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

TOMOYO Linux Cross Reference
Linux/mm/slub.c

Version: ~ [ linux-5.11-rc3 ] ~ [ linux-5.10.7 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.89 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.167 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.215 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.251 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.251 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ 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/notifier.h>
 23 #include <linux/seq_file.h>
 24 #include <linux/kasan.h>
 25 #include <linux/cpu.h>
 26 #include <linux/cpuset.h>
 27 #include <linux/mempolicy.h>
 28 #include <linux/ctype.h>
 29 #include <linux/debugobjects.h>
 30 #include <linux/kallsyms.h>
 31 #include <linux/memory.h>
 32 #include <linux/math64.h>
 33 #include <linux/fault-inject.h>
 34 #include <linux/stacktrace.h>
 35 #include <linux/prefetch.h>
 36 #include <linux/memcontrol.h>
 37 #include <linux/random.h>
 38 
 39 #include <trace/events/kmem.h>
 40 
 41 #include "internal.h"
 42 
 43 /*
 44  * Lock order:
 45  *   1. slab_mutex (Global Mutex)
 46  *   2. node->list_lock
 47  *   3. slab_lock(page) (Only on some arches and for debugging)
 48  *
 49  *   slab_mutex
 50  *
 51  *   The role of the slab_mutex is to protect the list of all the slabs
 52  *   and to synchronize major metadata changes to slab cache structures.
 53  *
 54  *   The slab_lock is only used for debugging and on arches that do not
 55  *   have the ability to do a cmpxchg_double. It only protects the second
 56  *   double word in the page struct. Meaning
 57  *      A. page->freelist       -> List of object free in a page
 58  *      B. page->counters       -> Counters of objects
 59  *      C. page->frozen         -> frozen state
 60  *
 61  *   If a slab is frozen then it is exempt from list management. It is not
 62  *   on any list. The processor that froze the slab is the one who can
 63  *   perform list operations on the page. Other processors may put objects
 64  *   onto the freelist but the processor that froze the slab is the only
 65  *   one that can retrieve the objects from the page's freelist.
 66  *
 67  *   The list_lock protects the partial and full list on each node and
 68  *   the partial slab counter. If taken then no new slabs may be added or
 69  *   removed from the lists nor make the number of partial slabs be modified.
 70  *   (Note that the total number of slabs is an atomic value that may be
 71  *   modified without taking the list lock).
 72  *
 73  *   The list_lock is a centralized lock and thus we avoid taking it as
 74  *   much as possible. As long as SLUB does not have to handle partial
 75  *   slabs, operations can continue without any centralized lock. F.e.
 76  *   allocating a long series of objects that fill up slabs does not require
 77  *   the list lock.
 78  *   Interrupts are disabled during allocation and deallocation in order to
 79  *   make the slab allocator safe to use in the context of an irq. In addition
 80  *   interrupts are disabled to ensure that the processor does not change
 81  *   while handling per_cpu slabs, due to kernel preemption.
 82  *
 83  * SLUB assigns one slab for allocation to each processor.
 84  * Allocations only occur from these slabs called cpu slabs.
 85  *
 86  * Slabs with free elements are kept on a partial list and during regular
 87  * operations no list for full slabs is used. If an object in a full slab is
 88  * freed then the slab will show up again on the partial lists.
 89  * We track full slabs for debugging purposes though because otherwise we
 90  * cannot scan all objects.
 91  *
 92  * Slabs are freed when they become empty. Teardown and setup is
 93  * minimal so we rely on the page allocators per cpu caches for
 94  * fast frees and allocs.
 95  *
 96  * Overloading of page flags that are otherwise used for LRU management.
 97  *
 98  * PageActive           The slab is frozen and exempt from list processing.
 99  *                      This means that the slab is dedicated to a purpose
100  *                      such as satisfying allocations for a specific
101  *                      processor. Objects may be freed in the slab while
102  *                      it is frozen but slab_free will then skip the usual
103  *                      list operations. It is up to the processor holding
104  *                      the slab to integrate the slab into the slab lists
105  *                      when the slab is no longer needed.
106  *
107  *                      One use of this flag is to mark slabs that are
108  *                      used for allocations. Then such a slab becomes a cpu
109  *                      slab. The cpu slab may be equipped with an additional
110  *                      freelist that allows lockless access to
111  *                      free objects in addition to the regular freelist
112  *                      that requires the slab lock.
113  *
114  * PageError            Slab requires special handling due to debug
115  *                      options set. This moves slab handling out of
116  *                      the fast path and disables lockless freelists.
117  */
118 
119 static inline int kmem_cache_debug(struct kmem_cache *s)
120 {
121 #ifdef CONFIG_SLUB_DEBUG
122         return unlikely(s->flags & SLAB_DEBUG_FLAGS);
123 #else
124         return 0;
125 #endif
126 }
127 
128 void *fixup_red_left(struct kmem_cache *s, void *p)
129 {
130         if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE)
131                 p += s->red_left_pad;
132 
133         return p;
134 }
135 
136 static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
137 {
138 #ifdef CONFIG_SLUB_CPU_PARTIAL
139         return !kmem_cache_debug(s);
140 #else
141         return false;
142 #endif
143 }
144 
145 /*
146  * Issues still to be resolved:
147  *
148  * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
149  *
150  * - Variable sizing of the per node arrays
151  */
152 
153 /* Enable to test recovery from slab corruption on boot */
154 #undef SLUB_RESILIENCY_TEST
155 
156 /* Enable to log cmpxchg failures */
157 #undef SLUB_DEBUG_CMPXCHG
158 
159 /*
160  * Mininum number of partial slabs. These will be left on the partial
161  * lists even if they are empty. kmem_cache_shrink may reclaim them.
162  */
163 #define MIN_PARTIAL 5
164 
165 /*
166  * Maximum number of desirable partial slabs.
167  * The existence of more partial slabs makes kmem_cache_shrink
168  * sort the partial list by the number of objects in use.
169  */
170 #define MAX_PARTIAL 10
171 
172 #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
173                                 SLAB_POISON | SLAB_STORE_USER)
174 
175 /*
176  * These debug flags cannot use CMPXCHG because there might be consistency
177  * issues when checking or reading debug information
178  */
179 #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
180                                 SLAB_TRACE)
181 
182 
183 /*
184  * Debugging flags that require metadata to be stored in the slab.  These get
185  * disabled when slub_debug=O is used and a cache's min order increases with
186  * metadata.
187  */
188 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
189 
190 #define OO_SHIFT        16
191 #define OO_MASK         ((1 << OO_SHIFT) - 1)
192 #define MAX_OBJS_PER_PAGE       32767 /* since page.objects is u15 */
193 
194 /* Internal SLUB flags */
195 /* Poison object */
196 #define __OBJECT_POISON         ((slab_flags_t __force)0x80000000U)
197 /* Use cmpxchg_double */
198 #define __CMPXCHG_DOUBLE        ((slab_flags_t __force)0x40000000U)
199 
200 /*
201  * Tracking user of a slab.
202  */
203 #define TRACK_ADDRS_COUNT 16
204 struct track {
205         unsigned long addr;     /* Called from address */
206 #ifdef CONFIG_STACKTRACE
207         unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
208 #endif
209         int cpu;                /* Was running on cpu */
210         int pid;                /* Pid context */
211         unsigned long when;     /* When did the operation occur */
212 };
213 
214 enum track_item { TRACK_ALLOC, TRACK_FREE };
215 
216 #ifdef CONFIG_SYSFS
217 static int sysfs_slab_add(struct kmem_cache *);
218 static int sysfs_slab_alias(struct kmem_cache *, const char *);
219 static void memcg_propagate_slab_attrs(struct kmem_cache *s);
220 static void sysfs_slab_remove(struct kmem_cache *s);
221 #else
222 static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
223 static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
224                                                         { return 0; }
225 static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
226 static inline void sysfs_slab_remove(struct kmem_cache *s) { }
227 #endif
228 
229 static inline void stat(const struct kmem_cache *s, enum stat_item si)
230 {
231 #ifdef CONFIG_SLUB_STATS
232         /*
233          * The rmw is racy on a preemptible kernel but this is acceptable, so
234          * avoid this_cpu_add()'s irq-disable overhead.
235          */
236         raw_cpu_inc(s->cpu_slab->stat[si]);
237 #endif
238 }
239 
240 /********************************************************************
241  *                      Core slab cache functions
242  *******************************************************************/
243 
244 /*
245  * Returns freelist pointer (ptr). With hardening, this is obfuscated
246  * with an XOR of the address where the pointer is held and a per-cache
247  * random number.
248  */
249 static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
250                                  unsigned long ptr_addr)
251 {
252 #ifdef CONFIG_SLAB_FREELIST_HARDENED
253         return (void *)((unsigned long)ptr ^ s->random ^ ptr_addr);
254 #else
255         return ptr;
256 #endif
257 }
258 
259 /* Returns the freelist pointer recorded at location ptr_addr. */
260 static inline void *freelist_dereference(const struct kmem_cache *s,
261                                          void *ptr_addr)
262 {
263         return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr),
264                             (unsigned long)ptr_addr);
265 }
266 
267 static inline void *get_freepointer(struct kmem_cache *s, void *object)
268 {
269         return freelist_dereference(s, object + s->offset);
270 }
271 
272 static void prefetch_freepointer(const struct kmem_cache *s, void *object)
273 {
274         if (object)
275                 prefetch(freelist_dereference(s, object + s->offset));
276 }
277 
278 static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
279 {
280         unsigned long freepointer_addr;
281         void *p;
282 
283         if (!debug_pagealloc_enabled())
284                 return get_freepointer(s, object);
285 
286         freepointer_addr = (unsigned long)object + s->offset;
287         probe_kernel_read(&p, (void **)freepointer_addr, sizeof(p));
288         return freelist_ptr(s, p, freepointer_addr);
289 }
290 
291 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
292 {
293         unsigned long freeptr_addr = (unsigned long)object + s->offset;
294 
295 #ifdef CONFIG_SLAB_FREELIST_HARDENED
296         BUG_ON(object == fp); /* naive detection of double free or corruption */
297 #endif
298 
299         *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
300 }
301 
302 /* Loop over all objects in a slab */
303 #define for_each_object(__p, __s, __addr, __objects) \
304         for (__p = fixup_red_left(__s, __addr); \
305                 __p < (__addr) + (__objects) * (__s)->size; \
306                 __p += (__s)->size)
307 
308 #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \
309         for (__p = fixup_red_left(__s, __addr), __idx = 1; \
310                 __idx <= __objects; \
311                 __p += (__s)->size, __idx++)
312 
313 /* Determine object index from a given position */
314 static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
315 {
316         return (p - addr) / s->size;
317 }
318 
319 static inline int order_objects(int order, unsigned long size, int reserved)
320 {
321         return ((PAGE_SIZE << order) - reserved) / size;
322 }
323 
324 static inline struct kmem_cache_order_objects oo_make(int order,
325                 unsigned long size, int reserved)
326 {
327         struct kmem_cache_order_objects x = {
328                 (order << OO_SHIFT) + order_objects(order, size, reserved)
329         };
330 
331         return x;
332 }
333 
334 static inline int oo_order(struct kmem_cache_order_objects x)
335 {
336         return x.x >> OO_SHIFT;
337 }
338 
339 static inline int oo_objects(struct kmem_cache_order_objects x)
340 {
341         return x.x & OO_MASK;
342 }
343 
344 /*
345  * Per slab locking using the pagelock
346  */
347 static __always_inline void slab_lock(struct page *page)
348 {
349         VM_BUG_ON_PAGE(PageTail(page), page);
350         bit_spin_lock(PG_locked, &page->flags);
351 }
352 
353 static __always_inline void slab_unlock(struct page *page)
354 {
355         VM_BUG_ON_PAGE(PageTail(page), page);
356         __bit_spin_unlock(PG_locked, &page->flags);
357 }
358 
359 static inline void set_page_slub_counters(struct page *page, unsigned long counters_new)
360 {
361         struct page tmp;
362         tmp.counters = counters_new;
363         /*
364          * page->counters can cover frozen/inuse/objects as well
365          * as page->_refcount.  If we assign to ->counters directly
366          * we run the risk of losing updates to page->_refcount, so
367          * be careful and only assign to the fields we need.
368          */
369         page->frozen  = tmp.frozen;
370         page->inuse   = tmp.inuse;
371         page->objects = tmp.objects;
372 }
373 
374 /* Interrupts must be disabled (for the fallback code to work right) */
375 static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
376                 void *freelist_old, unsigned long counters_old,
377                 void *freelist_new, unsigned long counters_new,
378                 const char *n)
379 {
380         VM_BUG_ON(!irqs_disabled());
381 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
382     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
383         if (s->flags & __CMPXCHG_DOUBLE) {
384                 if (cmpxchg_double(&page->freelist, &page->counters,
385                                    freelist_old, counters_old,
386                                    freelist_new, counters_new))
387                         return true;
388         } else
389 #endif
390         {
391                 slab_lock(page);
392                 if (page->freelist == freelist_old &&
393                                         page->counters == counters_old) {
394                         page->freelist = freelist_new;
395                         set_page_slub_counters(page, counters_new);
396                         slab_unlock(page);
397                         return true;
398                 }
399                 slab_unlock(page);
400         }
401 
402         cpu_relax();
403         stat(s, CMPXCHG_DOUBLE_FAIL);
404 
405 #ifdef SLUB_DEBUG_CMPXCHG
406         pr_info("%s %s: cmpxchg double redo ", n, s->name);
407 #endif
408 
409         return false;
410 }
411 
412 static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
413                 void *freelist_old, unsigned long counters_old,
414                 void *freelist_new, unsigned long counters_new,
415                 const char *n)
416 {
417 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
418     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
419         if (s->flags & __CMPXCHG_DOUBLE) {
420                 if (cmpxchg_double(&page->freelist, &page->counters,
421                                    freelist_old, counters_old,
422                                    freelist_new, counters_new))
423                         return true;
424         } else
425 #endif
426         {
427                 unsigned long flags;
428 
429                 local_irq_save(flags);
430                 slab_lock(page);
431                 if (page->freelist == freelist_old &&
432                                         page->counters == counters_old) {
433                         page->freelist = freelist_new;
434                         set_page_slub_counters(page, counters_new);
435                         slab_unlock(page);
436                         local_irq_restore(flags);
437                         return true;
438                 }
439                 slab_unlock(page);
440                 local_irq_restore(flags);
441         }
442 
443         cpu_relax();
444         stat(s, CMPXCHG_DOUBLE_FAIL);
445 
446 #ifdef SLUB_DEBUG_CMPXCHG
447         pr_info("%s %s: cmpxchg double redo ", n, s->name);
448 #endif
449 
450         return false;
451 }
452 
453 #ifdef CONFIG_SLUB_DEBUG
454 /*
455  * Determine a map of object in use on a page.
456  *
457  * Node listlock must be held to guarantee that the page does
458  * not vanish from under us.
459  */
460 static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
461 {
462         void *p;
463         void *addr = page_address(page);
464 
465         for (p = page->freelist; p; p = get_freepointer(s, p))
466                 set_bit(slab_index(p, s, addr), map);
467 }
468 
469 static inline int size_from_object(struct kmem_cache *s)
470 {
471         if (s->flags & SLAB_RED_ZONE)
472                 return s->size - s->red_left_pad;
473 
474         return s->size;
475 }
476 
477 static inline void *restore_red_left(struct kmem_cache *s, void *p)
478 {
479         if (s->flags & SLAB_RED_ZONE)
480                 p -= s->red_left_pad;
481 
482         return p;
483 }
484 
485 /*
486  * Debug settings:
487  */
488 #if defined(CONFIG_SLUB_DEBUG_ON)
489 static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
490 #else
491 static slab_flags_t slub_debug;
492 #endif
493 
494 static char *slub_debug_slabs;
495 static int disable_higher_order_debug;
496 
497 /*
498  * slub is about to manipulate internal object metadata.  This memory lies
499  * outside the range of the allocated object, so accessing it would normally
500  * be reported by kasan as a bounds error.  metadata_access_enable() is used
501  * to tell kasan that these accesses are OK.
502  */
503 static inline void metadata_access_enable(void)
504 {
505         kasan_disable_current();
506 }
507 
508 static inline void metadata_access_disable(void)
509 {
510         kasan_enable_current();
511 }
512 
513 /*
514  * Object debugging
515  */
516 
517 /* Verify that a pointer has an address that is valid within a slab page */
518 static inline int check_valid_pointer(struct kmem_cache *s,
519                                 struct page *page, void *object)
520 {
521         void *base;
522 
523         if (!object)
524                 return 1;
525 
526         base = page_address(page);
527         object = restore_red_left(s, object);
528         if (object < base || object >= base + page->objects * s->size ||
529                 (object - base) % s->size) {
530                 return 0;
531         }
532 
533         return 1;
534 }
535 
536 static void print_section(char *level, char *text, u8 *addr,
537                           unsigned int length)
538 {
539         metadata_access_enable();
540         print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
541                         length, 1);
542         metadata_access_disable();
543 }
544 
545 static struct track *get_track(struct kmem_cache *s, void *object,
546         enum track_item alloc)
547 {
548         struct track *p;
549 
550         if (s->offset)
551                 p = object + s->offset + sizeof(void *);
552         else
553                 p = object + s->inuse;
554 
555         return p + alloc;
556 }
557 
558 static void set_track(struct kmem_cache *s, void *object,
559                         enum track_item alloc, unsigned long addr)
560 {
561         struct track *p = get_track(s, object, alloc);
562 
563         if (addr) {
564 #ifdef CONFIG_STACKTRACE
565                 struct stack_trace trace;
566                 int i;
567 
568                 trace.nr_entries = 0;
569                 trace.max_entries = TRACK_ADDRS_COUNT;
570                 trace.entries = p->addrs;
571                 trace.skip = 3;
572                 metadata_access_enable();
573                 save_stack_trace(&trace);
574                 metadata_access_disable();
575 
576                 /* See rant in lockdep.c */
577                 if (trace.nr_entries != 0 &&
578                     trace.entries[trace.nr_entries - 1] == ULONG_MAX)
579                         trace.nr_entries--;
580 
581                 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
582                         p->addrs[i] = 0;
583 #endif
584                 p->addr = addr;
585                 p->cpu = smp_processor_id();
586                 p->pid = current->pid;
587                 p->when = jiffies;
588         } else
589                 memset(p, 0, sizeof(struct track));
590 }
591 
592 static void init_tracking(struct kmem_cache *s, void *object)
593 {
594         if (!(s->flags & SLAB_STORE_USER))
595                 return;
596 
597         set_track(s, object, TRACK_FREE, 0UL);
598         set_track(s, object, TRACK_ALLOC, 0UL);
599 }
600 
601 static void print_track(const char *s, struct track *t)
602 {
603         if (!t->addr)
604                 return;
605 
606         pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
607                s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
608 #ifdef CONFIG_STACKTRACE
609         {
610                 int i;
611                 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
612                         if (t->addrs[i])
613                                 pr_err("\t%pS\n", (void *)t->addrs[i]);
614                         else
615                                 break;
616         }
617 #endif
618 }
619 
620 static void print_tracking(struct kmem_cache *s, void *object)
621 {
622         if (!(s->flags & SLAB_STORE_USER))
623                 return;
624 
625         print_track("Allocated", get_track(s, object, TRACK_ALLOC));
626         print_track("Freed", get_track(s, object, TRACK_FREE));
627 }
628 
629 static void print_page_info(struct page *page)
630 {
631         pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
632                page, page->objects, page->inuse, page->freelist, page->flags);
633 
634 }
635 
636 static void slab_bug(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("=============================================================================\n");
645         pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
646         pr_err("-----------------------------------------------------------------------------\n\n");
647 
648         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
649         va_end(args);
650 }
651 
652 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
653 {
654         struct va_format vaf;
655         va_list args;
656 
657         va_start(args, fmt);
658         vaf.fmt = fmt;
659         vaf.va = &args;
660         pr_err("FIX %s: %pV\n", s->name, &vaf);
661         va_end(args);
662 }
663 
664 static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
665 {
666         unsigned int off;       /* Offset of last byte */
667         u8 *addr = page_address(page);
668 
669         print_tracking(s, p);
670 
671         print_page_info(page);
672 
673         pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
674                p, p - addr, get_freepointer(s, p));
675 
676         if (s->flags & SLAB_RED_ZONE)
677                 print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
678                               s->red_left_pad);
679         else if (p > addr + 16)
680                 print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
681 
682         print_section(KERN_ERR, "Object ", p,
683                       min_t(unsigned long, s->object_size, PAGE_SIZE));
684         if (s->flags & SLAB_RED_ZONE)
685                 print_section(KERN_ERR, "Redzone ", p + s->object_size,
686                         s->inuse - s->object_size);
687 
688         if (s->offset)
689                 off = s->offset + sizeof(void *);
690         else
691                 off = s->inuse;
692 
693         if (s->flags & SLAB_STORE_USER)
694                 off += 2 * sizeof(struct track);
695 
696         off += kasan_metadata_size(s);
697 
698         if (off != size_from_object(s))
699                 /* Beginning of the filler is the free pointer */
700                 print_section(KERN_ERR, "Padding ", p + off,
701                               size_from_object(s) - off);
702 
703         dump_stack();
704 }
705 
706 void object_err(struct kmem_cache *s, struct page *page,
707                         u8 *object, char *reason)
708 {
709         slab_bug(s, "%s", reason);
710         print_trailer(s, page, object);
711 }
712 
713 static void slab_err(struct kmem_cache *s, struct page *page,
714                         const char *fmt, ...)
