, iter, cmplen)) {
1319                         flags |= slub_debug;
1320                         break;
1321                 }
1322 
1323                 if (!*end)
1324                         break;
1325                 iter = end + 1;
1326         }
1327 
1328         return flags;
1329 }
1330 #else /* !CONFIG_SLUB_DEBUG */
1331 static inline void setup_object_debug(struct kmem_cache *s,
1332                         struct page *page, void *object) {}
1333 
1334 static inline int alloc_debug_processing(struct kmem_cache *s,
1335         struct page *page, void *object, unsigned long addr) { return 0; }
1336 
1337 static inline int free_debug_processing(
1338         struct kmem_cache *s, struct page *page,
1339         void *head, void *tail, int bulk_cnt,
1340         unsigned long addr) { return 0; }
1341 
1342 static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1343                         { return 1; }
1344 static inline int check_object(struct kmem_cache *s, struct page *page,
1345                         void *object, u8 val) { return 1; }
1346 static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1347                                         struct page *page) {}
1348 static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1349                                         struct page *page) {}
1350 slab_flags_t kmem_cache_flags(unsigned int object_size,
1351         slab_flags_t flags, const char *name,
1352         void (*ctor)(void *))
1353 {
1354         return flags;
1355 }
1356 #define slub_debug 0
1357 
1358 #define disable_higher_order_debug 0
1359 
1360 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1361                                                         { return 0; }
1362 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1363                                                         { return 0; }
1364 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1365                                                         int objects) {}
1366 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1367                                                         int objects) {}
1368 
1369 #endif /* CONFIG_SLUB_DEBUG */
1370 
1371 /*
1372  * Hooks for other subsystems that check memory allocations. In a typical
1373  * production configuration these hooks all should produce no code at all.
1374  */
1375 static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1376 {
1377         kmemleak_alloc(ptr, size, 1, flags);
1378         kasan_kmalloc_large(ptr, size, flags);
1379 }
1380 
1381 static __always_inline void kfree_hook(void *x)
1382 {
1383         kmemleak_free(x);
1384         kasan_kfree_large(x, _RET_IP_);
1385 }
1386 
1387 static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x)
1388 {
1389         kmemleak_free_recursive(x, s->flags);
1390 
1391         /*
1392          * Trouble is that we may no longer disable interrupts in the fast path
1393          * So in order to make the debug calls that expect irqs to be
1394          * disabled we need to disable interrupts temporarily.
1395          */
1396 #ifdef CONFIG_LOCKDEP
1397         {
1398                 unsigned long flags;
1399 
1400                 local_irq_save(flags);
1401                 debug_check_no_locks_freed(x, s->object_size);
1402                 local_irq_restore(flags);
1403         }
1404 #endif
1405         if (!(s->flags & SLAB_DEBUG_OBJECTS))
1406                 debug_check_no_obj_freed(x, s->object_size);
1407 
1408         /* KASAN might put x into memory quarantine, delaying its reuse */
1409         return kasan_slab_free(s, x, _RET_IP_);
1410 }
1411 
1412 static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1413                                            void **head, void **tail)
1414 {
1415 /*
1416  * Compiler cannot detect this function can be removed if slab_free_hook()
1417  * evaluates to nothing.  Thus, catch all relevant config debug options here.
1418  */
1419 #if defined(CONFIG_LOCKDEP)     ||              \
1420         defined(CONFIG_DEBUG_KMEMLEAK) ||       \
1421         defined(CONFIG_DEBUG_OBJECTS_FREE) ||   \
1422         defined(CONFIG_KASAN)
1423 
1424         void *object;
1425         void *next = *head;
1426         void *old_tail = *tail ? *tail : *head;
1427 
1428         /* Head and tail of the reconstructed freelist */
1429         *head = NULL;
1430         *tail = NULL;
1431 
1432         do {
1433                 object = next;
1434                 next = get_freepointer(s, object);
1435                 /* If object's reuse doesn't have to be delayed */
1436                 if (!slab_free_hook(s, object)) {
1437                         /* Move object to the new freelist */
1438                         set_freepointer(s, object, *head);
1439                         *head = object;
1440                         if (!*tail)
1441                                 *tail = object;
1442                 }
1443         } while (object != old_tail);
1444 
1445         if (*head == *tail)
1446                 *tail = NULL;
1447 
1448         return *head != NULL;
1449 #else
1450         return true;
1451 #endif
1452 }
1453 
1454 static void setup_object(struct kmem_cache *s, struct page *page,
1455                                 void *object)
1456 {
1457         setup_object_debug(s, page, object);
1458         kasan_init_slab_obj(s, object);
1459         if (unlikely(s->ctor)) {
1460                 kasan_unpoison_object_data(s, object);
1461                 s->ctor(object);
1462                 kasan_poison_object_data(s, object);
1463         }
1464 }
1465 
1466 /*
1467  * Slab allocation and freeing
1468  */
1469 static inline struct page *alloc_slab_page(struct kmem_cache *s,
1470                 gfp_t flags, int node, struct kmem_cache_order_objects oo)
1471 {
1472         struct page *page;
1473         unsigned int order = oo_order(oo);
1474 
1475         if (node == NUMA_NO_NODE)
1476                 page = alloc_pages(flags, order);
1477         else
1478                 page = __alloc_pages_node(node, flags, order);
1479 
1480         if (page && memcg_charge_slab(page, flags, order, s)) {
1481                 __free_pages(page, order);
1482                 page = NULL;
1483         }
1484 
1485         return page;
1486 }
1487 
1488 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1489 /* Pre-initialize the random sequence cache */
1490 static int init_cache_random_seq(struct kmem_cache *s)
1491 {
1492         unsigned int count = oo_objects(s->oo);
1493         int err;
1494 
1495         /* Bailout if already initialised */
1496         if (s->random_seq)
1497                 return 0;
1498 
1499         err = cache_random_seq_create(s, count, GFP_KERNEL);
1500         if (err) {
1501                 pr_err("SLUB: Unable to initialize free list for %s\n",
1502                         s->name);
1503                 return err;
1504         }
1505 
1506         /* Transform to an offset on the set of pages */
1507         if (s->random_seq) {
1508                 unsigned int i;
1509 
1510                 for (i = 0; i < count; i++)
1511                         s->random_seq[i] *= s->size;
1512         }
1513         return 0;
1514 }
1515 
1516 /* Initialize each random sequence freelist per cache */
1517 static void __init init_freelist_randomization(void)
1518 {
1519         struct kmem_cache *s;
1520 
1521         mutex_lock(&slab_mutex);
1522 
1523         list_for_each_entry(s, &slab_caches, list)
1524                 init_cache_random_seq(s);
1525 
1526         mutex_unlock(&slab_mutex);
1527 }
1528 
1529 /* Get the next entry on the pre-computed freelist randomized */
1530 static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1531                                 unsigned long *pos, void *start,
1532                                 unsigned long page_limit,
1533                                 unsigned long freelist_count)
1534 {
1535         unsigned int idx;
1536 
1537         /*
1538          * If the target page allocation failed, the number of objects on the
1539          * page might be smaller than the usual size defined by the cache.
1540          */
1541         do {
1542                 idx = s->random_seq[*pos];
1543                 *pos += 1;
1544                 if (*pos >= freelist_count)
1545                         *pos = 0;
1546         } while (unlikely(idx >= page_limit));
1547 
1548         return (char *)start + idx;
1549 }
1550 
1551 /* Shuffle the single linked freelist based on a random pre-computed sequence */
1552 static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1553 {
1554         void *start;
1555         void *cur;
1556         void *next;
1557         unsigned long idx, pos, page_limit, freelist_count;
1558 
1559         if (page->objects < 2 || !s->random_seq)
1560                 return false;
1561 
1562         freelist_count = oo_objects(s->oo);
1563         pos = get_random_int() % freelist_count;
1564 
1565         page_limit = page->objects * s->size;
1566         start = fixup_red_left(s, page_address(page));
1567 
1568         /* First entry is used as the base of the freelist */
1569         cur = next_freelist_entry(s, page, &pos, start, page_limit,
1570                                 freelist_count);
1571         page->freelist = cur;
1572 
1573         for (idx = 1; idx < page->objects; idx++) {
1574                 setup_object(s, page, cur);
1575                 next = next_freelist_entry(s, page, &pos, start, page_limit,
1576                         freelist_count);
1577                 set_freepointer(s, cur, next);
1578                 cur = next;
1579         }
1580         setup_object(s, page, cur);
1581         set_freepointer(s, cur, NULL);
1582 
1583         return true;
1584 }
1585 #else
1586 static inline int init_cache_random_seq(struct kmem_cache *s)
1587 {
1588         return 0;
1589 }
1590 static inline void init_freelist_randomization(void) { }
1591 static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1592 {
1593         return false;
1594 }
1595 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1596 
1597 static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1598 {
1599         struct page *page;
1600         struct kmem_cache_order_objects oo = s->oo;
1601         gfp_t alloc_gfp;
1602         void *start, *p;
1603         int idx, order;
1604         bool shuffle;
1605 
1606         flags &= gfp_allowed_mask;
1607 
1608         if (gfpflags_allow_blocking(flags))
1609                 local_irq_enable();
1610 
1611         flags |= s->allocflags;
1612 
1613         /*
1614          * Let the initial higher-order allocation fail under memory pressure
1615          * so we fall-back to the minimum order allocation.
1616          */
1617         alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1618         if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
1619                 alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
1620 
1621         page = alloc_slab_page(s, alloc_gfp, node, oo);
1622         if (unlikely(!page)) {
1623                 oo = s->min;
1624                 alloc_gfp = flags;
1625                 /*
1626                  * Allocation may have failed due to fragmentation.
1627                  * Try a lower order alloc if possible
1628                  */
1629                 page = alloc_slab_page(s, alloc_gfp, node, oo);
1630                 if (unlikely(!page))
1631                         goto out;
1632                 stat(s, ORDER_FALLBACK);
1633         }
1634 
1635         page->objects = oo_objects(oo);
1636 
1637         order = compound_order(page);
1638         page->slab_cache = s;
1639         __SetPageSlab(page);
1640         if (page_is_pfmemalloc(page))
1641                 SetPageSlabPfmemalloc(page);
1642 
1643         start = page_address(page);
1644 
1645         if (unlikely(s->flags & SLAB_POISON))
1646                 memset(start, POISON_INUSE, PAGE_SIZE << order);
1647 
1648         kasan_poison_slab(page);
1649 
1650         shuffle = shuffle_freelist(s, page);
1651 
1652         if (!shuffle) {
1653                 for_each_object_idx(p, idx, s, start, page->objects) {
1654                         setup_object(s, page, p);
1655                         if (likely(idx < page->objects))
1656                                 set_freepointer(s, p, p + s->size);
1657                         else
1658                                 set_freepointer(s, p, NULL);
1659                 }
1660                 page->freelist = fixup_red_left(s, start);
1661         }
1662 
1663         page->inuse = page->objects;
1664         page->frozen = 1;
1665 
1666 out:
1667         if (gfpflags_allow_blocking(flags))
1668                 local_irq_disable();
1669         if (!page)
1670                 return NULL;
1671 
1672         mod_lruvec_page_state(page,
1673                 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1674                 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1675                 1 << oo_order(oo));
1676 
1677         inc_slabs_node(s, page_to_nid(page), page->objects);
1678 
1679         return page;
1680 }
1681 
1682 static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1683 {
1684         if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
1685                 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1686                 flags &= ~GFP_SLAB_BUG_MASK;
1687                 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1688                                 invalid_mask, &invalid_mask, flags, &flags);
1689                 dump_stack();
1690         }
1691 
1692         return allocate_slab(s,
1693                 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1694 }
1695 
1696 static void __free_slab(struct kmem_cache *s, struct page *page)
1697 {
1698         int order = compound_order(page);
1699         int pages = 1 << order;
1700 
1701         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1702                 void *p;
1703 
1704                 slab_pad_check(s, page);
1705                 for_each_object(p, s, page_address(page),
1706                                                 page->objects)
1707                         check_object(s, page, p, SLUB_RED_INACTIVE);
1708         }
1709 
1710         mod_lruvec_page_state(page,
1711                 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1712                 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1713                 -pages);
1714 
1715         __ClearPageSlabPfmemalloc(page);
1716         __ClearPageSlab(page);
1717 
1718         page->mapping = NULL;
1719         if (current->reclaim_state)
1720                 current->reclaim_state->reclaimed_slab += pages;
1721         memcg_uncharge_slab(page, order, s);
1722         __free_pages(page, order);
1723 }
1724 
1725 static void rcu_free_slab(struct rcu_head *h)
1726 {
1727         struct page *page = container_of(h, struct page, rcu_head);
1728 
1729         __free_slab(page->slab_cache, page);
1730 }
1731 
1732 static void free_slab(struct kmem_cache *s, struct page *page)
1733 {
1734         if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) {
1735                 call_rcu(&page->rcu_head, rcu_free_slab);
1736         } else
1737                 __free_slab(s, page);
1738 }
1739 
1740 static void discard_slab(struct kmem_cache *s, struct page *page)
1741 {
1742         dec_slabs_node(s, page_to_nid(page), page->objects);
1743         free_slab(s, page);
1744 }
1745 
1746 /*
1747  * Management of partially allocated slabs.
1748  */
1749 static inline void
1750 __add_partial(struct kmem_cache_node *n, struct page *page, int tail)
1751 {
1752         n->nr_partial++;
1753         if (tail == DEACTIVATE_TO_TAIL)
1754                 list_add_tail(&page->lru, &n->partial);
1755         else
1756                 list_add(&page->lru, &n->partial);
1757 }
1758 
1759 static inline void add_partial(struct kmem_cache_node *n,
1760                                 struct page *page, int tail)
1761 {
1762         lockdep_assert_held(&n->list_lock);
1763         __add_partial(n, page, tail);
1764 }
1765 
1766 static inline void remove_partial(struct kmem_cache_node *n,
1767                                         struct page *page)
1768 {
1769         lockdep_assert_held(&n->list_lock);
1770         list_del(&page->lru);
1771         n->nr_partial--;
1772 }
1773 
1774 /*
1775  * Remove slab from the partial list, freeze it and
1776  * return the pointer to the freelist.
1777  *
1778  * Returns a list of objects or NULL if it fails.
1779  */
1780 static inline void *acquire_slab(struct kmem_cache *s,
1781                 struct kmem_cache_node *n, struct page *page,
1782                 int mode, int *objects)
1783 {
1784         void *freelist;
1785         unsigned long counters;
1786         struct page new;
1787 
1788         lockdep_assert_held(&n->list_lock);
1789 
1790         /*
1791          * Zap the freelist and set the frozen bit.
1792          * The old freelist is the list of objects for the
1793          * per cpu allocation list.
1794          */
1795         freelist = page->freelist;
1796         counters = page->counters;
1797         new.counters = counters;
1798         *objects = new.objects - new.inuse;
1799         if (mode) {
1800                 new.inuse = page->objects;
1801                 new.freelist = NULL;
1802         } else {
1803                 new.freelist = freelist;
1804         }
1805 
1806         VM_BUG_ON(new.frozen);
1807         new.frozen = 1;
1808 
1809         if (!__cmpxchg_double_slab(s, page,
1810                         freelist, counters,
1811                         new.freelist, new.counters,
1812                         "acquire_slab"))
1813                 return NULL;
1814 
1815         remove_partial(n, page);
1816         WARN_ON(!freelist);
1817         return freelist;
1818 }
1819 
1820 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
1821 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
1822 
1823 /*
1824  * Try to allocate a partial slab from a specific node.
1825  */
1826 static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1827                                 struct kmem_cache_cpu *c, gfp_t flags)
1828 {
1829         struct page *page, *page2;
1830         void *object = NULL;
1831         unsigned int available = 0;
1832         int objects;
1833 
1834         /*
1835          * Racy check. If we mistakenly see no partial slabs then we
1836          * just allocate an empty slab. If we mistakenly try to get a
1837          * partial slab and there is none available then get_partials()
1838          * will return NULL.
1839          */
1840         if (!n || !n->nr_partial)
1841                 return NULL;
1842 
1843         spin_lock(&n->list_lock);
1844         list_for_each_entry_safe(page, page2, &n->partial, lru) {
1845                 void *t;
1846 
1847                 if (!pfmemalloc_match(page, flags))
1848                         continue;
1849 
1850                 t = acquire_slab(s, n, page, object == NULL, &objects);
1851                 if (!t)
1852                         break;
1853 
1854                 available += objects;
1855                 if (!object) {
1856                         c->page = page;
1857                         stat(s, ALLOC_FROM_PARTIAL);
1858                         object = t;
1859                 } else {
1860                         put_cpu_partial(s, page, 0);
1861                         stat(s, CPU_PARTIAL_NODE);
1862                 }
1863                 if (!kmem_cache_has_cpu_partial(s)
1864                         || available > slub_cpu_partial(s) / 2)
1865                         break;
1866 
1867         }
1868         spin_unlock(&n->list_lock);
1869         return object;
1870 }
1871 
1872 /*
1873  * Get a page from somewhere. Search in increasing NUMA distances.
1874  */
1875 static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
1876                 struct kmem_cache_cpu *c)
1877 {
1878 #ifdef CONFIG_NUMA
1879         struct zonelist *zonelist;
1880         struct zoneref *z;
1881         struct zone *zone;
1882         enum zone_type high_zoneidx = gfp_zone(flags);
1883         void *object;
1884         unsigned int cpuset_mems_cookie;
1885 
1886         /*
1887          * The defrag ratio allows a configuration of the tradeoffs between
1888          * inter node defragmentation and node local allocations. A lower
1889          * defrag_ratio increases the tendency to do local allocations
1890          * instead of attempting to obtain partial slabs from other nodes.
1891          *
1892          * If the defrag_ratio is set to 0 then kmalloc() always
1893          * returns node local objects. If the ratio is higher then kmalloc()
1894          * may return off node objects because partial slabs are obtained
1895          * from other nodes and filled up.
1896          *
1897          * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
1898          * (which makes defrag_ratio = 1000) then every (well almost)
1899          * allocation will first attempt to defrag slab caches on other nodes.