715 {
716         va_list args;
717         char buf[100];
718 
719         va_start(args, fmt);
720         vsnprintf(buf, sizeof(buf), fmt, args);
721         va_end(args);
722         slab_bug(s, "%s", buf);
723         print_page_info(page);
724         dump_stack();
725 }
726 
727 static void init_object(struct kmem_cache *s, void *object, u8 val)
728 {
729         u8 *p = object;
730 
731         if (s->flags & SLAB_RED_ZONE)
732                 memset(p - s->red_left_pad, val, s->red_left_pad);
733 
734         if (s->flags & __OBJECT_POISON) {
735                 memset(p, POISON_FREE, s->object_size - 1);
736                 p[s->object_size - 1] = POISON_END;
737         }
738 
739         if (s->flags & SLAB_RED_ZONE)
740                 memset(p + s->object_size, val, s->inuse - s->object_size);
741 }
742 
743 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
744                                                 void *from, void *to)
745 {
746         slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
747         memset(from, data, to - from);
748 }
749 
750 static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
751                         u8 *object, char *what,
752                         u8 *start, unsigned int value, unsigned int bytes)
753 {
754         u8 *fault;
755         u8 *end;
756 
757         metadata_access_enable();
758         fault = memchr_inv(start, value, bytes);
759         metadata_access_disable();
760         if (!fault)
761                 return 1;
762 
763         end = start + bytes;
764         while (end > fault && end[-1] == value)
765                 end--;
766 
767         slab_bug(s, "%s overwritten", what);
768         pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
769                                         fault, end - 1, fault[0], value);
770         print_trailer(s, page, object);
771 
772         restore_bytes(s, what, value, fault, end);
773         return 0;
774 }
775 
776 /*
777  * Object layout:
778  *
779  * object address
780  *      Bytes of the object to be managed.
781  *      If the freepointer may overlay the object then the free
782  *      pointer is the first word of the object.
783  *
784  *      Poisoning uses 0x6b (POISON_FREE) and the last byte is
785  *      0xa5 (POISON_END)
786  *
787  * object + s->object_size
788  *      Padding to reach word boundary. This is also used for Redzoning.
789  *      Padding is extended by another word if Redzoning is enabled and
790  *      object_size == inuse.
791  *
792  *      We fill with 0xbb (RED_INACTIVE) for inactive objects and with
793  *      0xcc (RED_ACTIVE) for objects in use.
794  *
795  * object + s->inuse
796  *      Meta data starts here.
797  *
798  *      A. Free pointer (if we cannot overwrite object on free)
799  *      B. Tracking data for SLAB_STORE_USER
800  *      C. Padding to reach required alignment boundary or at mininum
801  *              one word if debugging is on to be able to detect writes
802  *              before the word boundary.
803  *
804  *      Padding is done using 0x5a (POISON_INUSE)
805  *
806  * object + s->size
807  *      Nothing is used beyond s->size.
808  *
809  * If slabcaches are merged then the object_size and inuse boundaries are mostly
810  * ignored. And therefore no slab options that rely on these boundaries
811  * may be used with merged slabcaches.
812  */
813 
814 static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
815 {
816         unsigned long off = s->inuse;   /* The end of info */
817 
818         if (s->offset)
819                 /* Freepointer is placed after the object. */
820                 off += sizeof(void *);
821 
822         if (s->flags & SLAB_STORE_USER)
823                 /* We also have user information there */
824                 off += 2 * sizeof(struct track);
825 
826         off += kasan_metadata_size(s);
827 
828         if (size_from_object(s) == off)
829                 return 1;
830 
831         return check_bytes_and_report(s, page, p, "Object padding",
832                         p + off, POISON_INUSE, size_from_object(s) - off);
833 }
834 
835 /* Check the pad bytes at the end of a slab page */
836 static int slab_pad_check(struct kmem_cache *s, struct page *page)
837 {
838         u8 *start;
839         u8 *fault;
840         u8 *end;
841         u8 *pad;
842         int length;
843         int remainder;
844 
845         if (!(s->flags & SLAB_POISON))
846                 return 1;
847 
848         start = page_address(page);
849         length = (PAGE_SIZE << compound_order(page)) - s->reserved;
850         end = start + length;
851         remainder = length % s->size;
852         if (!remainder)
853                 return 1;
854 
855         pad = end - remainder;
856         metadata_access_enable();
857         fault = memchr_inv(pad, POISON_INUSE, remainder);
858         metadata_access_disable();
859         if (!fault)
860                 return 1;
861         while (end > fault && end[-1] == POISON_INUSE)
862                 end--;
863 
864         slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
865         print_section(KERN_ERR, "Padding ", pad, remainder);
866 
867         restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
868         return 0;
869 }
870 
871 static int check_object(struct kmem_cache *s, struct page *page,
872                                         void *object, u8 val)
873 {
874         u8 *p = object;
875         u8 *endobject = object + s->object_size;
876 
877         if (s->flags & SLAB_RED_ZONE) {
878                 if (!check_bytes_and_report(s, page, object, "Redzone",
879                         object - s->red_left_pad, val, s->red_left_pad))
880                         return 0;
881 
882                 if (!check_bytes_and_report(s, page, object, "Redzone",
883                         endobject, val, s->inuse - s->object_size))
884                         return 0;
885         } else {
886                 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
887                         check_bytes_and_report(s, page, p, "Alignment padding",
888                                 endobject, POISON_INUSE,
889                                 s->inuse - s->object_size);
890                 }
891         }
892 
893         if (s->flags & SLAB_POISON) {
894                 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
895                         (!check_bytes_and_report(s, page, p, "Poison", p,
896                                         POISON_FREE, s->object_size - 1) ||
897                          !check_bytes_and_report(s, page, p, "Poison",
898                                 p + s->object_size - 1, POISON_END, 1)))
899                         return 0;
900                 /*
901                  * check_pad_bytes cleans up on its own.
902                  */
903                 check_pad_bytes(s, page, p);
904         }
905 
906         if (!s->offset && val == SLUB_RED_ACTIVE)
907                 /*
908                  * Object and freepointer overlap. Cannot check
909                  * freepointer while object is allocated.
910                  */
911                 return 1;
912 
913         /* Check free pointer validity */
914         if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
915                 object_err(s, page, p, "Freepointer corrupt");
916                 /*
917                  * No choice but to zap it and thus lose the remainder
918                  * of the free objects in this slab. May cause
919                  * another error because the object count is now wrong.
920                  */
921                 set_freepointer(s, p, NULL);
922                 return 0;
923         }
924         return 1;
925 }
926 
927 static int check_slab(struct kmem_cache *s, struct page *page)
928 {
929         int maxobj;
930 
931         VM_BUG_ON(!irqs_disabled());
932 
933         if (!PageSlab(page)) {
934                 slab_err(s, page, "Not a valid slab page");
935                 return 0;
936         }
937 
938         maxobj = order_objects(compound_order(page), s->size, s->reserved);
939         if (page->objects > maxobj) {
940                 slab_err(s, page, "objects %u > max %u",
941                         page->objects, maxobj);
942                 return 0;
943         }
944         if (page->inuse > page->objects) {
945                 slab_err(s, page, "inuse %u > max %u",
946                         page->inuse, page->objects);
947                 return 0;
948         }
949         /* Slab_pad_check fixes things up after itself */
950         slab_pad_check(s, page);
951         return 1;
952 }
953 
954 /*
955  * Determine if a certain object on a page is on the freelist. Must hold the
956  * slab lock to guarantee that the chains are in a consistent state.
957  */
958 static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
959 {
960         int nr = 0;
961         void *fp;
962         void *object = NULL;
963         int max_objects;
964 
965         fp = page->freelist;
966         while (fp && nr <= page->objects) {
967                 if (fp == search)
968                         return 1;
969                 if (!check_valid_pointer(s, page, fp)) {
970                         if (object) {
971                                 object_err(s, page, object,
972                                         "Freechain corrupt");
973                                 set_freepointer(s, object, NULL);
974                         } else {
975                                 slab_err(s, page, "Freepointer corrupt");
976                                 page->freelist = NULL;
977                                 page->inuse = page->objects;
978                                 slab_fix(s, "Freelist cleared");
979                                 return 0;
980                         }
981                         break;
982                 }
983                 object = fp;
984                 fp = get_freepointer(s, object);
985                 nr++;
986         }
987 
988         max_objects = order_objects(compound_order(page), s->size, s->reserved);
989         if (max_objects > MAX_OBJS_PER_PAGE)
990                 max_objects = MAX_OBJS_PER_PAGE;
991 
992         if (page->objects != max_objects) {
993                 slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
994                          page->objects, max_objects);
995                 page->objects = max_objects;
996                 slab_fix(s, "Number of objects adjusted.");
997         }
998         if (page->inuse != page->objects - nr) {
999                 slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
1000                          page->inuse, page->objects - nr);
1001                 page->inuse = page->objects - nr;
1002                 slab_fix(s, "Object count adjusted.");
1003         }
1004         return search == NULL;
1005 }
1006 
1007 static void trace(struct kmem_cache *s, struct page *page, void *object,
1008                                                                 int alloc)
1009 {
1010         if (s->flags & SLAB_TRACE) {
1011                 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
1012                         s->name,
1013                         alloc ? "alloc" : "free",
1014                         object, page->inuse,
1015                         page->freelist);
1016 
1017                 if (!alloc)
1018                         print_section(KERN_INFO, "Object ", (void *)object,
1019                                         s->object_size);
1020 
1021                 dump_stack();
1022         }
1023 }
1024 
1025 /*
1026  * Tracking of fully allocated slabs for debugging purposes.
1027  */
1028 static void add_full(struct kmem_cache *s,
1029         struct kmem_cache_node *n, struct page *page)
1030 {
1031         if (!(s->flags & SLAB_STORE_USER))
1032                 return;
1033 
1034         lockdep_assert_held(&n->list_lock);
1035         list_add(&page->lru, &n->full);
1036 }
1037 
1038 static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
1039 {
1040         if (!(s->flags & SLAB_STORE_USER))
1041                 return;
1042 
1043         lockdep_assert_held(&n->list_lock);
1044         list_del(&page->lru);
1045 }
1046 
1047 /* Tracking of the number of slabs for debugging purposes */
1048 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1049 {
1050         struct kmem_cache_node *n = get_node(s, node);
1051 
1052         return atomic_long_read(&n->nr_slabs);
1053 }
1054 
1055 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1056 {
1057         return atomic_long_read(&n->nr_slabs);
1058 }
1059 
1060 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
1061 {
1062         struct kmem_cache_node *n = get_node(s, node);
1063 
1064         /*
1065          * May be called early in order to allocate a slab for the
1066          * kmem_cache_node structure. Solve the chicken-egg
1067          * dilemma by deferring the increment of the count during
1068          * bootstrap (see early_kmem_cache_node_alloc).
1069          */
1070         if (likely(n)) {
1071                 atomic_long_inc(&n->nr_slabs);
1072                 atomic_long_add(objects, &n->total_objects);
1073         }
1074 }
1075 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
1076 {
1077         struct kmem_cache_node *n = get_node(s, node);
1078 
1079         atomic_long_dec(&n->nr_slabs);
1080         atomic_long_sub(objects, &n->total_objects);
1081 }
1082 
1083 /* Object debug checks for alloc/free paths */
1084 static void setup_object_debug(struct kmem_cache *s, struct page *page,
1085                                                                 void *object)
1086 {
1087         if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1088                 return;
1089 
1090         init_object(s, object, SLUB_RED_INACTIVE);
1091         init_tracking(s, object);
1092 }
1093 
1094 static inline int alloc_consistency_checks(struct kmem_cache *s,
1095                                         struct page *page,
1096                                         void *object, unsigned long addr)
1097 {
1098         if (!check_slab(s, page))
1099                 return 0;
1100 
1101         if (!check_valid_pointer(s, page, object)) {
1102                 object_err(s, page, object, "Freelist Pointer check fails");
1103                 return 0;
1104         }
1105 
1106         if (!check_object(s, page, object, SLUB_RED_INACTIVE))
1107                 return 0;
1108 
1109         return 1;
1110 }
1111 
1112 static noinline int alloc_debug_processing(struct kmem_cache *s,
1113                                         struct page *page,
1114                                         void *object, unsigned long addr)
1115 {
1116         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1117                 if (!alloc_consistency_checks(s, page, object, addr))
1118                         goto bad;
1119         }
1120 
1121         /* Success perform special debug activities for allocs */
1122         if (s->flags & SLAB_STORE_USER)
1123                 set_track(s, object, TRACK_ALLOC, addr);
1124         trace(s, page, object, 1);
1125         init_object(s, object, SLUB_RED_ACTIVE);
1126         return 1;
1127 
1128 bad:
1129         if (PageSlab(page)) {
1130                 /*
1131                  * If this is a slab page then lets do the best we can
1132                  * to avoid issues in the future. Marking all objects
1133                  * as used avoids touching the remaining objects.
1134                  */
1135                 slab_fix(s, "Marking all objects used");
1136                 page->inuse = page->objects;
1137                 page->freelist = NULL;
1138         }
1139         return 0;
1140 }
1141 
1142 static inline int free_consistency_checks(struct kmem_cache *s,
1143                 struct page *page, void *object, unsigned long addr)
1144 {
1145         if (!check_valid_pointer(s, page, object)) {
1146                 slab_err(s, page, "Invalid object pointer 0x%p", object);
1147                 return 0;
1148         }
1149 
1150         if (on_freelist(s, page, object)) {
1151                 object_err(s, page, object, "Object already free");
1152                 return 0;
1153         }
1154 
1155         if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1156                 return 0;
1157 
1158         if (unlikely(s != page->slab_cache)) {
1159                 if (!PageSlab(page)) {
1160                         slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1161                                  object);
1162                 } else if (!page->slab_cache) {
1163                         pr_err("SLUB <none>: no slab for object 0x%p.\n",
1164                                object);
1165                         dump_stack();
1166                 } else
1167                         object_err(s, page, object,
1168                                         "page slab pointer corrupt.");
1169                 return 0;
1170         }
1171         return 1;
1172 }
1173 
1174 /* Supports checking bulk free of a constructed freelist */
1175 static noinline int free_debug_processing(
1176         struct kmem_cache *s, struct page *page,
1177         void *head, void *tail, int bulk_cnt,
1178         unsigned long addr)
1179 {
1180         struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1181         void *object = head;
1182         int cnt = 0;
1183         unsigned long uninitialized_var(flags);
1184         int ret = 0;
1185 
1186         spin_lock_irqsave(&n->list_lock, flags);
1187         slab_lock(page);
1188 
1189         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1190                 if (!check_slab(s, page))
1191                         goto out;
1192         }
1193 
1194 next_object:
1195         cnt++;
1196 
1197         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1198                 if (!free_consistency_checks(s, page, object, addr))
1199                         goto out;
1200         }
1201 
1202         if (s->flags & SLAB_STORE_USER)
1203                 set_track(s, object, TRACK_FREE, addr);
1204         trace(s, page, object, 0);
1205         /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
1206         init_object(s, object, SLUB_RED_INACTIVE);
1207 
1208         /* Reached end of constructed freelist yet? */
1209         if (object != tail) {
1210                 object = get_freepointer(s, object);
1211                 goto next_object;
1212         }
1213         ret = 1;
1214 
1215 out:
1216         if (cnt != bulk_cnt)
1217                 slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1218                          bulk_cnt, cnt);
1219 
1220         slab_unlock(page);
1221         spin_unlock_irqrestore(&n->list_lock, flags);
1222         if (!ret)
1223                 slab_fix(s, "Object at 0x%p not freed", object);
1224         return ret;
1225 }
1226 
1227 static int __init setup_slub_debug(char *str)
1228 {
1229         slub_debug = DEBUG_DEFAULT_FLAGS;
1230         if (*str++ != '=' || !*str)
1231                 /*
1232                  * No options specified. Switch on full debugging.
1233                  */
1234                 goto out;
1235 
1236         if (*str == ',')
1237                 /*
1238                  * No options but restriction on slabs. This means full
1239                  * debugging for slabs matching a pattern.
1240                  */
1241                 goto check_slabs;
1242 
1243         slub_debug = 0;
1244         if (*str == '-')
1245                 /*
1246                  * Switch off all debugging measures.
1247                  */
1248                 goto out;
1249 
1250         /*
1251          * Determine which debug features should be switched on
1252          */
1253         for (; *str && *str != ','; str++) {
1254                 switch (tolower(*str)) {
1255                 case 'f':
1256                         slub_debug |= SLAB_CONSISTENCY_CHECKS;
1257                         break;
1258                 case 'z':
1259                         slub_debug |= SLAB_RED_ZONE;
1260                         break;
1261                 case 'p':
1262                         slub_debug |= SLAB_POISON;
1263                         break;
1264                 case 'u':
1265                         slub_debug |= SLAB_STORE_USER;
1266                         break;
1267                 case 't':
1268                         slub_debug |= SLAB_TRACE;
1269                         break;
1270                 case 'a':
1271                         slub_debug |= SLAB_FAILSLAB;
1272                         break;
1273                 case 'o':
1274                         /*
1275                          * Avoid enabling debugging on caches if its minimum
1276                          * order would increase as a result.
1277                          */
1278                         disable_higher_order_debug = 1;
1279                         break;
1280                 default:
1281                         pr_err("slub_debug option '%c' unknown. skipped\n",
1282                                *str);
1283                 }
1284         }
1285 
1286 check_slabs:
1287         if (*str == ',')
1288                 slub_debug_slabs = str + 1;
1289 out:
1290         return 1;
1291 }
1292 
1293 __setup("slub_debug", setup_slub_debug);
1294 
1295 slab_flags_t kmem_cache_flags(unsigned long object_size,
1296         slab_flags_t flags, const char *name,
1297         void (*ctor)(void *))
1298 {
1299         /*
1300          * Enable debugging if selected on the kernel commandline.
1301          */
1302         if (slub_debug && (!slub_debug_slabs || (name &&
1303                 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))))
1304                 flags |= slub_debug;
1305 
1306         return flags;
1307 }
1308 #else /* !CONFIG_SLUB_DEBUG */
1309 static inline void setup_object_debug(struct kmem_cache *s,
1310                         struct page *page, void *object) {}
1311 
1312 static inline int alloc_debug_processing(struct kmem_cache *s,
1313         struct page *page, void *object, unsigned long addr) { return 0; }
1314 
1315 static inline int free_debug_processing(
1316         struct kmem_cache *s, struct page *page,
1317         void *head, void *tail, int bulk_cnt,
1318         unsigned long addr) { return 0; }
1319 
1320 static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1321                         { return 1; }
1322 static inline int check_object(struct kmem_cache *s, struct page *page,
1323                         void *object, u8 val) { return 1; }
1324 static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1325                                         struct page *page) {}
1326 static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1327                                         struct page *page) {}
1328 slab_flags_t kmem_cache_flags(unsigned long object_size,
1329         slab_flags_t flags, const char *name,
1330         void (*ctor)(void *))
1331 {
1332         return flags;
1333 }
1334 #define slub_debug 0
1335 
1336 #define disable_higher_order_debug 0
1337 
1338 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1339                                                         { return 0; }
1340 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1341                                                         { return 0; }
1342 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1343                                                         int objects) {}
1344 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1345                                                         int objects) {}
1346 
1347 #endif /* CONFIG_SLUB_DEBUG */
1348 
1349 /*
1350  * Hooks for other subsystems that check memory allocations. In a typical
1351  * production configuration these hooks all should produce no code at all.
1352  */
1353 static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1354 {
1355         kmemleak_alloc(ptr, size, 1, flags);
1356         kasan_kmalloc_large(ptr, size, flags);
1357 }
1358 
1359 static __always_inline void kfree_hook(void *x)
1360 {
1361         kmemleak_free(x);
1362         kasan_kfree_large(x, _RET_IP_);
1363 }
1364 
1365 static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x)
1366 {
1367         kmemleak_free_recursive(x, s->flags);
1368 
1369         /*
1370          * Trouble is that we may no longer disable interrupts in the fast path
1371          * So in order to make the debug calls that expect irqs to be
1372          * disabled we need to disable interrupts temporarily.
1373          */
1374 #ifdef CONFIG_LOCKDEP
1375         {
1376                 unsigned long flags;
1377 
1378                 local_irq_save(flags);
1379                 debug_check_no_locks_freed(x, s->object_size);
1380                 local_irq_restore(flags);
1381         }
1382 #endif
1383         if (!(s->flags & SLAB_DEBUG_OBJECTS))
1384                 debug_check_no_obj_freed(x, s->object_size);
1385 
1386         /* KASAN might put x into memory quarantine, delaying its reuse */
1387         return kasan_slab_free(s, x, _RET_IP_);
1388 }
1389 
1390 static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1391                                            void **head, void **tail)
1392 {
1393 /*
1394  * Compiler cannot detect this function can be removed if slab_free_hook()
1395  * evaluates to nothing.  Thus, catch all relevant config debug options here.