1900          * This means scanning over all nodes to look for partial slabs which
1901          * may be expensive if we do it every time we are trying to find a slab
1902          * with available objects.
1903          */
1904         if (!s->remote_node_defrag_ratio ||
1905                         get_cycles() % 1024 > s->remote_node_defrag_ratio)
1906                 return NULL;
1907 
1908         do {
1909                 cpuset_mems_cookie = read_mems_allowed_begin();
1910                 zonelist = node_zonelist(mempolicy_slab_node(), flags);
1911                 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1912                         struct kmem_cache_node *n;
1913 
1914                         n = get_node(s, zone_to_nid(zone));
1915 
1916                         if (n && cpuset_zone_allowed(zone, flags) &&
1917                                         n->nr_partial > s->min_partial) {
1918                                 object = get_partial_node(s, n, c, flags);
1919                                 if (object) {
1920                                         /*
1921                                          * Don't check read_mems_allowed_retry()
1922                                          * here - if mems_allowed was updated in
1923                                          * parallel, that was a harmless race
1924                                          * between allocation and the cpuset
1925                                          * update
1926                                          */
1927                                         return object;
1928                                 }
1929                         }
1930                 }
1931         } while (read_mems_allowed_retry(cpuset_mems_cookie));
1932 #endif
1933         return NULL;
1934 }
1935 
1936 /*
1937  * Get a partial page, lock it and return it.
1938  */
1939 static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
1940                 struct kmem_cache_cpu *c)
1941 {
1942         void *object;
1943         int searchnode = node;
1944 
1945         if (node == NUMA_NO_NODE)
1946                 searchnode = numa_mem_id();
1947         else if (!node_present_pages(node))
1948                 searchnode = node_to_mem_node(node);
1949 
1950         object = get_partial_node(s, get_node(s, searchnode), c, flags);
1951         if (object || node != NUMA_NO_NODE)
1952                 return object;
1953 
1954         return get_any_partial(s, flags, c);
1955 }
1956 
1957 #ifdef CONFIG_PREEMPT
1958 /*
1959  * Calculate the next globally unique transaction for disambiguiation
1960  * during cmpxchg. The transactions start with the cpu number and are then
1961  * incremented by CONFIG_NR_CPUS.
1962  */
1963 #define TID_STEP  roundup_pow_of_two(CONFIG_NR_CPUS)
1964 #else
1965 /*
1966  * No preemption supported therefore also no need to check for
1967  * different cpus.
1968  */
1969 #define TID_STEP 1
1970 #endif
1971 
1972 static inline unsigned long next_tid(unsigned long tid)
1973 {
1974         return tid + TID_STEP;
1975 }
1976 
1977 static inline unsigned int tid_to_cpu(unsigned long tid)
1978 {
1979         return tid % TID_STEP;
1980 }
1981 
1982 static inline unsigned long tid_to_event(unsigned long tid)
1983 {
1984         return tid / TID_STEP;
1985 }
1986 
1987 static inline unsigned int init_tid(int cpu)
1988 {
1989         return cpu;
1990 }
1991 
1992 static inline void note_cmpxchg_failure(const char *n,
1993                 const struct kmem_cache *s, unsigned long tid)
1994 {
1995 #ifdef SLUB_DEBUG_CMPXCHG
1996         unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1997 
1998         pr_info("%s %s: cmpxchg redo ", n, s->name);
1999 
2000 #ifdef CONFIG_PREEMPT
2001         if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
2002                 pr_warn("due to cpu change %d -> %d\n",
2003                         tid_to_cpu(tid), tid_to_cpu(actual_tid));
2004         else
2005 #endif
2006         if (tid_to_event(tid) != tid_to_event(actual_tid))
2007                 pr_warn("due to cpu running other code. Event %ld->%ld\n",
2008                         tid_to_event(tid), tid_to_event(actual_tid));
2009         else
2010                 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
2011                         actual_tid, tid, next_tid(tid));
2012 #endif
2013         stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
2014 }
2015 
2016 static void init_kmem_cache_cpus(struct kmem_cache *s)
2017 {
2018         int cpu;
2019 
2020         for_each_possible_cpu(cpu)
2021                 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
2022 }
2023 
2024 /*
2025  * Remove the cpu slab
2026  */
2027 static void deactivate_slab(struct kmem_cache *s, struct page *page,
2028                                 void *freelist, struct kmem_cache_cpu *c)
2029 {
2030         enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
2031         struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2032         int lock = 0;
2033         enum slab_modes l = M_NONE, m = M_NONE;
2034         void *nextfree;
2035         int tail = DEACTIVATE_TO_HEAD;
2036         struct page new;
2037         struct page old;
2038 
2039         if (page->freelist) {
2040                 stat(s, DEACTIVATE_REMOTE_FREES);
2041                 tail = DEACTIVATE_TO_TAIL;
2042         }
2043 
2044         /*
2045          * Stage one: Free all available per cpu objects back
2046          * to the page freelist while it is still frozen. Leave the
2047          * last one.
2048          *
2049          * There is no need to take the list->lock because the page
2050          * is still frozen.
2051          */
2052         while (freelist && (nextfree = get_freepointer(s, freelist))) {
2053                 void *prior;
2054                 unsigned long counters;
2055 
2056                 do {
2057                         prior = page->freelist;
2058                         counters = page->counters;
2059                         set_freepointer(s, freelist, prior);
2060                         new.counters = counters;
2061                         new.inuse--;
2062                         VM_BUG_ON(!new.frozen);
2063 
2064                 } while (!__cmpxchg_double_slab(s, page,
2065                         prior, counters,
2066                         freelist, new.counters,
2067                         "drain percpu freelist"));
2068 
2069                 freelist = nextfree;
2070         }
2071 
2072         /*
2073          * Stage two: Ensure that the page is unfrozen while the
2074          * list presence reflects the actual number of objects
2075          * during unfreeze.
2076          *
2077          * We setup the list membership and then perform a cmpxchg
2078          * with the count. If there is a mismatch then the page
2079          * is not unfrozen but the page is on the wrong list.
2080          *
2081          * Then we restart the process which may have to remove
2082          * the page from the list that we just put it on again
2083          * because the number of objects in the slab may have
2084          * changed.
2085          */
2086 redo:
2087 
2088         old.freelist = page->freelist;
2089         old.counters = page->counters;
2090         VM_BUG_ON(!old.frozen);
2091 
2092         /* Determine target state of the slab */
2093         new.counters = old.counters;
2094         if (freelist) {
2095                 new.inuse--;
2096                 set_freepointer(s, freelist, old.freelist);
2097                 new.freelist = freelist;
2098         } else
2099                 new.freelist = old.freelist;
2100 
2101         new.frozen = 0;
2102 
2103         if (!new.inuse && n->nr_partial >= s->min_partial)
2104                 m = M_FREE;
2105         else if (new.freelist) {
2106                 m = M_PARTIAL;
2107                 if (!lock) {
2108                         lock = 1;
2109                         /*
2110                          * Taking the spinlock removes the possiblity
2111                          * that acquire_slab() will see a slab page that
2112                          * is frozen
2113                          */
2114                         spin_lock(&n->list_lock);
2115                 }
2116         } else {
2117                 m = M_FULL;
2118                 if (kmem_cache_debug(s) && !lock) {
2119                         lock = 1;
2120                         /*
2121                          * This also ensures that the scanning of full
2122                          * slabs from diagnostic functions will not see
2123                          * any frozen slabs.
2124                          */
2125                         spin_lock(&n->list_lock);
2126                 }
2127         }
2128 
2129         if (l != m) {
2130 
2131                 if (l == M_PARTIAL)
2132 
2133                         remove_partial(n, page);
2134 
2135                 else if (l == M_FULL)
2136 
2137                         remove_full(s, n, page);
2138 
2139                 if (m == M_PARTIAL) {
2140 
2141                         add_partial(n, page, tail);
2142                         stat(s, tail);
2143 
2144                 } else if (m == M_FULL) {
2145 
2146                         stat(s, DEACTIVATE_FULL);
2147                         add_full(s, n, page);
2148 
2149                 }
2150         }
2151 
2152         l = m;
2153         if (!__cmpxchg_double_slab(s, page,
2154                                 old.freelist, old.counters,
2155                                 new.freelist, new.counters,
2156                                 "unfreezing slab"))
2157                 goto redo;
2158 
2159         if (lock)
2160                 spin_unlock(&n->list_lock);
2161 
2162         if (m == M_FREE) {
2163                 stat(s, DEACTIVATE_EMPTY);
2164                 discard_slab(s, page);
2165                 stat(s, FREE_SLAB);
2166         }
2167 
2168         c->page = NULL;
2169         c->freelist = NULL;
2170 }
2171 
2172 /*
2173  * Unfreeze all the cpu partial slabs.
2174  *
2175  * This function must be called with interrupts disabled
2176  * for the cpu using c (or some other guarantee must be there
2177  * to guarantee no concurrent accesses).
2178  */
2179 static void unfreeze_partials(struct kmem_cache *s,
2180                 struct kmem_cache_cpu *c)
2181 {
2182 #ifdef CONFIG_SLUB_CPU_PARTIAL
2183         struct kmem_cache_node *n = NULL, *n2 = NULL;
2184         struct page *page, *discard_page = NULL;
2185 
2186         while ((page = c->partial)) {
2187                 struct page new;
2188                 struct page old;
2189 
2190                 c->partial = page->next;
2191 
2192                 n2 = get_node(s, page_to_nid(page));
2193                 if (n != n2) {
2194                         if (n)
2195                                 spin_unlock(&n->list_lock);
2196 
2197                         n = n2;
2198                         spin_lock(&n->list_lock);
2199                 }
2200 
2201                 do {
2202 
2203                         old.freelist = page->freelist;
2204                         old.counters = page->counters;
2205                         VM_BUG_ON(!old.frozen);
2206 
2207                         new.counters = old.counters;
2208                         new.freelist = old.freelist;
2209 
2210                         new.frozen = 0;
2211 
2212                 } while (!__cmpxchg_double_slab(s, page,
2213                                 old.freelist, old.counters,
2214                                 new.freelist, new.counters,
2215                                 "unfreezing slab"));
2216 
2217                 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
2218                         page->next = discard_page;
2219                         discard_page = page;
2220                 } else {
2221                         add_partial(n, page, DEACTIVATE_TO_TAIL);
2222                         stat(s, FREE_ADD_PARTIAL);
2223                 }
2224         }
2225 
2226         if (n)
2227                 spin_unlock(&n->list_lock);
2228 
2229         while (discard_page) {
2230                 page = discard_page;
2231                 discard_page = discard_page->next;
2232 
2233                 stat(s, DEACTIVATE_EMPTY);
2234                 discard_slab(s, page);
2235                 stat(s, FREE_SLAB);
2236         }
2237 #endif
2238 }
2239 
2240 /*
2241  * Put a page that was just frozen (in __slab_free) into a partial page
2242  * slot if available.
2243  *
2244  * If we did not find a slot then simply move all the partials to the
2245  * per node partial list.
2246  */
2247 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
2248 {
2249 #ifdef CONFIG_SLUB_CPU_PARTIAL
2250         struct page *oldpage;
2251         int pages;
2252         int pobjects;
2253 
2254         preempt_disable();
2255         do {
2256                 pages = 0;
2257                 pobjects = 0;
2258                 oldpage = this_cpu_read(s->cpu_slab->partial);
2259 
2260                 if (oldpage) {
2261                         pobjects = oldpage->pobjects;
2262                         pages = oldpage->pages;
2263                         if (drain && pobjects > s->cpu_partial) {
2264                                 unsigned long flags;
2265                                 /*
2266                                  * partial array is full. Move the existing
2267                                  * set to the per node partial list.
2268                                  */
2269                                 local_irq_save(flags);
2270                                 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2271                                 local_irq_restore(flags);
2272                                 oldpage = NULL;
2273                                 pobjects = 0;
2274                                 pages = 0;
2275                                 stat(s, CPU_PARTIAL_DRAIN);
2276                         }
2277                 }
2278 
2279                 pages++;
2280                 pobjects += page->objects - page->inuse;
2281 
2282                 page->pages = pages;
2283                 page->pobjects = pobjects;
2284                 page->next = oldpage;
2285 
2286         } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2287                                                                 != oldpage);
2288         if (unlikely(!s->cpu_partial)) {
2289                 unsigned long flags;
2290 
2291                 local_irq_save(flags);
2292                 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2293                 local_irq_restore(flags);
2294         }
2295         preempt_enable();
2296 #endif
2297 }
2298 
2299 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
2300 {
2301         stat(s, CPUSLAB_FLUSH);
2302         deactivate_slab(s, c->page, c->freelist, c);
2303 
2304         c->tid = next_tid(c->tid);
2305 }
2306 
2307 /*
2308  * Flush cpu slab.
2309  *
2310  * Called from IPI handler with interrupts disabled.
2311  */
2312 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
2313 {
2314         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2315 
2316         if (likely(c)) {
2317                 if (c->page)
2318                         flush_slab(s, c);
2319 
2320                 unfreeze_partials(s, c);
2321         }
2322 }
2323 
2324 static void flush_cpu_slab(void *d)
2325 {
2326         struct kmem_cache *s = d;
2327 
2328         __flush_cpu_slab(s, smp_processor_id());
2329 }
2330 
2331 static bool has_cpu_slab(int cpu, void *info)
2332 {
2333         struct kmem_cache *s = info;
2334         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2335 
2336         return c->page || slub_percpu_partial(c);
2337 }
2338 
2339 static void flush_all(struct kmem_cache *s)
2340 {
2341         on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
2342 }
2343 
2344 /*
2345  * Use the cpu notifier to insure that the cpu slabs are flushed when
2346  * necessary.
2347  */
2348 static int slub_cpu_dead(unsigned int cpu)
2349 {
2350         struct kmem_cache *s;
2351         unsigned long flags;
2352 
2353         mutex_lock(&slab_mutex);
2354         list_for_each_entry(s, &slab_caches, list) {
2355                 local_irq_save(flags);
2356                 __flush_cpu_slab(s, cpu);
2357                 local_irq_restore(flags);
2358         }
2359         mutex_unlock(&slab_mutex);
2360         return 0;
2361 }
2362 
2363 /*
2364  * Check if the objects in a per cpu structure fit numa
2365  * locality expectations.
2366  */
2367 static inline int node_match(struct page *page, int node)
2368 {
2369 #ifdef CONFIG_NUMA
2370         if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node))
2371                 return 0;
2372 #endif
2373         return 1;
2374 }
2375 
2376 #ifdef CONFIG_SLUB_DEBUG
2377 static int count_free(struct page *page)
2378 {
2379         return page->objects - page->inuse;
2380 }
2381 
2382 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2383 {
2384         return atomic_long_read(&n->total_objects);
2385 }
2386 #endif /* CONFIG_SLUB_DEBUG */
2387 
2388 #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
2389 static unsigned long count_partial(struct kmem_cache_node *n,
2390                                         int (*get_count)(struct page *))
2391 {
2392         unsigned long flags;
2393         unsigned long x = 0;
2394         struct page *page;
2395 
2396         spin_lock_irqsave(&n->list_lock, flags);
2397         list_for_each_entry(page, &n->partial, lru)
2398                 x += get_count(page);
2399         spin_unlock_irqrestore(&n->list_lock, flags);
2400         return x;
2401 }
2402 #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
2403 
2404 static noinline void
2405 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2406 {
2407 #ifdef CONFIG_SLUB_DEBUG
2408         static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2409                                       DEFAULT_RATELIMIT_BURST);
2410         int node;
2411         struct kmem_cache_node *n;
2412 
2413         if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2414                 return;
2415 
2416         pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2417                 nid, gfpflags, &gfpflags);
2418         pr_warn("  cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
2419                 s->name, s->object_size, s->size, oo_order(s->oo),
2420                 oo_order(s->min));
2421 
2422         if (oo_order(s->min) > get_order(s->object_size))
2423                 pr_warn("  %s debugging increased min order, use slub_debug=O to disable.\n",
2424                         s->name);
2425 
2426         for_each_kmem_cache_node(s, node, n) {
2427                 unsigned long nr_slabs;
2428                 unsigned long nr_objs;
2429                 unsigned long nr_free;
2430 
2431                 nr_free  = count_partial(n, count_free);
2432                 nr_slabs = node_nr_slabs(n);
2433                 nr_objs  = node_nr_objs(n);
2434 
2435                 pr_warn("  node %d: slabs: %ld, objs: %ld, free: %ld\n",
2436                         node, nr_slabs, nr_objs, nr_free);
2437         }
2438 #endif
2439 }
2440 
2441 static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2442                         int node, struct kmem_cache_cpu **pc)
2443 {
2444         void *freelist;
2445         struct kmem_cache_cpu *c = *pc;
2446         struct page *page;
2447 
2448         WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2449 
2450         freelist = get_partial(s, flags, node, c);
2451 
2452         if (freelist)
2453                 return freelist;
2454 
2455         page = new_slab(s, flags, node);
2456         if (page) {
2457                 c = raw_cpu_ptr(s->cpu_slab);
2458                 if (c->page)
2459                         flush_slab(s, c);
2460 
2461                 /*
2462                  * No other reference to the page yet so we can
2463                  * muck around with it freely without cmpxchg
2464                  */
2465                 freelist = page->freelist;
2466                 page->freelist = NULL;
2467 
2468                 stat(s, ALLOC_SLAB);
2469                 c->page = page;
2470                 *pc = c;
2471         } else
2472                 freelist = NULL;
2473 
2474         return freelist;
2475 }
2476 
2477 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2478 {
2479         if (unlikely(PageSlabPfmemalloc(page)))
2480                 return gfp_pfmemalloc_allowed(gfpflags);
2481 
2482         return true;
2483 }
2484 
2485 /*
2486  * Check the page->freelist of a page and either transfer the freelist to the
2487  * per cpu freelist or deactivate the page.
2488  *
2489  * The page is still frozen if the return value is not NULL.
2490  *
2491  * If this function returns NULL then the page has been unfrozen.
2492  *
2493  * This function must be called with interrupt disabled.
2494  */
2495 static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2496 {
2497         struct page new;
2498         unsigned long counters;
2499         void *freelist;
2500 
2501         do {
2502                 freelist = page->freelist;
2503                 counters = page->counters;
2504 
2505                 new.counters = counters;
2506                 VM_BUG_ON(!new.frozen);
2507 
2508                 new.inuse = page->objects;
2509                 new.frozen = freelist != NULL;
2510 
2511         } while (!__cmpxchg_double_slab(s, page,
2512                 freelist, counters,
2513                 NULL, new.counters,
2514                 "get_freelist"));
2515 
2516         return freelist;
2517 }
2518 
2519 /*
2520  * Slow path. The lockless freelist is empty or we need to perform
2521  * debugging duties.
2522  *
2523  * Processing is still very fast if new objects have been freed to the
2524  * regular freelist. In that case we simply take over the regular freelist
2525  * as the lockless freelist and zap the regular freelist.
2526  *
2527  * If that is not working then we fall back to the partial lists. We take the
2528  * first element of the freelist as the object to allocate now and move the
2529  * rest of the freelist to the lockless freelist.
2530  *
2531  * And if we were unable to get a new slab from the partial slab lists then
2532  * we need to allocate a new slab. This is the slowest path since it involves
2533  * a call to the page allocator and the setup of a new slab.
2534  *
2535  * Version of __slab_alloc to use when we know that interrupts are
2536  * already disabled (which is the case for bulk allocation).
2537  */
2538 static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2539                           unsigned long addr, struct kmem_cache_cpu *c)
2540 {
2541         void *freelist;
2542         struct page *page;
2543 
2544         page = c->page;
2545         if (!page)
2546                 goto new_slab;
2547 redo:
2548 
2549         if (unlikely(!node_match(page, node))) {
2550                 int searchnode = node;
2551 
2552                 if (node != NUMA_NO_NODE && !node_present_pages(node))
2553                         searchnode = node_to_mem_node(node);
2554 
2555                 if (unlikely(!node_match(page, searchnode))) {
2556                         stat(s, ALLOC_NODE_MISMATCH);
2557                         deactivate_slab(s, page, c->freelist, c);
2558                         goto new_slab;
2559                 }
2560         }
2561 
2562         /*
2563          * By rights, we should be searching for a slab page that was
2564          * PFMEMALLOC but right now, we are losing the pfmemalloc
2565          * information when the page leaves the per-cpu allocator
2566          */
2567         if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2568                 deactivate_slab(s, page, c->freelist, c);
2569                 goto new_slab;
2570         }
2571 
2572         /* must check again c->freelist in case of cpu migration or IRQ */
2573         freelist = c->freelist;
2574         if (freelist)
2575                 goto load_freelist;
2576 
2577         freelist = get_freelist(s, page);
2578 
2579         if (!freelist) {
2580                 c->page = NULL;
2581                 stat(s, DEACTIVATE_BYPASS);
2582                 goto new_slab;
2583         }
2584 
2585         stat(s, ALLOC_REFILL);
2586 
2587 load_freelist:
2588         /*
2589          * freelist is pointing to the list of objects to be used.
2590          * page is pointing to the page from which the objects are obtained.
2591          * That page must be frozen for per cpu allocations to work.
2592          */
2593         VM_BUG_ON(!c->page->frozen);
2594         c->freelist = get_freepointer(s, freelist);
2595         c->tid = next_tid(c->tid);
2596         return freelist;
2597 
2598 new_slab:
2599 
2600         if (slub_percpu_partial(c)) {
2601                 page = c->page = slub_percpu_partial(c);
2602                 slub_set_percpu_partial(c, page);
2603                 stat(s, CPU_PARTIAL_ALLOC);
2604                 goto redo;
2605         }
2606 
2607         freelist = new_slab_objects(s, gfpflags, node, &c);
2608 
2609         if (unlikely(!freelist)) {
2610                 slab_out_of_memory(s, gfpflags, node);
2611                 return NULL;
2612         }
2613 
2614         page = c->page;
2615         if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
2616                 goto load_freelist;
2617 
2618         /* Only entered in the debug case */
2619         if (kmem_cache_debug(s) &&
2620                         !alloc_debug_processing(s, page, freelist, addr))
2621                 goto new_slab;  /* Slab failed checks. Next slab needed */
2622 
2623         deactivate_slab(s, page, get_freepointer(s, freelist), c);
2624         return freelist;
2625 }
2626 
2627 /*
2628  * Another one that disabled interrupt and compensates for possible
2629  * cpu changes by refetching the per cpu area pointer.
2630  */
2631 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2632                           unsigned long addr, struct kmem_cache_cpu *c)
2633 {
2634         void *p;
2635         unsigned long flags;
2636 
2637         local_irq_save(flags);
2638 #ifdef CONFIG_PREEMPT
2639         /*
2640          * We may have been preempted and rescheduled on a different
2641          * cpu before disabling interrupts. Need to reload cpu area
2642          * pointer.
2643          */
2644         c = this_cpu_ptr(s->cpu_slab);
2645 #endif
2646 
2647         p = ___slab_alloc(s, gfpflags, node, addr, c);
2648         local_irq_restore(flags);
2649         return p;
2650 }
2651 
2652 /*
2653  * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2654  * have the fastpath folded into their functions. So no function call
2655  * overhead for requests that can be satisfied on the fastpath.
2656  *
2657  * The fastpath works by first checking if the lockless freelist can be used.
2658  * If not then __slab_alloc is called for slow processing.
2659  *
2660  * Otherwise we can simply pick the next object from the lockless free list.
2661  */
2662 static __always_inline void *slab_alloc_node(struct kmem_cache *s,
2663                 gfp_t gfpflags, int node, unsigned long addr)
2664 {
2665         void *object;
2666         struct kmem_cache_cpu *c;
2667         struct page *page;
2668         unsigned long tid;
2669 
2670         s = slab_pre_alloc_hook(s, gfpflags);
2671         if (!s)
2672                 return NULL;
2673 redo:
2674         /*
2675          * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2676          * enabled. We may switch back and forth between cpus while
2677          * reading from one cpu area. That does not matter as long
2678          * as we end up on the original cpu again when doing the cmpxchg.
2679          *
2680          * We should guarantee that tid and kmem_cache are retrieved on
2681          * the same cpu. It could be different if CONFIG_PREEMPT so we need
2682          * to check if it is matched or not.
2683          */
2684         do {
2685                 tid = this_cpu_read(s->cpu_slab->tid);
2686                 c = raw_cpu_ptr(s->cpu_slab);
2687         } while (IS_ENABLED(CONFIG_PREEMPT) &&
2688                  unlikely(tid != READ_ONCE(c->tid)));
2689 
2690         /*
2691          * Irqless object alloc/free algorithm used here depends on sequence
2692          * of fetching cpu_slab's data. tid should be fetched before anything
2693          * on c to guarantee that object and page associated with previous tid
2694          * won't be used with current tid. If we fetch tid first, object and
2695          * page could be one associated with next tid and our alloc/free
2696          * request will be failed. In this case, we will retry. So, no problem.
2697          */
2698         barrier();
2699 
2700         /*
2701          * The transaction ids are globally unique per cpu and per operation on
2702          * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2703          * occurs on the right processor and that there was no operation on the
2704          * linked list in between.
2705          */
2706 
2707         object = c->freelist;
2708         page = c->page;
2709         if (unlikely(!object || !node_match(page, node))) {
2710                 object = __slab_alloc(s, gfpflags, node, addr, c);
2711                 stat(s, ALLOC_SLOWPATH);
2712         } else {
2713                 void *next_object = get_freepointer_safe(s, object);
2714 
2715                 /*
2716                  * The cmpxchg will only match if there was no additional
2717                  * operation and if we are on the right processor.
2718                  *
2719                  * The cmpxchg does the following atomically (without lock
2720                  * semantics!)
2721                  * 1. Relocate first pointer to the current per cpu area.
2722                  * 2. Verify that tid and freelist have not been changed
2723                  * 3. If they were not changed replace tid and freelist
2724                  *
2725                  * Since this is without lock semantics the protection is only
2726                  * against code executing on this cpu *not* from access by
2727                  * other cpus.
2728                  */
2729                 if (unlikely(!this_cpu_cmpxchg_double(
2730                                 s->cpu_slab->freelist, s->cpu_slab->tid,
2731                                 object, tid,
2732                                 next_object, next_tid(tid)))) {
2733 
2734                         note_cmpxchg_failure("slab_alloc", s, tid);
2735                         goto redo;
2736                 }
2737                 prefetch_freepointer(s, next_object);
2738                 stat(s, ALLOC_FASTPATH);
2739         }
2740 
2741         if (unlikely(gfpflags & __GFP_ZERO) && object)
2742                 memset(object, 0, s->object_size);
2743 
2744         slab_post_alloc_hook(s, gfpflags, 1, &object);
2745 
2746         return object;
2747 }
2748 
2749 static __always_inline void *slab_alloc(struct kmem_cache *s,
2750                 gfp_t gfpflags, unsigned long addr)
2751 {
2752         return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2753 }
2754 
2755 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2756 {
2757         void *ret = slab_alloc(s, gfpflags, _RET_IP_);
2758 
2759         trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2760                                 s->size, gfpflags);
2761 
2762         return ret;
2763 }
2764 EXPORT_SYMBOL(kmem_cache_alloc);
2765 
2766 #ifdef CONFIG_TRACING
2767 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
2768 {
2769         void *ret = slab_alloc(s, gfpflags, _RET_IP_);
2770         trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2771         kasan_kmalloc(s, ret, size, gfpflags);
2772         return ret;
2773 }
2774 EXPORT_SYMBOL(kmem_cache_alloc_trace);
2775 #endif
2776 
2777 #ifdef CONFIG_NUMA
2778 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2779 {
2780         void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
2781 
2782         trace_kmem_cache_alloc_node(_RET_IP_, ret,
2783                                     s->object_size, s->size, gfpflags, node);
2784 
2785         return ret;
2786 }
2787 EXPORT_SYMBOL(kmem_cache_alloc_node);
2788 
2789 #ifdef CONFIG_TRACING
2790 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
2791                                     gfp_t gfpflags,
2792                                     int node, size_t size)
2793 {
2794         void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
2795 
2796         trace_kmalloc_node(_RET_IP_, ret,
2797                            size, s->size, gfpflags, node);
2798 
2799         kasan_kmalloc(s, ret, size, gfpflags);
2800         return ret;
2801 }
2802 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
2803 #endif
2804 #endif
2805 
2806 /*
2807  * Slow path handling. This may still be called frequently since objects
2808  * have a longer lifetime than the cpu slabs in most processing loads.
2809  *
2810  * So we still attempt to reduce cache line usage. Just take the slab
2811  * lock and free the item. If there is no additional partial page
2812  * handling required then we can return immediately.
2813  */
2814 static void __slab_free(struct kmem_cache *s, struct page *page,
2815                         void *head, void *tail, int cnt,
2816                         unsigned long addr)
2817 
2818 {
2819         void *prior;
2820         int was_frozen;
2821         struct page new;
2822         unsigned long counters;
2823         struct kmem_cache_node *n = NULL;
2824         unsigned long uninitialized_var(flags);
2825 
2826         stat(s, FREE_SLOWPATH);
2827 
2828         if (kmem_cache_debug(s) &&
2829             !free_debug_processing(s, page, head, tail, cnt, addr))
2830                 return;
2831 
2832         do {
2833                 if (unlikely(n)) {
2834                         spin_unlock_irqrestore(&n->list_lock, flags);
2835                         n = NULL;
2836                 }
2837                 prior = page->freelist;
2838                 counters = page->counters;
2839                 set_freepointer(s, tail, prior);
2840                 new.counters = counters;
2841                 was_frozen = new.frozen;
2842                 new.inuse -= cnt;
2843                 if ((!new.inuse || !prior) && !was_frozen) {
2844 
2845                         if (kmem_cache_has_cpu_partial(s) && !prior) {
2846 
2847                                 /*
2848                                  * Slab was on no list before and will be
2849                                  * partially empty
2850                                  * We can defer the list move and instead
2851                                  * freeze it.
2852                                  */
2853                                 new.frozen = 1;
2854 
2855                         } else { /* Needs to be taken off a list */
2856 
2857                                 n = get_node(s, page_to_nid(page));
2858                                 /*
2859                                  * Speculatively acquire the list_lock.
2860                                  * If the cmpxchg does not succeed then we may
2861                                  * drop the list_lock without any processing.
2862                                  *
2863                                  * Otherwise the list_lock will synchronize with
2864                                  * other processors updating the list of slabs.
2865                                  */
2866                                 spin_lock_irqsave(&n->list_lock, flags);
2867 
2868                         }
2869                 }
2870 
2871         } while (!cmpxchg_double_slab(s, page,
2872                 prior, counters,
2873                 head, new.counters,
2874                 "__slab_free"));
2875 
2876         if (likely(!n)) {
2877 
2878                 /*
2879                  * If we just froze the page then put it onto the
2880                  * per cpu partial list.
2881                  */
2882                 if (new.frozen && !was_frozen) {
2883                         put_cpu_partial(s, page, 1);
2884                         stat(s, CPU_PARTIAL_FREE);
2885                 }
2886                 /*
2887                  * The list lock was not taken therefore no list
2888                  * activity can be necessary.
2889                  */
2890                 if (was_frozen)
2891                         stat(s, FREE_FROZEN);
2892                 return;
2893         }
2894 
2895         if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
2896                 goto slab_empty;
2897 
2898         /*
2899          * Objects left in the slab. If it was not on the partial list before
2900          * then add it.
2901          */
2902         if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2903                 if (kmem_cache_debug(s))
2904                         remove_full(s, n, page);
2905                 add_partial(n, page, DEACTIVATE_TO_TAIL);
2906                 stat(s, FREE_ADD_PARTIAL);
2907         }
2908         spin_unlock_irqrestore(&n->list_lock, flags);
2909         return;
2910 
2911 slab_empty:
2912         if (prior) {
2913                 /*
2914                  * Slab on the partial list.
2915                  */
2916                 remove_partial(n, page);
2917                 stat(s, FREE_REMOVE_PARTIAL);
2918         } else {
2919                 /* Slab must be on the full list */
2920                 remove_full(s, n, page);
2921         }
2922 
2923         spin_unlock_irqrestore(&n->list_lock, flags);
2924         stat(s, FREE_SLAB);
2925         discard_slab(s, page);
2926 }
2927 
2928 /*
2929  * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2930  * can perform fastpath freeing without additional function calls.
2931  *
2932  * The fastpath is only possible if we are freeing to the current cpu slab
2933  * of this processor. This typically the case if we have just allocated
2934  * the item before.
2935  *
2936  * If fastpath is not possible then fall back to __slab_free where we deal
2937  * with all sorts of special processing.
2938  *
2939  * Bulk free of a freelist with several objects (all pointing to the
2940  * same page) possible by specifying head and tail ptr, plus objects
2941  * count (cnt). Bulk free indicated by tail pointer being set.
2942  */
2943 static __always_inline void do_slab_free(struct kmem_cache *s,
2944                                 struct page *page, void *head, void *tail,
2945                                 int cnt, unsigned long addr)
2946 {
2947         void *tail_obj = tail ? : head;
2948         struct kmem_cache_cpu *c;
2949         unsigned long tid;
2950 redo:
2951         /*
2952          * Determine the currently cpus per cpu slab.
2953          * The cpu may change afterward. However that does not matter since
2954          * data is retrieved via this pointer. If we are on the same cpu
2955          * during the cmpxchg then the free will succeed.
2956          */
2957         do {
2958                 tid = this_cpu_read(s->cpu_slab->tid);
2959                 c = raw_cpu_ptr(s->cpu_slab);
2960         } while (IS_ENABLED(CONFIG_PREEMPT) &&
2961                  unlikely(tid != READ_ONCE(c->tid)));
2962 
2963         /* Same with comment on barrier() in slab_alloc_node() */
2964         barrier();
2965 
2966         if (likely(page == c->page)) {
2967                 set_freepointer(s, tail_obj, c->freelist);
2968 
2969                 if (unlikely(!this_cpu_cmpxchg_double(
2970                                 s->cpu_slab->freelist, s->cpu_slab->tid,
2971                                 c->freelist, tid,
2972                                 head, next_tid(tid)))) {
2973 
2974                         note_cmpxchg_failure("slab_free", s, tid);
2975                         goto redo;
2976                 }
2977                 stat(s, FREE_FASTPATH);
2978         } else
2979                 __slab_free(s, page, head, tail_obj, cnt, addr);
2980 
2981 }
2982 
2983 static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
2984                                       void *head, void *tail, int cnt,
2985                                       unsigned long addr)
2986 {
2987         /*
2988          * With KASAN enabled slab_free_freelist_hook modifies the freelist
2989          * to remove objects, whose reuse must be delayed.
2990          */
2991         if (slab_free_freelist_hook(s, &head, &tail))
2992                 do_slab_free(s, page, head, tail, cnt, addr);
2993 }
2994 
2995 #ifdef CONFIG_KASAN
2996 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
2997 {
2998         do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
2999 }
3000 #endif
3001 
3002 void kmem_cache_free(struct kmem_cache *s, void *x)
3003 {
3004         s = cache_from_obj(s, x);
3005         if (!s)
3006                 return;
3007         slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
3008         trace_kmem_cache_free(_RET_IP_, x);
3009 }
3010 EXPORT_SYMBOL(kmem_cache_free);
3011 
3012 struct detached_freelist {
3013         struct page *page;
3014         void *tail;
3015         void *freelist;
3016         int cnt;
3017         struct kmem_cache *s;
3018 };
3019 
3020 /*
3021  * This function progressively scans the array with free objects (with
3022  * a limited look ahead) and extract objects belonging to the same
3023  * page.  It builds a detached freelist directly within the given
3024  * page/objects.  This can happen without any need for
3025  * synchronization, because the objects are owned by running process.
3026  * The freelist is build up as a single linked list in the objects.