1396  */
1397 #if defined(CONFIG_LOCKDEP)     ||              \
1398         defined(CONFIG_DEBUG_KMEMLEAK) ||       \
1399         defined(CONFIG_DEBUG_OBJECTS_FREE) ||   \
1400         defined(CONFIG_KASAN)
1401 
1402         void *object;
1403         void *next = *head;
1404         void *old_tail = *tail ? *tail : *head;
1405 
1406         /* Head and tail of the reconstructed freelist */
1407         *head = NULL;
1408         *tail = NULL;
1409 
1410         do {
1411                 object = next;
1412                 next = get_freepointer(s, object);
1413                 /* If object's reuse doesn't have to be delayed */
1414                 if (!slab_free_hook(s, object)) {
1415                         /* Move object to the new freelist */
1416                         set_freepointer(s, object, *head);
1417                         *head = object;
1418                         if (!*tail)
1419                                 *tail = object;
1420                 }
1421         } while (object != old_tail);
1422 
1423         if (*head == *tail)
1424                 *tail = NULL;
1425 
1426         return *head != NULL;
1427 #else
1428         return true;
1429 #endif
1430 }
1431 
1432 static void setup_object(struct kmem_cache *s, struct page *page,
1433                                 void *object)
1434 {
1435         setup_object_debug(s, page, object);
1436         kasan_init_slab_obj(s, object);
1437         if (unlikely(s->ctor)) {
1438                 kasan_unpoison_object_data(s, object);
1439                 s->ctor(object);
1440                 kasan_poison_object_data(s, object);
1441         }
1442 }
1443 
1444 /*
1445  * Slab allocation and freeing
1446  */
1447 static inline struct page *alloc_slab_page(struct kmem_cache *s,
1448                 gfp_t flags, int node, struct kmem_cache_order_objects oo)
1449 {
1450         struct page *page;
1451         int order = oo_order(oo);
1452 
1453         if (node == NUMA_NO_NODE)
1454                 page = alloc_pages(flags, order);
1455         else
1456                 page = __alloc_pages_node(node, flags, order);
1457 
1458         if (page && memcg_charge_slab(page, flags, order, s)) {
1459                 __free_pages(page, order);
1460                 page = NULL;
1461         }
1462 
1463         return page;
1464 }
1465 
1466 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1467 /* Pre-initialize the random sequence cache */
1468 static int init_cache_random_seq(struct kmem_cache *s)
1469 {
1470         int err;
1471         unsigned long i, count = oo_objects(s->oo);
1472 
1473         /* Bailout if already initialised */
1474         if (s->random_seq)
1475                 return 0;
1476 
1477         err = cache_random_seq_create(s, count, GFP_KERNEL);
1478         if (err) {
1479                 pr_err("SLUB: Unable to initialize free list for %s\n",
1480                         s->name);
1481                 return err;
1482         }
1483 
1484         /* Transform to an offset on the set of pages */
1485         if (s->random_seq) {
1486                 for (i = 0; i < count; i++)
1487                         s->random_seq[i] *= s->size;
1488         }
1489         return 0;
1490 }
1491 
1492 /* Initialize each random sequence freelist per cache */
1493 static void __init init_freelist_randomization(void)
1494 {
1495         struct kmem_cache *s;
1496 
1497         mutex_lock(&slab_mutex);
1498 
1499         list_for_each_entry(s, &slab_caches, list)
1500                 init_cache_random_seq(s);
1501 
1502         mutex_unlock(&slab_mutex);
1503 }
1504 
1505 /* Get the next entry on the pre-computed freelist randomized */
1506 static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1507                                 unsigned long *pos, void *start,
1508                                 unsigned long page_limit,
1509                                 unsigned long freelist_count)
1510 {
1511         unsigned int idx;
1512 
1513         /*
1514          * If the target page allocation failed, the number of objects on the
1515          * page might be smaller than the usual size defined by the cache.
1516          */
1517         do {
1518                 idx = s->random_seq[*pos];
1519                 *pos += 1;
1520                 if (*pos >= freelist_count)
1521                         *pos = 0;
1522         } while (unlikely(idx >= page_limit));
1523 
1524         return (char *)start + idx;
1525 }
1526 
1527 /* Shuffle the single linked freelist based on a random pre-computed sequence */
1528 static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1529 {
1530         void *start;
1531         void *cur;
1532         void *next;
1533         unsigned long idx, pos, page_limit, freelist_count;
1534 
1535         if (page->objects < 2 || !s->random_seq)
1536                 return false;
1537 
1538         freelist_count = oo_objects(s->oo);
1539         pos = get_random_int() % freelist_count;
1540 
1541         page_limit = page->objects * s->size;
1542         start = fixup_red_left(s, page_address(page));
1543 
1544         /* First entry is used as the base of the freelist */
1545         cur = next_freelist_entry(s, page, &pos, start, page_limit,
1546                                 freelist_count);
1547         page->freelist = cur;
1548 
1549         for (idx = 1; idx < page->objects; idx++) {
1550                 setup_object(s, page, cur);
1551                 next = next_freelist_entry(s, page, &pos, start, page_limit,
1552                         freelist_count);
1553                 set_freepointer(s, cur, next);
1554                 cur = next;
1555         }
1556         setup_object(s, page, cur);
1557         set_freepointer(s, cur, NULL);
1558 
1559         return true;
1560 }
1561 #else
1562 static inline int init_cache_random_seq(struct kmem_cache *s)
1563 {
1564         return 0;
1565 }
1566 static inline void init_freelist_randomization(void) { }
1567 static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1568 {
1569         return false;
1570 }
1571 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1572 
1573 static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1574 {
1575         struct page *page;
1576         struct kmem_cache_order_objects oo = s->oo;
1577         gfp_t alloc_gfp;
1578         void *start, *p;
1579         int idx, order;
1580         bool shuffle;
1581 
1582         flags &= gfp_allowed_mask;
1583 
1584         if (gfpflags_allow_blocking(flags))
1585                 local_irq_enable();
1586 
1587         flags |= s->allocflags;
1588 
1589         /*
1590          * Let the initial higher-order allocation fail under memory pressure
1591          * so we fall-back to the minimum order allocation.
1592          */
1593         alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1594         if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
1595                 alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
1596 
1597         page = alloc_slab_page(s, alloc_gfp, node, oo);
1598         if (unlikely(!page)) {
1599                 oo = s->min;
1600                 alloc_gfp = flags;
1601                 /*
1602                  * Allocation may have failed due to fragmentation.
1603                  * Try a lower order alloc if possible
1604                  */
1605                 page = alloc_slab_page(s, alloc_gfp, node, oo);
1606                 if (unlikely(!page))
1607                         goto out;
1608                 stat(s, ORDER_FALLBACK);
1609         }
1610 
1611         page->objects = oo_objects(oo);
1612 
1613         order = compound_order(page);
1614         page->slab_cache = s;
1615         __SetPageSlab(page);
1616         if (page_is_pfmemalloc(page))
1617                 SetPageSlabPfmemalloc(page);
1618 
1619         start = page_address(page);
1620 
1621         if (unlikely(s->flags & SLAB_POISON))
1622                 memset(start, POISON_INUSE, PAGE_SIZE << order);
1623 
1624         kasan_poison_slab(page);
1625 
1626         shuffle = shuffle_freelist(s, page);
1627 
1628         if (!shuffle) {
1629                 for_each_object_idx(p, idx, s, start, page->objects) {
1630                         setup_object(s, page, p);
1631                         if (likely(idx < page->objects))
1632                                 set_freepointer(s, p, p + s->size);
1633                         else
1634                                 set_freepointer(s, p, NULL);
1635                 }
1636                 page->freelist = fixup_red_left(s, start);
1637         }
1638 
1639         page->inuse = page->objects;
1640         page->frozen = 1;
1641 
1642 out:
1643         if (gfpflags_allow_blocking(flags))
1644                 local_irq_disable();
1645         if (!page)
1646                 return NULL;
1647 
1648         mod_lruvec_page_state(page,
1649                 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1650                 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1651                 1 << oo_order(oo));
1652 
1653         inc_slabs_node(s, page_to_nid(page), page->objects);
1654 
1655         return page;
1656 }
1657 
1658 static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1659 {
1660         if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
1661                 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1662                 flags &= ~GFP_SLAB_BUG_MASK;
1663                 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1664                                 invalid_mask, &invalid_mask, flags, &flags);
1665                 dump_stack();
1666         }
1667 
1668         return allocate_slab(s,
1669                 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1670 }
1671 
1672 static void __free_slab(struct kmem_cache *s, struct page *page)
1673 {
1674         int order = compound_order(page);
1675         int pages = 1 << order;
1676 
1677         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1678                 void *p;
1679 
1680                 slab_pad_check(s, page);
1681                 for_each_object(p, s, page_address(page),
1682                                                 page->objects)
1683                         check_object(s, page, p, SLUB_RED_INACTIVE);
1684         }
1685 
1686         mod_lruvec_page_state(page,
1687                 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1688                 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1689                 -pages);
1690 
1691         __ClearPageSlabPfmemalloc(page);
1692         __ClearPageSlab(page);
1693 
1694         page_mapcount_reset(page);
1695         if (current->reclaim_state)
1696                 current->reclaim_state->reclaimed_slab += pages;
1697         memcg_uncharge_slab(page, order, s);
1698         __free_pages(page, order);
1699 }
1700 
1701 #define need_reserve_slab_rcu                                           \
1702         (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1703 
1704 static void rcu_free_slab(struct rcu_head *h)
1705 {
1706         struct page *page;
1707 
1708         if (need_reserve_slab_rcu)
1709                 page = virt_to_head_page(h);
1710         else
1711                 page = container_of((struct list_head *)h, struct page, lru);
1712 
1713         __free_slab(page->slab_cache, page);
1714 }
1715 
1716 static void free_slab(struct kmem_cache *s, struct page *page)
1717 {
1718         if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) {
1719                 struct rcu_head *head;
1720 
1721                 if (need_reserve_slab_rcu) {
1722                         int order = compound_order(page);
1723                         int offset = (PAGE_SIZE << order) - s->reserved;
1724 
1725                         VM_BUG_ON(s->reserved != sizeof(*head));
1726                         head = page_address(page) + offset;
1727                 } else {
1728                         head = &page->rcu_head;
1729                 }
1730 
1731                 call_rcu(head, rcu_free_slab);
1732         } else
1733                 __free_slab(s, page);
1734 }
1735 
1736 static void discard_slab(struct kmem_cache *s, struct page *page)
1737 {
1738         dec_slabs_node(s, page_to_nid(page), page->objects);
1739         free_slab(s, page);
1740 }
1741 
1742 /*
1743  * Management of partially allocated slabs.
1744  */
1745 static inline void
1746 __add_partial(struct kmem_cache_node *n, struct page *page, int tail)
1747 {
1748         n->nr_partial++;
1749         if (tail == DEACTIVATE_TO_TAIL)
1750                 list_add_tail(&page->lru, &n->partial);
1751         else
1752                 list_add(&page->lru, &n->partial);
1753 }
1754 
1755 static inline void add_partial(struct kmem_cache_node *n,
1756                                 struct page *page, int tail)
1757 {
1758         lockdep_assert_held(&n->list_lock);
1759         __add_partial(n, page, tail);
1760 }
1761 
1762 static inline void remove_partial(struct kmem_cache_node *n,
1763                                         struct page *page)
1764 {
1765         lockdep_assert_held(&n->list_lock);
1766         list_del(&page->lru);
1767         n->nr_partial--;
1768 }
1769 
1770 /*
1771  * Remove slab from the partial list, freeze it and
1772  * return the pointer to the freelist.
1773  *
1774  * Returns a list of objects or NULL if it fails.
1775  */
1776 static inline void *acquire_slab(struct kmem_cache *s,
1777                 struct kmem_cache_node *n, struct page *page,
1778                 int mode, int *objects)
1779 {
1780         void *freelist;
1781         unsigned long counters;
1782         struct page new;
1783 
1784         lockdep_assert_held(&n->list_lock);
1785 
1786         /*
1787          * Zap the freelist and set the frozen bit.
1788          * The old freelist is the list of objects for the
1789          * per cpu allocation list.
1790          */
1791         freelist = page->freelist;
1792         counters = page->counters;
1793         new.counters = counters;
1794         *objects = new.objects - new.inuse;
1795         if (mode) {
1796                 new.inuse = page->objects;
1797                 new.freelist = NULL;
1798         } else {
1799                 new.freelist = freelist;
1800         }
1801 
1802         VM_BUG_ON(new.frozen);
1803         new.frozen = 1;
1804 
1805         if (!__cmpxchg_double_slab(s, page,
1806                         freelist, counters,
1807                         new.freelist, new.counters,
1808                         "acquire_slab"))
1809                 return NULL;
1810 
1811         remove_partial(n, page);
1812         WARN_ON(!freelist);
1813         return freelist;
1814 }
1815 
1816 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
1817 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
1818 
1819 /*
1820  * Try to allocate a partial slab from a specific node.
1821  */
1822 static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1823                                 struct kmem_cache_cpu *c, gfp_t flags)
1824 {
1825         struct page *page, *page2;
1826         void *object = NULL;
1827         int available = 0;
1828         int objects;
1829 
1830         /*
1831          * Racy check. If we mistakenly see no partial slabs then we
1832          * just allocate an empty slab. If we mistakenly try to get a
1833          * partial slab and there is none available then get_partials()
1834          * will return NULL.
1835          */
1836         if (!n || !n->nr_partial)
1837                 return NULL;
1838 
1839         spin_lock(&n->list_lock);
1840         list_for_each_entry_safe(page, page2, &n->partial, lru) {
1841                 void *t;
1842 
1843                 if (!pfmemalloc_match(page, flags))
1844                         continue;
1845 
1846                 t = acquire_slab(s, n, page, object == NULL, &objects);
1847                 if (!t)
1848                         break;
1849 
1850                 available += objects;
1851                 if (!object) {
1852                         c->page = page;
1853                         stat(s, ALLOC_FROM_PARTIAL);
1854                         object = t;
1855                 } else {
1856                         put_cpu_partial(s, page, 0);
1857                         stat(s, CPU_PARTIAL_NODE);
1858                 }
1859                 if (!kmem_cache_has_cpu_partial(s)
1860                         || available > slub_cpu_partial(s) / 2)
1861                         break;
1862 
1863         }
1864         spin_unlock(&n->list_lock);
1865         return object;
1866 }
1867 
1868 /*
1869  * Get a page from somewhere. Search in increasing NUMA distances.
1870  */
1871 static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
1872                 struct kmem_cache_cpu *c)
1873 {
1874 #ifdef CONFIG_NUMA
1875         struct zonelist *zonelist;
1876         struct zoneref *z;
1877         struct zone *zone;
1878         enum zone_type high_zoneidx = gfp_zone(flags);
1879         void *object;
1880         unsigned int cpuset_mems_cookie;
1881 
1882         /*
1883          * The defrag ratio allows a configuration of the tradeoffs between
1884          * inter node defragmentation and node local allocations. A lower
1885          * defrag_ratio increases the tendency to do local allocations
1886          * instead of attempting to obtain partial slabs from other nodes.
1887          *
1888          * If the defrag_ratio is set to 0 then kmalloc() always
1889          * returns node local objects. If the ratio is higher then kmalloc()
1890          * may return off node objects because partial slabs are obtained
1891          * from other nodes and filled up.
1892          *
1893          * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
1894          * (which makes defrag_ratio = 1000) then every (well almost)
1895          * allocation will first attempt to defrag slab caches on other nodes.
1896          * This means scanning over all nodes to look for partial slabs which
1897          * may be expensive if we do it every time we are trying to find a slab
1898          * with available objects.
1899          */
1900         if (!s->remote_node_defrag_ratio ||
1901                         get_cycles() % 1024 > s->remote_node_defrag_ratio)
1902                 return NULL;
1903 
1904         do {
1905                 cpuset_mems_cookie = read_mems_allowed_begin();
1906                 zonelist = node_zonelist(mempolicy_slab_node(), flags);
1907                 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1908                         struct kmem_cache_node *n;
1909 
1910                         n = get_node(s, zone_to_nid(zone));
1911 
1912                         if (n && cpuset_zone_allowed(zone, flags) &&
1913                                         n->nr_partial > s->min_partial) {
1914                                 object = get_partial_node(s, n, c, flags);
1915                                 if (object) {
1916                                         /*
1917                                          * Don't check read_mems_allowed_retry()
1918                                          * here - if mems_allowed was updated in
1919                                          * parallel, that was a harmless race
1920                                          * between allocation and the cpuset
1921                                          * update
1922                                          */
1923                                         return object;
1924                                 }
1925                         }
1926                 }
1927         } while (read_mems_allowed_retry(cpuset_mems_cookie));
1928 #endif
1929         return NULL;
1930 }
1931 
1932 /*
1933  * Get a partial page, lock it and return it.
1934  */
1935 static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
1936                 struct kmem_cache_cpu *c)
1937 {
1938         void *object;
1939         int searchnode = node;
1940 
1941         if (node == NUMA_NO_NODE)
1942                 searchnode = numa_mem_id();
1943         else if (!node_present_pages(node))
1944                 searchnode = node_to_mem_node(node);
1945 
1946         object = get_partial_node(s, get_node(s, searchnode), c, flags);
1947         if (object || node != NUMA_NO_NODE)
1948                 return object;
1949 
1950         return get_any_partial(s, flags, c);
1951 }
1952 
1953 #ifdef CONFIG_PREEMPT
1954 /*
1955  * Calculate the next globally unique transaction for disambiguiation
1956  * during cmpxchg. The transactions start with the cpu number and are then
1957  * incremented by CONFIG_NR_CPUS.
1958  */
1959 #define TID_STEP  roundup_pow_of_two(CONFIG_NR_CPUS)
1960 #else
1961 /*
1962  * No preemption supported therefore also no need to check for
1963  * different cpus.
1964  */
1965 #define TID_STEP 1
1966 #endif
1967 
1968 static inline unsigned long next_tid(unsigned long tid)
1969 {
1970         return tid + TID_STEP;
1971 }
1972 
1973 static inline unsigned int tid_to_cpu(unsigned long tid)
1974 {
1975         return tid % TID_STEP;
1976 }
1977 
1978 static inline unsigned long tid_to_event(unsigned long tid)
1979 {
1980         return tid / TID_STEP;
1981 }
1982 
1983 static inline unsigned int init_tid(int cpu)
1984 {
1985         return cpu;
1986 }
1987 
1988 static inline void note_cmpxchg_failure(const char *n,
1989                 const struct kmem_cache *s, unsigned long tid)
1990 {
1991 #ifdef SLUB_DEBUG_CMPXCHG
1992         unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1993 
1994         pr_info("%s %s: cmpxchg redo ", n, s->name);
1995 
1996 #ifdef CONFIG_PREEMPT
1997         if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1998                 pr_warn("due to cpu change %d -> %d\n",
1999                         tid_to_cpu(tid), tid_to_cpu(actual_tid));
2000         else
2001 #endif
2002         if (tid_to_event(tid) != tid_to_event(actual_tid))
2003                 pr_warn("due to cpu running other code. Event %ld->%ld\n",
2004                         tid_to_event(tid), tid_to_event(actual_tid));
2005         else
2006                 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
2007                         actual_tid, tid, next_tid(tid));
2008 #endif
2009         stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
2010 }
2011 
2012 static void init_kmem_cache_cpus(struct kmem_cache *s)
2013 {
2014         int cpu;
2015 
2016         for_each_possible_cpu(cpu)
2017                 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
2018 }
2019 
2020 /*
2021  * Remove the cpu slab
2022  */
2023 static void deactivate_slab(struct kmem_cache *s, struct page *page,
2024                                 void *freelist, struct kmem_cache_cpu *c)
2025 {
2026         enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
2027         struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2028         int lock = 0;
2029         enum slab_modes l = M_NONE, m = M_NONE;
2030         void *nextfree;
2031         int tail = DEACTIVATE_TO_HEAD;
2032         struct page new;
2033         struct page old;
2034 
2035         if (page->freelist) {
2036                 stat(s, DEACTIVATE_REMOTE_FREES);
2037                 tail = DEACTIVATE_TO_TAIL;
2038         }
2039 
2040         /*
2041          * Stage one: Free all available per cpu objects back
2042          * to the page freelist while it is still frozen. Leave the
2043          * last one.
2044          *
2045          * There is no need to take the list->lock because the page
2046          * is still frozen.
2047          */
2048         while (freelist && (nextfree = get_freepointer(s, freelist))) {
2049                 void *prior;
2050                 unsigned long counters;
2051 
2052                 do {
2053                         prior = page->freelist;
2054                         counters = page->counters;
2055                         set_freepointer(s, freelist, prior);
2056                         new.counters = counters;
2057                         new.inuse--;
2058                         VM_BUG_ON(!new.frozen);
2059 
2060                 } while (!__cmpxchg_double_slab(s, page,
2061                         prior, counters,
2062                         freelist, new.counters,
2063                         "drain percpu freelist"));
2064 
2065                 freelist = nextfree;
2066         }
2067 
2068         /*
2069          * Stage two: Ensure that the page is unfrozen while the
2070          * list presence reflects the actual number of objects
2071          * during unfreeze.