3027  * The idea is, that this detached freelist can then be bulk
3028  * transferred to the real freelist(s), but only requiring a single
3029  * synchronization primitive.  Look ahead in the array is limited due
3030  * to performance reasons.
3031  */
3032 static inline
3033 int build_detached_freelist(struct kmem_cache *s, size_t size,
3034                             void **p, struct detached_freelist *df)
3035 {
3036         size_t first_skipped_index = 0;
3037         int lookahead = 3;
3038         void *object;
3039         struct page *page;
3040 
3041         /* Always re-init detached_freelist */
3042         df->page = NULL;
3043 
3044         do {
3045                 object = p[--size];
3046                 /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
3047         } while (!object && size);
3048 
3049         if (!object)
3050                 return 0;
3051 
3052         page = virt_to_head_page(object);
3053         if (!s) {
3054                 /* Handle kalloc'ed objects */
3055                 if (unlikely(!PageSlab(page))) {
3056                         BUG_ON(!PageCompound(page));
3057                         kfree_hook(object);
3058                         __free_pages(page, compound_order(page));
3059                         p[size] = NULL; /* mark object processed */
3060                         return size;
3061                 }
3062                 /* Derive kmem_cache from object */
3063                 df->s = page->slab_cache;
3064         } else {
3065                 df->s = cache_from_obj(s, object); /* Support for memcg */
3066         }
3067 
3068         /* Start new detached freelist */
3069         df->page = page;
3070         set_freepointer(df->s, object, NULL);
3071         df->tail = object;
3072         df->freelist = object;
3073         p[size] = NULL; /* mark object processed */
3074         df->cnt = 1;
3075 
3076         while (size) {
3077                 object = p[--size];
3078                 if (!object)
3079                         continue; /* Skip processed objects */
3080 
3081                 /* df->page is always set at this point */
3082                 if (df->page == virt_to_head_page(object)) {
3083                         /* Opportunity build freelist */
3084                         set_freepointer(df->s, object, df->freelist);
3085                         df->freelist = object;
3086                         df->cnt++;
3087                         p[size] = NULL; /* mark object processed */
3088 
3089                         continue;
3090                 }
3091 
3092                 /* Limit look ahead search */
3093                 if (!--lookahead)
3094                         break;
3095 
3096                 if (!first_skipped_index)
3097                         first_skipped_index = size + 1;
3098         }
3099 
3100         return first_skipped_index;
3101 }
3102 
3103 /* Note that interrupts must be enabled when calling this function. */
3104 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
3105 {
3106         if (WARN_ON(!size))
3107                 return;
3108 
3109         do {
3110                 struct detached_freelist df;
3111 
3112                 size = build_detached_freelist(s, size, p, &df);
3113                 if (!df.page)
3114                         continue;
3115 
3116                 slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_);
3117         } while (likely(size));
3118 }
3119 EXPORT_SYMBOL(kmem_cache_free_bulk);
3120 
3121 /* Note that interrupts must be enabled when calling this function. */
3122 int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3123                           void **p)
3124 {
3125         struct kmem_cache_cpu *c;
3126         int i;
3127 
3128         /* memcg and kmem_cache debug support */
3129         s = slab_pre_alloc_hook(s, flags);
3130         if (unlikely(!s))
3131                 return false;
3132         /*
3133          * Drain objects in the per cpu slab, while disabling local
3134          * IRQs, which protects against PREEMPT and interrupts
3135          * handlers invoking normal fastpath.
3136          */
3137         local_irq_disable();
3138         c = this_cpu_ptr(s->cpu_slab);
3139 
3140         for (i = 0; i < size; i++) {
3141                 void *object = c->freelist;
3142 
3143                 if (unlikely(!object)) {
3144                         /*
3145                          * Invoking slow path likely have side-effect
3146                          * of re-populating per CPU c->freelist
3147                          */
3148                         p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
3149                                             _RET_IP_, c);
3150                         if (unlikely(!p[i]))
3151                                 goto error;
3152 
3153                         c = this_cpu_ptr(s->cpu_slab);
3154                         continue; /* goto for-loop */
3155                 }
3156                 c->freelist = get_freepointer(s, object);
3157                 p[i] = object;
3158         }
3159         c->tid = next_tid(c->tid);
3160         local_irq_enable();
3161 
3162         /* Clear memory outside IRQ disabled fastpath loop */
3163         if (unlikely(flags & __GFP_ZERO)) {
3164                 int j;
3165 
3166                 for (j = 0; j < i; j++)
3167                         memset(p[j], 0, s->object_size);
3168         }
3169 
3170         /* memcg and kmem_cache debug support */
3171         slab_post_alloc_hook(s, flags, size, p);
3172         return i;
3173 error:
3174         local_irq_enable();
3175         slab_post_alloc_hook(s, flags, i, p);
3176         __kmem_cache_free_bulk(s, i, p);
3177         return 0;
3178 }
3179 EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3180 
3181 
3182 /*
3183  * Object placement in a slab is made very easy because we always start at
3184  * offset 0. If we tune the size of the object to the alignment then we can
3185  * get the required alignment by putting one properly sized object after
3186  * another.
3187  *
3188  * Notice that the allocation order determines the sizes of the per cpu
3189  * caches. Each processor has always one slab available for allocations.
3190  * Increasing the allocation order reduces the number of times that slabs
3191  * must be moved on and off the partial lists and is therefore a factor in
3192  * locking overhead.
3193  */
3194 
3195 /*
3196  * Mininum / Maximum order of slab pages. This influences locking overhead
3197  * and slab fragmentation. A higher order reduces the number of partial slabs
3198  * and increases the number of allocations possible without having to
3199  * take the list_lock.
3200  */
3201 static unsigned int slub_min_order;
3202 static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
3203 static unsigned int slub_min_objects;
3204 
3205 /*
3206  * Calculate the order of allocation given an slab object size.
3207  *
3208  * The order of allocation has significant impact on performance and other
3209  * system components. Generally order 0 allocations should be preferred since
3210  * order 0 does not cause fragmentation in the page allocator. Larger objects
3211  * be problematic to put into order 0 slabs because there may be too much
3212  * unused space left. We go to a higher order if more than 1/16th of the slab
3213  * would be wasted.
3214  *
3215  * In order to reach satisfactory performance we must ensure that a minimum
3216  * number of objects is in one slab. Otherwise we may generate too much
3217  * activity on the partial lists which requires taking the list_lock. This is
3218  * less a concern for large slabs though which are rarely used.
3219  *
3220  * slub_max_order specifies the order where we begin to stop considering the
3221  * number of objects in a slab as critical. If we reach slub_max_order then
3222  * we try to keep the page order as low as possible. So we accept more waste
3223  * of space in favor of a small page order.
3224  *
3225  * Higher order allocations also allow the placement of more objects in a
3226  * slab and thereby reduce object handling overhead. If the user has
3227  * requested a higher mininum order then we start with that one instead of
3228  * the smallest order which will fit the object.
3229  */
3230 static inline unsigned int slab_order(unsigned int size,
3231                 unsigned int min_objects, unsigned int max_order,
3232                 unsigned int fract_leftover)
3233 {
3234         unsigned int min_order = slub_min_order;
3235         unsigned int order;
3236 
3237         if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
3238                 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
3239 
3240         for (order = max(min_order, (unsigned int)get_order(min_objects * size));
3241                         order <= max_order; order++) {
3242 
3243                 unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
3244                 unsigned int rem;
3245 
3246                 rem = slab_size % size;
3247 
3248                 if (rem <= slab_size / fract_leftover)
3249                         break;
3250         }
3251 
3252         return order;
3253 }
3254 
3255 static inline int calculate_order(unsigned int size)
3256 {
3257         unsigned int order;
3258         unsigned int min_objects;
3259         unsigned int max_objects;
3260 
3261         /*
3262          * Attempt to find best configuration for a slab. This
3263          * works by first attempting to generate a layout with
3264          * the best configuration and backing off gradually.
3265          *
3266          * First we increase the acceptable waste in a slab. Then
3267          * we reduce the minimum objects required in a slab.
3268          */
3269         min_objects = slub_min_objects;
3270         if (!min_objects)
3271                 min_objects = 4 * (fls(nr_cpu_ids) + 1);
3272         max_objects = order_objects(slub_max_order, size);
3273         min_objects = min(min_objects, max_objects);
3274 
3275         while (min_objects > 1) {
3276                 unsigned int fraction;
3277 
3278                 fraction = 16;
3279                 while (fraction >= 4) {
3280                         order = slab_order(size, min_objects,
3281                                         slub_max_order, fraction);
3282                         if (order <= slub_max_order)
3283                                 return order;
3284                         fraction /= 2;
3285                 }
3286                 min_objects--;
3287         }
3288 
3289         /*
3290          * We were unable to place multiple objects in a slab. Now
3291          * lets see if we can place a single object there.
3292          */
3293         order = slab_order(size, 1, slub_max_order, 1);
3294         if (order <= slub_max_order)
3295                 return order;
3296 
3297         /*
3298          * Doh this slab cannot be placed using slub_max_order.
3299          */
3300         order = slab_order(size, 1, MAX_ORDER, 1);
3301         if (order < MAX_ORDER)
3302                 return order;
3303         return -ENOSYS;
3304 }
3305 
3306 static void
3307 init_kmem_cache_node(struct kmem_cache_node *n)
3308 {
3309         n->nr_partial = 0;
3310         spin_lock_init(&n->list_lock);
3311         INIT_LIST_HEAD(&n->partial);
3312 #ifdef CONFIG_SLUB_DEBUG
3313         atomic_long_set(&n->nr_slabs, 0);
3314         atomic_long_set(&n->total_objects, 0);
3315         INIT_LIST_HEAD(&n->full);
3316 #endif
3317 }
3318 
3319 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
3320 {
3321         BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
3322                         KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
3323 
3324         /*
3325          * Must align to double word boundary for the double cmpxchg
3326          * instructions to work; see __pcpu_double_call_return_bool().
3327          */
3328         s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3329                                      2 * sizeof(void *));
3330 
3331         if (!s->cpu_slab)
3332                 return 0;
3333 
3334         init_kmem_cache_cpus(s);
3335 
3336         return 1;
3337 }
3338 
3339 static struct kmem_cache *kmem_cache_node;
3340 
3341 /*
3342  * No kmalloc_node yet so do it by hand. We know that this is the first
3343  * slab on the node for this slabcache. There are no concurrent accesses
3344  * possible.
3345  *
3346  * Note that this function only works on the kmem_cache_node
3347  * when allocating for the kmem_cache_node. This is used for bootstrapping
3348  * memory on a fresh node that has no slab structures yet.
3349  */
3350 static void early_kmem_cache_node_alloc(int node)
3351 {
3352         struct page *page;
3353         struct kmem_cache_node *n;
3354 
3355         BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
3356 
3357         page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
3358 
3359         BUG_ON(!page);
3360         if (page_to_nid(page) != node) {
3361                 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3362                 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
3363         }
3364 
3365         n = page->freelist;
3366         BUG_ON(!n);
3367         page->freelist = get_freepointer(kmem_cache_node, n);
3368         page->inuse = 1;
3369         page->frozen = 0;
3370         kmem_cache_node->node[node] = n;
3371 #ifdef CONFIG_SLUB_DEBUG
3372         init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
3373         init_tracking(kmem_cache_node, n);
3374 #endif
3375         kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node),
3376                       GFP_KERNEL);
3377         init_kmem_cache_node(n);
3378         inc_slabs_node(kmem_cache_node, node, page->objects);
3379 
3380         /*
3381          * No locks need to be taken here as it has just been
3382          * initialized and there is no concurrent access.
3383          */
3384         __add_partial(n, page, DEACTIVATE_TO_HEAD);
3385 }
3386 
3387 static void free_kmem_cache_nodes(struct kmem_cache *s)
3388 {
3389         int node;
3390         struct kmem_cache_node *n;
3391 
3392         for_each_kmem_cache_node(s, node, n) {
3393                 s->node[node] = NULL;
3394                 kmem_cache_free(kmem_cache_node, n);
3395         }
3396 }
3397 
3398 void __kmem_cache_release(struct kmem_cache *s)
3399 {
3400         cache_random_seq_destroy(s);
3401         free_percpu(s->cpu_slab);
3402         free_kmem_cache_nodes(s);
3403 }
3404 
3405 static int init_kmem_cache_nodes(struct kmem_cache *s)
3406 {
3407         int node;
3408 
3409         for_each_node_state(node, N_NORMAL_MEMORY) {
3410                 struct kmem_cache_node *n;
3411 
3412                 if (slab_state == DOWN) {
3413                         early_kmem_cache_node_alloc(node);
3414                         continue;
3415                 }
3416                 n = kmem_cache_alloc_node(kmem_cache_node,
3417                                                 GFP_KERNEL, node);
3418 
3419                 if (!n) {
3420                         free_kmem_cache_nodes(s);
3421                         return 0;
3422                 }
3423 
3424                 init_kmem_cache_node(n);
3425                 s->node[node] = n;
3426         }
3427         return 1;
3428 }
3429 
3430 static void set_min_partial(struct kmem_cache *s, unsigned long min)
3431 {
3432         if (min < MIN_PARTIAL)
3433                 min = MIN_PARTIAL;
3434         else if (min > MAX_PARTIAL)
3435                 min = MAX_PARTIAL;
3436         s->min_partial = min;
3437 }
3438 
3439 static void set_cpu_partial(struct kmem_cache *s)
3440 {
3441 #ifdef CONFIG_SLUB_CPU_PARTIAL
3442         /*
3443          * cpu_partial determined the maximum number of objects kept in the
3444          * per cpu partial lists of a processor.
3445          *
3446          * Per cpu partial lists mainly contain slabs that just have one
3447          * object freed. If they are used for allocation then they can be
3448          * filled up again with minimal effort. The slab will never hit the
3449          * per node partial lists and therefore no locking will be required.
3450          *
3451          * This setting also determines
3452          *
3453          * A) The number of objects from per cpu partial slabs dumped to the
3454          *    per node list when we reach the limit.
3455          * B) The number of objects in cpu partial slabs to extract from the
3456          *    per node list when we run out of per cpu objects. We only fetch
3457          *    50% to keep some capacity around for frees.
3458          */
3459         if (!kmem_cache_has_cpu_partial(s))
3460                 s->cpu_partial = 0;
3461         else if (s->size >= PAGE_SIZE)
3462                 s->cpu_partial = 2;
3463         else if (s->size >= 1024)
3464                 s->cpu_partial = 6;
3465         else if (s->size >= 256)
3466                 s->cpu_partial = 13;
3467         else
3468                 s->cpu_partial = 30;
3469 #endif
3470 }
3471 
3472 /*
3473  * calculate_sizes() determines the order and the distribution of data within
3474  * a slab object.
3475  */
3476 static int calculate_sizes(struct kmem_cache *s, int forced_order)
3477 {
3478         slab_flags_t flags = s->flags;
3479         unsigned int size = s->object_size;
3480         unsigned int order;
3481 
3482         /*
3483          * Round up object size to the next word boundary. We can only
3484          * place the free pointer at word boundaries and this determines
3485          * the possible location of the free pointer.
3486          */
3487         size = ALIGN(size, sizeof(void *));
3488 
3489 #ifdef CONFIG_SLUB_DEBUG
3490         /*
3491          * Determine if we can poison the object itself. If the user of
3492          * the slab may touch the object after free or before allocation
3493          * then we should never poison the object itself.
3494          */
3495         if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
3496                         !s->ctor)
3497                 s->flags |= __OBJECT_POISON;
3498         else
3499                 s->flags &= ~__OBJECT_POISON;
3500 
3501 
3502         /*
3503          * If we are Redzoning then check if there is some space between the
3504          * end of the object and the free pointer. If not then add an
3505          * additional word to have some bytes to store Redzone information.
3506          */
3507         if ((flags & SLAB_RED_ZONE) && size == s->object_size)
3508                 size += sizeof(void *);
3509 #endif
3510 
3511         /*
3512          * With that we have determined the number of bytes in actual use
3513          * by the object. This is the potential offset to the free pointer.
3514          */
3515         s->inuse = size;
3516 
3517         if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
3518                 s->ctor)) {
3519                 /*
3520                  * Relocate free pointer after the object if it is not
3521                  * permitted to overwrite the first word of the object on
3522                  * kmem_cache_free.
3523                  *
3524                  * This is the case if we do RCU, have a constructor or
3525                  * destructor or are poisoning the objects.
3526                  */
3527                 s->offset = size;
3528                 size += sizeof(void *);
3529         }
3530 
3531 #ifdef CONFIG_SLUB_DEBUG
3532         if (flags & SLAB_STORE_USER)
3533                 /*
3534                  * Need to store information about allocs and frees after
3535                  * the object.
3536                  */
3537                 size += 2 * sizeof(struct track);
3538 #endif
3539 
3540         kasan_cache_create(s, &size, &s->flags);
3541 #ifdef CONFIG_SLUB_DEBUG
3542         if (flags & SLAB_RED_ZONE) {
3543                 /*
3544                  * Add some empty padding so that we can catch
3545                  * overwrites from earlier objects rather than let
3546                  * tracking information or the free pointer be
3547                  * corrupted if a user writes before the start
3548                  * of the object.
3549                  */
3550                 size += sizeof(void *);
3551 
3552                 s->red_left_pad = sizeof(void *);
3553                 s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3554                 size += s->red_left_pad;
3555         }
3556 #endif
3557 
3558         /*
3559          * SLUB stores one object immediately after another beginning from
3560          * offset 0. In order to align the objects we have to simply size
3561          * each object to conform to the alignment.
3562          */
3563         size = ALIGN(size, s->align);
3564         s->size = size;
3565         if (forced_order >= 0)
3566                 order = forced_order;
3567         else
3568                 order = calculate_order(size);
3569 
3570         if ((int)order < 0)
3571                 return 0;
3572 
3573         s->allocflags = 0;
3574         if (order)
3575                 s->allocflags |= __GFP_COMP;
3576 
3577         if (s->flags & SLAB_CACHE_DMA)
3578                 s->allocflags |= GFP_DMA;
3579 
3580         if (s->flags & SLAB_RECLAIM_ACCOUNT)
3581                 s->allocflags |= __GFP_RECLAIMABLE;
3582 
3583         /*
3584          * Determine the number of objects per slab
3585          */
3586         s->oo = oo_make(order, size);
3587         s->min = oo_make(get_order(size), size);
3588         if (oo_objects(s->oo) > oo_objects(s->max))
3589                 s->max = s->oo;
3590 
3591         return !!oo_objects(s->oo);
3592 }
3593 
3594 static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
3595 {
3596         s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
3597 #ifdef CONFIG_SLAB_FREELIST_HARDENED
3598         s->random = get_random_long();
3599 #endif
3600 
3601         if (!calculate_sizes(s, -1))
3602                 goto error;
3603         if (disable_higher_order_debug) {
3604                 /*
3605                  * Disable debugging flags that store metadata if the min slab
3606                  * order increased.
3607                  */
3608                 if (get_order(s->size) > get_order(s->object_size)) {
3609                         s->flags &= ~DEBUG_METADATA_FLAGS;
3610                         s->offset = 0;
3611                         if (!calculate_sizes(s, -1))
3612                                 goto error;
3613                 }
3614         }
3615 
3616 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3617     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3618         if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
3619                 /* Enable fast mode */
3620                 s->flags |= __CMPXCHG_DOUBLE;
3621 #endif
3622 
3623         /*
3624          * The larger the object size is, the more pages we want on the partial
3625          * list to avoid pounding the page allocator excessively.
3626          */
3627         set_min_partial(s, ilog2(s->size) / 2);
3628 
3629         set_cpu_partial(s);
3630 
3631 #ifdef CONFIG_NUMA
3632         s->remote_node_defrag_ratio = 1000;
3633 #endif
3634 
3635         /* Initialize the pre-computed randomized freelist if slab is up */
3636         if (slab_state >= UP) {
3637                 if (init_cache_random_seq(s))
3638                         goto error;
3639         }
3640 
3641         if (!init_kmem_cache_nodes(s))
3642                 goto error;
3643 
3644         if (alloc_kmem_cache_cpus(s))
3645                 return 0;
3646 
3647         free_kmem_cache_nodes(s);
3648 error:
3649         if (flags & SLAB_PANIC)
3650                 panic("Cannot create slab %s size=%u realsize=%u order=%u offset=%u flags=%lx\n",
3651                       s->name, s->size, s->size,
3652                       oo_order(s->oo), s->offset, (unsigned long)flags);
3653         return -EINVAL;
3654 }
3655 
3656 static void list_slab_objects(struct kmem_cache *s, struct page *page,
3657                                                         const char *text)
3658 {
3659 #ifdef CONFIG_SLUB_DEBUG
3660         void *addr = page_address(page);
3661         void *p;
3662         unsigned long *map = bitmap_zalloc(page->objects, GFP_ATOMIC);
3663         if (!map)
3664                 return;
3665         slab_err(s, page, text, s->name);
3666         slab_lock(page);
3667 
3668         get_map(s, page, map);
3669         for_each_object(p, s, addr, page->objects) {
3670 
3671                 if (!test_bit(slab_index(p, s, addr), map)) {
3672                         pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
3673                         print_tracking(s, p);
3674                 }
3675         }
3676         slab_unlock(page);
3677         bitmap_free(map);
3678 #endif
3679 }
3680 
3681 /*
3682  * Attempt to free all partial slabs on a node.
3683  * This is called from __kmem_cache_shutdown(). We must take list_lock
3684  * because sysfs file might still access partial list after the shutdowning.
3685  */
3686 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
3687 {
3688         LIST_HEAD(discard);
3689         struct page *page, *h;
3690 
3691         BUG_ON(irqs_disabled());
3692         spin_lock_irq(&n->list_lock);
3693         list_for_each_entry_safe(page, h, &n->partial, lru) {
3694                 if (!page->inuse) {
3695                         remove_partial(n, page);
3696                         list_add(&page->lru, &discard);
3697                 } else {
3698                         list_slab_objects(s, page,
3699                         "Objects remaining in %s on __kmem_cache_shutdown()");
3700                 }
3701         }
3702         spin_unlock_irq(&n->list_lock);
3703 
3704         list_for_each_entry_safe(page, h, &discard, lru)
3705                 discard_slab(s, page);
3706 }
3707 
3708 bool __kmem_cache_empty(struct kmem_cache *s)
3709 {
3710         int node;
3711         struct kmem_cache_node *n;
3712 
3713         for_each_kmem_cache_node(s, node, n)
3714                 if (n->nr_partial || slabs_node(s, node))
3715                         return false;
3716         return true;
3717 }
3718 
3719 /*
3720  * Release all resources used by a slab cache.
3721  */
3722 int __kmem_cache_shutdown(struct kmem_cache *s)
3723 {
3724         int node;
3725         struct kmem_cache_node *n;
3726 
3727         flush_all(s);
3728         /* Attempt to free all objects */
3729         for_each_kmem_cache_node(s, node, n) {
3730                 free_partial(s, n);
3731                 if (n->nr_partial || slabs_node(s, node))
3732                         return 1;
3733         }
3734         sysfs_slab_remove(s);
3735         return 0;
3736 }
3737 
3738 /********************************************************************
3739  *              Kmalloc subsystem
3740  *******************************************************************/
3741 
3742 static int __init setup_slub_min_order(char *str)
3743 {
3744         get_option(&str, (int *)&slub_min_order);
3745 
3746         return 1;
3747 }
3748 
3749 __setup("slub_min_order=", setup_slub_min_order);
3750 
3751 static int __init setup_slub_max_order(char *str)
3752 {
3753         get_option(&str, (int *)&slub_max_order);
3754         slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1);
3755 
3756         return 1;
3757 }
3758 
3759 __setup("slub_max_order=", setup_slub_max_order);
3760 
3761 static int __init setup_slub_min_objects(char *str)
3762 {
3763         get_option(&str, (int *)&slub_min_objects);
3764 
3765         return 1;
3766 }
3767 
3768 __setup("slub_min_objects=", setup_slub_min_objects);
3769 
3770 void *__kmalloc(size_t size, gfp_t flags)
3771 {
3772         struct kmem_cache *s;
3773         void *ret;
3774 
3775         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
3776                 return kmalloc_large(size, flags);
3777 
3778         s = kmalloc_slab(size, flags);
3779 
3780         if (unlikely(ZERO_OR_NULL_PTR(s)))
3781                 return s;
3782 
3783         ret = slab_alloc(s, flags, _RET_IP_);
3784 
3785         trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
3786 
3787         kasan_kmalloc(s, ret, size, flags);
3788 
3789         return ret;
3790 }
3791 EXPORT_SYMBOL(__kmalloc);
3792 
3793 #ifdef CONFIG_NUMA
3794 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3795 {
3796         struct page *page;
3797         void *ptr = NULL;
3798 
3799         flags |= __GFP_COMP;
3800         page = alloc_pages_node(node, flags, get_order(size));
3801         if (page)
3802                 ptr = page_address(page);
3803 
3804         kmalloc_large_node_hook(ptr, size, flags);
3805         return ptr;
3806 }
3807 
3808 void *__kmalloc_node(size_t size, gfp_t flags, int node)
3809 {
3810         struct kmem_cache *s;
3811         void *ret;
3812 
3813         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
3814                 ret = kmalloc_large_node(size, flags, node);
3815 
3816                 trace_kmalloc_node(_RET_IP_, ret,
3817                                    size, PAGE_SIZE << get_order(size),
3818                                    flags, node);
3819 
3820                 return ret;
3821         }
3822 
3823         s = kmalloc_slab(size, flags);
3824 
3825         if (unlikely(ZERO_OR_NULL_PTR(s)))
3826                 return s;
3827 
3828         ret = slab_alloc_node(s, flags, node, _RET_IP_);
3829 
3830         trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
3831 
3832         kasan_kmalloc(s, ret, size, flags);
3833 
3834         return ret;
3835 }
3836 EXPORT_SYMBOL(__kmalloc_node);
3837 #endif
3838 
3839 #ifdef CONFIG_HARDENED_USERCOPY
3840 /*
3841  * Rejects incorrectly sized objects and objects that are to be copied
3842  * to/from userspace but do not fall entirely within the containing slab
3843  * cache's usercopy region.
3844  *
3845  * Returns NULL if check passes, otherwise const char * to name of cache
3846  * to indicate an error.
3847  */
3848 void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
3849                          bool to_user)
3850 {
3851         struct kmem_cache *s;
3852         unsigned int offset;
3853         size_t object_size;
3854 
3855         /* Find object and usable object size. */
3856         s = page->slab_cache;
3857 
3858         /* Reject impossible pointers. */
3859         if (ptr < page_address(page))
3860                 usercopy_abort("SLUB object not in SLUB page?!", NULL,
3861                                to_user, 0, n);
3862 
3863         /* Find offset within object. */
3864         offset = (ptr - page_address(page)) % s->size;
3865 
3866         /* Adjust for redzone and reject if within the redzone. */
3867         if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) {
3868                 if (offset < s->red_left_pad)
3869                         usercopy_abort("SLUB object in left red zone",
3870                                        s->name, to_user, offset, n);
3871                 offset -= s->red_left_pad;
3872         }
3873 
3874         /* Allow address range falling entirely within usercopy region. */
3875         if (offset >= s->useroffset &&
3876             offset - s->useroffset <= s->usersize &&
3877             n <= s->useroffset - offset + s->usersize)
3878                 return;
3879 
3880         /*
3881          * If the copy is still within the allocated object, produce
3882          * a warning instead of rejecting the copy. This is intended
3883          * to be a temporary method to find any missing usercopy
3884          * whitelists.
3885          */
3886         object_size = slab_ksize(s);
3887         if (usercopy_fallback &&
3888             offset <= object_size && n <= object_size - offset) {
3889                 usercopy_warn("SLUB object", s->name, to_user, offset, n);
3890                 return;
3891         }
3892 
3893         usercopy_abort("SLUB object", s->name, to_user, offset, n);
3894 }
3895 #endif /* CONFIG_HARDENED_USERCOPY */
3896 
3897 static size_t __ksize(const void *object)
3898 {
3899         struct page *page;
3900 
3901         if (unlikely(object == ZERO_SIZE_PTR))
3902                 return 0;
3903 
3904         page = virt_to_head_page(object);
3905 
3906         if (unlikely(!PageSlab(page))) {
3907                 WARN_ON(!PageCompound(page));
3908                 return PAGE_SIZE << compound_order(page);
3909         }
3910 
3911         return slab_ksize(page->slab_cache);
3912 }
3913 
3914 size_t ksize(const void *object)
3915 {
3916         size_t size = __ksize(object);
3917         /* We assume that ksize callers could use whole allocated area,
3918          * so we need to unpoison this area.
3919          */
3920         kasan_unpoison_shadow(object, size);
3921         return size;
3922 }
3923 EXPORT_SYMBOL(ksize);
3924 
3925 void kfree(const void *x)
3926 {
3927         struct page *page;
3928         void *object = (void *)x;
3929 
3930         trace_kfree(_RET_IP_, x);
3931 
3932         if (unlikely(ZERO_OR_NULL_PTR(x)))
3933                 return;
3934 
3935         page = virt_to_head_page(x);
3936         if (unlikely(!PageSlab(page))) {
3937                 BUG_ON(!PageCompound(page));
3938                 kfree_hook(object);
3939                 __free_pages(page, compound_order(page));
3940                 return;
3941         }
3942         slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
3943 }
3944 EXPORT_SYMBOL(kfree);
3945 
3946 #define SHRINK_PROMOTE_MAX 32
3947 
3948 /*
3949  * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3950  * up most to the head of the partial lists. New allocations will then
3951  * fill those up and thus they can be removed from the partial lists.
3952  *
3953  * The slabs with the least items are placed last. This results in them
3954  * being allocated from last increasing the chance that the last objects
3955  * are freed in them.
3956  */
3957 int __kmem_cache_shrink(struct kmem_cache *s)
3958 {
3959         int node;
3960         int i;
3961         struct kmem_cache_node *n;
3962         struct page *page;
3963         struct page *t;
3964         struct list_head discard;
3965         struct list_head promote[SHRINK_PROMOTE_MAX];
3966         unsigned long flags;
3967         int ret = 0;
3968 
3969         flush_all(s);
3970         for_each_kmem_cache_node(s, node, n) {
3971                 INIT_LIST_HEAD(&discard);
3972                 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
3973                         INIT_LIST_HEAD(promote + i);
3974 
3975                 spin_lock_irqsave(&n->list_lock, flags);
3976 
3977                 /*
3978                  * Build lists of slabs to discard or promote.
3979                  *
3980                  * Note that concurrent frees may occur while we hold the
3981                  * list_lock. page->inuse here is the upper limit.
3982                  */
3983                 list_for_each_entry_safe(page, t, &n->partial, lru) {
3984                         int free = page->objects - page->inuse;
3985 
3986                         /* Do not reread page->inuse */
3987                         barrier();
3988 
3989                         /* We do not keep full slabs on the list */
3990                         BUG_ON(free <= 0);
3991 
3992                         if (free == page->objects) {
3993                                 list_move(&page->lru, &discard);
3994                                 n->nr_partial--;
3995                         } else if (free <= SHRINK_PROMOTE_MAX)
3996                                 list_move(&page->lru, promote + free - 1);
3997                 }
3998 
3999                 /*
4000                  * Promote the slabs filled up most to the head of the
4001                  * partial list.
4002                  */
4003                 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
4004                         list_splice(promote + i, &n->partial);
4005 
4006                 spin_unlock_irqrestore(&n->list_lock, flags);
4007 
4008                 /* Release empty slabs */
4009                 list_for_each_entry_safe(page, t, &discard, lru)
4010                         discard_slab(s, page);
4011 
4012                 if (slabs_node(s, node))
4013                         ret = 1;
4014         }
4015 
4016         return ret;
4017 }
4018 
4019 #ifdef CONFIG_MEMCG
4020 static void kmemcg_cache_deact_after_rcu(struct kmem_cache *s)
4021 {
4022         /*
4023          * Called with all the locks held after a sched RCU grace period.
4024          * Even if @s becomes empty after shrinking, we can't know that @s
4025          * doesn't have allocations already in-flight and thus can't
4026          * destroy @s until the associated memcg is released.
4027          *
4028          * However, let's remove the sysfs files for empty caches here.
4029          * Each cache has a lot of interface files which aren't
4030          * particularly useful for empty draining caches; otherwise, we can
4031          * easily end up with millions of unnecessary sysfs files on
4032          * systems which have a lot of memory and transient cgroups.
4033          */
4034         if (!__kmem_cache_shrink(s))
4035                 sysfs_slab_remove(s);
4036 }
4037 
4038 void __kmemcg_cache_deactivate(struct kmem_cache *s)
4039 {
4040         /*
4041          * Disable empty slabs caching. Used to avoid pinning offline
4042          * memory cgroups by kmem pages that can be freed.
4043          */
4044         slub_set_cpu_partial(s, 0);
4045         s->min_partial = 0;
4046 
4047         /*
4048          * s->cpu_partial is checked locklessly (see put_cpu_partial), so
4049          * we have to make sure the change is visible before shrinking.
4050          */
4051         slab_deactivate_memcg_cache_rcu_sched(s, kmemcg_cache_deact_after_rcu);
4052 }
4053 #endif
4054 
4055 static int slab_mem_going_offline_callback(void *arg)
4056 {
4057         struct kmem_cache *s;
4058 
4059         mutex_lock(&slab_mutex);
4060         list_for_each_entry(s, &slab_caches, list)
4061                 __kmem_cache_shrink(s);
4062         mutex_unlock(&slab_mutex);
4063 
4064         return 0;
4065 }
4066 
4067 static void slab_mem_offline_callback(void *arg)
4068 {
4069         struct kmem_cache_node *n;
4070         struct kmem_cache *s;
4071         struct memory_notify *marg = arg;
4072         int offline_node;
4073 
4074         offline_node = marg->status_change_nid_normal;
4075 
4076         /*
4077          * If the node still has available memory. we need kmem_cache_node
4078          * for it yet.
4079          */
4080         if (offline_node < 0)
4081                 return;
4082 
4083         mutex_lock(&slab_mutex);
4084         list_for_each_entry(s, &slab_caches, list) {
4085                 n = get_node(s, offline_node);
4086                 if (n) {
4087                         /*
4088                          * if n->nr_slabs > 0, slabs still exist on the node
4089                          * that is going down. We were unable to free them,
4090                          * and offline_pages() function shouldn't call this
4091                          * callback. So, we must fail.
4092                          */
4093                         BUG_ON(slabs_node(s, offline_node));
4094 
4095                         s->node[offline_node] = NULL;
4096                         kmem_cache_free(kmem_cache_node, n);
4097                 }
4098         }
4099         mutex_unlock(&slab_mutex);
4100 }
4101 
4102 static int slab_mem_going_online_callback(void *arg)
4103 {
4104         struct kmem_cache_node *n;
4105         struct kmem_cache *s;
4106         struct memory_notify *marg = arg;
4107         int nid = marg->status_change_nid_normal;
4108         int ret = 0;
4109 
4110         /*
4111          * If the node's memory is already available, then kmem_cache_node is
4112          * already created. Nothing to do.
4113          */
4114         if (nid < 0)
4115                 return 0;
4116 
4117         /*
4118          * We are bringing a node online. No memory is available yet. We must
4119          * allocate a kmem_cache_node structure in order to bring the node
4120          * online.
4121          */
4122         mutex_lock(&slab_mutex);
4123         list_for_each_entry(s, &slab_caches, list) {
4124                 /*
4125                  * XXX: kmem_cache_alloc_node will fallback to other nodes
4126                  *      since memory is not yet available from the node that
4127                  *      is brought up.