2072          *
2073          * We setup the list membership and then perform a cmpxchg
2074          * with the count. If there is a mismatch then the page
2075          * is not unfrozen but the page is on the wrong list.
2076          *
2077          * Then we restart the process which may have to remove
2078          * the page from the list that we just put it on again
2079          * because the number of objects in the slab may have
2080          * changed.
2081          */
2082 redo:
2083 
2084         old.freelist = page->freelist;
2085         old.counters = page->counters;
2086         VM_BUG_ON(!old.frozen);
2087 
2088         /* Determine target state of the slab */
2089         new.counters = old.counters;
2090         if (freelist) {
2091                 new.inuse--;
2092                 set_freepointer(s, freelist, old.freelist);
2093                 new.freelist = freelist;
2094         } else
2095                 new.freelist = old.freelist;
2096 
2097         new.frozen = 0;
2098 
2099         if (!new.inuse && n->nr_partial >= s->min_partial)
2100                 m = M_FREE;
2101         else if (new.freelist) {
2102                 m = M_PARTIAL;
2103                 if (!lock) {
2104                         lock = 1;
2105                         /*
2106                          * Taking the spinlock removes the possiblity
2107                          * that acquire_slab() will see a slab page that
2108                          * is frozen
2109                          */
2110                         spin_lock(&n->list_lock);
2111                 }
2112         } else {
2113                 m = M_FULL;
2114                 if (kmem_cache_debug(s) && !lock) {
2115                         lock = 1;
2116                         /*
2117                          * This also ensures that the scanning of full
2118                          * slabs from diagnostic functions will not see
2119                          * any frozen slabs.
2120                          */
2121                         spin_lock(&n->list_lock);
2122                 }
2123         }
2124 
2125         if (l != m) {
2126 
2127                 if (l == M_PARTIAL)
2128 
2129                         remove_partial(n, page);
2130 
2131                 else if (l == M_FULL)
2132 
2133                         remove_full(s, n, page);
2134 
2135                 if (m == M_PARTIAL) {
2136 
2137                         add_partial(n, page, tail);
2138                         stat(s, tail);
2139 
2140                 } else if (m == M_FULL) {
2141 
2142                         stat(s, DEACTIVATE_FULL);
2143                         add_full(s, n, page);
2144 
2145                 }
2146         }
2147 
2148         l = m;
2149         if (!__cmpxchg_double_slab(s, page,
2150                                 old.freelist, old.counters,
2151                                 new.freelist, new.counters,
2152                                 "unfreezing slab"))
2153                 goto redo;
2154 
2155         if (lock)
2156                 spin_unlock(&n->list_lock);
2157 
2158         if (m == M_FREE) {
2159                 stat(s, DEACTIVATE_EMPTY);
2160                 discard_slab(s, page);
2161                 stat(s, FREE_SLAB);
2162         }
2163 
2164         c->page = NULL;
2165         c->freelist = NULL;
2166 }
2167 
2168 /*
2169  * Unfreeze all the cpu partial slabs.
2170  *
2171  * This function must be called with interrupts disabled
2172  * for the cpu using c (or some other guarantee must be there
2173  * to guarantee no concurrent accesses).
2174  */
2175 static void unfreeze_partials(struct kmem_cache *s,
2176                 struct kmem_cache_cpu *c)
2177 {
2178 #ifdef CONFIG_SLUB_CPU_PARTIAL
2179         struct kmem_cache_node *n = NULL, *n2 = NULL;
2180         struct page *page, *discard_page = NULL;
2181 
2182         while ((page = c->partial)) {
2183                 struct page new;
2184                 struct page old;
2185 
2186                 c->partial = page->next;
2187 
2188                 n2 = get_node(s, page_to_nid(page));
2189                 if (n != n2) {
2190                         if (n)
2191                                 spin_unlock(&n->list_lock);
2192 
2193                         n = n2;
2194                         spin_lock(&n->list_lock);
2195                 }
2196 
2197                 do {
2198 
2199                         old.freelist = page->freelist;
2200                         old.counters = page->counters;
2201                         VM_BUG_ON(!old.frozen);
2202 
2203                         new.counters = old.counters;
2204                         new.freelist = old.freelist;
2205 
2206                         new.frozen = 0;
2207 
2208                 } while (!__cmpxchg_double_slab(s, page,
2209                                 old.freelist, old.counters,
2210                                 new.freelist, new.counters,
2211                                 "unfreezing slab"));
2212 
2213                 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
2214                         page->next = discard_page;
2215                         discard_page = page;
2216                 } else {
2217                         add_partial(n, page, DEACTIVATE_TO_TAIL);
2218                         stat(s, FREE_ADD_PARTIAL);
2219                 }
2220         }
2221 
2222         if (n)
2223                 spin_unlock(&n->list_lock);
2224 
2225         while (discard_page) {
2226                 page = discard_page;
2227                 discard_page = discard_page->next;
2228 
2229                 stat(s, DEACTIVATE_EMPTY);
2230                 discard_slab(s, page);
2231                 stat(s, FREE_SLAB);
2232         }
2233 #endif
2234 }
2235 
2236 /*
2237  * Put a page that was just frozen (in __slab_free) into a partial page
2238  * slot if available.
2239  *
2240  * If we did not find a slot then simply move all the partials to the
2241  * per node partial list.
2242  */
2243 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
2244 {
2245 #ifdef CONFIG_SLUB_CPU_PARTIAL
2246         struct page *oldpage;
2247         int pages;
2248         int pobjects;
2249 
2250         preempt_disable();
2251         do {
2252                 pages = 0;
2253                 pobjects = 0;
2254                 oldpage = this_cpu_read(s->cpu_slab->partial);
2255 
2256                 if (oldpage) {
2257                         pobjects = oldpage->pobjects;
2258                         pages = oldpage->pages;
2259                         if (drain && pobjects > s->cpu_partial) {
2260                                 unsigned long flags;
2261                                 /*
2262                                  * partial array is full. Move the existing
2263                                  * set to the per node partial list.
2264                                  */
2265                                 local_irq_save(flags);
2266                                 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2267                                 local_irq_restore(flags);
2268                                 oldpage = NULL;
2269                                 pobjects = 0;
2270                                 pages = 0;
2271                                 stat(s, CPU_PARTIAL_DRAIN);
2272                         }
2273                 }
2274 
2275                 pages++;
2276                 pobjects += page->objects - page->inuse;
2277 
2278                 page->pages = pages;
2279                 page->pobjects = pobjects;
2280                 page->next = oldpage;
2281 
2282         } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2283                                                                 != oldpage);
2284         if (unlikely(!s->cpu_partial)) {
2285                 unsigned long flags;
2286 
2287                 local_irq_save(flags);
2288                 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2289                 local_irq_restore(flags);
2290         }
2291         preempt_enable();
2292 #endif
2293 }
2294 
2295 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
2296 {
2297         stat(s, CPUSLAB_FLUSH);
2298         deactivate_slab(s, c->page, c->freelist, c);
2299 
2300         c->tid = next_tid(c->tid);
2301 }
2302 
2303 /*
2304  * Flush cpu slab.
2305  *
2306  * Called from IPI handler with interrupts disabled.
2307  */
2308 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
2309 {
2310         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2311 
2312         if (likely(c)) {
2313                 if (c->page)
2314                         flush_slab(s, c);
2315 
2316                 unfreeze_partials(s, c);
2317         }
2318 }
2319 
2320 static void flush_cpu_slab(void *d)
2321 {
2322         struct kmem_cache *s = d;
2323 
2324         __flush_cpu_slab(s, smp_processor_id());
2325 }
2326 
2327 static bool has_cpu_slab(int cpu, void *info)
2328 {
2329         struct kmem_cache *s = info;
2330         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2331 
2332         return c->page || slub_percpu_partial(c);
2333 }
2334 
2335 static void flush_all(struct kmem_cache *s)
2336 {
2337         on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
2338 }
2339 
2340 /*
2341  * Use the cpu notifier to insure that the cpu slabs are flushed when
2342  * necessary.
2343  */
2344 static int slub_cpu_dead(unsigned int cpu)
2345 {
2346         struct kmem_cache *s;
2347         unsigned long flags;
2348 
2349         mutex_lock(&slab_mutex);
2350         list_for_each_entry(s, &slab_caches, list) {
2351                 local_irq_save(flags);
2352                 __flush_cpu_slab(s, cpu);
2353                 local_irq_restore(flags);
2354         }
2355         mutex_unlock(&slab_mutex);
2356         return 0;
2357 }
2358 
2359 /*
2360  * Check if the objects in a per cpu structure fit numa
2361  * locality expectations.
2362  */
2363 static inline int node_match(struct page *page, int node)
2364 {
2365 #ifdef CONFIG_NUMA
2366         if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node))
2367                 return 0;
2368 #endif
2369         return 1;
2370 }
2371 
2372 #ifdef CONFIG_SLUB_DEBUG
2373 static int count_free(struct page *page)
2374 {
2375         return page->objects - page->inuse;
2376 }
2377 
2378 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2379 {
2380         return atomic_long_read(&n->total_objects);
2381 }
2382 #endif /* CONFIG_SLUB_DEBUG */
2383 
2384 #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
2385 static unsigned long count_partial(struct kmem_cache_node *n,
2386                                         int (*get_count)(struct page *))
2387 {
2388         unsigned long flags;
2389         unsigned long x = 0;
2390         struct page *page;
2391 
2392         spin_lock_irqsave(&n->list_lock, flags);
2393         list_for_each_entry(page, &n->partial, lru)
2394                 x += get_count(page);
2395         spin_unlock_irqrestore(&n->list_lock, flags);
2396         return x;
2397 }
2398 #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
2399 
2400 static noinline void
2401 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2402 {
2403 #ifdef CONFIG_SLUB_DEBUG
2404         static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2405                                       DEFAULT_RATELIMIT_BURST);
2406         int node;
2407         struct kmem_cache_node *n;
2408 
2409         if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2410                 return;
2411 
2412         pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2413                 nid, gfpflags, &gfpflags);
2414         pr_warn("  cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n",
2415                 s->name, s->object_size, s->size, oo_order(s->oo),
2416                 oo_order(s->min));
2417 
2418         if (oo_order(s->min) > get_order(s->object_size))
2419                 pr_warn("  %s debugging increased min order, use slub_debug=O to disable.\n",
2420                         s->name);
2421 
2422         for_each_kmem_cache_node(s, node, n) {
2423                 unsigned long nr_slabs;
2424                 unsigned long nr_objs;
2425                 unsigned long nr_free;
2426 
2427                 nr_free  = count_partial(n, count_free);
2428                 nr_slabs = node_nr_slabs(n);
2429                 nr_objs  = node_nr_objs(n);
2430 
2431                 pr_warn("  node %d: slabs: %ld, objs: %ld, free: %ld\n",
2432                         node, nr_slabs, nr_objs, nr_free);
2433         }
2434 #endif
2435 }
2436 
2437 static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2438                         int node, struct kmem_cache_cpu **pc)
2439 {
2440         void *freelist;
2441         struct kmem_cache_cpu *c = *pc;
2442         struct page *page;
2443 
2444         freelist = get_partial(s, flags, node, c);
2445 
2446         if (freelist)
2447                 return freelist;
2448 
2449         page = new_slab(s, flags, node);
2450         if (page) {
2451                 c = raw_cpu_ptr(s->cpu_slab);
2452                 if (c->page)
2453                         flush_slab(s, c);
2454 
2455                 /*
2456                  * No other reference to the page yet so we can
2457                  * muck around with it freely without cmpxchg
2458                  */
2459                 freelist = page->freelist;
2460                 page->freelist = NULL;
2461 
2462                 stat(s, ALLOC_SLAB);
2463                 c->page = page;
2464                 *pc = c;
2465         } else
2466                 freelist = NULL;
2467 
2468         return freelist;
2469 }
2470 
2471 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2472 {
2473         if (unlikely(PageSlabPfmemalloc(page)))
2474                 return gfp_pfmemalloc_allowed(gfpflags);
2475 
2476         return true;
2477 }
2478 
2479 /*
2480  * Check the page->freelist of a page and either transfer the freelist to the
2481  * per cpu freelist or deactivate the page.
2482  *
2483  * The page is still frozen if the return value is not NULL.
2484  *
2485  * If this function returns NULL then the page has been unfrozen.
2486  *
2487  * This function must be called with interrupt disabled.
2488  */
2489 static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2490 {
2491         struct page new;
2492         unsigned long counters;
2493         void *freelist;
2494 
2495         do {
2496                 freelist = page->freelist;
2497                 counters = page->counters;
2498 
2499                 new.counters = counters;
2500                 VM_BUG_ON(!new.frozen);
2501 
2502                 new.inuse = page->objects;
2503                 new.frozen = freelist != NULL;
2504 
2505         } while (!__cmpxchg_double_slab(s, page,
2506                 freelist, counters,
2507                 NULL, new.counters,
2508                 "get_freelist"));
2509 
2510         return freelist;
2511 }
2512 
2513 /*
2514  * Slow path. The lockless freelist is empty or we need to perform
2515  * debugging duties.
2516  *
2517  * Processing is still very fast if new objects have been freed to the
2518  * regular freelist. In that case we simply take over the regular freelist
2519  * as the lockless freelist and zap the regular freelist.
2520  *
2521  * If that is not working then we fall back to the partial lists. We take the
2522  * first element of the freelist as the object to allocate now and move the
2523  * rest of the freelist to the lockless freelist.
2524  *
2525  * And if we were unable to get a new slab from the partial slab lists then
2526  * we need to allocate a new slab. This is the slowest path since it involves
2527  * a call to the page allocator and the setup of a new slab.
2528  *
2529  * Version of __slab_alloc to use when we know that interrupts are
2530  * already disabled (which is the case for bulk allocation).
2531  */
2532 static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2533                           unsigned long addr, struct kmem_cache_cpu *c)
2534 {
2535         void *freelist;
2536         struct page *page;
2537 
2538         page = c->page;
2539         if (!page)
2540                 goto new_slab;
2541 redo:
2542 
2543         if (unlikely(!node_match(page, node))) {
2544                 int searchnode = node;
2545 
2546                 if (node != NUMA_NO_NODE && !node_present_pages(node))
2547                         searchnode = node_to_mem_node(node);
2548 
2549                 if (unlikely(!node_match(page, searchnode))) {
2550                         stat(s, ALLOC_NODE_MISMATCH);
2551                         deactivate_slab(s, page, c->freelist, c);
2552                         goto new_slab;
2553                 }
2554         }
2555 
2556         /*
2557          * By rights, we should be searching for a slab page that was
2558          * PFMEMALLOC but right now, we are losing the pfmemalloc
2559          * information when the page leaves the per-cpu allocator
2560          */
2561         if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2562                 deactivate_slab(s, page, c->freelist, c);
2563                 goto new_slab;
2564         }
2565 
2566         /* must check again c->freelist in case of cpu migration or IRQ */
2567         freelist = c->freelist;
2568         if (freelist)
2569                 goto load_freelist;
2570 
2571         freelist = get_freelist(s, page);
2572 
2573         if (!freelist) {
2574                 c->page = NULL;
2575                 stat(s, DEACTIVATE_BYPASS);
2576                 goto new_slab;
2577         }
2578 
2579         stat(s, ALLOC_REFILL);
2580 
2581 load_freelist:
2582         /*
2583          * freelist is pointing to the list of objects to be used.
2584          * page is pointing to the page from which the objects are obtained.
2585          * That page must be frozen for per cpu allocations to work.
2586          */
2587         VM_BUG_ON(!c->page->frozen);
2588         c->freelist = get_freepointer(s, freelist);
2589         c->tid = next_tid(c->tid);
2590         return freelist;
2591 
2592 new_slab:
2593 
2594         if (slub_percpu_partial(c)) {
2595                 page = c->page = slub_percpu_partial(c);
2596                 slub_set_percpu_partial(c, page);
2597                 stat(s, CPU_PARTIAL_ALLOC);
2598                 goto redo;
2599         }
2600 
2601         freelist = new_slab_objects(s, gfpflags, node, &c);
2602 
2603         if (unlikely(!freelist)) {
2604                 slab_out_of_memory(s, gfpflags, node);
2605                 return NULL;
2606         }
2607 
2608         page = c->page;
2609         if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
2610                 goto load_freelist;
2611 
2612         /* Only entered in the debug case */
2613         if (kmem_cache_debug(s) &&
2614                         !alloc_debug_processing(s, page, freelist, addr))
2615                 goto new_slab;  /* Slab failed checks. Next slab needed */
2616 
2617         deactivate_slab(s, page, get_freepointer(s, freelist), c);
2618         return freelist;
2619 }
2620 
2621 /*
2622  * Another one that disabled interrupt and compensates for possible
2623  * cpu changes by refetching the per cpu area pointer.
2624  */
2625 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2626                           unsigned long addr, struct kmem_cache_cpu *c)
2627 {
2628         void *p;
2629         unsigned long flags;
2630 
2631         local_irq_save(flags);
2632 #ifdef CONFIG_PREEMPT
2633         /*
2634          * We may have been preempted and rescheduled on a different
2635          * cpu before disabling interrupts. Need to reload cpu area
2636          * pointer.
2637          */
2638         c = this_cpu_ptr(s->cpu_slab);
2639 #endif
2640 
2641         p = ___slab_alloc(s, gfpflags, node, addr, c);
2642         local_irq_restore(flags);
2643         return p;
2644 }
2645 
2646 /*
2647  * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2648  * have the fastpath folded into their functions. So no function call
2649  * overhead for requests that can be satisfied on the fastpath.
2650  *
2651  * The fastpath works by first checking if the lockless freelist can be used.
2652  * If not then __slab_alloc is called for slow processing.
2653  *
2654  * Otherwise we can simply pick the next object from the lockless free list.
2655  */
2656 static __always_inline void *slab_alloc_node(struct kmem_cache *s,
2657                 gfp_t gfpflags, int node, unsigned long addr)
2658 {
2659         void *object;
2660         struct kmem_cache_cpu *c;
2661         struct page *page;
2662         unsigned long tid;
2663 
2664         s = slab_pre_alloc_hook(s, gfpflags);
2665         if (!s)
2666                 return NULL;
2667 redo:
2668         /*
2669          * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2670          * enabled. We may switch back and forth between cpus while
2671          * reading from one cpu area. That does not matter as long
2672          * as we end up on the original cpu again when doing the cmpxchg.
2673          *
2674          * We should guarantee that tid and kmem_cache are retrieved on
2675          * the same cpu. It could be different if CONFIG_PREEMPT so we need
2676          * to check if it is matched or not.
2677          */
2678         do {
2679                 tid = this_cpu_read(s->cpu_slab->tid);
2680                 c = raw_cpu_ptr(s->cpu_slab);
2681         } while (IS_ENABLED(CONFIG_PREEMPT) &&
2682                  unlikely(tid != READ_ONCE(c->tid)));
2683 
2684         /*
2685          * Irqless object alloc/free algorithm used here depends on sequence
2686          * of fetching cpu_slab's data. tid should be fetched before anything
2687          * on c to guarantee that object and page associated with previous tid
2688          * won't be used with current tid. If we fetch tid first, object and
2689          * page could be one associated with next tid and our alloc/free
2690          * request will be failed. In this case, we will retry. So, no problem.
2691          */
2692         barrier();
2693 
2694         /*
2695          * The transaction ids are globally unique per cpu and per operation on
2696          * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2697          * occurs on the right processor and that there was no operation on the
2698          * linked list in between.
2699          */
2700 
2701         object = c->freelist;
2702         page = c->page;
2703         if (unlikely(!object || !node_match(page, node))) {
2704                 object = __slab_alloc(s, gfpflags, node, addr, c);
2705                 stat(s, ALLOC_SLOWPATH);
2706         } else {
2707                 void *next_object = get_freepointer_safe(s, object);
2708 
2709                 /*
2710                  * The cmpxchg will only match if there was no additional
2711                  * operation and if we are on the right processor.
2712                  *
2713                  * The cmpxchg does the following atomically (without lock
2714                  * semantics!)
2715                  * 1. Relocate first pointer to the current per cpu area.
2716                  * 2. Verify that tid and freelist have not been changed
2717                  * 3. If they were not changed replace tid and freelist
2718                  *
2719                  * Since this is without lock semantics the protection is only
2720                  * against code executing on this cpu *not* from access by
2721                  * other cpus.
2722                  */
2723                 if (unlikely(!this_cpu_cmpxchg_double(
2724                                 s->cpu_slab->freelist, s->cpu_slab->tid,
2725                                 object, tid,
2726                                 next_object, next_tid(tid)))) {
2727 
2728                         note_cmpxchg_failure("slab_alloc", s, tid);
2729                         goto redo;
2730                 }
2731                 prefetch_freepointer(s, next_object);
2732                 stat(s, ALLOC_FASTPATH);
2733         }
2734 
2735         if (unlikely(gfpflags & __GFP_ZERO) && object)
2736                 memset(object, 0, s->object_size);
2737 
2738         slab_post_alloc_hook(s, gfpflags, 1, &object);
2739 
2740         return object;
2741 }
2742 
2743 static __always_inline void *slab_alloc(struct kmem_cache *s,
2744                 gfp_t gfpflags, unsigned long addr)
2745 {
2746         return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2747 }
2748 
2749 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2750 {
2751         void *ret = slab_alloc(s, gfpflags, _RET_IP_);
2752 
2753         trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2754                                 s->size, gfpflags);
2755 
2756         return ret;
2757 }
2758 EXPORT_SYMBOL(kmem_cache_alloc);
2759 
2760 #ifdef CONFIG_TRACING
2761 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
2762 {
2763         void *ret = slab_alloc(s, gfpflags, _RET_IP_);
2764         trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2765         kasan_kmalloc(s, ret, size, gfpflags);
2766         return ret;
2767 }
2768 EXPORT_SYMBOL(kmem_cache_alloc_trace);
2769 #endif
2770 
2771 #ifdef CONFIG_NUMA
2772 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2773 {
2774         void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
2775 
2776         trace_kmem_cache_alloc_node(_RET_IP_, ret,
2777                                     s->object_size, s->size, gfpflags, node);
2778 
2779         return ret;
2780 }
2781 EXPORT_SYMBOL(kmem_cache_alloc_node);
2782 
2783 #ifdef CONFIG_TRACING
2784 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
2785                                     gfp_t gfpflags,
2786                                     int node, size_t size)
2787 {
2788         void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
2789 
2790         trace_kmalloc_node(_RET_IP_, ret,
2791                            size, s->size, gfpflags, node);
2792 
2793         kasan_kmalloc(s, ret, size, gfpflags);
2794         return ret;
2795 }
2796 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
2797 #endif
2798 #endif
2799 
2800 /*
2801  * Slow path handling. This may still be called frequently since objects
2802  * have a longer lifetime than the cpu slabs in most processing loads.
2803  *
2804  * So we still attempt to reduce cache line usage. Just take the slab
2805  * lock and free the item. If there is no additional partial page
2806  * handling required then we can return immediately.
2807  */
2808 static void __slab_free(struct kmem_cache *s, struct page *page,
2809                         void *head, void *tail, int cnt,
2810                         unsigned long addr)
2811 
2812 {
2813         void *prior;
2814         int was_frozen;
2815         struct page new;
2816         unsigned long counters;
2817         struct kmem_cache_node *n = NULL;
2818         unsigned long uninitialized_var(flags);
2819 
2820         stat(s, FREE_SLOWPATH);
2821 
2822         if (kmem_cache_debug(s) &&
2823             !free_debug_processing(s, page, head, tail, cnt, addr))
2824                 return;
2825 
2826         do {
2827                 if (unlikely(n)) {
2828                         spin_unlock_irqrestore(&n->list_lock, flags);
2829                         n = NULL;
2830                 }
2831                 prior = page->freelist;
2832                 counters = page->counters;
2833                 set_freepointer(s, tail, prior);
2834                 new.counters = counters;
2835                 was_frozen = new.frozen;
2836                 new.inuse -= cnt;
2837                 if ((!new.inuse || !prior) && !was_frozen) {
2838 
2839                         if (kmem_cache_has_cpu_partial(s) && !prior) {
2840 
2841                                 /*
2842                                  * Slab was on no list before and will be
2843                                  * partially empty
2844                                  * We can defer the list move and instead
2845                                  * freeze it.
2846                                  */
2847                                 new.frozen = 1;
2848 
2849                         } else { /* Needs to be taken off a list */
2850 
2851                                 n = get_node(s, page_to_nid(page));
2852                                 /*
2853                                  * Speculatively acquire the list_lock.
2854                                  * If the cmpxchg does not succeed then we may
2855                                  * drop the list_lock without any processing.
2856                                  *
2857                                  * Otherwise the list_lock will synchronize with
2858                                  * other processors updating the list of slabs.
2859                                  */
2860                                 spin_lock_irqsave(&n->list_lock, flags);
2861 
2862                         }
2863                 }
2864 
2865         } while (!cmpxchg_double_slab(s, page,
2866                 prior, counters,
2867                 head, new.counters,
2868                 "__slab_free"));
2869 
2870         if (likely(!n)) {
2871 
2872                 /*
2873                  * If we just froze the page then put it onto the
2874                  * per cpu partial list.
2875                  */
2876                 if (new.frozen && !was_frozen) {
2877                         put_cpu_partial(s, page, 1);
2878                         stat(s, CPU_PARTIAL_FREE);
2879                 }
2880                 /*
2881                  * The list lock was not taken therefore no list
2882                  * activity can be necessary.
2883                  */
2884                 if (was_frozen)
2885                         stat(s, FREE_FROZEN);
2886                 return;
2887         }
2888 
2889         if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
2890                 goto slab_empty;
2891 
2892         /*
2893          * Objects left in the slab. If it was not on the partial list before
2894          * then add it.
2895          */
2896         if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2897                 if (kmem_cache_debug(s))
2898                         remove_full(s, n, page);
2899                 add_partial(n, page, DEACTIVATE_TO_TAIL);
2900                 stat(s, FREE_ADD_PARTIAL);
2901         }
2902         spin_unlock_irqrestore(&n->list_lock, flags);
2903         return;
2904 
2905 slab_empty:
2906         if (prior) {
2907                 /*
2908                  * Slab on the partial list.
2909                  */
2910                 remove_partial(n, page);
2911                 stat(s, FREE_REMOVE_PARTIAL);
2912         } else {
2913                 /* Slab must be on the full list */
2914                 remove_full(s, n, page);
2915         }
2916 
2917         spin_unlock_irqrestore(&n->list_lock, flags);
2918         stat(s, FREE_SLAB);
2919         discard_slab(s, page);
2920 }
2921 
2922 /*
2923  * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2924  * can perform fastpath freeing without additional function calls.
2925  *
2926  * The fastpath is only possible if we are freeing to the current cpu slab
2927  * of this processor. This typically the case if we have just allocated
2928  * the item before.
2929  *
2930  * If fastpath is not possible then fall back to __slab_free where we deal
2931  * with all sorts of special processing.
2932  *
2933  * Bulk free of a freelist with several objects (all pointing to the
2934  * same page) possible by specifying head and tail ptr, plus objects
2935  * count (cnt). Bulk free indicated by tail pointer being set.
2936  */
2937 static __always_inline void do_slab_free(struct kmem_cache *s,
2938                                 struct page *page, void *head, void *tail,
2939                                 int cnt, unsigned long addr)
2940 {
2941         void *tail_obj = tail ? : head;
2942         struct kmem_cache_cpu *c;
2943         unsigned long tid;
2944 redo:
2945         /*
2946          * Determine the currently cpus per cpu slab.
2947          * The cpu may change afterward. However that does not matter since
2948          * data is retrieved via this pointer. If we are on the same cpu
2949          * during the cmpxchg then the free will succeed.
2950          */
2951         do {
2952                 tid = this_cpu_read(s->cpu_slab->tid);
2953                 c = raw_cpu_ptr(s->cpu_slab);
2954         } while (IS_ENABLED(CONFIG_PREEMPT) &&
2955                  unlikely(tid != READ_ONCE(c->tid)));
2956 
2957         /* Same with comment on barrier() in slab_alloc_node() */
2958         barrier();
2959 
2960         if (likely(page == c->page)) {
2961                 set_freepointer(s, tail_obj, c->freelist);
2962 
2963                 if (unlikely(!this_cpu_cmpxchg_double(
2964                                 s->cpu_slab->freelist, s->cpu_slab->tid,
2965                                 c->freelist, tid,
2966                                 head, next_tid(tid)))) {
2967 
2968                         note_cmpxchg_failure("slab_free", s, tid);
2969                         goto redo;
2970                 }
2971                 stat(s, FREE_FASTPATH);
2972         } else
2973                 __slab_free(s, page, head, tail_obj, cnt, addr);
2974 
2975 }
2976 
2977 static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
2978                                       void *head, void *tail, int cnt,
2979                                       unsigned long addr)
2980 {
2981         /*
2982          * With KASAN enabled slab_free_freelist_hook modifies the freelist
2983          * to remove objects, whose reuse must be delayed.
2984          */
2985         if (slab_free_freelist_hook(s, &head, &tail))
2986                 do_slab_free(s, page, head, tail, cnt, addr);
2987 }
2988 
2989 #ifdef CONFIG_KASAN
2990 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
2991 {
2992         do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
2993 }
2994 #endif
2995 
2996 void kmem_cache_free(struct kmem_cache *s, void *x)
2997 {
2998         s = cache_from_obj(s, x);
2999         if (!s)
3000                 return;
3001         slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
3002         trace_kmem_cache_free(_RET_IP_, x);
3003 }
3004 EXPORT_SYMBOL(kmem_cache_free);
3005 
3006 struct detached_freelist {
3007         struct page *page;
3008         void *tail;
3009         void *freelist;
3010         int cnt;
3011         struct kmem_cache *s;
3012 };
3013 
3014 /*
3015  * This function progressively scans the array with free objects (with
3016  * a limited look ahead) and extract objects belonging to the same
3017  * page.  It builds a detached freelist directly within the given
3018  * page/objects.  This can happen without any need for
3019  * synchronization, because the objects are owned by running process.
3020  * The freelist is build up as a single linked list in the objects.
3021  * The idea is, that this detached freelist can then be bulk
3022  * transferred to the real freelist(s), but only requiring a single
3023  * synchronization primitive.  Look ahead in the array is limited due
3024  * to performance reasons.
3025  */
3026 static inline
3027 int build_detached_freelist(struct kmem_cache *s, size_t size,
3028                             void **p, struct detached_freelist *df)
3029 {
3030         size_t first_skipped_index = 0;
3031         int lookahead = 3;
3032         void *object;
3033         struct page *page;
3034 
3035         /* Always re-init detached_freelist */
3036         df->page = NULL;
3037 
3038         do {
3039                 object = p[--size];
3040                 /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
3041         } while (!object && size);
3042 
3043         if (!object)
3044                 return 0;
3045 
3046         page = virt_to_head_page(object);
3047         if (!s) {
3048                 /* Handle kalloc'ed objects */
3049                 if (unlikely(!PageSlab(page))) {
3050                         BUG_ON(!PageCompound(page));
3051                         kfree_hook(object);
3052                         __free_pages(page, compound_order(page));
3053                         p[size] = NULL; /* mark object processed */
3054                         return size;
3055                 }
3056                 /* Derive kmem_cache from object */
3057                 df->s = page->slab_cache;
3058         } else {
3059                 df->s = cache_from_obj(s, object); /* Support for memcg */
3060         }
3061 
3062         /* Start new detached freelist */
3063         df->page = page;
3064         set_freepointer(df->s, object, NULL);
3065         df->tail = object;
3066         df->freelist = object;
3067         p[size] = NULL; /* mark object processed */
3068         df->cnt = 1;
3069 
3070         while (size) {
3071                 object = p[--size];
3072                 if (!object)
3073                         continue; /* Skip processed objects */
3074 
3075                 /* df->page is always set at this point */
3076                 if (df->page == virt_to_head_page(object)) {
3077                         /* Opportunity build freelist */
3078                         set_freepointer(df->s, object, df->freelist);
3079                         df->freelist = object;
3080                         df->cnt++;
3081                         p[size] = NULL; /* mark object processed */
3082 
3083                         continue;
3084                 }
3085 
3086                 /* Limit look ahead search */
3087                 if (!--lookahead)
3088                         break;
3089 
3090                 if (!first_skipped_index)
3091                         first_skipped_index = size + 1;
3092         }
3093 
3094         return first_skipped_index;
3095 }
3096 
3097 /* Note that interrupts must be enabled when calling this function. */
3098 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
3099 {
3100         if (WARN_ON(!size))
3101                 return;
3102 
3103         do {
3104                 struct detached_freelist df;
3105 
3106                 size = build_detached_freelist(s, size, p, &df);
3107                 if (!df.page)
3108                         continue;
3109 
3110                 slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_);
3111         } while (likely(size));
3112 }
3113 EXPORT_SYMBOL(kmem_cache_free_bulk);
3114 
3115 /* Note that interrupts must be enabled when calling this function. */
3116 int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3117                           void **p)
3118 {
3119         struct kmem_cache_cpu *c;
3120         int i;
3121 
3122         /* memcg and kmem_cache debug support */
3123         s = slab_pre_alloc_hook(s, flags);
3124         if (unlikely(!s))
3125                 return false;
3126         /*
3127          * Drain objects in the per cpu slab, while disabling local
3128          * IRQs, which protects against PREEMPT and interrupts
3129          * handlers invoking normal fastpath.
3130          */
3131         local_irq_disable();
3132         c = this_cpu_ptr(s->cpu_slab);
3133 
3134         for (i = 0; i < size; i++) {
3135                 void *object = c->freelist;
3136 
3137                 if (unlikely(!object)) {
3138                         /*
3139                          * Invoking slow path likely have side-effect
3140                          * of re-populating per CPU c->freelist
3141                          */
3142                         p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
3143                                             _RET_IP_, c);
3144                         if (unlikely(!p[i]))
3145                                 goto error;
3146 
3147                         c = this_cpu_ptr(s->cpu_slab);
3148                         continue; /* goto for-loop */
3149                 }
3150                 c->freelist = get_freepointer(s, object);
3151                 p[i] = object;
3152         }
3153         c->tid = next_tid(c->tid);
3154         local_irq_enable();
3155 
3156         /* Clear memory outside IRQ disabled fastpath loop */
3157         if (unlikely(flags & __GFP_ZERO)) {
3158                 int j;
3159 
3160                 for (j = 0; j < i; j++)
3161                         memset(p[j], 0, s->object_size);
3162         }
3163 
3164         /* memcg and kmem_cache debug support */
3165         slab_post_alloc_hook(s, flags, size, p);
3166         return i;
3167 error:
3168         local_irq_enable();
3169         slab_post_alloc_hook(s, flags, i, p);
3170         __kmem_cache_free_bulk(s, i, p);
3171         return 0;
3172 }
3173 EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3174 
3175 
3176 /*
3177  * Object placement in a slab is made very easy because we always start at
3178  * offset 0. If we tune the size of the object to the alignment then we can
3179  * get the required alignment by putting one properly sized object after
3180  * another.
3181  *
3182  * Notice that the allocation order determines the sizes of the per cpu
3183  * caches. Each processor has always one slab available for allocations.
3184  * Increasing the allocation order reduces the number of times that slabs
3185  * must be moved on and off the partial lists and is therefore a factor in
3186  * locking overhead.
3187  */
3188 
3189 /*
3190  * Mininum / Maximum order of slab pages. This influences locking overhead
3191  * and slab fragmentation. A higher order reduces the number of partial slabs
3192  * and increases the number of allocations possible without having to
3193  * take the list_lock.
3194  */
3195 static int slub_min_order;
3196 static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
3197 static int slub_min_objects;
3198 
3199 /*
3200  * Calculate the order of allocation given an slab object size.
3201  *
3202  * The order of allocation has significant impact on performance and other
3203  * system components. Generally order 0 allocations should be preferred since
3204  * order 0 does not cause fragmentation in the page allocator. Larger objects
3205  * be problematic to put into order 0 slabs because there may be too much
3206  * unused space left. We go to a higher order if more than 1/16th of the slab
3207  * would be wasted.
3208  *
3209  * In order to reach satisfactory performance we must ensure that a minimum
3210  * number of objects is in one slab. Otherwise we may generate too much
3211  * activity on the partial lists which requires taking the list_lock. This is
3212  * less a concern for large slabs though which are rarely used.
3213  *
3214  * slub_max_order specifies the order where we begin to stop considering the
3215  * number of objects in a slab as critical. If we reach slub_max_order then
3216  * we try to keep the page order as low as possible. So we accept more waste
3217  * of space in favor of a small page order.
3218  *
3219  * Higher order allocations also allow the placement of more objects in a
3220  * slab and thereby reduce object handling overhead. If the user has
3221  * requested a higher mininum order then we start with that one instead of
3222  * the smallest order which will fit the object.
3223  */
3224 static inline int slab_order(int size, int min_objects,
3225                                 int max_order, int fract_leftover, int reserved)
3226 {
3227         int order;
3228         int rem;
3229         int min_order = slub_min_order;
3230 
3231         if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
3232                 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
3233 
3234         for (order = max(min_order, get_order(min_objects * size + reserved));
3235                         order <= max_order; order++) {
3236 
3237                 unsigned long slab_size = PAGE_SIZE << order;
3238 
3239                 rem = (slab_size - reserved) % size;
3240 
3241                 if (rem <= slab_size / fract_leftover)
3242                         break;
3243         }
3244 
3245         return order;
3246 }
3247 
3248 static inline int calculate_order(int size, int reserved)
3249 {
3250         int order;
3251         int min_objects;
3252         int fraction;
3253         int max_objects;
3254 
3255         /*
3256          * Attempt to find best configuration for a slab. This
3257          * works by first attempting to generate a layout with
3258          * the best configuration and backing off gradually.
3259          *
3260          * First we increase the acceptable waste in a slab. Then
3261          * we reduce the minimum objects required in a slab.
3262          */
3263         min_objects = slub_min_objects;
3264         if (!min_objects)
3265                 min_objects = 4 * (fls(nr_cpu_ids) + 1);
3266         max_objects = order_objects(slub_max_order, size, reserved);
3267         min_objects = min(min_objects, max_objects);
3268 
3269         while (min_objects > 1) {
3270                 fraction = 16;
3271                 while (fraction >= 4) {
3272                         order = slab_order(size, min_objects,
3273                                         slub_max_order, fraction, reserved);
3274                         if (order <= slub_max_order)
3275                                 return order;
3276                         fraction /= 2;
3277                 }
3278                 min_objects--;
3279         }
3280 
3281         /*
3282          * We were unable to place multiple objects in a slab. Now
3283          * lets see if we can place a single object there.
3284          */
3285         order = slab_order(size, 1, slub_max_order, 1, reserved);
3286         if (order <= slub_max_order)
3287                 return order;
3288 
3289         /*
3290          * Doh this slab cannot be placed using slub_max_order.
3291          */
3292         order = slab_order(size, 1, MAX_ORDER, 1, reserved);
3293         if (order < MAX_ORDER)
3294                 return order;
3295         return -ENOSYS;
3296 }
3297 
3298 static void
3299 init_kmem_cache_node(struct kmem_cache_node *n)
3300 {
3301         n->nr_partial = 0;
3302         spin_lock_init(&n->list_lock);
3303         INIT_LIST_HEAD(&n->partial);
3304 #ifdef CONFIG_SLUB_DEBUG
3305         atomic_long_set(&n->nr_slabs, 0);
3306         atomic_long_set(&n->total_objects, 0);
3307         INIT_LIST_HEAD(&n->full);
3308 #endif
3309 }
3310 
3311 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
3312 {
3313         BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
3314                         KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
3315 
3316         /*
3317          * Must align to double word boundary for the double cmpxchg
3318          * instructions to work; see __pcpu_double_call_return_bool().
3319          */
3320         s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3321                                      2 * sizeof(void *));
3322 
3323         if (!s->cpu_slab)
3324                 return 0;
3325 
3326         init_kmem_cache_cpus(s);
3327 
3328         return 1;
3329 }
3330 
3331 static struct kmem_cache *kmem_cache_node;
3332 
3333 /*
3334  * No kmalloc_node yet so do it by hand. We know that this is the first
3335  * slab on the node for this slabcache. There are no concurrent accesses
3336  * possible.
3337  *
3338  * Note that this function only works on the kmem_cache_node
3339  * when allocating for the kmem_cache_node. This is used for bootstrapping
3340  * memory on a fresh node that has no slab structures yet.
3341  */
3342 static void early_kmem_cache_node_alloc(int node)
3343 {
3344         struct page *page;
3345         struct kmem_cache_node *n;
3346 
3347         BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
3348 
3349         page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
3350 
3351         BUG_ON(!page);
3352         if (page_to_nid(page) != node) {
3353                 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3354                 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
3355         }
3356 
3357         n = page->freelist;
3358         BUG_ON(!n);
3359         page->freelist = get_freepointer(kmem_cache_node, n);
3360         page->inuse = 1;
3361         page->frozen = 0;
3362         kmem_cache_node->node[node] = n;
3363 #ifdef CONFIG_SLUB_DEBUG
3364         init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
3365         init_tracking(kmem_cache_node, n);
3366 #endif
3367         kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node),
3368                       GFP_KERNEL);
3369         init_kmem_cache_node(n);
3370         inc_slabs_node(kmem_cache_node, node, page->objects);
3371 
3372         /*
3373          * No locks need to be taken here as it has just been
3374          * initialized and there is no concurrent access.
3375          */
3376         __add_partial(n, page, DEACTIVATE_TO_HEAD);
3377 }
3378 
3379 static void free_kmem_cache_nodes(struct kmem_cache *s)
3380 {
3381         int node;
3382         struct kmem_cache_node *n;
3383 
3384         for_each_kmem_cache_node(s, node, n) {
3385                 s->node[node] = NULL;
3386                 kmem_cache_free(kmem_cache_node, n);
3387         }
3388 }
3389 
3390 void __kmem_cache_release(struct kmem_cache *s)
3391 {
3392         cache_random_seq_destroy(s);
3393         free_percpu(s->cpu_slab);
3394         free_kmem_cache_nodes(s);
3395 }
3396 
3397 static int init_kmem_cache_nodes(struct kmem_cache *s)
3398 {
3399         int node;
3400 
3401         for_each_node_state(node, N_NORMAL_MEMORY) {
3402                 struct kmem_cache_node *n;
3403 
3404                 if (slab_state == DOWN) {
3405                         early_kmem_cache_node_alloc(node);
3406                         continue;
3407                 }
3408                 n = kmem_cache_alloc_node(kmem_cache_node,
3409                                                 GFP_KERNEL, node);
3410 
3411                 if (!n) {
3412                         free_kmem_cache_nodes(s);
3413                         return 0;
3414                 }
3415 
3416                 init_kmem_cache_node(n);
3417                 s->node[node] = n;
3418         }
3419         return 1;
3420 }
3421 
3422 static void set_min_partial(struct kmem_cache *s, unsigned long min)
3423 {
3424         if (min < MIN_PARTIAL)
3425                 min = MIN_PARTIAL;
3426         else if (min > MAX_PARTIAL)
3427                 min = MAX_PARTIAL;
3428         s->min_partial = min;
3429 }
3430 
3431 static void set_cpu_partial(struct kmem_cache *s)
3432 {
3433 #ifdef CONFIG_SLUB_CPU_PARTIAL
3434         /*
3435          * cpu_partial determined the maximum number of objects kept in the
3436          * per cpu partial lists of a processor.
3437          *
3438          * Per cpu partial lists mainly contain slabs that just have one
3439          * object freed. If they are used for allocation then they can be
3440          * filled up again with minimal effort. The slab will never hit the
3441          * per node partial lists and therefore no locking will be required.
3442          *
3443          * This setting also determines
3444          *
3445          * A) The number of objects from per cpu partial slabs dumped to the
3446          *    per node list when we reach the limit.
3447          * B) The number of objects in cpu partial slabs to extract from the
3448          *    per node list when we run out of per cpu objects. We only fetch
3449          *    50% to keep some capacity around for frees.
3450          */
3451         if (!kmem_cache_has_cpu_partial(s))
3452                 s->cpu_partial = 0;
3453         else if (s->size >= PAGE_SIZE)
3454                 s->cpu_partial = 2;
3455         else if (s->size >= 1024)
3456                 s->cpu_partial = 6;
3457         else if (s->size >= 256)
3458                 s->cpu_partial = 13;
3459         else
3460                 s->cpu_partial = 30;
3461 #endif
3462 }
3463 
3464 /*
3465  * calculate_sizes() determines the order and the distribution of data within
3466  * a slab object.
3467  */
3468 static int calculate_sizes(struct kmem_cache *s, int forced_order)
3469 {
3470         slab_flags_t flags = s->flags;
3471         size_t size = s->object_size;
3472         int order;
3473 
3474         /*
3475          * Round up object size to the next word boundary. We can only
3476          * place the free pointer at word boundaries and this determines
3477          * the possible location of the free pointer.
3478          */
3479         size = ALIGN(size, sizeof(void *));
3480 
3481 #ifdef CONFIG_SLUB_DEBUG
3482         /*
3483          * Determine if we can poison the object itself. If the user of
3484          * the slab may touch the object after free or before allocation
3485          * then we should never poison the object itself.
3486          */
3487         if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
3488                         !s->ctor)
3489                 s->flags |= __OBJECT_POISON;
3490         else
3491                 s->flags &= ~__OBJECT_POISON;
3492 
3493 
3494         /*
3495          * If we are Redzoning then check if there is some space between the
3496          * end of the object and the free pointer. If not then add an
3497          * additional word to have some bytes to store Redzone information.
3498          */
3499         if ((flags & SLAB_RED_ZONE) && size == s->object_size)
3500                 size += sizeof(void *);
3501 #endif
3502 
3503         /*
3504          * With that we have determined the number of bytes in actual use
3505          * by the object. This is the potential offset to the free pointer.
3506          */
3507         s->inuse = size;
3508 
3509         if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
3510                 s->ctor)) {
3511                 /*
3512                  * Relocate free pointer after the object if it is not
3513                  * permitted to overwrite the first word of the object on
3514                  * kmem_cache_free.
3515                  *
3516                  * This is the case if we do RCU, have a constructor or
3517                  * destructor or are poisoning the objects.
3518                  */
3519                 s->offset = size;
3520                 size += sizeof(void *);
3521         }
3522 
3523 #ifdef CONFIG_SLUB_DEBUG
3524         if (flags & SLAB_STORE_USER)
3525                 /*
3526                  * Need to store information about allocs and frees after
3527                  * the object.
3528                  */
3529                 size += 2 * sizeof(struct track);
3530 #endif
3531 
3532         kasan_cache_create(s, &size, &s->flags);
3533 #ifdef CONFIG_SLUB_DEBUG
3534         if (flags & SLAB_RED_ZONE) {
3535                 /*
3536                  * Add some empty padding so that we can catch
3537                  * overwrites from earlier objects rather than let
3538                  * tracking information or the free pointer be
3539                  * corrupted if a user writes before the start
3540                  * of the object.
3541                  */
3542                 size += sizeof(void *);
3543 
3544                 s->red_left_pad = sizeof(void *);
3545                 s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3546                 size += s->red_left_pad;
3547         }
3548 #endif
3549 
3550         /*
3551          * SLUB stores one object immediately after another beginning from
3552          * offset 0. In order to align the objects we have to simply size
3553          * each object to conform to the alignment.
3554          */
3555         size = ALIGN(size, s->align);
3556         s->size = size;
3557         if (forced_order >= 0)
3558                 order = forced_order;
3559         else
3560                 order = calculate_order(size, s->reserved);
3561 
3562         if (order < 0)
3563                 return 0;
3564 
3565         s->allocflags = 0;
3566         if (order)
3567                 s->allocflags |= __GFP_COMP;
3568 
3569         if (s->flags & SLAB_CACHE_DMA)
3570                 s->allocflags |= GFP_DMA;
3571 
3572         if (s->flags & SLAB_RECLAIM_ACCOUNT)
3573                 s->allocflags |= __GFP_RECLAIMABLE;
3574 
3575         /*
3576          * Determine the number of objects per slab
3577          */
3578         s->oo = oo_make(order, size, s->reserved);
3579         s->min = oo_make(get_order(size), size, s->reserved);
3580         if (oo_objects(s->oo) > oo_objects(s->max))
3581                 s->max = s->oo;
3582 
3583         return !!oo_objects(s->oo);
3584 }
3585 
3586 static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
3587 {
3588         s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
3589         s->reserved = 0;
3590 #ifdef CONFIG_SLAB_FREELIST_HARDENED
3591         s->random = get_random_long();
3592 #endif
3593 
3594         if (need_reserve_slab_rcu && (s->flags & SLAB_TYPESAFE_BY_RCU))
3595                 s->reserved = sizeof(struct rcu_head);
3596 
3597         if (!calculate_sizes(s, -1))
3598                 goto error;
3599         if (disable_higher_order_debug) {
3600                 /*
3601                  * Disable debugging flags that store metadata if the min slab
3602                  * order increased.
3603                  */
3604                 if (get_order(s->size) > get_order(s->object_size)) {
3605                         s->flags &= ~DEBUG_METADATA_FLAGS;
3606                         s->offset = 0;
3607                         if (!calculate_sizes(s, -1))
3608                                 goto error;
3609                 }
3610         }
3611 
3612 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3613     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3614         if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
3615                 /* Enable fast mode */
3616                 s->flags |= __CMPXCHG_DOUBLE;
3617 #endif
3618 
3619         /*
3620          * The larger the object size is, the more pages we want on the partial
3621          * list to avoid pounding the page allocator excessively.
3622          */
3623         set_min_partial(s, ilog2(s->size) / 2);
3624 
3625         set_cpu_partial(s);
3626 
3627 #ifdef CONFIG_NUMA
3628         s->remote_node_defrag_ratio = 1000;
3629 #endif
3630 
3631         /* Initialize the pre-computed randomized freelist if slab is up */
3632         if (slab_state >= UP) {
3633                 if (init_cache_random_seq(s))
3634                         goto error;
3635         }
3636 
3637         if (!init_kmem_cache_nodes(s))
3638                 goto error;
3639 
3640         if (alloc_kmem_cache_cpus(s))
3641                 return 0;
3642 
3643         free_kmem_cache_nodes(s);
3644 error:
3645         if (flags & SLAB_PANIC)
3646                 panic("Cannot create slab %s size=%lu realsize=%u order=%u offset=%u flags=%lx\n",
3647                       s->name, (unsigned long)s->size, s->size,
3648                       oo_order(s->oo), s->offset, (unsigned long)flags);
3649         return -EINVAL;
3650 }
3651 
3652 static void list_slab_objects(struct kmem_cache *s, struct page *page,
3653                                                         const char *text)
3654 {
3655 #ifdef CONFIG_SLUB_DEBUG
3656         void *addr = page_address(page);
3657         void *p;
3658         unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3659                                      sizeof(long), GFP_ATOMIC);
3660         if (!map)
3661                 return;
3662         slab_err(s, page, text, s->name);
3663         slab_lock(page);
3664 
3665         get_map(s, page, map);
3666         for_each_object(p, s, addr, page->objects) {
3667 
3668                 if (!test_bit(slab_index(p, s, addr), map)) {
3669                         pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
3670                         print_tracking(s, p);
3671                 }
3672         }
3673         slab_unlock(page);
3674         kfree(map);
3675 #endif
3676 }
3677 
3678 /*
3679  * Attempt to free all partial slabs on a node.
3680  * This is called from __kmem_cache_shutdown(). We must take list_lock
3681  * because sysfs file might still access partial list after the shutdowning.
3682  */
3683 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
3684 {
3685         LIST_HEAD(discard);
3686         struct page *page, *h;
3687 
3688         BUG_ON(irqs_disabled());
3689         spin_lock_irq(&n->list_lock);
3690         list_for_each_entry_safe(page, h, &n->partial, lru) {
3691                 if (!page->inuse) {
3692                         remove_partial(n, page);
3693                         list_add(&page->lru, &discard);
3694                 } else {
3695                         list_slab_objects(s, page,
3696                         "Objects remaining in %s on __kmem_cache_shutdown()");
3697                 }
3698         }
3699         spin_unlock_irq(&n->list_lock);
3700 
3701         list_for_each_entry_safe(page, h, &discard, lru)
3702                 discard_slab(s, page);
3703 }
3704 
3705 /*
3706  * Release all resources used by a slab cache.
3707  */
3708 int __kmem_cache_shutdown(struct kmem_cache *s)
3709 {
3710         int node;
3711         struct kmem_cache_node *n;
3712 
3713         flush_all(s);
3714         /* Attempt to free all objects */
3715         for_each_kmem_cache_node(s, node, n) {
3716                 free_partial(s, n);
3717                 if (n->nr_partial || slabs_node(s, node))
3718                         return 1;
3719         }
3720         sysfs_slab_remove(s);
3721         return 0;
3722 }
3723 
3724 /********************************************************************
3725  *              Kmalloc subsystem
3726  *******************************************************************/
3727 
3728 static int __init setup_slub_min_order(char *str)
3729 {
3730         get_option(&str, &slub_min_order);
3731 
3732         return 1;
3733 }
3734 
3735 __setup("slub_min_order=", setup_slub_min_order);
3736 
3737 static int __init setup_slub_max_order(char *str)
3738 {
3739         get_option(&str, &slub_max_order);
3740         slub_max_order = min(slub_max_order, MAX_ORDER - 1);
3741 
3742         return 1;
3743 }
3744 
3745 __setup("slub_max_order=", setup_slub_max_order);
3746 
3747 static int __init setup_slub_min_objects(char *str)
3748 {
3749         get_option(&str, &slub_min_objects);
3750 
3751         return 1;
3752 }
3753 
3754 __setup("slub_min_objects=", setup_slub_min_objects);
3755 
3756 void *__kmalloc(size_t size, gfp_t flags)
3757 {
3758         struct kmem_cache *s;
3759         void *ret;
3760 
3761         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
3762                 return kmalloc_large(size, flags);
3763 
3764         s = kmalloc_slab(size, flags);
3765 
3766         if (unlikely(ZERO_OR_NULL_PTR(s)))
3767                 return s;
3768 
3769         ret = slab_alloc(s, flags, _RET_IP_);
3770 
3771         trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
3772 
3773         kasan_kmalloc(s, ret, size, flags);
3774 
3775         return ret;
3776 }
3777 EXPORT_SYMBOL(__kmalloc);
3778 
3779 #ifdef CONFIG_NUMA
3780 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3781 {
3782         struct page *page;
3783         void *ptr = NULL;
3784 
3785         flags |= __GFP_COMP;
3786         page = alloc_pages_node(node, flags, get_order(size));
3787         if (page)
3788                 ptr = page_address(page);
3789 
3790         kmalloc_large_node_hook(ptr, size, flags);
3791         return ptr;
3792 }
3793 
3794 void *__kmalloc_node(size_t size, gfp_t flags, int node)
3795 {
3796         struct kmem_cache *s;
3797         void *ret;
3798 
3799         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
3800                 ret = kmalloc_large_node(size, flags, node);
3801 
3802                 trace_kmalloc_node(_RET_IP_, ret,
3803                                    size, PAGE_SIZE << get_order(size),
3804                                    flags, node);
3805 
3806                 return ret;
3807         }
3808 
3809         s = kmalloc_slab(size, flags);
3810 
3811         if (unlikely(ZERO_OR_NULL_PTR(s)))
3812                 return s;
3813 
3814         ret = slab_alloc_node(s, flags, node, _RET_IP_);
3815 
3816         trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
3817 
3818         kasan_kmalloc(s, ret, size, flags);
3819 
3820         return ret;
3821 }
3822 EXPORT_SYMBOL(__kmalloc_node);
3823 #endif
3824 
3825 #ifdef CONFIG_HARDENED_USERCOPY
3826 /*
3827  * Rejects incorrectly sized objects and objects that are to be copied
3828  * to/from userspace but do not fall entirely within the containing slab
3829  * cache's usercopy region.
3830  *
3831  * Returns NULL if check passes, otherwise const char * to name of cache
3832  * to indicate an error.
3833  */
3834 void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
3835                          bool to_user)
3836 {
3837         struct kmem_cache *s;
3838         unsigned long offset;
3839         size_t object_size;
3840 
3841         /* Find object and usable object size. */
3842         s = page->slab_cache;
3843 
3844         /* Reject impossible pointers. */
3845         if (ptr < page_address(page))
3846                 usercopy_abort("SLUB object not in SLUB page?!", NULL,
3847                                to_user, 0, n);
3848 
3849         /* Find offset within object. */
3850         offset = (ptr - page_address(page)) % s->size;
3851 
3852         /* Adjust for redzone and reject if within the redzone. */
3853         if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) {
3854                 if (offset < s->red_left_pad)
3855                         usercopy_abort("SLUB object in left red zone",
3856                                        s->name, to_user, offset, n);
3857                 offset -= s->red_left_pad;
3858         }
3859 
3860         /* Allow address range falling entirely within usercopy region. */
3861         if (offset >= s->useroffset &&
3862             offset - s->useroffset <= s->usersize &&
3863             n <= s->useroffset - offset + s->usersize)
3864                 return;
3865 
3866         /*
3867          * If the copy is still within the allocated object, produce
3868          * a warning instead of rejecting the copy. This is intended
3869          * to be a temporary method to find any missing usercopy
3870          * whitelists.
3871          */
3872         object_size = slab_ksize(s);
3873         if (usercopy_fallback &&
3874             offset <= object_size && n <= object_size - offset) {
3875                 usercopy_warn("SLUB object", s->name, to_user, offset, n);
3876                 return;
3877         }
3878 
3879         usercopy_abort("SLUB object", s->name, to_user, offset, n);
3880 }
3881 #endif /* CONFIG_HARDENED_USERCOPY */
3882 
3883 static size_t __ksize(const void *object)
3884 {
3885         struct page *page;
3886 
3887         if (unlikely(object == ZERO_SIZE_PTR))
3888                 return 0;
3889 
3890         page = virt_to_head_page(object);
3891 
3892         if (unlikely(!PageSlab(page))) {
3893                 WARN_ON(!PageCompound(page));
3894                 return PAGE_SIZE << compound_order(page);
3895         }
3896 
3897         return slab_ksize(page->slab_cache);
3898 }
3899 
3900 size_t ksize(const void *object)
3901 {
3902         size_t size = __ksize(object);
3903         /* We assume that ksize callers could use whole allocated area,
3904          * so we need to unpoison this area.
3905          */
3906         kasan_unpoison_shadow(object, size);
3907         return size;
3908 }
3909 EXPORT_SYMBOL(ksize);
3910 
3911 void kfree(const void *x)
3912 {
3913         struct page *page;
3914         void *object = (void *)x;
3915 
3916         trace_kfree(_RET_IP_, x);
3917 
3918         if (unlikely(ZERO_OR_NULL_PTR(x)))
3919                 return;
3920 
3921         page = virt_to_head_page(x);
3922         if (unlikely(!PageSlab(page))) {
3923                 BUG_ON(!PageCompound(page));
3924                 kfree_hook(object);
3925                 __free_pages(page, compound_order(page));
3926                 return;
3927         }
3928         slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
3929 }
3930 EXPORT_SYMBOL(kfree);
3931 
3932 #define SHRINK_PROMOTE_MAX 32
3933 
3934 /*
3935  * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3936  * up most to the head of the partial lists. New allocations will then
3937  * fill those up and thus they can be removed from the partial lists.
3938  *
3939  * The slabs with the least items are placed last. This results in them
3940  * being allocated from last increasing the chance that the last objects
3941  * are freed in them.
3942  */
3943 int __kmem_cache_shrink(struct kmem_cache *s)
3944 {
3945         int node;
3946         int i;
3947         struct kmem_cache_node *n;
3948         struct page *page;
3949         struct page *t;
3950         struct list_head discard;
3951         struct list_head promote[SHRINK_PROMOTE_MAX];
3952         unsigned long flags;
3953         int ret = 0;
3954 
3955         flush_all(s);
3956         for_each_kmem_cache_node(s, node, n) {
3957                 INIT_LIST_HEAD(&discard);
3958                 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
3959                         INIT_LIST_HEAD(promote + i);
3960 
3961                 spin_lock_irqsave(&n->list_lock, flags);
3962 
3963                 /*
3964                  * Build lists of slabs to discard or promote.
3965                  *
3966                  * Note that concurrent frees may occur while we hold the
3967                  * list_lock. page->inuse here is the upper limit.
3968                  */
3969                 list_for_each_entry_safe(page, t, &n->partial, lru) {
3970                         int free = page->objects - page->inuse;
3971 
3972                         /* Do not reread page->inuse */
3973                         barrier();
3974 
3975                         /* We do not keep full slabs on the list */
3976                         BUG_ON(free <= 0);
3977 
3978                         if (free == page->objects) {
3979                                 list_move(&page->lru, &discard);
3980                                 n->nr_partial--;
3981                         } else if (free <= SHRINK_PROMOTE_MAX)
3982                                 list_move(&page->lru, promote + free - 1);
3983                 }
3984 
3985                 /*
3986                  * Promote the slabs filled up most to the head of the
3987                  * partial list.
3988                  */
3989                 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
3990                         list_splice(promote + i, &n->partial);
3991 
3992                 spin_unlock_irqrestore(&n->list_lock, flags);
3993 
3994                 /* Release empty slabs */
3995                 list_for_each_entry_safe(page, t, &discard, lru)
3996                         discard_slab(s, page);
3997 
3998                 if (slabs_node(s, node))
3999                         ret = 1;
4000         }
4001 
4002         return ret;
4003 }
4004 
4005 #ifdef CONFIG_MEMCG
4006 static void kmemcg_cache_deact_after_rcu(struct kmem_cache *s)
4007 {
4008         /*
4009          * Called with all the locks held after a sched RCU grace period.
4010          * Even if @s becomes empty after shrinking, we can't know that @s
4011          * doesn't have allocations already in-flight and thus can't
4012          * destroy @s until the associated memcg is released.
4013          *
4014          * However, let's remove the sysfs files for empty caches here.
4015          * Each cache has a lot of interface files which aren't
4016          * particularly useful for empty draining caches; otherwise, we can
4017          * easily end up with millions of unnecessary sysfs files on
4018          * systems which have a lot of memory and transient cgroups.
4019          */
4020         if (!__kmem_cache_shrink(s))
4021                 sysfs_slab_remove(s);
4022 }
4023 
4024 void __kmemcg_cache_deactivate(struct kmem_cache *s)
4025 {
4026         /*
4027          * Disable empty slabs caching. Used to avoid pinning offline
4028          * memory cgroups by kmem pages that can be freed.
4029          */
4030         slub_set_cpu_partial(s, 0);
4031         s->min_partial = 0;
4032 
4033         /*
4034          * s->cpu_partial is checked locklessly (see put_cpu_partial), so
4035          * we have to make sure the change is visible before shrinking.
4036          */
4037         slab_deactivate_memcg_cache_rcu_sched(s, kmemcg_cache_deact_after_rcu);
4038 }
4039 #endif
4040 
4041 static int slab_mem_going_offline_callback(void *arg)
4042 {
4043         struct kmem_cache *s;
4044 
4045         mutex_lock(&slab_mutex);
4046         list_for_each_entry(s, &slab_caches, list)
4047                 __kmem_cache_shrink(s);
4048         mutex_unlock(&slab_mutex);
4049 
4050         return 0;
4051 }
4052 
4053 static void slab_mem_offline_callback(void *arg)
4054 {
4055         struct kmem_cache_node *n;
4056         struct kmem_cache *s;
4057         struct memory_notify *marg = arg;
4058         int offline_node;
4059 
4060         offline_node = marg->status_change_nid_normal;
4061 
4062         /*
4063          * If the node still has available memory. we need kmem_cache_node
4064          * for it yet.
4065          */
4066         if (offline_node < 0)
4067                 return;
4068 
4069         mutex_lock(&slab_mutex);
4070         list_for_each_entry(s, &slab_caches, list) {
4071                 n = get_node(s, offline_node);
4072                 if (n) {
4073                         /*
4074                          * if n->nr_slabs > 0, slabs still exist on the node
4075                          * that is going down. We were unable to free them,
4076                          * and offline_pages() function shouldn't call this
4077                          * callback. So, we must fail.
4078                          */
4079                         BUG_ON(slabs_node(s, offline_node));
4080 
4081                         s->node[offline_node] = NULL;
4082                         kmem_cache_free(kmem_cache_node, n);
4083                 }
4084         }
4085         mutex_unlock(&slab_mutex);
4086 }
4087 
4088 static int slab_mem_going_online_callback(void *arg)
4089 {
4090         struct kmem_cache_node *n;
4091         struct kmem_cache *s;
4092         struct memory_notify *marg = arg;
4093         int nid = marg->status_change_nid_normal;
4094         int ret = 0;
4095 
4096         /*
4097          * If the node's memory is already available, then kmem_cache_node is
4098          * already created. Nothing to do.
4099          */
4100         if (nid < 0)
4101                 return 0;
4102 
4103         /*
4104          * We are bringing a node online. No memory is available yet. We must
4105          * allocate a kmem_cache_node structure in order to bring the node
4106          * online.
4107          */
4108         mutex_lock(&slab_mutex);
4109         list_for_each_entry(s, &slab_caches, list) {
4110                 /*
4111                  * XXX: kmem_cache_alloc_node will fallback to other nodes
4112                  *      since memory is not yet available from the node that
4113                  *      is brought up.
4114                  */
4115                 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
4116                 if (!n) {
4117                         ret = -ENOMEM;
4118                         goto out;
4119                 }
4120                 init_kmem_cache_node(n);
4121                 s->node[nid] = n;
4122         }
4123 out:
4124         mutex_unlock(&slab_mutex);
4125         return ret;
4126 }
4127 
4128 static int slab_memory_callback(struct notifier_block *self,
4129                                 unsigned long action, void *arg)
4130 {
4131         int ret = 0;
4132 
4133         switch (action) {
4134         case MEM_GOING_ONLINE:
4135                 ret = slab_mem_going_online_callback(arg);
4136                 break;
4137         case MEM_GOING_OFFLINE:
4138                 ret = slab_mem_going_offline_callback(arg);
4139                 break;
4140         case MEM_OFFLINE:
4141         case MEM_CANCEL_ONLINE:
4142                 slab_mem_offline_callback(arg);
4143                 break;
4144         case MEM_ONLINE:
4145         case MEM_CANCEL_OFFLINE:
4146                 break;
4147         }
4148         if (ret)
4149                 ret = notifier_from_errno(ret);
4150         else
4151                 ret = NOTIFY_OK;
4152         return ret;
4153 }
4154 
4155 static struct notifier_block slab_memory_callback_nb = {
4156         .notifier_call = slab_memory_callback,
4157         .priority = SLAB_CALLBACK_PRI,
4158 };
4159 
4160 /********************************************************************
4161  *                      Basic setup of slabs
4162  *******************************************************************/
4163 
4164 /*
4165  * Used for early kmem_cache structures that were allocated using
4166  * the page allocator. Allocate them properly then fix up the pointers
4167  * that may be pointing to the wrong kmem_cache structure.
4168  */
4169 
4170 static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
4171 {
4172         int node;
4173         struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
4174         struct kmem_cache_node *n;
4175 
4176         memcpy(s, static_cache, kmem_cache->object_size);
4177 
4178         /*
4179          * This runs very early, and only the boot processor is supposed to be
4180          * up.  Even if it weren't true, IRQs are not up so we couldn't fire
4181          * IPIs around.
4182          */
4183         __flush_cpu_slab(s, smp_processor_id());
4184         for_each_kmem_cache_node(s, node, n) {
4185                 struct page *p;
4186 
4187                 list_for_each_entry(p, &n->partial, lru)
4188                         p->slab_cache = s;
4189 
4190 #ifdef CONFIG_SLUB_DEBUG
4191                 list_for_each_entry(p, &n->full, lru)
4192                         p->slab_cache = s;
4193 #endif
4194         }
4195         slab_init_memcg_params(s);
4196         list_add(&s->list, &slab_caches);
4197         memcg_link_cache(s);
4198         return s;
4199 }
4200 
4201 void __init kmem_cache_init(void)
4202 {
4203         static __initdata struct kmem_cache boot_kmem_cache,
4204                 boot_kmem_cache_node;
4205 
4206         if (debug_guardpage_minorder())
4207                 slub_max_order = 0;
4208 
4209         kmem_cache_node = &boot_kmem_cache_node;
4210         kmem_cache = &boot_kmem_cache;
4211 
4212         create_boot_cache(kmem_cache_node, "kmem_cache_node",
4213                 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
4214 
4215         register_hotmemory_notifier(&slab_memory_callback_nb);
4216 
4217         /* Able to allocate the per node structures */
4218         slab_state = PARTIAL;
4219 
4220         create_boot_cache(kmem_cache, "kmem_cache",
4221                         offsetof(struct kmem_cache, node) +
4222                                 nr_node_ids * sizeof(struct kmem_cache_node *),
4223                        SLAB_HWCACHE_ALIGN, 0, 0);
4224 
4225         kmem_cache = bootstrap(&boot_kmem_cache);
4226 
4227         /*
4228          * Allocate kmem_cache_node properly from the kmem_cache slab.
4229          * kmem_cache_node is separately allocated so no need to
4230          * update any list pointers.
4231          */
4232         kmem_cache_node = bootstrap(&boot_kmem_cache_node);
4233 
4234         /* Now we can use the kmem_cache to allocate kmalloc slabs */
4235         setup_kmalloc_cache_index_table();
4236         create_kmalloc_caches(0);
4237 
4238         /* Setup random freelists for each cache */
4239         init_freelist_randomization();
4240 
4241         cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4242                                   slub_cpu_dead);
4243 
4244         pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%u, Nodes=%d\n",
4245                 cache_line_size(),
4246                 slub_min_order, slub_max_order, slub_min_objects,
4247                 nr_cpu_ids, nr_node_ids);
4248 }
4249 
4250 void __init kmem_cache_init_late(void)
4251 {
4252 }
4253 
4254 struct kmem_cache *
4255 __kmem_cache_alias(const char *name, size_t size, size_t align,
4256                    slab_flags_t flags, void (*ctor)(void *))
4257 {
4258         struct kmem_cache *s, *c;
4259 
4260         s = find_mergeable(size, align, flags, name, ctor);
4261         if (s) {
4262                 s->refcount++;
4263 
4264                 /*
4265                  * Adjust the object sizes so that we clear
4266                  * the complete object on kzalloc.
4267                  */
4268                 s->object_size = max(s->object_size, (int)size);
4269                 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
4270 
4271                 for_each_memcg_cache(c, s) {
4272                         c->object_size = s->object_size;
4273                         c->inuse = max_t(int, c->inuse,
4274                                          ALIGN(size, sizeof(void *)));
4275                 }
4276 
4277                 if (sysfs_slab_alias(s, name)) {
4278                         s->refcount--;
4279                         s = NULL;
4280                 }
4281         }
4282 
4283         return s;
4284 }
4285 
4286 int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
4287 {
4288         int err;
4289 
4290         err = kmem_cache_open(s, flags);
4291         if (err)
4292                 return err;
4293 
4294         /* Mutex is not taken during early boot */
4295         if (slab_state <= UP)
4296                 return 0;
4297 
4298         memcg_propagate_slab_attrs(s);
4299         err = sysfs_slab_add(s);
4300         if (err)
4301                 __kmem_cache_release(s);
4302 
4303         return err;
4304 }
4305 
4306 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
4307 {
4308         struct kmem_cache *s;
4309         void *ret;
4310 
4311         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
4312                 return kmalloc_large(size, gfpflags);
4313 
4314         s = kmalloc_slab(size, gfpflags);
4315 
4316         if (unlikely(ZERO_OR_NULL_PTR(s)))
4317                 return s;
4318 
4319         ret = slab_alloc(s, gfpflags, caller);
4320 
4321         /* Honor the call site pointer we received. */
4322         trace_kmalloc(caller, ret, size, s->size, gfpflags);
4323 
4324         return ret;
4325 }
4326 
4327 #ifdef CONFIG_NUMA
4328 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
4329                                         int node, unsigned long caller)
4330 {
4331         struct kmem_cache *s;
4332         void *ret;
4333 
4334         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
4335                 ret = kmalloc_large_node(size, gfpflags, node);
4336 
4337                 trace_kmalloc_node(caller, ret,
4338                                    size, PAGE_SIZE << get_order(size),
4339                                    gfpflags, node);
4340 
4341                 return ret;
4342         }
4343 
4344         s = kmalloc_slab(size, gfpflags);
4345 
4346         if (unlikely(ZERO_OR_NULL_PTR(s)))
4347                 return s;
4348 
4349         ret = slab_alloc_node(s, gfpflags, node, caller);
4350 
4351         /* Honor the call site pointer we received. */
4352         trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
4353 
4354         return ret;
4355 }
4356 #endif
4357 
4358 #ifdef CONFIG_SYSFS
4359 static int count_inuse(struct page *page)
4360 {
4361         return page->inuse;
4362 }
4363 
4364 static int count_total(struct page *page)
4365 {
4366         return page->objects;
4367 }
4368 #endif
4369 
4370 #ifdef CONFIG_SLUB_DEBUG
4371 static int validate_slab(struct kmem_cache *s, struct page *page,
4372                                                 unsigned long *map)
4373 {
4374         void *p;
4375         void *addr = page_address(page);
4376 
4377         if (!check_slab(s, page) ||
4378                         !on_freelist(s, page, NULL))
4379                 return 0;
4380 
4381         /* Now we know that a valid freelist exists */
4382         bitmap_zero(map, page->objects);
4383 
4384         get_map(s, page, map);
4385         for_each_object(p, s, addr, page->objects) {
4386                 if (test_bit(slab_index(p, s, addr), map))
4387                         if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4388                                 return 0;
4389         }
4390 
4391         for_each_object(p, s, addr, page->objects)
4392                 if (!test_bit(slab_index(p, s, addr), map))
4393                         if (!check_object(s, page, p, SLUB_RED_ACTIVE))
4394                                 return 0;
4395         return 1;
4396 }
4397 
4398 static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4399                                                 unsigned long *map)
4400 {
4401         slab_lock(page);
4402         validate_slab(s, page, map);
4403         slab_unlock(page);
4404 }
4405 
4406 static int validate_slab_node(struct kmem_cache *s,
4407                 struct kmem_cache_node *n, unsigned long *map)
4408 {
4409         unsigned long count = 0;
4410         struct page *page;
4411         unsigned long flags;
4412 
4413         spin_lock_irqsave(&n->list_lock, flags);
4414 
4415         list_for_each_entry(page, &n->partial, lru) {
4416                 validate_slab_slab(s, page, map);
4417                 count++;
4418         }
4419         if (count != n->nr_partial)
4420                 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4421                        s->name, count, n->nr_partial);
4422 
4423         if (!(s->flags & SLAB_STORE_USER))
4424                 goto out;
4425 
4426         list_for_each_entry(page, &n->full, lru) {
4427                 validate_slab_slab(s, page, map);
4428                 count++;
4429         }
4430         if (count != atomic_long_read(&n->nr_slabs))
4431                 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4432                        s->name, count, atomic_long_read(&n->nr_slabs));
4433 
4434 out:
4435         spin_unlock_irqrestore(&n->list_lock, flags);
4436         return count;
4437 }
4438 
4439 static long validate_slab_cache(struct kmem_cache *s)
4440 {
4441         int node;
4442         unsigned long count = 0;
4443         unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4444                                 sizeof(unsigned long), GFP_KERNEL);
4445         struct kmem_cache_node *n;
4446 
4447         if (!map)
4448                 return -ENOMEM;
4449 
4450         flush_all(s);
4451         for_each_kmem_cache_node(s, node, n)
4452                 count += validate_slab_node(s, n, map);
4453         kfree(map);
4454         return count;
4455 }
4456 /*
4457  * Generate lists of code addresses where slabcache objects are allocated
4458  * and freed.
4459  */
4460 
4461 struct location {
4462         unsigned long count;
4463         unsigned long addr;
4464         long long sum_time;
4465         long min_time;
4466         long max_time;
4467         long min_pid;
4468         long max_pid;
4469         DECLARE_BITMAP(cpus, NR_CPUS);
4470         nodemask_t nodes;
4471 };
4472 
4473 struct loc_track {
4474         unsigned long max;
4475         unsigned long count;
4476         struct location *loc;
4477 };
4478 
4479 static void free_loc_track(struct loc_track *t)
4480 {
4481         if (t->max)
4482                 free_pages((unsigned long)t->loc,
4483                         get_order(sizeof(struct location) * t->max));
4484 }
4485 
4486 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
4487 {
4488         struct location *l;
4489         int order;
4490 
4491         order = get_order(sizeof(struct location) * max);
4492 
4493         l = (void *)__get_free_pages(flags, order);
4494         if (!l)
4495                 return 0;
4496 
4497         if (t->count) {
4498                 memcpy(l, t->loc, sizeof(struct location) * t->count);
4499                 free_loc_track(t);
4500         }
4501         t->max = max;
4502         t->loc = l;
4503         return 1;
4504 }
4505 
4506 static int add_location(struct loc_track *t, struct kmem_cache *s,
4507                                 const struct track *track)
4508 {
4509         long start, end, pos;
4510         struct location *l;
4511         unsigned long caddr;
4512         unsigned long age = jiffies - track->when;
4513 
4514         start = -1;
4515         end = t->count;
4516 
4517         for ( ; ; ) {
4518                 pos = start + (end - start + 1) / 2;
4519 
4520                 /*
4521                  * There is nothing at "end". If we end up there
4522                  * we need to add something to before end.
4523                  */
4524                 if (pos == end)
4525                         break;
4526 
4527                 caddr = t->loc[pos].addr;
4528                 if (track->addr == caddr) {
4529 
4530                         l = &t->loc[pos];
4531                         l->count++;
4532                         if (track->when) {
4533                                 l->sum_time += age;
4534                                 if (age < l->min_time)
4535                                         l->min_time = age;
4536                                 if (age > l->max_time)
4537                                         l->max_time = age;
4538 
4539                                 if (track->pid < l->min_pid)
4540                                         l->min_pid = track->pid;
4541                                 if (track->pid > l->max_pid)
4542                                         l->max_pid = track->pid;
4543 
4544                                 cpumask_set_cpu(track->cpu,
4545                                                 to_cpumask(l->cpus));
4546                         }
4547                         node_set(page_to_nid(virt_to_page(track)), l->nodes);
4548                         return 1;
4549                 }
4550 
4551                 if (track->addr < caddr)
4552                         end = pos;
4553                 else
4554                         start = pos;
4555         }
4556 
4557         /*
4558          * Not found. Insert new tracking element.
4559          */
4560         if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
4561                 return 0;
4562 
4563         l = t->loc + pos;
4564         if (pos < t->count)
4565                 memmove(l + 1, l,
4566                         (t->count - pos) * sizeof(struct location));
4567         t->count++;
4568         l->count = 1;
4569         l->addr = track->addr;
4570         l->sum_time = age;
4571         l->min_time = age;
4572         l->max_time = age;
4573         l->min_pid = track->pid;
4574         l->max_pid = track->pid;
4575         cpumask_clear(to_cpumask(l->cpus));
4576         cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
4577         nodes_clear(l->nodes);
4578         node_set(page_to_nid(virt_to_page(track)), l->nodes);
4579         return 1;
4580 }
4581 
4582 static void process_slab(struct loc_track *t, struct kmem_cache *s,
4583                 struct page *page, enum track_item alloc,
4584                 unsigned long *map)
4585 {
4586         void *addr = page_address(page);
4587         void *p;
4588 
4589         bitmap_zero(map, page->objects);
4590         get_map(s, page, map);
4591 
4592         for_each_object(p, s, addr, page->objects)
4593                 if (!test_bit(slab_index(p, s, addr), map))
4594                         add_location(t, s, get_track(s, p, alloc));
4595 }
4596 
4597 static int list_locations(struct kmem_cache *s, char *buf,
4598                                         enum track_item alloc)
4599 {
4600         int len = 0;
4601         unsigned long i;
4602         struct loc_track t = { 0, 0, NULL };
4603         int node;
4604         unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4605                                      sizeof(unsigned long), GFP_KERNEL);
4606         struct kmem_cache_node *n;
4607 
4608         if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4609                                      GFP_KERNEL)) {
4610                 kfree(map);
4611                 return sprintf(buf, "Out of memory\n");
4612         }
4613         /* Push back cpu slabs */
4614         flush_all(s);
4615 
4616         for_each_kmem_cache_node(s, node, n) {
4617                 unsigned long flags;
4618                 struct page *page;
4619 
4620                 if (!atomic_long_read(&n->nr_slabs))
4621                         continue;
4622 
4623                 spin_lock_irqsave(&n->list_lock, flags);
4624                 list_for_each_entry(page, &n->partial, lru)
4625                         process_slab(&t, s, page, alloc, map);
4626                 list_for_each_entry(page, &n->full, lru)
4627                         process_slab(&t, s, page, alloc, map);
4628                 spin_unlock_irqrestore(&n->list_lock, flags);
4629         }
4630 
4631         for (i = 0; i < t.count; i++) {
4632                 struct location *l = &t.loc[i];
4633 
4634                 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
4635                         break;
4636                 len += sprintf(buf + len, "%7ld ", l->count);
4637 
4638                 if (l->addr)
4639                         len += sprintf(buf + len, "%pS", (void *)l->addr);
4640                 else
4641                         len += sprintf(buf + len, "<not-available>");
4642 
4643                 if (l->sum_time != l->min_time) {
4644                         len += sprintf(buf + len, " age=%ld/%ld/%ld",
4645                                 l->min_time,
4646                                 (long)div_u64(l->sum_time, l->count),
4647                                 l->max_time);
4648                 } else
4649                         len += sprintf(buf + len, " age=%ld",
4650                                 l->min_time);
4651 
4652                 if (l->min_pid != l->max_pid)
4653                         len += sprintf(buf + len, " pid=%ld-%ld",
4654                                 l->min_pid, l->max_pid);
4655                 else
4656                         len += sprintf(buf + len, " pid=%ld",
4657                                 l->min_pid);
4658 
4659                 if (num_online_cpus() > 1 &&
4660                                 !cpumask_empty(to_cpumask(l->cpus)) &&
4661                                 len < PAGE_SIZE - 60)
4662                         len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4663                                          " cpus=%*pbl",
4664                                          cpumask_pr_args(to_cpumask(l->cpus)));
4665 
4666                 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
4667                                 len < PAGE_SIZE - 60)
4668                         len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4669                                          " nodes=%*pbl",
4670                                          nodemask_pr_args(&l->nodes));
4671 
4672                 len += sprintf(buf + len, "\n");
4673         }
4674 
4675         free_loc_track(&t);
4676         kfree(map);
4677         if (!t.count)
4678                 len += sprintf(buf, "No data\n");
4679         return len;
4680 }
4681 #endif
4682 
4683 #ifdef SLUB_RESILIENCY_TEST
4684 static void __init resiliency_test(void)
4685 {
4686         u8 *p;
4687 
4688         BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
4689 
4690         pr_err("SLUB resiliency testing\n");
4691         pr_err("-----------------------\n");
4692         pr_err("A. Corruption after allocation\n");
4693 
4694         p = kzalloc(16, GFP_KERNEL);
4695         p[16] = 0x12;
4696         pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4697                p + 16);
4698 
4699         validate_slab_cache(kmalloc_caches[4]);
4700 
4701         /* Hmmm... The next two are dangerous */
4702         p = kzalloc(32, GFP_KERNEL);
4703         p[32 + sizeof(void *)] = 0x34;
4704         pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4705                p);
4706         pr_err("If allocated object is overwritten then not detectable\n\n");
4707 
4708         validate_slab_cache(kmalloc_caches[5]);
4709         p = kzalloc(64, GFP_KERNEL);
4710         p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4711         *p = 0x56;
4712         pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4713                p);
4714         pr_err("If allocated object is overwritten then not detectable\n\n");
4715         validate_slab_cache(kmalloc_caches[6]);
4716 
4717         pr_err("\nB. Corruption after free\n");
4718         p = kzalloc(128, GFP_KERNEL);
4719         kfree(p);
4720         *p = 0x78;
4721         pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
4722         validate_slab_cache(kmalloc_caches[7]);
4723 
4724         p = kzalloc(256, GFP_KERNEL);
4725         kfree(p);
4726         p[50] = 0x9a;
4727         pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
4728         validate_slab_cache(kmalloc_caches[8]);
4729 
4730         p = kzalloc(512, GFP_KERNEL);
4731         kfree(p);
4732         p[512] = 0xab;
4733         pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
4734         validate_slab_cache(kmalloc_caches[9]);
4735 }
4736 #else
4737 #ifdef CONFIG_SYSFS
4738 static void resiliency_test(void) {};
4739 #endif
4740 #endif
4741 
4742 #ifdef CONFIG_SYSFS
4743 enum slab_stat_type {
4744         SL_ALL,                 /* All slabs */
4745         SL_PARTIAL,             /* Only partially allocated slabs */
4746         SL_CPU,                 /* Only slabs used for cpu caches */
4747         SL_OBJECTS,             /* Determine allocated objects not slabs */
4748         SL_TOTAL                /* Determine object capacity not slabs */
4749 };
4750 
4751 #define SO_ALL          (1 << SL_ALL)
4752 #define SO_PARTIAL      (1 << SL_PARTIAL)
4753 #define SO_CPU          (1 << SL_CPU)
4754 #define SO_OBJECTS      (1 << SL_OBJECTS)
4755 #define SO_TOTAL        (1 << SL_TOTAL)
4756 
4757 #ifdef CONFIG_MEMCG
4758 static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON);
4759 
4760 static int __init setup_slub_memcg_sysfs(char *str)
4761 {
4762         int v;
4763 
4764         if (get_option(&str, &v) > 0)
4765                 memcg_sysfs_enabled = v;
4766 
4767         return 1;
4768 }
4769 
4770 __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs);
4771 #endif
4772 
4773 static ssize_t show_slab_objects(struct kmem_cache *s,
4774                             char *buf, unsigned long flags)
4775 {
4776         unsigned long total = 0;
4777         int node;
4778         int x;
4779         unsigned long *nodes;
4780 
4781         nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
4782         if (!nodes)
4783                 return -ENOMEM;
4784 
4785         if (flags & SO_CPU) {
4786                 int cpu;
4787 
4788                 for_each_possible_cpu(cpu) {
4789                         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4790                                                                cpu);
4791                         int node;
4792                         struct page *page;
4793 
4794                         page = READ_ONCE(c->page);
4795                         if (!page)
4796                                 continue;
4797 
4798                         node = page_to_nid(page);
4799                         if (flags & SO_TOTAL)
4800                                 x = page->objects;
4801                         else if (flags & SO_OBJECTS)
4802                                 x = page->inuse;
4803                         else
4804                                 x = 1;
4805 
4806                         total += x;
4807                         nodes[node] += x;
4808 
4809                         page = slub_percpu_partial_read_once(c);
4810                         if (page) {
4811                                 node = page_to_nid(page);
4812                                 if (flags & SO_TOTAL)
4813                                         WARN_ON_ONCE(1);
4814                                 else if (flags & SO_OBJECTS)
4815                                         WARN_ON_ONCE(1);
4816                                 else
4817                                         x = page->pages;
4818                                 total += x;
4819                                 nodes[node] += x;
4820                         }
4821                 }
4822         }
4823 
4824         get_online_mems();
4825 #ifdef CONFIG_SLUB_DEBUG
4826         if (flags & SO_ALL) {
4827                 struct kmem_cache_node *n;
4828 
4829                 for_each_kmem_cache_node(s, node, n) {
4830 
4831                         if (flags & SO_TOTAL)
4832                                 x = atomic_long_read(&n->total_objects);
4833                         else if (flags & SO_OBJECTS)
4834                                 x = atomic_long_read(&n->total_objects) -
4835                                         count_partial(n, count_free);
4836                         else
4837                                 x = atomic_long_read(&n->nr_slabs);
4838                         total += x;
4839                         nodes[node] += x;
4840                 }
4841 
4842         } else
4843 #endif
4844         if (flags & SO_PARTIAL) {
4845                 struct kmem_cache_node *n;
4846 
4847                 for_each_kmem_cache_node(s, node, n) {
4848                         if (flags & SO_TOTAL)
4849                                 x = count_partial(n, count_total);
4850                         else if (flags & SO_OBJECTS)
4851                                 x = count_partial(n, count_inuse);
4852                         else
4853                                 x = n->nr_partial;
4854                         total += x;
4855                         nodes[node] += x;
4856                 }
4857         }
4858         x = sprintf(buf, "%lu", total);
4859 #ifdef CONFIG_NUMA
4860         for (node = 0; node < nr_node_ids; node++)
4861                 if (nodes[node])
4862                         x += sprintf(buf + x, " N%d=%lu",
4863                                         node, nodes[node]);
4864 #endif
4865         put_online_mems();
4866         kfree(nodes);
4867         return x + sprintf(buf + x, "\n");
4868 }
4869 
4870 #ifdef CONFIG_SLUB_DEBUG
4871 static int any_slab_objects(struct kmem_cache *s)
4872 {
4873         int node;
4874         struct kmem_cache_node *n;
4875 
4876         for_each_kmem_cache_node(s, node, n)
4877                 if (atomic_long_read(&n->total_objects))
4878                         return 1;
4879 
4880         return 0;
4881 }
4882 #endif
4883 
4884 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4885 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4886 
4887 struct slab_attribute {
4888         struct attribute attr;
4889         ssize_t (*show)(struct kmem_cache *s, char *buf);
4890         ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4891 };
4892 
4893 #define SLAB_ATTR_RO(_name) \
4894         static struct slab_attribute _name##_attr = \
4895         __ATTR(_name, 0400, _name##_show, NULL)
4896 
4897 #define SLAB_ATTR(_name) \
4898         static struct slab_attribute _name##_attr =  \
4899         __ATTR(_name, 0600, _name##_show, _name##_store)
4900 
4901 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4902 {
4903         return sprintf(buf, "%d\n", s->size);
4904 }
4905 SLAB_ATTR_RO(slab_size);
4906 
4907 static ssize_t align_show(struct kmem_cache *s, char *buf)
4908 {
4909         return sprintf(buf, "%d\n", s->align);
4910 }
4911 SLAB_ATTR_RO(align);
4912 
4913 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4914 {
4915         return sprintf(buf, "%d\n", s->object_size);
4916 }
4917 SLAB_ATTR_RO(object_size);
4918 
4919 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4920 {
4921         return sprintf(buf, "%d\n", oo_objects(s->oo));
4922 }
4923 SLAB_ATTR_RO(objs_per_slab);
4924 
4925 static ssize_t order_store(struct kmem_cache *s,
4926                                 const char *buf, size_t length)
4927 {
4928         unsigned long order;
4929         int err;
4930 
4931         err = kstrtoul(buf, 10, &order);
4932         if (err)
4933                 return err;
4934 
4935         if (order > slub_max_order || order < slub_min_order)
4936                 return -EINVAL;
4937 
4938         calculate_sizes(s, order);
4939         return length;
4940 }
4941 
4942 static ssize_t order_show(struct kmem_cache *s, char *buf)
4943 {
4944         return sprintf(buf, "%d\n", oo_order(s->oo));
4945 }
4946 SLAB_ATTR(order);
4947 
4948 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4949 {
4950         return sprintf(buf, "%lu\n", s->min_partial);
4951 }
4952 
4953 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4954                                  size_t length)
4955 {
4956         unsigned long min;
4957         int err;
4958 
4959         err = kstrtoul(buf, 10, &min);
4960         if (err)
4961                 return err;
4962 
4963         set_min_partial(s, min);
4964         return length;
4965 }
4966 SLAB_ATTR(min_partial);
4967 
4968 static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4969 {
4970         return sprintf(buf, "%u\n", slub_cpu_partial(s));
4971 }
4972 
4973 static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4974                                  size_t length)
4975 {
4976         unsigned long objects;
4977         int err;
4978 
4979         err = kstrtoul(buf, 10, &objects);
4980         if (err)
4981                 return err;
4982         if (objects && !kmem_cache_has_cpu_partial(s))
4983                 return -EINVAL;
4984 
4985         slub_set_cpu_partial(s, objects);
4986         flush_all(s);
4987         return length;
4988 }
4989 SLAB_ATTR(cpu_partial);
4990 
4991 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4992 {
4993         if (!s->ctor)
4994                 return 0;
4995         return sprintf(buf, "%pS\n", s->ctor);
4996 }
4997 SLAB_ATTR_RO(ctor);
4998 
4999 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5000 {
5001         return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
5002 }
5003 SLAB_ATTR_RO(aliases);
5004 
5005 static ssize_t partial_show(struct kmem_cache *s, char *buf)
5006 {
5007         return show_slab_objects(s, buf, SO_PARTIAL);
5008 }
5009 SLAB_ATTR_RO(partial);
5010 
5011 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5012 {
5013         return show_slab_objects(s, buf, SO_CPU);
5014 }
5015 SLAB_ATTR_RO(cpu_slabs);
5016 
5017 static ssize_t objects_show(struct kmem_cache *s, char *buf)
5018 {
5019         return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
5020 }
5021 SLAB_ATTR_RO(objects);
5022 
5023 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5024 {
5025         return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5026 }
5027 SLAB_ATTR_RO(objects_partial);
5028 
5029 static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5030 {
5031         int objects = 0;
5032         int pages = 0;
5033         int cpu;
5034         int len;
5035 
5036         for_each_online_cpu(cpu) {
5037                 struct page *page;
5038 
5039                 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5040 
5041                 if (page) {
5042                         pages += page->pages;
5043                         objects += page->pobjects;
5044                 }
5045         }
5046 
5047         len = sprintf(buf, "%d(%d)", objects, pages);
5048 
5049 #ifdef CONFIG_SMP
5050         for_each_online_cpu(cpu) {
5051                 struct page *page;
5052 
5053                 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5054 
5055                 if (page && len < PAGE_SIZE - 20)
5056                         len += sprintf(buf + len, " C%d=%d(%d)", cpu,
5057                                 page->pobjects, page->pages);
5058         }
5059 #endif
5060         return len + sprintf(buf + len, "\n");
5061 }
5062 SLAB_ATTR_RO(slabs_cpu_partial);
5063 
5064 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5065 {
5066         return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5067 }
5068 
5069 static ssize_t reclaim_account_store(struct kmem_cache *s,
5070                                 const char *buf, size_t length)
5071 {
5072         s->flags &= ~SLAB_RECLAIM_ACCOUNT;
5073         if (buf[0] == '1')
5074                 s->flags |= SLAB_RECLAIM_ACCOUNT;
5075         return length;
5076 }
5077 SLAB_ATTR(reclaim_account);
5078 
5079 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5080 {
5081         return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
5082 }
5083 SLAB_ATTR_RO(hwcache_align);
5084 
5085 #ifdef CONFIG_ZONE_DMA
5086 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5087 {
5088         return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5089 }
5090 SLAB_ATTR_RO(cache_dma);
5091 #endif
5092 
5093 static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5094 {
5095         return sprintf(buf, "%zu\n", s->usersize);
5096 }
5097 SLAB_ATTR_RO(usersize);
5098 
5099 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5100 {
5101         return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
5102 }
5103 SLAB_ATTR_RO(destroy_by_rcu);
5104 
5105 static ssize_t reserved_show(struct kmem_cache *s, char *buf)
5106 {
5107         return sprintf(buf, "%d\n", s->reserved);
5108 }
5109 SLAB_ATTR_RO(reserved);
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, "%d\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 long ratio;
5309         int err;
5310 
5311         err = kstrtoul(buf, 10, &ratio);
5312         if (err)
5313                 return err;
5314 
5315         if (ratio <= 100)
5316                 s->remote_node_defrag_ratio = ratio * 10;
5317 
5318         return length;
5319 }
5320 SLAB_ATTR(remote_node_defrag_ratio);
5321 #endif
5322 
5323 #ifdef CONFIG_SLUB_STATS
5324 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5325 {
5326         unsigned long sum  = 0;
5327         int cpu;
5328         int len;
5329         int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
5330 
5331         if (!data)
5332                 return -ENOMEM;
5333 
5334         for_each_online_cpu(cpu) {
5335                 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
5336 
5337                 data[cpu] = x;
5338                 sum += x;
5339         }
5340 
5341         len = sprintf(buf, "%lu", sum);
5342 
5343 #ifdef CONFIG_SMP
5344         for_each_online_cpu(cpu) {
5345                 if (data[cpu] && len < PAGE_SIZE - 20)
5346                         len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
5347         }
5348 #endif
5349         kfree(data);
5350         return len + sprintf(buf + len, "\n");
5351 }
5352 
5353 static void clear_stat(struct kmem_cache *s, enum stat_item si)
5354 {
5355         int cpu;
5356 
5357         for_each_online_cpu(cpu)
5358                 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
5359 }
5360 
5361 #define STAT_ATTR(si, text)                                     \
5362 static ssize_t text##_show(struct kmem_cache *s, char *buf)     \
5363 {                                                               \
5364         return show_stat(s, buf, si);                           \
5365 }                                                               \
5366 static ssize_t text##_store(struct kmem_cache *s,               \
5367                                 const char *buf, size_t length) \
5368 {                                                               \
5369         if (buf[0] != '')                                      \
5370                 return -EINVAL;                                 \
5371         clear_stat(s, si);                                      \
5372         return length;                                          \
5373 }                                                               \
5374 SLAB_ATTR(text);                                                \
5375 
5376 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5377 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5378 STAT_ATTR(FREE_FASTPATH, free_fastpath);
5379 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5380 STAT_ATTR(FREE_FROZEN, free_frozen);
5381 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5382 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5383 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5384 STAT_ATTR(ALLOC_SLAB, alloc_slab);
5385 STAT_ATTR(ALLOC_REFILL, alloc_refill);
5386 STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
5387 STAT_ATTR(FREE_SLAB, free_slab);
5388 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5389 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5390 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5391 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5392 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5393 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
5394 STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
5395 STAT_ATTR(ORDER_FALLBACK, order_fallback);
5396 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5397 STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
5398 STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5399 STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
5400 STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5401 STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
5402 #endif
5403 
5404 static struct attribute *slab_attrs[] = {
5405         &slab_size_attr.attr,
5406         &object_size_attr.attr,
5407         &objs_per_slab_attr.attr,
5408         &order_attr.attr,
5409         &min_partial_attr.attr,
5410         &cpu_partial_attr.attr,
5411         &objects_attr.attr,
5412         &objects_partial_attr.attr,
5413         &partial_attr.attr,
5414         &cpu_slabs_attr.attr,
5415         &ctor_attr.attr,
5416         &aliases_attr.attr,
5417         &align_attr.attr,
5418         &hwcache_align_attr.attr,
5419         &reclaim_account_attr.attr,
5420         &destroy_by_rcu_attr.attr,
5421         &shrink_attr.attr,
5422         &reserved_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, "%07d", 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         kobject_del(&s->kobj);
5702 out:
5703         kobject_put(&s->kobj);
5704 }
5705 
5706 static int sysfs_slab_add(struct kmem_cache *s)
5707 {
5708         int err;
5709         const char *name;
5710         struct kset *kset = cache_kset(s);
5711         int unmergeable = slab_unmergeable(s);
5712 
5713         INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn);
5714 
5715         if (!kset) {
5716                 kobject_init(&s->kobj, &slab_ktype);
5717                 return 0;
5718         }
5719 
5720         if (!unmergeable && disable_higher_order_debug &&
5721                         (slub_debug & DEBUG_METADATA_FLAGS))
5722                 unmergeable = 1;
5723 
5724         if (unmergeable) {
5725                 /*
5726                  * Slabcache can never be merged so we can use the name proper.
5727                  * This is typically the case for debug situations. In that
5728                  * case we can catch duplicate names easily.
5729                  */
5730                 sysfs_remove_link(&slab_kset->kobj, s->name);
5731                 name = s->name;
5732         } else {
5733                 /*
5734                  * Create a unique name for the slab as a target
5735                  * for the symlinks.
5736                  */
5737                 name = create_unique_id(s);
5738         }
5739 
5740         s->kobj.kset = kset;
5741         err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
5742         if (err)
5743                 goto out;
5744 
5745         err = sysfs_create_group(&s->kobj, &slab_attr_group);
5746         if (err)
5747                 goto out_del_kobj;
5748 
5749 #ifdef CONFIG_MEMCG
5750         if (is_root_cache(s) && memcg_sysfs_enabled) {
5751                 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5752                 if (!s->memcg_kset) {
5753                         err = -ENOMEM;
5754                         goto out_del_kobj;
5755                 }
5756         }
5757 #endif
5758 
5759         kobject_uevent(&s->kobj, KOBJ_ADD);
5760         if (!unmergeable) {
5761                 /* Setup first alias */
5762                 sysfs_slab_alias(s, s->name);
5763         }
5764 out:
5765         if (!unmergeable)
5766                 kfree(name);
5767         return err;
5768 out_del_kobj:
5769         kobject_del(&s->kobj);
5770         goto out;
5771 }
5772 
5773 static void sysfs_slab_remove(struct kmem_cache *s)
5774 {
5775         if (slab_state < FULL)
5776                 /*
5777                  * Sysfs has not been setup yet so no need to remove the
5778                  * cache from sysfs.
5779                  */
5780                 return;
5781 
5782         kobject_get(&s->kobj);
5783         schedule_work(&s->kobj_remove_work);
5784 }
5785 
5786 void sysfs_slab_release(struct kmem_cache *s)
5787 {
5788         if (slab_state >= FULL)
5789                 kobject_put(&s->kobj);
5790 }
5791 
5792 /*
5793  * Need to buffer aliases during bootup until sysfs becomes
5794  * available lest we lose that information.
5795  */
5796 struct saved_alias {
5797         struct kmem_cache *s;
5798         const char *name;
5799         struct saved_alias *next;
5800 };
5801 
5802 static struct saved_alias *alias_list;
5803 
5804 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5805 {
5806         struct saved_alias *al;
5807 
5808         if (slab_state == FULL) {
5809                 /*
5810                  * If we have a leftover link then remove it.
5811                  */
5812                 sysfs_remove_link(&slab_kset->kobj, name);
5813                 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
5814         }
5815 
5816         al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5817         if (!al)
5818                 return -ENOMEM;
5819 
5820         al->s = s;
5821         al->name = name;
5822         al->next = alias_list;
5823         alias_list = al;
5824         return 0;
5825 }
5826 
5827 static int __init slab_sysfs_init(void)
5828 {
5829         struct kmem_cache *s;
5830         int err;
5831 
5832         mutex_lock(&slab_mutex);
5833 
5834         slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
5835         if (!slab_kset) {
5836                 mutex_unlock(&slab_mutex);
5837                 pr_err("Cannot register slab subsystem.\n");
5838                 return -ENOSYS;
5839         }
5840 
5841         slab_state = FULL;
5842 
5843         list_for_each_entry(s, &slab_caches, list) {
5844                 err = sysfs_slab_add(s);
5845                 if (err)
5846                         pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5847                                s->name);
5848         }
5849 
5850         while (alias_list) {
5851                 struct saved_alias *al = alias_list;
5852 
5853                 alias_list = alias_list->next;
5854                 err = sysfs_slab_alias(al->s, al->name);
5855                 if (err)
5856                         pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5857                                al->name);
5858                 kfree(al);
5859         }
5860 
5861         mutex_unlock(&slab_mutex);
5862         resiliency_test();
5863         return 0;
5864 }
5865 
5866 __initcall(slab_sysfs_init);
5867 #endif /* CONFIG_SYSFS */
5868 
5869 /*
5870  * The /proc/slabinfo ABI
5871  */
5872 #ifdef CONFIG_SLUB_DEBUG
5873 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
5874 {
5875         unsigned long nr_slabs = 0;
5876         unsigned long nr_objs = 0;
5877         unsigned long nr_free = 0;
5878         int node;
5879         struct kmem_cache_node *n;
5880 
5881         for_each_kmem_cache_node(s, node, n) {
5882                 nr_slabs += node_nr_slabs(n);
5883                 nr_objs += node_nr_objs(n);
5884                 nr_free += count_partial(n, count_free);
5885         }
5886 
5887         sinfo->active_objs = nr_objs - nr_free;
5888         sinfo->num_objs = nr_objs;
5889         sinfo->active_slabs = nr_slabs;
5890         sinfo->num_slabs = nr_slabs;
5891         sinfo->objects_per_slab = oo_objects(s->oo);
5892         sinfo->cache_order = oo_order(s->oo);
5893 }
5894 
5895 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
5896 {
5897 }
5898 
5899 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5900                        size_t count, loff_t *ppos)
5901 {
5902         return -EIO;
5903 }
5904 #endif /* CONFIG_SLUB_DEBUG */
5905 

~ [ 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