4128                  */
4129                 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
4130                 if (!n) {
4131                         ret = -ENOMEM;
4132                         goto out;
4133                 }
4134                 init_kmem_cache_node(n);
4135                 s->node[nid] = n;
4136         }
4137 out:
4138         mutex_unlock(&slab_mutex);
4139         return ret;
4140 }
4141 
4142 static int slab_memory_callback(struct notifier_block *self,
4143                                 unsigned long action, void *arg)
4144 {
4145         int ret = 0;
4146 
4147         switch (action) {
4148         case MEM_GOING_ONLINE:
4149                 ret = slab_mem_going_online_callback(arg);
4150                 break;
4151         case MEM_GOING_OFFLINE:
4152                 ret = slab_mem_going_offline_callback(arg);
4153                 break;
4154         case MEM_OFFLINE:
4155         case MEM_CANCEL_ONLINE:
4156                 slab_mem_offline_callback(arg);
4157                 break;
4158         case MEM_ONLINE:
4159         case MEM_CANCEL_OFFLINE:
4160                 break;
4161         }
4162         if (ret)
4163                 ret = notifier_from_errno(ret);
4164         else
4165                 ret = NOTIFY_OK;
4166         return ret;
4167 }
4168 
4169 static struct notifier_block slab_memory_callback_nb = {
4170         .notifier_call = slab_memory_callback,
4171         .priority = SLAB_CALLBACK_PRI,
4172 };
4173 
4174 /********************************************************************
4175  *                      Basic setup of slabs
4176  *******************************************************************/
4177 
4178 /*
4179  * Used for early kmem_cache structures that were allocated using
4180  * the page allocator. Allocate them properly then fix up the pointers
4181  * that may be pointing to the wrong kmem_cache structure.
4182  */
4183 
4184 static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
4185 {
4186         int node;
4187         struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
4188         struct kmem_cache_node *n;
4189 
4190         memcpy(s, static_cache, kmem_cache->object_size);
4191 
4192         /*
4193          * This runs very early, and only the boot processor is supposed to be
4194          * up.  Even if it weren't true, IRQs are not up so we couldn't fire
4195          * IPIs around.
4196          */
4197         __flush_cpu_slab(s, smp_processor_id());
4198         for_each_kmem_cache_node(s, node, n) {
4199                 struct page *p;
4200 
4201                 list_for_each_entry(p, &n->partial, lru)
4202                         p->slab_cache = s;
4203 
4204 #ifdef CONFIG_SLUB_DEBUG
4205                 list_for_each_entry(p, &n->full, lru)
4206                         p->slab_cache = s;
4207 #endif
4208         }
4209         slab_init_memcg_params(s);
4210         list_add(&s->list, &slab_caches);
4211         memcg_link_cache(s);
4212         return s;
4213 }
4214 
4215 void __init kmem_cache_init(void)
4216 {
4217         static __initdata struct kmem_cache boot_kmem_cache,
4218                 boot_kmem_cache_node;
4219 
4220         if (debug_guardpage_minorder())
4221                 slub_max_order = 0;
4222 
4223         kmem_cache_node = &boot_kmem_cache_node;
4224         kmem_cache = &boot_kmem_cache;
4225 
4226         create_boot_cache(kmem_cache_node, "kmem_cache_node",
4227                 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
4228 
4229         register_hotmemory_notifier(&slab_memory_callback_nb);
4230 
4231         /* Able to allocate the per node structures */
4232         slab_state = PARTIAL;
4233 
4234         create_boot_cache(kmem_cache, "kmem_cache",
4235                         offsetof(struct kmem_cache, node) +
4236                                 nr_node_ids * sizeof(struct kmem_cache_node *),
4237                        SLAB_HWCACHE_ALIGN, 0, 0);
4238 
4239         kmem_cache = bootstrap(&boot_kmem_cache);
4240         kmem_cache_node = bootstrap(&boot_kmem_cache_node);
4241 
4242         /* Now we can use the kmem_cache to allocate kmalloc slabs */
4243         setup_kmalloc_cache_index_table();
4244         create_kmalloc_caches(0);
4245 
4246         /* Setup random freelists for each cache */
4247         init_freelist_randomization();
4248 
4249         cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4250                                   slub_cpu_dead);
4251 
4252         pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%d\n",
4253                 cache_line_size(),
4254                 slub_min_order, slub_max_order, slub_min_objects,
4255                 nr_cpu_ids, nr_node_ids);
4256 }
4257 
4258 void __init kmem_cache_init_late(void)
4259 {
4260 }
4261 
4262 struct kmem_cache *
4263 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
4264                    slab_flags_t flags, void (*ctor)(void *))
4265 {
4266         struct kmem_cache *s, *c;
4267 
4268         s = find_mergeable(size, align, flags, name, ctor);
4269         if (s) {
4270                 s->refcount++;
4271 
4272                 /*
4273                  * Adjust the object sizes so that we clear
4274                  * the complete object on kzalloc.
4275                  */
4276                 s->object_size = max(s->object_size, size);
4277                 s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
4278 
4279                 for_each_memcg_cache(c, s) {
4280                         c->object_size = s->object_size;
4281                         c->inuse = max(c->inuse, ALIGN(size, sizeof(void *)));
4282                 }
4283 
4284                 if (sysfs_slab_alias(s, name)) {
4285                         s->refcount--;
4286                         s = NULL;
4287                 }
4288         }
4289 
4290         return s;
4291 }
4292 
4293 int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
4294 {
4295         int err;
4296 
4297         err = kmem_cache_open(s, flags);
4298         if (err)
4299                 return err;
4300 
4301         /* Mutex is not taken during early boot */
4302         if (slab_state <= UP)
4303                 return 0;
4304 
4305         memcg_propagate_slab_attrs(s);
4306         err = sysfs_slab_add(s);
4307         if (err)
4308                 __kmem_cache_release(s);
4309 
4310         return err;
4311 }
4312 
4313 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
4314 {
4315         struct kmem_cache *s;
4316         void *ret;
4317 
4318         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
4319                 return kmalloc_large(size, gfpflags);
4320 
4321         s = kmalloc_slab(size, gfpflags);
4322 
4323         if (unlikely(ZERO_OR_NULL_PTR(s)))
4324                 return s;
4325 
4326         ret = slab_alloc(s, gfpflags, caller);
4327 
4328         /* Honor the call site pointer we received. */
4329         trace_kmalloc(caller, ret, size, s->size, gfpflags);
4330 
4331         return ret;
4332 }
4333 
4334 #ifdef CONFIG_NUMA
4335 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
4336                                         int node, unsigned long caller)
4337 {
4338         struct kmem_cache *s;
4339         void *ret;
4340 
4341         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
4342                 ret = kmalloc_large_node(size, gfpflags, node);
4343 
4344                 trace_kmalloc_node(caller, ret,
4345                                    size, PAGE_SIZE << get_order(size),
4346                                    gfpflags, node);
4347 
4348                 return ret;
4349         }
4350 
4351         s = kmalloc_slab(size, gfpflags);
4352 
4353         if (unlikely(ZERO_OR_NULL_PTR(s)))
4354                 return s;
4355 
4356         ret = slab_alloc_node(s, gfpflags, node, caller);
4357 
4358         /* Honor the call site pointer we received. */
4359         trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
4360 
4361         return ret;
4362 }
4363 #endif
4364 
4365 #ifdef CONFIG_SYSFS
4366 static int count_inuse(struct page *page)
4367 {
4368         return page->inuse;
4369 }
4370 
4371 static int count_total(struct page *page)
4372 {
4373         return page->objects;
4374 }
4375 #endif
4376 
4377 #ifdef CONFIG_SLUB_DEBUG
4378 static int validate_slab(struct kmem_cache *s, struct page *page,
4379                                                 unsigned long *map)
4380 {
4381         void *p;
4382         void *addr = page_address(page);
4383 
4384         if (!check_slab(s, page) ||
4385                         !on_freelist(s, page, NULL))
4386                 return 0;
4387 
4388         /* Now we know that a valid freelist exists */
4389         bitmap_zero(map, page->objects);
4390 
4391         get_map(s, page, map);
4392         for_each_object(p, s, addr, page->objects) {
4393                 if (test_bit(slab_index(p, s, addr), map))
4394                         if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4395                                 return 0;
4396         }
4397 
4398         for_each_object(p, s, addr, page->objects)
4399                 if (!test_bit(slab_index(p, s, addr), map))
4400                         if (!check_object(s, page, p, SLUB_RED_ACTIVE))
4401                                 return 0;
4402         return 1;
4403 }
4404 
4405 static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4406                                                 unsigned long *map)
4407 {
4408         slab_lock(page);
4409         validate_slab(s, page, map);
4410         slab_unlock(page);
4411 }
4412 
4413 static int validate_slab_node(struct kmem_cache *s,
4414                 struct kmem_cache_node *n, unsigned long *map)
4415 {
4416         unsigned long count = 0;
4417         struct page *page;
4418         unsigned long flags;
4419 
4420         spin_lock_irqsave(&n->list_lock, flags);
4421 
4422         list_for_each_entry(page, &n->partial, lru) {
4423                 validate_slab_slab(s, page, map);
4424                 count++;
4425         }
4426         if (count != n->nr_partial)
4427                 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4428                        s->name, count, n->nr_partial);
4429 
4430         if (!(s->flags & SLAB_STORE_USER))
4431                 goto out;
4432 
4433         list_for_each_entry(page, &n->full, lru) {
4434                 validate_slab_slab(s, page, map);
4435                 count++;
4436         }
4437         if (count != atomic_long_read(&n->nr_slabs))
4438                 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4439                        s->name, count, atomic_long_read(&n->nr_slabs));
4440 
4441 out:
4442         spin_unlock_irqrestore(&n->list_lock, flags);
4443         return count;
4444 }
4445 
4446 static long validate_slab_cache(struct kmem_cache *s)
4447 {
4448         int node;
4449         unsigned long count = 0;
4450         struct kmem_cache_node *n;
4451         unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL);
4452 
4453         if (!map)
4454                 return -ENOMEM;
4455 
4456         flush_all(s);
4457         for_each_kmem_cache_node(s, node, n)
4458                 count += validate_slab_node(s, n, map);
4459         bitmap_free(map);
4460         return count;
4461 }
4462 /*
4463  * Generate lists of code addresses where slabcache objects are allocated
4464  * and freed.
4465  */
4466 
4467 struct location {
4468         unsigned long count;
4469         unsigned long addr;
4470         long long sum_time;
4471         long min_time;
4472         long max_time;
4473         long min_pid;
4474         long max_pid;
4475         DECLARE_BITMAP(cpus, NR_CPUS);
4476         nodemask_t nodes;
4477 };
4478 
4479 struct loc_track {
4480         unsigned long max;
4481         unsigned long count;
4482         struct location *loc;
4483 };
4484 
4485 static void free_loc_track(struct loc_track *t)
4486 {
4487         if (t->max)
4488                 free_pages((unsigned long)t->loc,
4489                         get_order(sizeof(struct location) * t->max));
4490 }
4491 
4492 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
4493 {
4494         struct location *l;
4495         int order;
4496 
4497         order = get_order(sizeof(struct location) * max);
4498 
4499         l = (void *)__get_free_pages(flags, order);
4500         if (!l)
4501                 return 0;
4502 
4503         if (t->count) {
4504                 memcpy(l, t->loc, sizeof(struct location) * t->count);
4505                 free_loc_track(t);
4506         }
4507         t->max = max;
4508         t->loc = l;
4509         return 1;
4510 }
4511 
4512 static int add_location(struct loc_track *t, struct kmem_cache *s,
4513                                 const struct track *track)
4514 {
4515         long start, end, pos;
4516         struct location *l;
4517         unsigned long caddr;
4518         unsigned long age = jiffies - track->when;
4519 
4520         start = -1;
4521         end = t->count;
4522 
4523         for ( ; ; ) {
4524                 pos = start + (end - start + 1) / 2;
4525 
4526                 /*
4527                  * There is nothing at "end". If we end up there
4528                  * we need to add something to before end.
4529                  */
4530                 if (pos == end)
4531                         break;
4532 
4533                 caddr = t->loc[pos].addr;
4534                 if (track->addr == caddr) {
4535 
4536                         l = &t->loc[pos];
4537                         l->count++;
4538                         if (track->when) {
4539                                 l->sum_time += age;
4540                                 if (age < l->min_time)
4541                                         l->min_time = age;
4542                                 if (age > l->max_time)
4543                                         l->max_time = age;
4544 
4545                                 if (track->pid < l->min_pid)
4546                                         l->min_pid = track->pid;
4547                                 if (track->pid > l->max_pid)
4548                                         l->max_pid = track->pid;
4549 
4550                                 cpumask_set_cpu(track->cpu,
4551                                                 to_cpumask(l->cpus));
4552                         }
4553                         node_set(page_to_nid(virt_to_page(track)), l->nodes);
4554                         return 1;
4555                 }
4556 
4557                 if (track->addr < caddr)
4558                         end = pos;
4559                 else
4560                         start = pos;
4561         }
4562 
4563         /*
4564          * Not found. Insert new tracking element.
4565          */
4566         if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
4567                 return 0;
4568 
4569         l = t->loc + pos;
4570         if (pos < t->count)
4571                 memmove(l + 1, l,
4572                         (t->count - pos) * sizeof(struct location));
4573         t->count++;
4574         l->count = 1;
4575         l->addr = track->addr;
4576         l->sum_time = age;
4577         l->min_time = age;
4578         l->max_time = age;
4579         l->min_pid = track->pid;
4580         l->max_pid = track->pid;
4581         cpumask_clear(to_cpumask(l->cpus));
4582         cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
4583         nodes_clear(l->nodes);
4584         node_set(page_to_nid(virt_to_page(track)), l->nodes);
4585         return 1;
4586 }
4587 
4588 static void process_slab(struct loc_track *t, struct kmem_cache *s,
4589                 struct page *page, enum track_item alloc,
4590                 unsigned long *map)
4591 {
4592         void *addr = page_address(page);
4593         void *p;
4594 
4595         bitmap_zero(map, page->objects);
4596         get_map(s, page, map);
4597 
4598         for_each_object(p, s, addr, page->objects)
4599                 if (!test_bit(slab_index(p, s, addr), map))
4600                         add_location(t, s, get_track(s, p, alloc));
4601 }
4602 
4603 static int list_locations(struct kmem_cache *s, char *buf,
4604                                         enum track_item alloc)
4605 {
4606         int len = 0;
4607         unsigned long i;
4608         struct loc_track t = { 0, 0, NULL };
4609         int node;
4610         struct kmem_cache_node *n;
4611         unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL);
4612 
4613         if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4614                                      GFP_KERNEL)) {
4615                 bitmap_free(map);
4616                 return sprintf(buf, "Out of memory\n");
4617         }
4618         /* Push back cpu slabs */
4619         flush_all(s);
4620 
4621         for_each_kmem_cache_node(s, node, n) {
4622                 unsigned long flags;
4623                 struct page *page;
4624 
4625                 if (!atomic_long_read(&n->nr_slabs))
4626                         continue;
4627 
4628                 spin_lock_irqsave(&n->list_lock, flags);
4629                 list_for_each_entry(page, &n->partial, lru)
4630                         process_slab(&t, s, page, alloc, map);
4631                 list_for_each_entry(page, &n->full, lru)
4632                         process_slab(&t, s, page, alloc, map);
4633                 spin_unlock_irqrestore(&n->list_lock, flags);
4634         }
4635 
4636         for (i = 0; i < t.count; i++) {
4637                 struct location *l = &t.loc[i];
4638 
4639                 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
4640                         break;
4641                 len += sprintf(buf + len, "%7ld ", l->count);
4642 
4643                 if (l->addr)
4644                         len += sprintf(buf + len, "%pS", (void *)l->addr);
4645                 else
4646                         len += sprintf(buf + len, "<not-available>");
4647 
4648                 if (l->sum_time != l->min_time) {
4649                         len += sprintf(buf + len, " age=%ld/%ld/%ld",
4650                                 l->min_time,
4651                                 (long)div_u64(l->sum_time, l->count),
4652                                 l->max_time);
4653                 } else
4654                         len += sprintf(buf + len, " age=%ld",
4655                                 l->min_time);
4656 
4657                 if (l->min_pid != l->max_pid)
4658                         len += sprintf(buf + len, " pid=%ld-%ld",
4659                                 l->min_pid, l->max_pid);
4660                 else
4661                         len += sprintf(buf + len, " pid=%ld",
4662                                 l->min_pid);
4663 
4664                 if (num_online_cpus() > 1 &&
4665                                 !cpumask_empty(to_cpumask(l->cpus)) &&
4666                                 len < PAGE_SIZE - 60)
4667                         len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4668                                          " cpus=%*pbl",
4669                                          cpumask_pr_args(to_cpumask(l->cpus)));
4670 
4671                 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
4672                                 len < PAGE_SIZE - 60)
4673                         len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4674                                          " nodes=%*pbl",
4675                                          nodemask_pr_args(&l->nodes));
4676 
4677                 len += sprintf(buf + len, "\n");
4678         }
4679 
4680         free_loc_track(&t);
4681         bitmap_free(map);
4682         if (!t.count)
4683                 len += sprintf(buf, "No data\n");
4684         return len;
4685 }
4686 #endif
4687 
4688 #ifdef SLUB_RESILIENCY_TEST
4689 static void __init resiliency_test(void)
4690 {
4691         u8 *p;
4692         int type = KMALLOC_NORMAL;
4693 
4694         BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
4695 
4696         pr_err("SLUB resiliency testing\n");
4697         pr_err("-----------------------\n");
4698         pr_err("A. Corruption after allocation\n");
4699 
4700         p = kzalloc(16, GFP_KERNEL);
4701         p[16] = 0x12;
4702         pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4703                p + 16);
4704 
4705         validate_slab_cache(kmalloc_caches[type][4]);
4706 
4707         /* Hmmm... The next two are dangerous */
4708         p = kzalloc(32, GFP_KERNEL);
4709         p[32 + sizeof(void *)] = 0x34;
4710         pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4711                p);
4712         pr_err("If allocated object is overwritten then not detectable\n\n");
4713 
4714         validate_slab_cache(kmalloc_caches[type][5]);
4715         p = kzalloc(64, GFP_KERNEL);
4716         p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4717         *p = 0x56;
4718         pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4719                p);
4720         pr_err("If allocated object is overwritten then not detectable\n\n");
4721         validate_slab_cache(kmalloc_caches[type][6]);
4722 
4723         pr_err("\nB. Corruption after free\n");
4724         p = kzalloc(128, GFP_KERNEL);
4725         kfree(p);
4726         *p = 0x78;
4727         pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
4728         validate_slab_cache(kmalloc_caches[type][7]);
4729 
4730         p = kzalloc(256, GFP_KERNEL);
4731         kfree(p);
4732         p[50] = 0x9a;
4733         pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
4734         validate_slab_cache(kmalloc_caches[type][8]);
4735 
4736         p = kzalloc(512, GFP_KERNEL);
4737         kfree(p);
4738         p[512] = 0xab;
4739         pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
4740         validate_slab_cache(kmalloc_caches[type][9]);
4741 }
4742 #else
4743 #ifdef CONFIG_SYSFS
4744 static void resiliency_test(void) {};
4745 #endif
4746 #endif
4747 
4748 #ifdef CONFIG_SYSFS
4749 enum slab_stat_type {
4750         SL_ALL,                 /* All slabs */
4751         SL_PARTIAL,             /* Only partially allocated slabs */
4752         SL_CPU,                 /* Only slabs used for cpu caches */
4753         SL_OBJECTS,             /* Determine allocated objects not slabs */
4754         SL_TOTAL                /* Determine object capacity not slabs */
4755 };
4756 
4757 #define SO_ALL          (1 << SL_ALL)
4758 #define SO_PARTIAL      (1 << SL_PARTIAL)
4759 #define SO_CPU          (1 << SL_CPU)
4760 #define SO_OBJECTS      (1 << SL_OBJECTS)
4761 #define SO_TOTAL        (1 << SL_TOTAL)
4762 
4763 #ifdef CONFIG_MEMCG
4764 static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON);
4765 
4766 static int __init setup_slub_memcg_sysfs(char *str)
4767 {
4768         int v;
4769 
4770         if (get_option(&str, &v) > 0)
4771                 memcg_sysfs_enabled = v;
4772 
4773         return 1;
4774 }
4775 
4776 __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs);
4777 #endif
4778 
4779 static ssize_t show_slab_objects(struct kmem_cache *s,
4780                             char *buf, unsigned long flags)
4781 {
4782         unsigned long total = 0;
4783         int node;
4784         int x;
4785         unsigned long *nodes;
4786 
4787         nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
4788         if (!nodes)
4789                 return -ENOMEM;
4790 
4791         if (flags & SO_CPU) {
4792                 int cpu;
4793 
4794                 for_each_possible_cpu(cpu) {
4795                         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4796                                                                cpu);
4797                         int node;
4798                         struct page *page;
4799 
4800                         page = READ_ONCE(c->page);
4801                         if (!page)
4802                                 continue;
4803 
4804                         node = page_to_nid(page);
4805                         if (flags & SO_TOTAL)
4806                                 x = page->objects;
4807                         else if (flags & SO_OBJECTS)
4808                                 x = page->inuse;
4809                         else
4810                                 x = 1;
4811 
4812                         total += x;
4813                         nodes[node] += x;
4814 
4815                         page = slub_percpu_partial_read_once(c);
4816                         if (page) {
4817                                 node = page_to_nid(page);
4818                                 if (flags & SO_TOTAL)
4819                                         WARN_ON_ONCE(1);
4820                                 else if (flags & SO_OBJECTS)
4821                                         WARN_ON_ONCE(1);
4822                                 else
4823                                         x = page->pages;
4824                                 total += x;
4825                                 nodes[node] += x;
4826                         }
4827                 }
4828         }
4829 
4830         get_online_mems();
4831 #ifdef CONFIG_SLUB_DEBUG
4832         if (flags & SO_ALL) {
4833                 struct kmem_cache_node *n;
4834 
4835                 for_each_kmem_cache_node(s, node, n) {
4836 
4837                         if (flags & SO_TOTAL)
4838                                 x = atomic_long_read(&n->total_objects);
4839                         else if (flags & SO_OBJECTS)
4840                                 x = atomic_long_read(&n->total_objects) -
4841                                         count_partial(n, count_free);
4842                         else
4843                                 x = atomic_long_read(&n->nr_slabs);
4844                         total += x;
4845                         nodes[node] += x;
4846                 }
4847 
4848         } else
4849 #endif
4850         if (flags & SO_PARTIAL) {
4851                 struct kmem_cache_node *n;
4852 
4853                 for_each_kmem_cache_node(s, node, n) {
4854                         if (flags & SO_TOTAL)
4855                                 x = count_partial(n, count_total);
4856                         else if (flags & SO_OBJECTS)
4857                                 x = count_partial(n, count_inuse);
4858                         else
4859                                 x = n->nr_partial;
4860                         total += x;
4861                         nodes[node] += x;
4862                 }
4863         }
4864         x = sprintf(buf, "%lu", total);
4865 #ifdef CONFIG_NUMA
4866         for (node = 0; node < nr_node_ids; node++)
4867                 if (nodes[node])
4868                         x += sprintf(buf + x, " N%d=%lu",
4869                                         node, nodes[node]);
4870 #endif
4871         put_online_mems();
4872         kfree(nodes);
4873         return x + sprintf(buf + x, "\n");
4874 }
4875 
4876 #ifdef CONFIG_SLUB_DEBUG
4877 static int any_slab_objects(struct kmem_cache *s)
4878 {
4879         int node;
4880         struct kmem_cache_node *n;
4881 
4882         for_each_kmem_cache_node(s, node, n)
4883                 if (atomic_long_read(&n->total_objects))
4884                         return 1;
4885 
4886         return 0;
4887 }
4888 #endif
4889 
4890 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4891 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4892 
4893 struct slab_attribute {
4894         struct attribute attr;
4895         ssize_t (*show)(struct kmem_cache *s, char *buf);
4896         ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4897 };
4898 
4899 #define SLAB_ATTR_RO(_name) \
4900         static struct slab_attribute _name##_attr = \
4901         __ATTR(_name, 0400, _name##_show, NULL)
4902 
4903 #define SLAB_ATTR(_name) \
4904         static struct slab_attribute _name##_attr =  \
4905         __ATTR(_name, 0600, _name##_show, _name##_store)
4906 
4907 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4908 {
4909         return sprintf(buf, "%u\n", s->size);
4910 }
4911 SLAB_ATTR_RO(slab_size);
4912 
4913 static ssize_t align_show(struct kmem_cache *s, char *buf)
4914 {
4915         return sprintf(buf, "%u\n", s->align);
4916 }
4917 SLAB_ATTR_RO(align);
4918 
4919 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4920 {
4921         return sprintf(buf, "%u\n", s->object_size);
4922 }
4923 SLAB_ATTR_RO(object_size);
4924 
4925 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4926 {
4927         return sprintf(buf, "%u\n", oo_objects(s->oo));
4928 }
4929 SLAB_ATTR_RO(objs_per_slab);
4930 
4931 static ssize_t order_store(struct kmem_cache *s,
4932                                 const char *buf, size_t length)
4933 {
4934         unsigned int order;
4935         int err;
4936 
4937         err = kstrtouint(buf, 10, &order);
4938         if (err)
4939                 return err;
4940 
4941         if (order > slub_max_order || order < slub_min_order)
4942                 return -EINVAL;
4943 
4944         calculate_sizes(s, order);
4945         return length;
4946 }
4947 
4948 static ssize_t order_show(struct kmem_cache *s, char *buf)
4949 {
4950         return sprintf(buf, "%u\n", oo_order(s->oo));
4951 }
4952 SLAB_ATTR(order);
4953 
4954 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4955 {
4956         return sprintf(buf, "%lu\n", s->min_partial);
4957 }
4958 
4959 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4960                                  size_t length)
4961 {
4962         unsigned long min;
4963         int err;
4964 
4965         err = kstrtoul(buf, 10, &min);
4966         if (err)
4967                 return err;
4968 
4969         set_min_partial(s, min);
4970         return length;
4971 }
4972 SLAB_ATTR(min_partial);
4973 
4974 static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4975 {
4976         return sprintf(buf, "%u\n", slub_cpu_partial(s));
4977 }
4978 
4979 static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4980                                  size_t length)
4981 {
4982         unsigned int objects;
4983         int err;
4984 
4985         err = kstrtouint(buf, 10, &objects);
4986         if (err)
4987                 return err;
4988         if (objects && !kmem_cache_has_cpu_partial(s))
4989                 return -EINVAL;
4990 
4991         slub_set_cpu_partial(s, objects);
4992         flush_all(s);
4993         return length;
4994 }
4995 SLAB_ATTR(cpu_partial);
4996 
4997 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4998 {
4999         if (!s->ctor)
5000                 return 0;
5001         return sprintf(buf, "%pS\n", s->ctor);
5002 }
5003 SLAB_ATTR_RO(ctor);
5004 
5005 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5006 {
5007         return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
5008 }
5009 SLAB_ATTR_RO(aliases);
5010 
5011 static ssize_t partial_show(struct kmem_cache *s, char *buf)
5012 {
5013         return show_slab_objects(s, buf, SO_PARTIAL);
5014 }
5015 SLAB_ATTR_RO(partial);
5016 
5017 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5018 {
5019         return show_slab_objects(s, buf, SO_CPU);
5020 }
5021 SLAB_ATTR_RO(cpu_slabs);
5022 
5023 static ssize_t objects_show(struct kmem_cache *s, char *buf)
5024 {
5025         return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
5026 }
5027 SLAB_ATTR_RO(objects);
5028 
5029 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5030 {
5031         return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5032 }
5033 SLAB_ATTR_RO(objects_partial);
5034 
5035 static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5036 {
5037         int objects = 0;
5038         int pages = 0;
5039         int cpu;
5040         int len;
5041 
5042         for_each_online_cpu(cpu) {
5043                 struct page *page;
5044 
5045                 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5046 
5047                 if (page) {
5048                         pages += page->pages;
5049                         objects += page->pobjects;
5050                 }
5051         }
5052 
5053         len = sprintf(buf, "%d(%d)", objects, pages);
5054 
5055 #ifdef CONFIG_SMP
5056         for_each_online_cpu(cpu) {
5057                 struct page *page;
5058 
5059                 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5060 
5061                 if (page && len < PAGE_SIZE - 20)
5062                         len += sprintf(buf + len, " C%d=%d(%d)", cpu,
5063                                 page->pobjects, page->pages);
5064         }
5065 #endif
5066         return len + sprintf(buf + len, "\n");
5067 }
5068 SLAB_ATTR_RO(slabs_cpu_partial);
5069 
5070 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5071 {
5072         return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5073 }
5074 
5075 static ssize_t reclaim_account_store(struct kmem_cache *s,
5076                                 const char *buf, size_t length)
5077 {
5078         s->flags &= ~SLAB_RECLAIM_ACCOUNT;
5079         if (buf[0] == '1')
5080                 s->flags |= SLAB_RECLAIM_ACCOUNT;
5081         return length;
5082 }
5083 SLAB_ATTR(reclaim_account);
5084 
5085 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5086 {
5087         return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
5088 }
5089 SLAB_ATTR_RO(hwcache_align);
5090 
5091 #ifdef CONFIG_ZONE_DMA
5092 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5093 {
5094         return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5095 }
5096 SLAB_ATTR_RO(cache_dma);
5097 #endif
5098 
5099 static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5100 {
5101         return sprintf(buf, "%u\n", s->usersize);
5102 }
5103 SLAB_ATTR_RO(usersize);
5104 
5105 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5106 {
5107         return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
5108 }
5109 SLAB_ATTR_RO(destroy_by_rcu);
5110 
5111 #ifdef CONFIG_SLUB_DEBUG
5112 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5113 {
5114         return show_slab_objects(s, buf, SO_ALL);
5115 }
5116 SLAB_ATTR_RO(slabs);
5117 
5118 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5119 {
5120         return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5121 }
5122 SLAB_ATTR_RO(total_objects);
5123 
5124 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5125 {
5126         return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
5127 }
5128 
5129 static ssize_t sanity_checks_store(struct kmem_cache *s,
5130                                 const char *buf, size_t length)
5131 {
5132         s->flags &= ~SLAB_CONSISTENCY_CHECKS;
5133         if (buf[0] == '1') {
5134                 s->flags &= ~__CMPXCHG_DOUBLE;
5135                 s->flags |= SLAB_CONSISTENCY_CHECKS;
5136         }
5137         return length;
5138 }
5139 SLAB_ATTR(sanity_checks);
5140 
5141 static ssize_t trace_show(struct kmem_cache *s, char *buf)
5142 {
5143         return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5144 }
5145 
5146 static ssize_t trace_store(struct kmem_cache *s, const char *buf,
5147                                                         size_t length)
5148 {
5149         /*
5150          * Tracing a merged cache is going to give confusing results
5151          * as well as cause other issues like converting a mergeable
5152          * cache into an umergeable one.
5153          */
5154         if (s->refcount > 1)
5155                 return -EINVAL;
5156 
5157         s->flags &= ~SLAB_TRACE;
5158         if (buf[0] == '1') {
5159                 s->flags &= ~__CMPXCHG_DOUBLE;
5160                 s->flags |= SLAB_TRACE;
5161         }
5162         return length;
5163 }
5164 SLAB_ATTR(trace);
5165 
5166 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5167 {
5168         return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5169 }
5170 
5171 static ssize_t red_zone_store(struct kmem_cache *s,
5172                                 const char *buf, size_t length)
5173 {
5174         if (any_slab_objects(s))
5175                 return -EBUSY;
5176 
5177         s->flags &= ~SLAB_RED_ZONE;
5178         if (buf[0] == '1') {
5179                 s->flags |= SLAB_RED_ZONE;
5180         }
5181         calculate_sizes(s, -1);
5182         return length;
5183 }
5184 SLAB_ATTR(red_zone);
5185 
5186 static ssize_t poison_show(struct kmem_cache *s, char *buf)
5187 {
5188         return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
5189 }
5190 
5191 static ssize_t poison_store(struct kmem_cache *s,
5192                                 const char *buf, size_t length)
5193 {
5194         if (any_slab_objects(s))
5195                 return -EBUSY;
5196 
5197         s->flags &= ~SLAB_POISON;
5198         if (buf[0] == '1') {
5199                 s->flags |= SLAB_POISON;
5200         }
5201         calculate_sizes(s, -1);
5202         return length;
5203 }
5204 SLAB_ATTR(poison);
5205 
5206 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5207 {
5208         return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5209 }
5210 
5211 static ssize_t store_user_store(struct kmem_cache *s,
5212                                 const char *buf, size_t length)
5213 {
5214         if (any_slab_objects(s))
5215                 return -EBUSY;
5216 
5217         s->flags &= ~SLAB_STORE_USER;
5218         if (buf[0] == '1') {
5219                 s->flags &= ~__CMPXCHG_DOUBLE;
5220                 s->flags |= SLAB_STORE_USER;
5221         }
5222         calculate_sizes(s, -1);
5223         return length;
5224 }
5225 SLAB_ATTR(store_user);
5226 
5227 static ssize_t validate_show(struct kmem_cache *s, char *buf)
5228 {
5229         return 0;
5230 }
5231 
5232 static ssize_t validate_store(struct kmem_cache *s,
5233                         const char *buf, size_t length)
5234 {
5235         int ret = -EINVAL;
5236 
5237         if (buf[0] == '1') {
5238                 ret = validate_slab_cache(s);
5239                 if (ret >= 0)
5240                         ret = length;
5241         }
5242         return ret;
5243 }
5244 SLAB_ATTR(validate);
5245 
5246 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
5247 {
5248         if (!(s->flags & SLAB_STORE_USER))
5249                 return -ENOSYS;
5250         return list_locations(s, buf, TRACK_ALLOC);
5251 }
5252 SLAB_ATTR_RO(alloc_calls);
5253 
5254 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
5255 {
5256         if (!(s->flags & SLAB_STORE_USER))
5257                 return -ENOSYS;
5258         return list_locations(s, buf, TRACK_FREE);
5259 }
5260 SLAB_ATTR_RO(free_calls);
5261 #endif /* CONFIG_SLUB_DEBUG */
5262 
5263 #ifdef CONFIG_FAILSLAB
5264 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5265 {
5266         return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5267 }
5268 
5269 static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5270                                                         size_t length)
5271 {
5272         if (s->refcount > 1)
5273                 return -EINVAL;
5274 
5275         s->flags &= ~SLAB_FAILSLAB;
5276         if (buf[0] == '1')
5277                 s->flags |= SLAB_FAILSLAB;
5278         return length;
5279 }
5280 SLAB_ATTR(failslab);
5281 #endif
5282 
5283 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5284 {
5285         return 0;
5286 }
5287 
5288 static ssize_t shrink_store(struct kmem_cache *s,
5289                         const char *buf, size_t length)
5290 {
5291         if (buf[0] == '1')
5292                 kmem_cache_shrink(s);
5293         else
5294                 return -EINVAL;
5295         return length;
5296 }
5297 SLAB_ATTR(shrink);
5298 
5299 #ifdef CONFIG_NUMA
5300 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
5301 {
5302         return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10);
5303 }
5304 
5305 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
5306                                 const char *buf, size_t length)
5307 {
5308         unsigned int ratio;
5309         int err;
5310 
5311         err = kstrtouint(buf, 10, &ratio);
5312         if (err)
5313                 return err;
5314         if (ratio > 100)
5315                 return -ERANGE;
5316 
5317         s->remote_node_defrag_ratio = ratio * 10;
5318 
5319         return length;
5320 }
5321 SLAB_ATTR(remote_node_defrag_ratio);
5322 #endif
5323 
5324 #ifdef CONFIG_SLUB_STATS
5325 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5326 {
5327         unsigned long sum  = 0;
5328         int cpu;
5329         int len;
5330         int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
5331 
5332         if (!data)
5333                 return -ENOMEM;
5334 
5335         for_each_online_cpu(cpu) {
5336                 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
5337 
5338                 data[cpu] = x;
5339                 sum += x;
5340         }
5341 
5342         len = sprintf(buf, "%lu", sum);
5343 
5344 #ifdef CONFIG_SMP
5345         for_each_online_cpu(cpu) {
5346                 if (data[cpu] && len < PAGE_SIZE - 20)
5347                         len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
5348         }
5349 #endif
5350         kfree(data);
5351         return len + sprintf(buf + len, "\n");
5352 }
5353 
5354 static void clear_stat(struct kmem_cache *s, enum stat_item si)
5355 {
5356         int cpu;
5357 
5358         for_each_online_cpu(cpu)
5359                 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
5360 }
5361 
5362 #define STAT_ATTR(si, text)                                     \
5363 static ssize_t text##_show(struct kmem_cache *s, char *buf)     \
5364 {                                                               \
5365         return show_stat(s, buf, si);                           \
5366 }                                                               \
5367 static ssize_t text##_store(struct kmem_cache *s,               \
5368                                 const char *buf, size_t length) \
5369 {                                                               \
5370         if (buf[0] != '')                                      \
5371                 return -EINVAL;                                 \
5372         clear_stat(s, si);                                      \
5373         return length;                                          \
5374 }                                                               \
5375 SLAB_ATTR(text);                                                \
5376 
5377 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5378 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5379 STAT_ATTR(FREE_FASTPATH, free_fastpath);
5380 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5381 STAT_ATTR(FREE_FROZEN, free_frozen);
5382 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5383 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5384 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5385 STAT_ATTR(ALLOC_SLAB, alloc_slab);
5386 STAT_ATTR(ALLOC_REFILL, alloc_refill);
5387 STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
5388 STAT_ATTR(FREE_SLAB, free_slab);
5389 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5390 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5391 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5392 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5393 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5394 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
5395 STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
5396 STAT_ATTR(ORDER_FALLBACK, order_fallback);
5397 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5398 STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
5399 STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5400 STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
5401 STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5402 STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
5403 #endif
5404 
5405 static struct attribute *slab_attrs[] = {
5406         &slab_size_attr.attr,
5407         &object_size_attr.attr,
5408         &objs_per_slab_attr.attr,
5409         &order_attr.attr,
5410         &min_partial_attr.attr,
5411         &cpu_partial_attr.attr,
5412         &objects_attr.attr,
5413         &objects_partial_attr.attr,
5414         &partial_attr.attr,
5415         &cpu_slabs_attr.attr,
5416         &ctor_attr.attr,
5417         &aliases_attr.attr,
5418         &align_attr.attr,
5419         &hwcache_align_attr.attr,
5420         &reclaim_account_attr.attr,
5421         &destroy_by_rcu_attr.attr,
5422         &shrink_attr.attr,
5423         &slabs_cpu_partial_attr.attr,
5424 #ifdef CONFIG_SLUB_DEBUG
5425         &total_objects_attr.attr,
5426         &slabs_attr.attr,
5427         &sanity_checks_attr.attr,
5428         &trace_attr.attr,
5429         &red_zone_attr.attr,
5430         &poison_attr.attr,
5431         &store_user_attr.attr,
5432         &validate_attr.attr,
5433         &alloc_calls_attr.attr,
5434         &free_calls_attr.attr,
5435 #endif
5436 #ifdef CONFIG_ZONE_DMA
5437         &cache_dma_attr.attr,
5438 #endif
5439 #ifdef CONFIG_NUMA
5440         &remote_node_defrag_ratio_attr.attr,
5441 #endif
5442 #ifdef CONFIG_SLUB_STATS
5443         &alloc_fastpath_attr.attr,
5444         &alloc_slowpath_attr.attr,
5445         &free_fastpath_attr.attr,
5446         &free_slowpath_attr.attr,
5447         &free_frozen_attr.attr,
5448         &free_add_partial_attr.attr,
5449         &free_remove_partial_attr.attr,
5450         &alloc_from_partial_attr.attr,
5451         &alloc_slab_attr.attr,
5452         &alloc_refill_attr.attr,
5453         &alloc_node_mismatch_attr.attr,
5454         &free_slab_attr.attr,
5455         &cpuslab_flush_attr.attr,
5456         &deactivate_full_attr.attr,
5457         &deactivate_empty_attr.attr,
5458         &deactivate_to_head_attr.attr,
5459         &deactivate_to_tail_attr.attr,
5460         &deactivate_remote_frees_attr.attr,
5461         &deactivate_bypass_attr.attr,
5462         &order_fallback_attr.attr,
5463         &cmpxchg_double_fail_attr.attr,
5464         &cmpxchg_double_cpu_fail_attr.attr,
5465         &cpu_partial_alloc_attr.attr,
5466         &cpu_partial_free_attr.attr,
5467         &cpu_partial_node_attr.attr,
5468         &cpu_partial_drain_attr.attr,
5469 #endif
5470 #ifdef CONFIG_FAILSLAB
5471         &failslab_attr.attr,
5472 #endif
5473         &usersize_attr.attr,
5474 
5475         NULL
5476 };
5477 
5478 static const struct attribute_group slab_attr_group = {
5479         .attrs = slab_attrs,
5480 };
5481 
5482 static ssize_t slab_attr_show(struct kobject *kobj,
5483                                 struct attribute *attr,
5484                                 char *buf)
5485 {
5486         struct slab_attribute *attribute;
5487         struct kmem_cache *s;
5488         int err;
5489 
5490         attribute = to_slab_attr(attr);
5491         s = to_slab(kobj);
5492 
5493         if (!attribute->show)
5494                 return -EIO;
5495 
5496         err = attribute->show(s, buf);
5497 
5498         return err;
5499 }
5500 
5501 static ssize_t slab_attr_store(struct kobject *kobj,
5502                                 struct attribute *attr,
5503                                 const char *buf, size_t len)
5504 {
5505         struct slab_attribute *attribute;
5506         struct kmem_cache *s;
5507         int err;
5508 
5509         attribute = to_slab_attr(attr);
5510         s = to_slab(kobj);
5511 
5512         if (!attribute->store)
5513                 return -EIO;
5514 
5515         err = attribute->store(s, buf, len);
5516 #ifdef CONFIG_MEMCG
5517         if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
5518                 struct kmem_cache *c;
5519 
5520                 mutex_lock(&slab_mutex);
5521                 if (s->max_attr_size < len)
5522                         s->max_attr_size = len;
5523 
5524                 /*
5525                  * This is a best effort propagation, so this function's return
5526                  * value will be determined by the parent cache only. This is
5527                  * basically because not all attributes will have a well
5528                  * defined semantics for rollbacks - most of the actions will
5529                  * have permanent effects.
5530                  *
5531                  * Returning the error value of any of the children that fail
5532                  * is not 100 % defined, in the sense that users seeing the
5533                  * error code won't be able to know anything about the state of
5534                  * the cache.
5535                  *
5536                  * Only returning the error code for the parent cache at least
5537                  * has well defined semantics. The cache being written to
5538                  * directly either failed or succeeded, in which case we loop
5539                  * through the descendants with best-effort propagation.
5540                  */
5541                 for_each_memcg_cache(c, s)
5542                         attribute->store(c, buf, len);
5543                 mutex_unlock(&slab_mutex);
5544         }
5545 #endif
5546         return err;
5547 }
5548 
5549 static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5550 {
5551 #ifdef CONFIG_MEMCG
5552         int i;
5553         char *buffer = NULL;
5554         struct kmem_cache *root_cache;
5555 
5556         if (is_root_cache(s))
5557                 return;
5558 
5559         root_cache = s->memcg_params.root_cache;
5560 
5561         /*
5562          * This mean this cache had no attribute written. Therefore, no point
5563          * in copying default values around
5564          */
5565         if (!root_cache->max_attr_size)
5566                 return;
5567 
5568         for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5569                 char mbuf[64];
5570                 char *buf;
5571                 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
5572                 ssize_t len;
5573 
5574                 if (!attr || !attr->store || !attr->show)
5575                         continue;
5576 
5577                 /*
5578                  * It is really bad that we have to allocate here, so we will
5579                  * do it only as a fallback. If we actually allocate, though,
5580                  * we can just use the allocated buffer until the end.
5581                  *
5582                  * Most of the slub attributes will tend to be very small in
5583                  * size, but sysfs allows buffers up to a page, so they can
5584                  * theoretically happen.
5585                  */
5586                 if (buffer)
5587                         buf = buffer;
5588                 else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf))
5589                         buf = mbuf;
5590                 else {
5591                         buffer = (char *) get_zeroed_page(GFP_KERNEL);
5592                         if (WARN_ON(!buffer))
5593                                 continue;
5594                         buf = buffer;
5595                 }
5596 
5597                 len = attr->show(root_cache, buf);
5598                 if (len > 0)
5599                         attr->store(s, buf, len);
5600         }
5601 
5602         if (buffer)
5603                 free_page((unsigned long)buffer);
5604 #endif
5605 }
5606 
5607 static void kmem_cache_release(struct kobject *k)
5608 {
5609         slab_kmem_cache_release(to_slab(k));
5610 }
5611 
5612 static const struct sysfs_ops slab_sysfs_ops = {
5613         .show = slab_attr_show,
5614         .store = slab_attr_store,
5615 };
5616 
5617 static struct kobj_type slab_ktype = {
5618         .sysfs_ops = &slab_sysfs_ops,
5619         .release = kmem_cache_release,
5620 };
5621 
5622 static int uevent_filter(struct kset *kset, struct kobject *kobj)
5623 {
5624         struct kobj_type *ktype = get_ktype(kobj);
5625 
5626         if (ktype == &slab_ktype)
5627                 return 1;
5628         return 0;
5629 }
5630 
5631 static const struct kset_uevent_ops slab_uevent_ops = {
5632         .filter = uevent_filter,
5633 };
5634 
5635 static struct kset *slab_kset;
5636 
5637 static inline struct kset *cache_kset(struct kmem_cache *s)
5638 {
5639 #ifdef CONFIG_MEMCG
5640         if (!is_root_cache(s))
5641                 return s->memcg_params.root_cache->memcg_kset;
5642 #endif
5643         return slab_kset;
5644 }
5645 
5646 #define ID_STR_LENGTH 64
5647 
5648 /* Create a unique string id for a slab cache:
5649  *
5650  * Format       :[flags-]size
5651  */
5652 static char *create_unique_id(struct kmem_cache *s)
5653 {
5654         char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5655         char *p = name;
5656 
5657         BUG_ON(!name);
5658 
5659         *p++ = ':';
5660         /*
5661          * First flags affecting slabcache operations. We will only
5662          * get here for aliasable slabs so we do not need to support
5663          * too many flags. The flags here must cover all flags that
5664          * are matched during merging to guarantee that the id is
5665          * unique.
5666          */
5667         if (s->flags & SLAB_CACHE_DMA)
5668                 *p++ = 'd';
5669         if (s->flags & SLAB_RECLAIM_ACCOUNT)
5670                 *p++ = 'a';
5671         if (s->flags & SLAB_CONSISTENCY_CHECKS)
5672                 *p++ = 'F';
5673         if (s->flags & SLAB_ACCOUNT)
5674                 *p++ = 'A';
5675         if (p != name + 1)
5676                 *p++ = '-';
5677         p += sprintf(p, "%07u", s->size);
5678 
5679         BUG_ON(p > name + ID_STR_LENGTH - 1);
5680         return name;
5681 }
5682 
5683 static void sysfs_slab_remove_workfn(struct work_struct *work)
5684 {
5685         struct kmem_cache *s =
5686                 container_of(work, struct kmem_cache, kobj_remove_work);
5687 
5688         if (!s->kobj.state_in_sysfs)
5689                 /*
5690                  * For a memcg cache, this may be called during
5691                  * deactivation and again on shutdown.  Remove only once.
5692                  * A cache is never shut down before deactivation is
5693                  * complete, so no need to worry about synchronization.
5694                  */
5695                 goto out;
5696 
5697 #ifdef CONFIG_MEMCG
5698         kset_unregister(s->memcg_kset);
5699 #endif
5700         kobject_uevent(&s->kobj, KOBJ_REMOVE);
5701 out:
5702         kobject_put(&s->kobj);
5703 }
5704 
5705 static int sysfs_slab_add(struct kmem_cache *s)
5706 {
5707         int err;
5708         const char *name;
5709         struct kset *kset = cache_kset(s);
5710         int unmergeable = slab_unmergeable(s);
5711 
5712         INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn);
5713 
5714         if (!kset) {
5715                 kobject_init(&s->kobj, &slab_ktype);
5716                 return 0;
5717         }
5718 
5719         if (!unmergeable && disable_higher_order_debug &&
5720                         (slub_debug & DEBUG_METADATA_FLAGS))
5721                 unmergeable = 1;
5722 
5723         if (unmergeable) {
5724                 /*
5725                  * Slabcache can never be merged so we can use the name proper.
5726                  * This is typically the case for debug situations. In that
5727                  * case we can catch duplicate names easily.
5728                  */
5729                 sysfs_remove_link(&slab_kset->kobj, s->name);
5730                 name = s->name;
5731         } else {
5732                 /*
5733                  * Create a unique name for the slab as a target
5734                  * for the symlinks.
5735                  */
5736                 name = create_unique_id(s);
5737         }
5738 
5739         s->kobj.kset = kset;
5740         err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
5741         if (err)
5742                 goto out;
5743 
5744         err = sysfs_create_group(&s->kobj, &slab_attr_group);
5745         if (err)
5746                 goto out_del_kobj;
5747 
5748 #ifdef CONFIG_MEMCG
5749         if (is_root_cache(s) && memcg_sysfs_enabled) {
5750                 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5751                 if (!s->memcg_kset) {
5752                         err = -ENOMEM;
5753                         goto out_del_kobj;
5754                 }
5755         }
5756 #endif
5757 
5758         kobject_uevent(&s->kobj, KOBJ_ADD);
5759         if (!unmergeable) {
5760                 /* Setup first alias */
5761                 sysfs_slab_alias(s, s->name);
5762         }
5763 out:
5764         if (!unmergeable)
5765                 kfree(name);
5766         return err;
5767 out_del_kobj:
5768         kobject_del(&s->kobj);
5769         goto out;
5770 }
5771 
5772 static void sysfs_slab_remove(struct kmem_cache *s)
5773 {
5774         if (slab_state < FULL)
5775                 /*
5776                  * Sysfs has not been setup yet so no need to remove the
5777                  * cache from sysfs.
5778                  */
5779                 return;
5780 
5781         kobject_get(&s->kobj);
5782         schedule_work(&s->kobj_remove_work);
5783 }
5784 
5785 void sysfs_slab_unlink(struct kmem_cache *s)
5786 {
5787         if (slab_state >= FULL)
5788                 kobject_del(&s->kobj);
5789 }
5790 
5791 void sysfs_slab_release(struct kmem_cache *s)
5792 {
5793         if (slab_state >= FULL)
5794                 kobject_put(&s->kobj);
5795 }
5796 
5797 /*
5798  * Need to buffer aliases during bootup until sysfs becomes
5799  * available lest we lose that information.
5800  */
5801 struct saved_alias {
5802         struct kmem_cache *s;
5803         const char *name;
5804         struct saved_alias *next;
5805 };
5806 
5807 static struct saved_alias *alias_list;
5808 
5809 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5810 {
5811         struct saved_alias *al;
5812 
5813         if (slab_state == FULL) {
5814                 /*
5815                  * If we have a leftover link then remove it.
5816                  */
5817                 sysfs_remove_link(&slab_kset->kobj, name);
5818                 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
5819         }
5820 
5821         al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5822         if (!al)
5823                 return -ENOMEM;
5824 
5825         al->s = s;
5826         al->name = name;
5827         al->next = alias_list;
5828         alias_list = al;
5829         return 0;
5830 }
5831 
5832 static int __init slab_sysfs_init(void)
5833 {
5834         struct kmem_cache *s;
5835         int err;
5836 
5837         mutex_lock(&slab_mutex);
5838 
5839         slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
5840         if (!slab_kset) {
5841                 mutex_unlock(&slab_mutex);
5842                 pr_err("Cannot register slab subsystem.\n");
5843                 return -ENOSYS;
5844         }
5845 
5846         slab_state = FULL;
5847 
5848         list_for_each_entry(s, &slab_caches, list) {
5849                 err = sysfs_slab_add(s);
5850                 if (err)
5851                         pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5852                                s->name);
5853         }
5854 
5855         while (alias_list) {
5856                 struct saved_alias *al = alias_list;
5857 
5858                 alias_list = alias_list->next;
5859                 err = sysfs_slab_alias(al->s, al->name);
5860                 if (err)
5861                         pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5862                                al->name);
5863                 kfree(al);
5864         }
5865 
5866         mutex_unlock(&slab_mutex);
5867         resiliency_test();
5868         return 0;
5869 }
5870 
5871 __initcall(slab_sysfs_init);
5872 #endif /* CONFIG_SYSFS */
5873 
5874 /*
5875  * The /proc/slabinfo ABI
5876  */
5877 #ifdef CONFIG_SLUB_DEBUG
5878 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
5879 {
5880         unsigned long nr_slabs = 0;
5881         unsigned long nr_objs = 0;
5882         unsigned long nr_free = 0;
5883         int node;
5884         struct kmem_cache_node *n;
5885 
5886         for_each_kmem_cache_node(s, node, n) {
5887                 nr_slabs += node_nr_slabs(n);
5888                 nr_objs += node_nr_objs(n);
5889                 nr_free += count_partial(n, count_free);
5890         }
5891 
5892         sinfo->active_objs = nr_objs - nr_free;
5893         sinfo->num_objs = nr_objs;
5894         sinfo->active_slabs = nr_slabs;
5895         sinfo->num_slabs = nr_slabs;
5896         sinfo->objects_per_slab = oo_objects(s->oo);
5897         sinfo->cache_order = oo_order(s->oo);
5898 }
5899 
5900 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
5901 {
5902 }
5903 
5904 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5905                        size_t count, loff_t *ppos)
5906 {
5907         return -EIO;
5908 }
5909 #endif /* CONFIG_SLUB_DEBUG */
5910 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp