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

TOMOYO Linux Cross Reference
Linux/mm/slub.c

Version: ~ [ linux-5.13-rc2 ] ~ [ linux-5.12.4 ] ~ [ linux-5.11.21 ] ~ [ linux-5.10.37 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.119 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.190 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.232 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.268 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.268 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * SLUB: A slab allocator that limits cache line use instead of queuing
  3  * objects in per cpu and per node lists.
  4  *
  5  * The allocator synchronizes using per slab locks or atomic operatios
  6  * and only uses a centralized lock to manage a pool of partial slabs.
  7  *
  8  * (C) 2007 SGI, Christoph Lameter
  9  * (C) 2011 Linux Foundation, Christoph Lameter
 10  */
 11 
 12 #include <linux/mm.h>
 13 #include <linux/swap.h> /* struct reclaim_state */
 14 #include <linux/module.h>
 15 #include <linux/bit_spinlock.h>
 16 #include <linux/interrupt.h>
 17 #include <linux/bitops.h>
 18 #include <linux/slab.h>
 19 #include "slab.h"
 20 #include <linux/proc_fs.h>
 21 #include <linux/notifier.h>
 22 #include <linux/seq_file.h>
 23 #include <linux/kasan.h>
 24 #include <linux/kmemcheck.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 
 38 #include <trace/events/kmem.h>
 39 
 40 #include "internal.h"
 41 
 42 /*
 43  * Lock order:
 44  *   1. slab_mutex (Global Mutex)
 45  *   2. node->list_lock
 46  *   3. slab_lock(page) (Only on some arches and for debugging)
 47  *
 48  *   slab_mutex
 49  *
 50  *   The role of the slab_mutex is to protect the list of all the slabs
 51  *   and to synchronize major metadata changes to slab cache structures.
 52  *
 53  *   The slab_lock is only used for debugging and on arches that do not
 54  *   have the ability to do a cmpxchg_double. It only protects the second
 55  *   double word in the page struct. Meaning
 56  *      A. page->freelist       -> List of object free in a page
 57  *      B. page->counters       -> Counters of objects
 58  *      C. page->frozen         -> frozen state
 59  *
 60  *   If a slab is frozen then it is exempt from list management. It is not
 61  *   on any list. The processor that froze the slab is the one who can
 62  *   perform list operations on the page. Other processors may put objects
 63  *   onto the freelist but the processor that froze the slab is the only
 64  *   one that can retrieve the objects from the page's freelist.
 65  *
 66  *   The list_lock protects the partial and full list on each node and
 67  *   the partial slab counter. If taken then no new slabs may be added or
 68  *   removed from the lists nor make the number of partial slabs be modified.
 69  *   (Note that the total number of slabs is an atomic value that may be
 70  *   modified without taking the list lock).
 71  *
 72  *   The list_lock is a centralized lock and thus we avoid taking it as
 73  *   much as possible. As long as SLUB does not have to handle partial
 74  *   slabs, operations can continue without any centralized lock. F.e.
 75  *   allocating a long series of objects that fill up slabs does not require
 76  *   the list lock.
 77  *   Interrupts are disabled during allocation and deallocation in order to
 78  *   make the slab allocator safe to use in the context of an irq. In addition
 79  *   interrupts are disabled to ensure that the processor does not change
 80  *   while handling per_cpu slabs, due to kernel preemption.
 81  *
 82  * SLUB assigns one slab for allocation to each processor.
 83  * Allocations only occur from these slabs called cpu slabs.
 84  *
 85  * Slabs with free elements are kept on a partial list and during regular
 86  * operations no list for full slabs is used. If an object in a full slab is
 87  * freed then the slab will show up again on the partial lists.
 88  * We track full slabs for debugging purposes though because otherwise we
 89  * cannot scan all objects.
 90  *
 91  * Slabs are freed when they become empty. Teardown and setup is
 92  * minimal so we rely on the page allocators per cpu caches for
 93  * fast frees and allocs.
 94  *
 95  * Overloading of page flags that are otherwise used for LRU management.
 96  *
 97  * PageActive           The slab is frozen and exempt from list processing.
 98  *                      This means that the slab is dedicated to a purpose
 99  *                      such as satisfying allocations for a specific
100  *                      processor. Objects may be freed in the slab while
101  *                      it is frozen but slab_free will then skip the usual
102  *                      list operations. It is up to the processor holding
103  *                      the slab to integrate the slab into the slab lists
104  *                      when the slab is no longer needed.
105  *
106  *                      One use of this flag is to mark slabs that are
107  *                      used for allocations. Then such a slab becomes a cpu
108  *                      slab. The cpu slab may be equipped with an additional
109  *                      freelist that allows lockless access to
110  *                      free objects in addition to the regular freelist
111  *                      that requires the slab lock.
112  *
113  * PageError            Slab requires special handling due to debug
114  *                      options set. This moves slab handling out of
115  *                      the fast path and disables lockless freelists.
116  */
117 
118 static inline int kmem_cache_debug(struct kmem_cache *s)
119 {
120 #ifdef CONFIG_SLUB_DEBUG
121         return unlikely(s->flags & SLAB_DEBUG_FLAGS);
122 #else
123         return 0;
124 #endif
125 }
126 
127 void *fixup_red_left(struct kmem_cache *s, void *p)
128 {
129         if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE)
130                 p += s->red_left_pad;
131 
132         return p;
133 }
134 
135 static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
136 {
137 #ifdef CONFIG_SLUB_CPU_PARTIAL
138         return !kmem_cache_debug(s);
139 #else
140         return false;
141 #endif
142 }
143 
144 /*
145  * Issues still to be resolved:
146  *
147  * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
148  *
149  * - Variable sizing of the per node arrays
150  */
151 
152 /* Enable to test recovery from slab corruption on boot */
153 #undef SLUB_RESILIENCY_TEST
154 
155 /* Enable to log cmpxchg failures */
156 #undef SLUB_DEBUG_CMPXCHG
157 
158 /*
159  * Mininum number of partial slabs. These will be left on the partial
160  * lists even if they are empty. kmem_cache_shrink may reclaim them.
161  */
162 #define MIN_PARTIAL 5
163 
164 /*
165  * Maximum number of desirable partial slabs.
166  * The existence of more partial slabs makes kmem_cache_shrink
167  * sort the partial list by the number of objects in use.
168  */
169 #define MAX_PARTIAL 10
170 
171 #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
172                                 SLAB_POISON | SLAB_STORE_USER)
173 
174 /*
175  * These debug flags cannot use CMPXCHG because there might be consistency
176  * issues when checking or reading debug information
177  */
178 #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
179                                 SLAB_TRACE)
180 
181 
182 /*
183  * Debugging flags that require metadata to be stored in the slab.  These get
184  * disabled when slub_debug=O is used and a cache's min order increases with
185  * metadata.
186  */
187 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
188 
189 #define OO_SHIFT        16
190 #define OO_MASK         ((1 << OO_SHIFT) - 1)
191 #define MAX_OBJS_PER_PAGE       32767 /* since page.objects is u15 */
192 
193 /* Internal SLUB flags */
194 #define __OBJECT_POISON         0x80000000UL /* Poison object */
195 #define __CMPXCHG_DOUBLE        0x40000000UL /* Use cmpxchg_double */
196 
197 #ifdef CONFIG_SMP
198 static struct notifier_block slab_notifier;
199 #endif
200 
201 /*
202  * Tracking user of a slab.
203  */
204 #define TRACK_ADDRS_COUNT 16
205 struct track {
206         unsigned long addr;     /* Called from address */
207 #ifdef CONFIG_STACKTRACE
208         unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
209 #endif
210         int cpu;                /* Was running on cpu */
211         int pid;                /* Pid context */
212         unsigned long when;     /* When did the operation occur */
213 };
214 
215 enum track_item { TRACK_ALLOC, TRACK_FREE };
216 
217 #ifdef CONFIG_SYSFS
218 static int sysfs_slab_add(struct kmem_cache *);
219 static int sysfs_slab_alias(struct kmem_cache *, const char *);
220 static void memcg_propagate_slab_attrs(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 #endif
227 
228 static inline void stat(const struct kmem_cache *s, enum stat_item si)
229 {
230 #ifdef CONFIG_SLUB_STATS
231         /*
232          * The rmw is racy on a preemptible kernel but this is acceptable, so
233          * avoid this_cpu_add()'s irq-disable overhead.
234          */
235         raw_cpu_inc(s->cpu_slab->stat[si]);
236 #endif
237 }
238 
239 /********************************************************************
240  *                      Core slab cache functions
241  *******************************************************************/
242 
243 static inline void *get_freepointer(struct kmem_cache *s, void *object)
244 {
245         return *(void **)(object + s->offset);
246 }
247 
248 static void prefetch_freepointer(const struct kmem_cache *s, void *object)
249 {
250         prefetch(object + s->offset);
251 }
252 
253 static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
254 {
255         void *p;
256 
257         if (!debug_pagealloc_enabled())
258                 return get_freepointer(s, object);
259 
260         probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
261         return p;
262 }
263 
264 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
265 {
266         *(void **)(object + s->offset) = fp;
267 }
268 
269 /* Loop over all objects in a slab */
270 #define for_each_object(__p, __s, __addr, __objects) \
271         for (__p = fixup_red_left(__s, __addr); \
272                 __p < (__addr) + (__objects) * (__s)->size; \
273                 __p += (__s)->size)
274 
275 #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \
276         for (__p = fixup_red_left(__s, __addr), __idx = 1; \
277                 __idx <= __objects; \
278                 __p += (__s)->size, __idx++)
279 
280 /* Determine object index from a given position */
281 static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
282 {
283         return (p - addr) / s->size;
284 }
285 
286 static inline int order_objects(int order, unsigned long size, int reserved)
287 {
288         return ((PAGE_SIZE << order) - reserved) / size;
289 }
290 
291 static inline struct kmem_cache_order_objects oo_make(int order,
292                 unsigned long size, int reserved)
293 {
294         struct kmem_cache_order_objects x = {
295                 (order << OO_SHIFT) + order_objects(order, size, reserved)
296         };
297 
298         return x;
299 }
300 
301 static inline int oo_order(struct kmem_cache_order_objects x)
302 {
303         return x.x >> OO_SHIFT;
304 }
305 
306 static inline int oo_objects(struct kmem_cache_order_objects x)
307 {
308         return x.x & OO_MASK;
309 }
310 
311 /*
312  * Per slab locking using the pagelock
313  */
314 static __always_inline void slab_lock(struct page *page)
315 {
316         VM_BUG_ON_PAGE(PageTail(page), page);
317         bit_spin_lock(PG_locked, &page->flags);
318 }
319 
320 static __always_inline void slab_unlock(struct page *page)
321 {
322         VM_BUG_ON_PAGE(PageTail(page), page);
323         __bit_spin_unlock(PG_locked, &page->flags);
324 }
325 
326 static inline void set_page_slub_counters(struct page *page, unsigned long counters_new)
327 {
328         struct page tmp;
329         tmp.counters = counters_new;
330         /*
331          * page->counters can cover frozen/inuse/objects as well
332          * as page->_refcount.  If we assign to ->counters directly
333          * we run the risk of losing updates to page->_refcount, so
334          * be careful and only assign to the fields we need.
335          */
336         page->frozen  = tmp.frozen;
337         page->inuse   = tmp.inuse;
338         page->objects = tmp.objects;
339 }
340 
341 /* Interrupts must be disabled (for the fallback code to work right) */
342 static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
343                 void *freelist_old, unsigned long counters_old,
344                 void *freelist_new, unsigned long counters_new,
345                 const char *n)
346 {
347         VM_BUG_ON(!irqs_disabled());
348 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
349     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
350         if (s->flags & __CMPXCHG_DOUBLE) {
351                 if (cmpxchg_double(&page->freelist, &page->counters,
352                                    freelist_old, counters_old,
353                                    freelist_new, counters_new))
354                         return true;
355         } else
356 #endif
357         {
358                 slab_lock(page);
359                 if (page->freelist == freelist_old &&
360                                         page->counters == counters_old) {
361                         page->freelist = freelist_new;
362                         set_page_slub_counters(page, counters_new);
363                         slab_unlock(page);
364                         return true;
365                 }
366                 slab_unlock(page);
367         }
368 
369         cpu_relax();
370         stat(s, CMPXCHG_DOUBLE_FAIL);
371 
372 #ifdef SLUB_DEBUG_CMPXCHG
373         pr_info("%s %s: cmpxchg double redo ", n, s->name);
374 #endif
375 
376         return false;
377 }
378 
379 static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
380                 void *freelist_old, unsigned long counters_old,
381                 void *freelist_new, unsigned long counters_new,
382                 const char *n)
383 {
384 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
385     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
386         if (s->flags & __CMPXCHG_DOUBLE) {
387                 if (cmpxchg_double(&page->freelist, &page->counters,
388                                    freelist_old, counters_old,
389                                    freelist_new, counters_new))
390                         return true;
391         } else
392 #endif
393         {
394                 unsigned long flags;
395 
396                 local_irq_save(flags);
397                 slab_lock(page);
398                 if (page->freelist == freelist_old &&
399                                         page->counters == counters_old) {
400                         page->freelist = freelist_new;
401                         set_page_slub_counters(page, counters_new);
402                         slab_unlock(page);
403                         local_irq_restore(flags);
404                         return true;
405                 }
406                 slab_unlock(page);
407                 local_irq_restore(flags);
408         }
409 
410         cpu_relax();
411         stat(s, CMPXCHG_DOUBLE_FAIL);
412 
413 #ifdef SLUB_DEBUG_CMPXCHG
414         pr_info("%s %s: cmpxchg double redo ", n, s->name);
415 #endif
416 
417         return false;
418 }
419 
420 #ifdef CONFIG_SLUB_DEBUG
421 /*
422  * Determine a map of object in use on a page.
423  *
424  * Node listlock must be held to guarantee that the page does
425  * not vanish from under us.
426  */
427 static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
428 {
429         void *p;
430         void *addr = page_address(page);
431 
432         for (p = page->freelist; p; p = get_freepointer(s, p))
433                 set_bit(slab_index(p, s, addr), map);
434 }
435 
436 static inline int size_from_object(struct kmem_cache *s)
437 {
438         if (s->flags & SLAB_RED_ZONE)
439                 return s->size - s->red_left_pad;
440 
441         return s->size;
442 }
443 
444 static inline void *restore_red_left(struct kmem_cache *s, void *p)
445 {
446         if (s->flags & SLAB_RED_ZONE)
447                 p -= s->red_left_pad;
448 
449         return p;
450 }
451 
452 /*
453  * Debug settings:
454  */
455 #if defined(CONFIG_SLUB_DEBUG_ON)
456 static int slub_debug = DEBUG_DEFAULT_FLAGS;
457 #else
458 static int slub_debug;
459 #endif
460 
461 static char *slub_debug_slabs;
462 static int disable_higher_order_debug;
463 
464 /*
465  * slub is about to manipulate internal object metadata.  This memory lies
466  * outside the range of the allocated object, so accessing it would normally
467  * be reported by kasan as a bounds error.  metadata_access_enable() is used
468  * to tell kasan that these accesses are OK.
469  */
470 static inline void metadata_access_enable(void)
471 {
472         kasan_disable_current();
473 }
474 
475 static inline void metadata_access_disable(void)
476 {
477         kasan_enable_current();
478 }
479 
480 /*
481  * Object debugging
482  */
483 
484 /* Verify that a pointer has an address that is valid within a slab page */
485 static inline int check_valid_pointer(struct kmem_cache *s,
486                                 struct page *page, void *object)
487 {
488         void *base;
489 
490         if (!object)
491                 return 1;
492 
493         base = page_address(page);
494         object = restore_red_left(s, object);
495         if (object < base || object >= base + page->objects * s->size ||
496                 (object - base) % s->size) {
497                 return 0;
498         }
499 
500         return 1;
501 }
502 
503 static void print_section(char *text, u8 *addr, unsigned int length)
504 {
505         metadata_access_enable();
506         print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
507                         length, 1);
508         metadata_access_disable();
509 }
510 
511 static struct track *get_track(struct kmem_cache *s, void *object,
512         enum track_item alloc)
513 {
514         struct track *p;
515 
516         if (s->offset)
517                 p = object + s->offset + sizeof(void *);
518         else
519                 p = object + s->inuse;
520 
521         return p + alloc;
522 }
523 
524 static void set_track(struct kmem_cache *s, void *object,
525                         enum track_item alloc, unsigned long addr)
526 {
527         struct track *p = get_track(s, object, alloc);
528 
529         if (addr) {
530 #ifdef CONFIG_STACKTRACE
531                 struct stack_trace trace;
532                 int i;
533 
534                 trace.nr_entries = 0;
535                 trace.max_entries = TRACK_ADDRS_COUNT;
536                 trace.entries = p->addrs;
537                 trace.skip = 3;
538                 metadata_access_enable();
539                 save_stack_trace(&trace);
540                 metadata_access_disable();
541 
542                 /* See rant in lockdep.c */
543                 if (trace.nr_entries != 0 &&
544                     trace.entries[trace.nr_entries - 1] == ULONG_MAX)
545                         trace.nr_entries--;
546 
547                 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
548                         p->addrs[i] = 0;
549 #endif
550                 p->addr = addr;
551                 p->cpu = smp_processor_id();
552                 p->pid = current->pid;
553                 p->when = jiffies;
554         } else
555                 memset(p, 0, sizeof(struct track));
556 }
557 
558 static void init_tracking(struct kmem_cache *s, void *object)
559 {
560         if (!(s->flags & SLAB_STORE_USER))
561                 return;
562 
563         set_track(s, object, TRACK_FREE, 0UL);
564         set_track(s, object, TRACK_ALLOC, 0UL);
565 }
566 
567 static void print_track(const char *s, struct track *t)
568 {
569         if (!t->addr)
570                 return;
571 
572         pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
573                s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
574 #ifdef CONFIG_STACKTRACE
575         {
576                 int i;
577                 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
578                         if (t->addrs[i])
579                                 pr_err("\t%pS\n", (void *)t->addrs[i]);
580                         else
581                                 break;
582         }
583 #endif
584 }
585 
586 static void print_tracking(struct kmem_cache *s, void *object)
587 {
588         if (!(s->flags & SLAB_STORE_USER))
589                 return;
590 
591         print_track("Allocated", get_track(s, object, TRACK_ALLOC));
592         print_track("Freed", get_track(s, object, TRACK_FREE));
593 }
594 
595 static void print_page_info(struct page *page)
596 {
597         pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
598                page, page->objects, page->inuse, page->freelist, page->flags);
599 
600 }
601 
602 static void slab_bug(struct kmem_cache *s, char *fmt, ...)
603 {
604         struct va_format vaf;
605         va_list args;
606 
607         va_start(args, fmt);
608         vaf.fmt = fmt;
609         vaf.va = &args;
610         pr_err("=============================================================================\n");
611         pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
612         pr_err("-----------------------------------------------------------------------------\n\n");
613 
614         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
615         va_end(args);
616 }
617 
618 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
619 {
620         struct va_format vaf;
621         va_list args;
622 
623         va_start(args, fmt);
624         vaf.fmt = fmt;
625         vaf.va = &args;
626         pr_err("FIX %s: %pV\n", s->name, &vaf);
627         va_end(args);
628 }
629 
630 static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
631 {
632         unsigned int off;       /* Offset of last byte */
633         u8 *addr = page_address(page);
634 
635         print_tracking(s, p);
636 
637         print_page_info(page);
638 
639         pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
640                p, p - addr, get_freepointer(s, p));
641 
642         if (s->flags & SLAB_RED_ZONE)
643                 print_section("Redzone ", p - s->red_left_pad, s->red_left_pad);
644         else if (p > addr + 16)
645                 print_section("Bytes b4 ", p - 16, 16);
646 
647         print_section("Object ", p, min_t(unsigned long, s->object_size,
648                                 PAGE_SIZE));
649         if (s->flags & SLAB_RED_ZONE)
650                 print_section("Redzone ", p + s->object_size,
651                         s->inuse - s->object_size);
652 
653         if (s->offset)
654                 off = s->offset + sizeof(void *);
655         else
656                 off = s->inuse;
657 
658         if (s->flags & SLAB_STORE_USER)
659                 off += 2 * sizeof(struct track);
660 
661         off += kasan_metadata_size(s);
662 
663         if (off != size_from_object(s))
664                 /* Beginning of the filler is the free pointer */
665                 print_section("Padding ", p + off, size_from_object(s) - off);
666 
667         dump_stack();
668 }
669 
670 void object_err(struct kmem_cache *s, struct page *page,
671                         u8 *object, char *reason)
672 {
673         slab_bug(s, "%s", reason);
674         print_trailer(s, page, object);
675 }
676 
677 static void slab_err(struct kmem_cache *s, struct page *page,
678                         const char *fmt, ...)
679 {
680         va_list args;
681         char buf[100];
682 
683         va_start(args, fmt);
684         vsnprintf(buf, sizeof(buf), fmt, args);
685         va_end(args);
686         slab_bug(s, "%s", buf);
687         print_page_info(page);
688         dump_stack();
689 }
690 
691 static void init_object(struct kmem_cache *s, void *object, u8 val)
692 {
693         u8 *p = object;
694 
695         if (s->flags & SLAB_RED_ZONE)
696                 memset(p - s->red_left_pad, val, s->red_left_pad);
697 
698         if (s->flags & __OBJECT_POISON) {
699                 memset(p, POISON_FREE, s->object_size - 1);
700                 p[s->object_size - 1] = POISON_END;
701         }
702 
703         if (s->flags & SLAB_RED_ZONE)
704                 memset(p + s->object_size, val, s->inuse - s->object_size);
705 }
706 
707 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
708                                                 void *from, void *to)
709 {
710         slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
711         memset(from, data, to - from);
712 }
713 
714 static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
715                         u8 *object, char *what,
716                         u8 *start, unsigned int value, unsigned int bytes)
717 {
718         u8 *fault;
719         u8 *end;
720 
721         metadata_access_enable();
722         fault = memchr_inv(start, value, bytes);
723         metadata_access_disable();
724         if (!fault)
725                 return 1;
726 
727         end = start + bytes;
728         while (end > fault && end[-1] == value)
729                 end--;
730 
731         slab_bug(s, "%s overwritten", what);
732         pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
733                                         fault, end - 1, fault[0], value);
734         print_trailer(s, page, object);
735 
736         restore_bytes(s, what, value, fault, end);
737         return 0;
738 }
739 
740 /*
741  * Object layout:
742  *
743  * object address
744  *      Bytes of the object to be managed.
745  *      If the freepointer may overlay the object then the free
746  *      pointer is the first word of the object.
747  *
748  *      Poisoning uses 0x6b (POISON_FREE) and the last byte is
749  *      0xa5 (POISON_END)
750  *
751  * object + s->object_size
752  *      Padding to reach word boundary. This is also used for Redzoning.
753  *      Padding is extended by another word if Redzoning is enabled and
754  *      object_size == inuse.
755  *
756  *      We fill with 0xbb (RED_INACTIVE) for inactive objects and with
757  *      0xcc (RED_ACTIVE) for objects in use.
758  *
759  * object + s->inuse
760  *      Meta data starts here.
761  *
762  *      A. Free pointer (if we cannot overwrite object on free)
763  *      B. Tracking data for SLAB_STORE_USER
764  *      C. Padding to reach required alignment boundary or at mininum
765  *              one word if debugging is on to be able to detect writes
766  *              before the word boundary.
767  *
768  *      Padding is done using 0x5a (POISON_INUSE)
769  *
770  * object + s->size
771  *      Nothing is used beyond s->size.
772  *
773  * If slabcaches are merged then the object_size and inuse boundaries are mostly
774  * ignored. And therefore no slab options that rely on these boundaries
775  * may be used with merged slabcaches.
776  */
777 
778 static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
779 {
780         unsigned long off = s->inuse;   /* The end of info */
781 
782         if (s->offset)
783                 /* Freepointer is placed after the object. */
784                 off += sizeof(void *);
785 
786         if (s->flags & SLAB_STORE_USER)
787                 /* We also have user information there */
788                 off += 2 * sizeof(struct track);
789 
790         off += kasan_metadata_size(s);
791 
792         if (size_from_object(s) == off)
793                 return 1;
794 
795         return check_bytes_and_report(s, page, p, "Object padding",
796                         p + off, POISON_INUSE, size_from_object(s) - off);
797 }
798 
799 /* Check the pad bytes at the end of a slab page */
800 static int slab_pad_check(struct kmem_cache *s, struct page *page)
801 {
802         u8 *start;
803         u8 *fault;
804         u8 *end;
805         int length;
806         int remainder;
807 
808         if (!(s->flags & SLAB_POISON))
809                 return 1;
810 
811         start = page_address(page);
812         length = (PAGE_SIZE << compound_order(page)) - s->reserved;
813         end = start + length;
814         remainder = length % s->size;
815         if (!remainder)
816                 return 1;
817 
818         metadata_access_enable();
819         fault = memchr_inv(end - remainder, POISON_INUSE, remainder);
820         metadata_access_disable();
821         if (!fault)
822                 return 1;
823         while (end > fault && end[-1] == POISON_INUSE)
824                 end--;
825 
826         slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
827         print_section("Padding ", end - remainder, remainder);
828 
829         restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
830         return 0;
831 }
832 
833 static int check_object(struct kmem_cache *s, struct page *page,
834                                         void *object, u8 val)
835 {
836         u8 *p = object;
837         u8 *endobject = object + s->object_size;
838 
839         if (s->flags & SLAB_RED_ZONE) {
840                 if (!check_bytes_and_report(s, page, object, "Redzone",
841                         object - s->red_left_pad, val, s->red_left_pad))
842                         return 0;
843 
844                 if (!check_bytes_and_report(s, page, object, "Redzone",
845                         endobject, val, s->inuse - s->object_size))
846                         return 0;
847         } else {
848                 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
849                         check_bytes_and_report(s, page, p, "Alignment padding",
850                                 endobject, POISON_INUSE,
851                                 s->inuse - s->object_size);
852                 }
853         }
854 
855         if (s->flags & SLAB_POISON) {
856                 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
857                         (!check_bytes_and_report(s, page, p, "Poison", p,
858                                         POISON_FREE, s->object_size - 1) ||
859                          !check_bytes_and_report(s, page, p, "Poison",
860                                 p + s->object_size - 1, POISON_END, 1)))
861                         return 0;
862                 /*
863                  * check_pad_bytes cleans up on its own.
864                  */
865                 check_pad_bytes(s, page, p);
866         }
867 
868         if (!s->offset && val == SLUB_RED_ACTIVE)
869                 /*
870                  * Object and freepointer overlap. Cannot check
871                  * freepointer while object is allocated.
872                  */
873                 return 1;
874 
875         /* Check free pointer validity */
876         if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
877                 object_err(s, page, p, "Freepointer corrupt");
878                 /*
879                  * No choice but to zap it and thus lose the remainder
880                  * of the free objects in this slab. May cause
881                  * another error because the object count is now wrong.
882                  */
883                 set_freepointer(s, p, NULL);
884                 return 0;
885         }
886         return 1;
887 }
888 
889 static int check_slab(struct kmem_cache *s, struct page *page)
890 {
891         int maxobj;
892 
893         VM_BUG_ON(!irqs_disabled());
894 
895         if (!PageSlab(page)) {
896                 slab_err(s, page, "Not a valid slab page");
897                 return 0;
898         }
899 
900         maxobj = order_objects(compound_order(page), s->size, s->reserved);
901         if (page->objects > maxobj) {
902                 slab_err(s, page, "objects %u > max %u",
903                         page->objects, maxobj);
904                 return 0;
905         }
906         if (page->inuse > page->objects) {
907                 slab_err(s, page, "inuse %u > max %u",
908                         page->inuse, page->objects);
909                 return 0;
910         }
911         /* Slab_pad_check fixes things up after itself */
912         slab_pad_check(s, page);
913         return 1;
914 }
915 
916 /*
917  * Determine if a certain object on a page is on the freelist. Must hold the
918  * slab lock to guarantee that the chains are in a consistent state.
919  */
920 static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
921 {
922         int nr = 0;
923         void *fp;
924         void *object = NULL;
925         int max_objects;
926 
927         fp = page->freelist;
928         while (fp && nr <= page->objects) {
929                 if (fp == search)
930                         return 1;
931                 if (!check_valid_pointer(s, page, fp)) {
932                         if (object) {
933                                 object_err(s, page, object,
934                                         "Freechain corrupt");
935                                 set_freepointer(s, object, NULL);
936                         } else {
937                                 slab_err(s, page, "Freepointer corrupt");
938                                 page->freelist = NULL;
939                                 page->inuse = page->objects;
940                                 slab_fix(s, "Freelist cleared");
941                                 return 0;
942                         }
943                         break;
944                 }
945                 object = fp;
946                 fp = get_freepointer(s, object);
947                 nr++;
948         }
949 
950         max_objects = order_objects(compound_order(page), s->size, s->reserved);
951         if (max_objects > MAX_OBJS_PER_PAGE)
952                 max_objects = MAX_OBJS_PER_PAGE;
953 
954         if (page->objects != max_objects) {
955                 slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
956                          page->objects, max_objects);
957                 page->objects = max_objects;
958                 slab_fix(s, "Number of objects adjusted.");
959         }
960         if (page->inuse != page->objects - nr) {
961                 slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
962                          page->inuse, page->objects - nr);
963                 page->inuse = page->objects - nr;
964                 slab_fix(s, "Object count adjusted.");
965         }
966         return search == NULL;
967 }
968 
969 static void trace(struct kmem_cache *s, struct page *page, void *object,
970                                                                 int alloc)
971 {
972         if (s->flags & SLAB_TRACE) {
973                 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
974                         s->name,
975                         alloc ? "alloc" : "free",
976                         object, page->inuse,
977                         page->freelist);
978 
979                 if (!alloc)
980                         print_section("Object ", (void *)object,
981                                         s->object_size);
982 
983                 dump_stack();
984         }
985 }
986 
987 /*
988  * Tracking of fully allocated slabs for debugging purposes.
989  */
990 static void add_full(struct kmem_cache *s,
991         struct kmem_cache_node *n, struct page *page)
992 {
993         if (!(s->flags & SLAB_STORE_USER))
994                 return;
995 
996         lockdep_assert_held(&n->list_lock);
997         list_add(&page->lru, &n->full);
998 }
999 
1000 static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
1001 {
1002         if (!(s->flags & SLAB_STORE_USER))
1003                 return;
1004 
1005         lockdep_assert_held(&n->list_lock);
1006         list_del(&page->lru);
1007 }
1008 
1009 /* Tracking of the number of slabs for debugging purposes */
1010 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1011 {
1012         struct kmem_cache_node *n = get_node(s, node);
1013 
1014         return atomic_long_read(&n->nr_slabs);
1015 }
1016 
1017 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1018 {
1019         return atomic_long_read(&n->nr_slabs);
1020 }
1021 
1022 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
1023 {
1024         struct kmem_cache_node *n = get_node(s, node);
1025 
1026         /*
1027          * May be called early in order to allocate a slab for the
1028          * kmem_cache_node structure. Solve the chicken-egg
1029          * dilemma by deferring the increment of the count during
1030          * bootstrap (see early_kmem_cache_node_alloc).
1031          */
1032         if (likely(n)) {
1033                 atomic_long_inc(&n->nr_slabs);
1034                 atomic_long_add(objects, &n->total_objects);
1035         }
1036 }
1037 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
1038 {
1039         struct kmem_cache_node *n = get_node(s, node);
1040 
1041         atomic_long_dec(&n->nr_slabs);
1042         atomic_long_sub(objects, &n->total_objects);
1043 }
1044 
1045 /* Object debug checks for alloc/free paths */
1046 static void setup_object_debug(struct kmem_cache *s, struct page *page,
1047                                                                 void *object)
1048 {
1049         if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1050                 return;
1051 
1052         init_object(s, object, SLUB_RED_INACTIVE);
1053         init_tracking(s, object);
1054 }
1055 
1056 static inline int alloc_consistency_checks(struct kmem_cache *s,
1057                                         struct page *page,
1058                                         void *object, unsigned long addr)
1059 {
1060         if (!check_slab(s, page))
1061                 return 0;
1062 
1063         if (!check_valid_pointer(s, page, object)) {
1064                 object_err(s, page, object, "Freelist Pointer check fails");
1065                 return 0;
1066         }
1067 
1068         if (!check_object(s, page, object, SLUB_RED_INACTIVE))
1069                 return 0;
1070 
1071         return 1;
1072 }
1073 
1074 static noinline int alloc_debug_processing(struct kmem_cache *s,
1075                                         struct page *page,
1076                                         void *object, unsigned long addr)
1077 {
1078         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1079                 if (!alloc_consistency_checks(s, page, object, addr))
1080                         goto bad;
1081         }
1082 
1083         /* Success perform special debug activities for allocs */
1084         if (s->flags & SLAB_STORE_USER)
1085                 set_track(s, object, TRACK_ALLOC, addr);
1086         trace(s, page, object, 1);
1087         init_object(s, object, SLUB_RED_ACTIVE);
1088         return 1;
1089 
1090 bad:
1091         if (PageSlab(page)) {
1092                 /*
1093                  * If this is a slab page then lets do the best we can
1094                  * to avoid issues in the future. Marking all objects
1095                  * as used avoids touching the remaining objects.
1096                  */
1097                 slab_fix(s, "Marking all objects used");
1098                 page->inuse = page->objects;
1099                 page->freelist = NULL;
1100         }
1101         return 0;
1102 }
1103 
1104 static inline int free_consistency_checks(struct kmem_cache *s,
1105                 struct page *page, void *object, unsigned long addr)
1106 {
1107         if (!check_valid_pointer(s, page, object)) {
1108                 slab_err(s, page, "Invalid object pointer 0x%p", object);
1109                 return 0;
1110         }
1111 
1112         if (on_freelist(s, page, object)) {
1113                 object_err(s, page, object, "Object already free");
1114                 return 0;
1115         }
1116 
1117         if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1118                 return 0;
1119 
1120         if (unlikely(s != page->slab_cache)) {
1121                 if (!PageSlab(page)) {
1122                         slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1123                                  object);
1124                 } else if (!page->slab_cache) {
1125                         pr_err("SLUB <none>: no slab for object 0x%p.\n",
1126                                object);
1127                         dump_stack();
1128                 } else
1129                         object_err(s, page, object,
1130                                         "page slab pointer corrupt.");
1131                 return 0;
1132         }
1133         return 1;
1134 }
1135 
1136 /* Supports checking bulk free of a constructed freelist */
1137 static noinline int free_debug_processing(
1138         struct kmem_cache *s, struct page *page,
1139         void *head, void *tail, int bulk_cnt,
1140         unsigned long addr)
1141 {
1142         struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1143         void *object = head;
1144         int cnt = 0;
1145         unsigned long uninitialized_var(flags);
1146         int ret = 0;
1147 
1148         spin_lock_irqsave(&n->list_lock, flags);
1149         slab_lock(page);
1150 
1151         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1152                 if (!check_slab(s, page))
1153                         goto out;
1154         }
1155 
1156 next_object:
1157         cnt++;
1158 
1159         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1160                 if (!free_consistency_checks(s, page, object, addr))
1161                         goto out;
1162         }
1163 
1164         if (s->flags & SLAB_STORE_USER)
1165                 set_track(s, object, TRACK_FREE, addr);
1166         trace(s, page, object, 0);
1167         /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
1168         init_object(s, object, SLUB_RED_INACTIVE);
1169 
1170         /* Reached end of constructed freelist yet? */
1171         if (object != tail) {
1172                 object = get_freepointer(s, object);
1173                 goto next_object;
1174         }
1175         ret = 1;
1176 
1177 out:
1178         if (cnt != bulk_cnt)
1179                 slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1180                          bulk_cnt, cnt);
1181 
1182         slab_unlock(page);
1183         spin_unlock_irqrestore(&n->list_lock, flags);
1184         if (!ret)
1185                 slab_fix(s, "Object at 0x%p not freed", object);
1186         return ret;
1187 }
1188 
1189 static int __init setup_slub_debug(char *str)
1190 {
1191         slub_debug = DEBUG_DEFAULT_FLAGS;
1192         if (*str++ != '=' || !*str)
1193                 /*
1194                  * No options specified. Switch on full debugging.
1195                  */
1196                 goto out;
1197 
1198         if (*str == ',')
1199                 /*
1200                  * No options but restriction on slabs. This means full
1201                  * debugging for slabs matching a pattern.
1202                  */
1203                 goto check_slabs;
1204 
1205         slub_debug = 0;
1206         if (*str == '-')
1207                 /*
1208                  * Switch off all debugging measures.
1209                  */
1210                 goto out;
1211 
1212         /*
1213          * Determine which debug features should be switched on
1214          */
1215         for (; *str && *str != ','; str++) {
1216                 switch (tolower(*str)) {
1217                 case 'f':
1218                         slub_debug |= SLAB_CONSISTENCY_CHECKS;
1219                         break;
1220                 case 'z':
1221                         slub_debug |= SLAB_RED_ZONE;
1222                         break;
1223                 case 'p':
1224                         slub_debug |= SLAB_POISON;
1225                         break;
1226                 case 'u':
1227                         slub_debug |= SLAB_STORE_USER;
1228                         break;
1229                 case 't':
1230                         slub_debug |= SLAB_TRACE;
1231                         break;
1232                 case 'a':
1233                         slub_debug |= SLAB_FAILSLAB;
1234                         break;
1235                 case 'o':
1236                         /*
1237                          * Avoid enabling debugging on caches if its minimum
1238                          * order would increase as a result.
1239                          */
1240                         disable_higher_order_debug = 1;
1241                         break;
1242                 default:
1243                         pr_err("slub_debug option '%c' unknown. skipped\n",
1244                                *str);
1245                 }
1246         }
1247 
1248 check_slabs:
1249         if (*str == ',')
1250                 slub_debug_slabs = str + 1;
1251 out:
1252         return 1;
1253 }
1254 
1255 __setup("slub_debug", setup_slub_debug);
1256 
1257 unsigned long kmem_cache_flags(unsigned long object_size,
1258         unsigned long flags, const char *name,
1259         void (*ctor)(void *))
1260 {
1261         /*
1262          * Enable debugging if selected on the kernel commandline.
1263          */
1264         if (slub_debug && (!slub_debug_slabs || (name &&
1265                 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))))
1266                 flags |= slub_debug;
1267 
1268         return flags;
1269 }
1270 #else /* !CONFIG_SLUB_DEBUG */
1271 static inline void setup_object_debug(struct kmem_cache *s,
1272                         struct page *page, void *object) {}
1273 
1274 static inline int alloc_debug_processing(struct kmem_cache *s,
1275         struct page *page, void *object, unsigned long addr) { return 0; }
1276 
1277 static inline int free_debug_processing(
1278         struct kmem_cache *s, struct page *page,
1279         void *head, void *tail, int bulk_cnt,
1280         unsigned long addr) { return 0; }
1281 
1282 static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1283                         { return 1; }
1284 static inline int check_object(struct kmem_cache *s, struct page *page,
1285                         void *object, u8 val) { return 1; }
1286 static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1287                                         struct page *page) {}
1288 static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1289                                         struct page *page) {}
1290 unsigned long kmem_cache_flags(unsigned long object_size,
1291         unsigned long flags, const char *name,
1292         void (*ctor)(void *))
1293 {
1294         return flags;
1295 }
1296 #define slub_debug 0
1297 
1298 #define disable_higher_order_debug 0
1299 
1300 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1301                                                         { return 0; }
1302 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1303                                                         { return 0; }
1304 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1305                                                         int objects) {}
1306 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1307                                                         int objects) {}
1308 
1309 #endif /* CONFIG_SLUB_DEBUG */
1310 
1311 /*
1312  * Hooks for other subsystems that check memory allocations. In a typical
1313  * production configuration these hooks all should produce no code at all.
1314  */
1315 static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1316 {
1317         kmemleak_alloc(ptr, size, 1, flags);
1318         kasan_kmalloc_large(ptr, size, flags);
1319 }
1320 
1321 static inline void kfree_hook(const void *x)
1322 {
1323         kmemleak_free(x);
1324         kasan_kfree_large(x);
1325 }
1326 
1327 static inline void *slab_free_hook(struct kmem_cache *s, void *x)
1328 {
1329         void *freeptr;
1330 
1331         kmemleak_free_recursive(x, s->flags);
1332 
1333         /*
1334          * Trouble is that we may no longer disable interrupts in the fast path
1335          * So in order to make the debug calls that expect irqs to be
1336          * disabled we need to disable interrupts temporarily.
1337          */
1338 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
1339         {
1340                 unsigned long flags;
1341 
1342                 local_irq_save(flags);
1343                 kmemcheck_slab_free(s, x, s->object_size);
1344                 debug_check_no_locks_freed(x, s->object_size);
1345                 local_irq_restore(flags);
1346         }
1347 #endif
1348         if (!(s->flags & SLAB_DEBUG_OBJECTS))
1349                 debug_check_no_obj_freed(x, s->object_size);
1350 
1351         freeptr = get_freepointer(s, x);
1352         /*
1353          * kasan_slab_free() may put x into memory quarantine, delaying its
1354          * reuse. In this case the object's freelist pointer is changed.
1355          */
1356         kasan_slab_free(s, x);
1357         return freeptr;
1358 }
1359 
1360 static inline void slab_free_freelist_hook(struct kmem_cache *s,
1361                                            void *head, void *tail)
1362 {
1363 /*
1364  * Compiler cannot detect this function can be removed if slab_free_hook()
1365  * evaluates to nothing.  Thus, catch all relevant config debug options here.
1366  */
1367 #if defined(CONFIG_KMEMCHECK) ||                \
1368         defined(CONFIG_LOCKDEP) ||              \
1369         defined(CONFIG_DEBUG_KMEMLEAK) ||       \
1370         defined(CONFIG_DEBUG_OBJECTS_FREE) ||   \
1371         defined(CONFIG_KASAN)
1372 
1373         void *object = head;
1374         void *tail_obj = tail ? : head;
1375         void *freeptr;
1376 
1377         do {
1378                 freeptr = slab_free_hook(s, object);
1379         } while ((object != tail_obj) && (object = freeptr));
1380 #endif
1381 }
1382 
1383 static void setup_object(struct kmem_cache *s, struct page *page,
1384                                 void *object)
1385 {
1386         setup_object_debug(s, page, object);
1387         kasan_init_slab_obj(s, object);
1388         if (unlikely(s->ctor)) {
1389                 kasan_unpoison_object_data(s, object);
1390                 s->ctor(object);
1391                 kasan_poison_object_data(s, object);
1392         }
1393 }
1394 
1395 /*
1396  * Slab allocation and freeing
1397  */
1398 static inline struct page *alloc_slab_page(struct kmem_cache *s,
1399                 gfp_t flags, int node, struct kmem_cache_order_objects oo)
1400 {
1401         struct page *page;
1402         int order = oo_order(oo);
1403 
1404         flags |= __GFP_NOTRACK;
1405 
1406         if (node == NUMA_NO_NODE)
1407                 page = alloc_pages(flags, order);
1408         else
1409                 page = __alloc_pages_node(node, flags, order);
1410 
1411         if (page && memcg_charge_slab(page, flags, order, s)) {
1412                 __free_pages(page, order);
1413                 page = NULL;
1414         }
1415 
1416         return page;
1417 }
1418 
1419 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1420 /* Pre-initialize the random sequence cache */
1421 static int init_cache_random_seq(struct kmem_cache *s)
1422 {
1423         int err;
1424         unsigned long i, count = oo_objects(s->oo);
1425 
1426         err = cache_random_seq_create(s, count, GFP_KERNEL);
1427         if (err) {
1428                 pr_err("SLUB: Unable to initialize free list for %s\n",
1429                         s->name);
1430                 return err;
1431         }
1432 
1433         /* Transform to an offset on the set of pages */
1434         if (s->random_seq) {
1435                 for (i = 0; i < count; i++)
1436                         s->random_seq[i] *= s->size;
1437         }
1438         return 0;
1439 }
1440 
1441 /* Initialize each random sequence freelist per cache */
1442 static void __init init_freelist_randomization(void)
1443 {
1444         struct kmem_cache *s;
1445 
1446         mutex_lock(&slab_mutex);
1447 
1448         list_for_each_entry(s, &slab_caches, list)
1449                 init_cache_random_seq(s);
1450 
1451         mutex_unlock(&slab_mutex);
1452 }
1453 
1454 /* Get the next entry on the pre-computed freelist randomized */
1455 static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1456                                 unsigned long *pos, void *start,
1457                                 unsigned long page_limit,
1458                                 unsigned long freelist_count)
1459 {
1460         unsigned int idx;
1461 
1462         /*
1463          * If the target page allocation failed, the number of objects on the
1464          * page might be smaller than the usual size defined by the cache.
1465          */
1466         do {
1467                 idx = s->random_seq[*pos];
1468                 *pos += 1;
1469                 if (*pos >= freelist_count)
1470                         *pos = 0;
1471         } while (unlikely(idx >= page_limit));
1472 
1473         return (char *)start + idx;
1474 }
1475 
1476 /* Shuffle the single linked freelist based on a random pre-computed sequence */
1477 static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1478 {
1479         void *start;
1480         void *cur;
1481         void *next;
1482         unsigned long idx, pos, page_limit, freelist_count;
1483 
1484         if (page->objects < 2 || !s->random_seq)
1485                 return false;
1486 
1487         freelist_count = oo_objects(s->oo);
1488         pos = get_random_int() % freelist_count;
1489 
1490         page_limit = page->objects * s->size;
1491         start = fixup_red_left(s, page_address(page));
1492 
1493         /* First entry is used as the base of the freelist */
1494         cur = next_freelist_entry(s, page, &pos, start, page_limit,
1495                                 freelist_count);
1496         page->freelist = cur;
1497 
1498         for (idx = 1; idx < page->objects; idx++) {
1499                 setup_object(s, page, cur);
1500                 next = next_freelist_entry(s, page, &pos, start, page_limit,
1501                         freelist_count);
1502                 set_freepointer(s, cur, next);
1503                 cur = next;
1504         }
1505         setup_object(s, page, cur);
1506         set_freepointer(s, cur, NULL);
1507 
1508         return true;
1509 }
1510 #else
1511 static inline int init_cache_random_seq(struct kmem_cache *s)
1512 {
1513         return 0;
1514 }
1515 static inline void init_freelist_randomization(void) { }
1516 static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1517 {
1518         return false;
1519 }
1520 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1521 
1522 static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1523 {
1524         struct page *page;
1525         struct kmem_cache_order_objects oo = s->oo;
1526         gfp_t alloc_gfp;
1527         void *start, *p;
1528         int idx, order;
1529         bool shuffle;
1530 
1531         flags &= gfp_allowed_mask;
1532 
1533         if (gfpflags_allow_blocking(flags))
1534                 local_irq_enable();
1535 
1536         flags |= s->allocflags;
1537 
1538         /*
1539          * Let the initial higher-order allocation fail under memory pressure
1540          * so we fall-back to the minimum order allocation.
1541          */
1542         alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1543         if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
1544                 alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
1545 
1546         page = alloc_slab_page(s, alloc_gfp, node, oo);
1547         if (unlikely(!page)) {
1548                 oo = s->min;
1549                 alloc_gfp = flags;
1550                 /*
1551                  * Allocation may have failed due to fragmentation.
1552                  * Try a lower order alloc if possible
1553                  */
1554                 page = alloc_slab_page(s, alloc_gfp, node, oo);
1555                 if (unlikely(!page))
1556                         goto out;
1557                 stat(s, ORDER_FALLBACK);
1558         }
1559 
1560         if (kmemcheck_enabled &&
1561             !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
1562                 int pages = 1 << oo_order(oo);
1563 
1564                 kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node);
1565 
1566                 /*
1567                  * Objects from caches that have a constructor don't get
1568                  * cleared when they're allocated, so we need to do it here.
1569                  */
1570                 if (s->ctor)
1571                         kmemcheck_mark_uninitialized_pages(page, pages);
1572                 else
1573                         kmemcheck_mark_unallocated_pages(page, pages);
1574         }
1575 
1576         page->objects = oo_objects(oo);
1577 
1578         order = compound_order(page);
1579         page->slab_cache = s;
1580         __SetPageSlab(page);
1581         if (page_is_pfmemalloc(page))
1582                 SetPageSlabPfmemalloc(page);
1583 
1584         start = page_address(page);
1585 
1586         if (unlikely(s->flags & SLAB_POISON))
1587                 memset(start, POISON_INUSE, PAGE_SIZE << order);
1588 
1589         kasan_poison_slab(page);
1590 
1591         shuffle = shuffle_freelist(s, page);
1592 
1593         if (!shuffle) {
1594                 for_each_object_idx(p, idx, s, start, page->objects) {
1595                         setup_object(s, page, p);
1596                         if (likely(idx < page->objects))
1597                                 set_freepointer(s, p, p + s->size);
1598                         else
1599                                 set_freepointer(s, p, NULL);
1600                 }
1601                 page->freelist = fixup_red_left(s, start);
1602         }
1603 
1604         page->inuse = page->objects;
1605         page->frozen = 1;
1606 
1607 out:
1608         if (gfpflags_allow_blocking(flags))
1609                 local_irq_disable();
1610         if (!page)
1611                 return NULL;
1612 
1613         mod_zone_page_state(page_zone(page),
1614                 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1615                 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1616                 1 << oo_order(oo));
1617 
1618         inc_slabs_node(s, page_to_nid(page), page->objects);
1619 
1620         return page;
1621 }
1622 
1623 static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1624 {
1625         if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
1626                 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1627                 flags &= ~GFP_SLAB_BUG_MASK;
1628                 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1629                                 invalid_mask, &invalid_mask, flags, &flags);
1630         }
1631 
1632         return allocate_slab(s,
1633                 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1634 }
1635 
1636 static void __free_slab(struct kmem_cache *s, struct page *page)
1637 {
1638         int order = compound_order(page);
1639         int pages = 1 << order;
1640 
1641         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1642                 void *p;
1643 
1644                 slab_pad_check(s, page);
1645                 for_each_object(p, s, page_address(page),
1646                                                 page->objects)
1647                         check_object(s, page, p, SLUB_RED_INACTIVE);
1648         }
1649 
1650         kmemcheck_free_shadow(page, compound_order(page));
1651 
1652         mod_zone_page_state(page_zone(page),
1653                 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1654                 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1655                 -pages);
1656 
1657         __ClearPageSlabPfmemalloc(page);
1658         __ClearPageSlab(page);
1659 
1660         page_mapcount_reset(page);
1661         if (current->reclaim_state)
1662                 current->reclaim_state->reclaimed_slab += pages;
1663         memcg_uncharge_slab(page, order, s);
1664         __free_pages(page, order);
1665 }
1666 
1667 #define need_reserve_slab_rcu                                           \
1668         (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1669 
1670 static void rcu_free_slab(struct rcu_head *h)
1671 {
1672         struct page *page;
1673 
1674         if (need_reserve_slab_rcu)
1675                 page = virt_to_head_page(h);
1676         else
1677                 page = container_of((struct list_head *)h, struct page, lru);
1678 
1679         __free_slab(page->slab_cache, page);
1680 }
1681 
1682 static void free_slab(struct kmem_cache *s, struct page *page)
1683 {
1684         if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1685                 struct rcu_head *head;
1686 
1687                 if (need_reserve_slab_rcu) {
1688                         int order = compound_order(page);
1689                         int offset = (PAGE_SIZE << order) - s->reserved;
1690 
1691                         VM_BUG_ON(s->reserved != sizeof(*head));
1692                         head = page_address(page) + offset;
1693                 } else {
1694                         head = &page->rcu_head;
1695                 }
1696 
1697                 call_rcu(head, rcu_free_slab);
1698         } else
1699                 __free_slab(s, page);
1700 }
1701 
1702 static void discard_slab(struct kmem_cache *s, struct page *page)
1703 {
1704         dec_slabs_node(s, page_to_nid(page), page->objects);
1705         free_slab(s, page);
1706 }
1707 
1708 /*
1709  * Management of partially allocated slabs.
1710  */
1711 static inline void
1712 __add_partial(struct kmem_cache_node *n, struct page *page, int tail)
1713 {
1714         n->nr_partial++;
1715         if (tail == DEACTIVATE_TO_TAIL)
1716                 list_add_tail(&page->lru, &n->partial);
1717         else
1718                 list_add(&page->lru, &n->partial);
1719 }
1720 
1721 static inline void add_partial(struct kmem_cache_node *n,
1722                                 struct page *page, int tail)
1723 {
1724         lockdep_assert_held(&n->list_lock);
1725         __add_partial(n, page, tail);
1726 }
1727 
1728 static inline void remove_partial(struct kmem_cache_node *n,
1729                                         struct page *page)
1730 {
1731         lockdep_assert_held(&n->list_lock);
1732         list_del(&page->lru);
1733         n->nr_partial--;
1734 }
1735 
1736 /*
1737  * Remove slab from the partial list, freeze it and
1738  * return the pointer to the freelist.
1739  *
1740  * Returns a list of objects or NULL if it fails.
1741  */
1742 static inline void *acquire_slab(struct kmem_cache *s,
1743                 struct kmem_cache_node *n, struct page *page,
1744                 int mode, int *objects)
1745 {
1746         void *freelist;
1747         unsigned long counters;
1748         struct page new;
1749 
1750         lockdep_assert_held(&n->list_lock);
1751 
1752         /*
1753          * Zap the freelist and set the frozen bit.
1754          * The old freelist is the list of objects for the
1755          * per cpu allocation list.
1756          */
1757         freelist = page->freelist;
1758         counters = page->counters;
1759         new.counters = counters;
1760         *objects = new.objects - new.inuse;
1761         if (mode) {
1762                 new.inuse = page->objects;
1763                 new.freelist = NULL;
1764         } else {
1765                 new.freelist = freelist;
1766         }
1767 
1768         VM_BUG_ON(new.frozen);
1769         new.frozen = 1;
1770 
1771         if (!__cmpxchg_double_slab(s, page,
1772                         freelist, counters,
1773                         new.freelist, new.counters,
1774                         "acquire_slab"))
1775                 return NULL;
1776 
1777         remove_partial(n, page);
1778         WARN_ON(!freelist);
1779         return freelist;
1780 }
1781 
1782 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
1783 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
1784 
1785 /*
1786  * Try to allocate a partial slab from a specific node.
1787  */
1788 static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1789                                 struct kmem_cache_cpu *c, gfp_t flags)
1790 {
1791         struct page *page, *page2;
1792         void *object = NULL;
1793         int available = 0;
1794         int objects;
1795 
1796         /*
1797          * Racy check. If we mistakenly see no partial slabs then we
1798          * just allocate an empty slab. If we mistakenly try to get a
1799          * partial slab and there is none available then get_partials()
1800          * will return NULL.
1801          */
1802         if (!n || !n->nr_partial)
1803                 return NULL;
1804 
1805         spin_lock(&n->list_lock);
1806         list_for_each_entry_safe(page, page2, &n->partial, lru) {
1807                 void *t;
1808 
1809                 if (!pfmemalloc_match(page, flags))
1810                         continue;
1811 
1812                 t = acquire_slab(s, n, page, object == NULL, &objects);
1813                 if (!t)
1814                         break;
1815 
1816                 available += objects;
1817                 if (!object) {
1818                         c->page = page;
1819                         stat(s, ALLOC_FROM_PARTIAL);
1820                         object = t;
1821                 } else {
1822                         put_cpu_partial(s, page, 0);
1823                         stat(s, CPU_PARTIAL_NODE);
1824                 }
1825                 if (!kmem_cache_has_cpu_partial(s)
1826                         || available > s->cpu_partial / 2)
1827                         break;
1828 
1829         }
1830         spin_unlock(&n->list_lock);
1831         return object;
1832 }
1833 
1834 /*
1835  * Get a page from somewhere. Search in increasing NUMA distances.
1836  */
1837 static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
1838                 struct kmem_cache_cpu *c)
1839 {
1840 #ifdef CONFIG_NUMA
1841         struct zonelist *zonelist;
1842         struct zoneref *z;
1843         struct zone *zone;
1844         enum zone_type high_zoneidx = gfp_zone(flags);
1845         void *object;
1846         unsigned int cpuset_mems_cookie;
1847 
1848         /*
1849          * The defrag ratio allows a configuration of the tradeoffs between
1850          * inter node defragmentation and node local allocations. A lower
1851          * defrag_ratio increases the tendency to do local allocations
1852          * instead of attempting to obtain partial slabs from other nodes.
1853          *
1854          * If the defrag_ratio is set to 0 then kmalloc() always
1855          * returns node local objects. If the ratio is higher then kmalloc()
1856          * may return off node objects because partial slabs are obtained
1857          * from other nodes and filled up.
1858          *
1859          * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
1860          * (which makes defrag_ratio = 1000) then every (well almost)
1861          * allocation will first attempt to defrag slab caches on other nodes.
1862          * This means scanning over all nodes to look for partial slabs which
1863          * may be expensive if we do it every time we are trying to find a slab
1864          * with available objects.
1865          */
1866         if (!s->remote_node_defrag_ratio ||
1867                         get_cycles() % 1024 > s->remote_node_defrag_ratio)
1868                 return NULL;
1869 
1870         do {
1871                 cpuset_mems_cookie = read_mems_allowed_begin();
1872                 zonelist = node_zonelist(mempolicy_slab_node(), flags);
1873                 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1874                         struct kmem_cache_node *n;
1875 
1876                         n = get_node(s, zone_to_nid(zone));
1877 
1878                         if (n && cpuset_zone_allowed(zone, flags) &&
1879                                         n->nr_partial > s->min_partial) {
1880                                 object = get_partial_node(s, n, c, flags);
1881                                 if (object) {
1882                                         /*
1883                                          * Don't check read_mems_allowed_retry()
1884                                          * here - if mems_allowed was updated in
1885                                          * parallel, that was a harmless race
1886                                          * between allocation and the cpuset
1887                                          * update
1888                                          */
1889                                         return object;
1890                                 }
1891                         }
1892                 }
1893         } while (read_mems_allowed_retry(cpuset_mems_cookie));
1894 #endif
1895         return NULL;
1896 }
1897 
1898 /*
1899  * Get a partial page, lock it and return it.
1900  */
1901 static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
1902                 struct kmem_cache_cpu *c)
1903 {
1904         void *object;
1905         int searchnode = node;
1906 
1907         if (node == NUMA_NO_NODE)
1908                 searchnode = numa_mem_id();
1909         else if (!node_present_pages(node))
1910                 searchnode = node_to_mem_node(node);
1911 
1912         object = get_partial_node(s, get_node(s, searchnode), c, flags);
1913         if (object || node != NUMA_NO_NODE)
1914                 return object;
1915 
1916         return get_any_partial(s, flags, c);
1917 }
1918 
1919 #ifdef CONFIG_PREEMPT
1920 /*
1921  * Calculate the next globally unique transaction for disambiguiation
1922  * during cmpxchg. The transactions start with the cpu number and are then
1923  * incremented by CONFIG_NR_CPUS.
1924  */
1925 #define TID_STEP  roundup_pow_of_two(CONFIG_NR_CPUS)
1926 #else
1927 /*
1928  * No preemption supported therefore also no need to check for
1929  * different cpus.
1930  */
1931 #define TID_STEP 1
1932 #endif
1933 
1934 static inline unsigned long next_tid(unsigned long tid)
1935 {
1936         return tid + TID_STEP;
1937 }
1938 
1939 static inline unsigned int tid_to_cpu(unsigned long tid)
1940 {
1941         return tid % TID_STEP;
1942 }
1943 
1944 static inline unsigned long tid_to_event(unsigned long tid)
1945 {
1946         return tid / TID_STEP;
1947 }
1948 
1949 static inline unsigned int init_tid(int cpu)
1950 {
1951         return cpu;
1952 }
1953 
1954 static inline void note_cmpxchg_failure(const char *n,
1955                 const struct kmem_cache *s, unsigned long tid)
1956 {
1957 #ifdef SLUB_DEBUG_CMPXCHG
1958         unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1959 
1960         pr_info("%s %s: cmpxchg redo ", n, s->name);
1961 
1962 #ifdef CONFIG_PREEMPT
1963         if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1964                 pr_warn("due to cpu change %d -> %d\n",
1965                         tid_to_cpu(tid), tid_to_cpu(actual_tid));
1966         else
1967 #endif
1968         if (tid_to_event(tid) != tid_to_event(actual_tid))
1969                 pr_warn("due to cpu running other code. Event %ld->%ld\n",
1970                         tid_to_event(tid), tid_to_event(actual_tid));
1971         else
1972                 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
1973                         actual_tid, tid, next_tid(tid));
1974 #endif
1975         stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
1976 }
1977 
1978 static void init_kmem_cache_cpus(struct kmem_cache *s)
1979 {
1980         int cpu;
1981 
1982         for_each_possible_cpu(cpu)
1983                 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
1984 }
1985 
1986 /*
1987  * Remove the cpu slab
1988  */
1989 static void deactivate_slab(struct kmem_cache *s, struct page *page,
1990                                 void *freelist)
1991 {
1992         enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
1993         struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1994         int lock = 0;
1995         enum slab_modes l = M_NONE, m = M_NONE;
1996         void *nextfree;
1997         int tail = DEACTIVATE_TO_HEAD;
1998         struct page new;
1999         struct page old;
2000 
2001         if (page->freelist) {
2002                 stat(s, DEACTIVATE_REMOTE_FREES);
2003                 tail = DEACTIVATE_TO_TAIL;
2004         }
2005 
2006         /*
2007          * Stage one: Free all available per cpu objects back
2008          * to the page freelist while it is still frozen. Leave the
2009          * last one.
2010          *
2011          * There is no need to take the list->lock because the page
2012          * is still frozen.
2013          */
2014         while (freelist && (nextfree = get_freepointer(s, freelist))) {
2015                 void *prior;
2016                 unsigned long counters;
2017 
2018                 do {
2019                         prior = page->freelist;
2020                         counters = page->counters;
2021                         set_freepointer(s, freelist, prior);
2022                         new.counters = counters;
2023                         new.inuse--;
2024                         VM_BUG_ON(!new.frozen);
2025 
2026                 } while (!__cmpxchg_double_slab(s, page,
2027                         prior, counters,
2028                         freelist, new.counters,
2029                         "drain percpu freelist"));
2030 
2031                 freelist = nextfree;
2032         }
2033 
2034         /*
2035          * Stage two: Ensure that the page is unfrozen while the
2036          * list presence reflects the actual number of objects
2037          * during unfreeze.
2038          *
2039          * We setup the list membership and then perform a cmpxchg
2040          * with the count. If there is a mismatch then the page
2041          * is not unfrozen but the page is on the wrong list.
2042          *
2043          * Then we restart the process which may have to remove
2044          * the page from the list that we just put it on again
2045          * because the number of objects in the slab may have
2046          * changed.
2047          */
2048 redo:
2049 
2050         old.freelist = page->freelist;
2051         old.counters = page->counters;
2052         VM_BUG_ON(!old.frozen);
2053 
2054         /* Determine target state of the slab */
2055         new.counters = old.counters;
2056         if (freelist) {
2057                 new.inuse--;
2058                 set_freepointer(s, freelist, old.freelist);
2059                 new.freelist = freelist;
2060         } else
2061                 new.freelist = old.freelist;
2062 
2063         new.frozen = 0;
2064 
2065         if (!new.inuse && n->nr_partial >= s->min_partial)
2066                 m = M_FREE;
2067         else if (new.freelist) {
2068                 m = M_PARTIAL;
2069                 if (!lock) {
2070                         lock = 1;
2071                         /*
2072                          * Taking the spinlock removes the possiblity
2073                          * that acquire_slab() will see a slab page that
2074                          * is frozen
2075                          */
2076                         spin_lock(&n->list_lock);
2077                 }
2078         } else {
2079                 m = M_FULL;
2080                 if (kmem_cache_debug(s) && !lock) {
2081                         lock = 1;
2082                         /*
2083                          * This also ensures that the scanning of full
2084                          * slabs from diagnostic functions will not see
2085                          * any frozen slabs.
2086                          */
2087                         spin_lock(&n->list_lock);
2088                 }
2089         }
2090 
2091         if (l != m) {
2092 
2093                 if (l == M_PARTIAL)
2094 
2095                         remove_partial(n, page);
2096 
2097                 else if (l == M_FULL)
2098 
2099                         remove_full(s, n, page);
2100 
2101                 if (m == M_PARTIAL) {
2102 
2103                         add_partial(n, page, tail);
2104                         stat(s, tail);
2105 
2106                 } else if (m == M_FULL) {
2107 
2108                         stat(s, DEACTIVATE_FULL);
2109                         add_full(s, n, page);
2110 
2111                 }
2112         }
2113 
2114         l = m;
2115         if (!__cmpxchg_double_slab(s, page,
2116                                 old.freelist, old.counters,
2117                                 new.freelist, new.counters,
2118                                 "unfreezing slab"))
2119                 goto redo;
2120 
2121         if (lock)
2122                 spin_unlock(&n->list_lock);
2123 
2124         if (m == M_FREE) {
2125                 stat(s, DEACTIVATE_EMPTY);
2126                 discard_slab(s, page);
2127                 stat(s, FREE_SLAB);
2128         }
2129 }
2130 
2131 /*
2132  * Unfreeze all the cpu partial slabs.
2133  *
2134  * This function must be called with interrupts disabled
2135  * for the cpu using c (or some other guarantee must be there
2136  * to guarantee no concurrent accesses).
2137  */
2138 static void unfreeze_partials(struct kmem_cache *s,
2139                 struct kmem_cache_cpu *c)
2140 {
2141 #ifdef CONFIG_SLUB_CPU_PARTIAL
2142         struct kmem_cache_node *n = NULL, *n2 = NULL;
2143         struct page *page, *discard_page = NULL;
2144 
2145         while ((page = c->partial)) {
2146                 struct page new;
2147                 struct page old;
2148 
2149                 c->partial = page->next;
2150 
2151                 n2 = get_node(s, page_to_nid(page));
2152                 if (n != n2) {
2153                         if (n)
2154                                 spin_unlock(&n->list_lock);
2155 
2156                         n = n2;
2157                         spin_lock(&n->list_lock);
2158                 }
2159 
2160                 do {
2161 
2162                         old.freelist = page->freelist;
2163                         old.counters = page->counters;
2164                         VM_BUG_ON(!old.frozen);
2165 
2166                         new.counters = old.counters;
2167                         new.freelist = old.freelist;
2168 
2169                         new.frozen = 0;
2170 
2171                 } while (!__cmpxchg_double_slab(s, page,
2172                                 old.freelist, old.counters,
2173                                 new.freelist, new.counters,
2174                                 "unfreezing slab"));
2175 
2176                 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
2177                         page->next = discard_page;
2178                         discard_page = page;
2179                 } else {
2180                         add_partial(n, page, DEACTIVATE_TO_TAIL);
2181                         stat(s, FREE_ADD_PARTIAL);
2182                 }
2183         }
2184 
2185         if (n)
2186                 spin_unlock(&n->list_lock);
2187 
2188         while (discard_page) {
2189                 page = discard_page;
2190                 discard_page = discard_page->next;
2191 
2192                 stat(s, DEACTIVATE_EMPTY);
2193                 discard_slab(s, page);
2194                 stat(s, FREE_SLAB);
2195         }
2196 #endif
2197 }
2198 
2199 /*
2200  * Put a page that was just frozen (in __slab_free) into a partial page
2201  * slot if available. This is done without interrupts disabled and without
2202  * preemption disabled. The cmpxchg is racy and may put the partial page
2203  * onto a random cpus partial slot.
2204  *
2205  * If we did not find a slot then simply move all the partials to the
2206  * per node partial list.
2207  */
2208 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
2209 {
2210 #ifdef CONFIG_SLUB_CPU_PARTIAL
2211         struct page *oldpage;
2212         int pages;
2213         int pobjects;
2214 
2215         preempt_disable();
2216         do {
2217                 pages = 0;
2218                 pobjects = 0;
2219                 oldpage = this_cpu_read(s->cpu_slab->partial);
2220 
2221                 if (oldpage) {
2222                         pobjects = oldpage->pobjects;
2223                         pages = oldpage->pages;
2224                         if (drain && pobjects > s->cpu_partial) {
2225                                 unsigned long flags;
2226                                 /*
2227                                  * partial array is full. Move the existing
2228                                  * set to the per node partial list.
2229                                  */
2230                                 local_irq_save(flags);
2231                                 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2232                                 local_irq_restore(flags);
2233                                 oldpage = NULL;
2234                                 pobjects = 0;
2235                                 pages = 0;
2236                                 stat(s, CPU_PARTIAL_DRAIN);
2237                         }
2238                 }
2239 
2240                 pages++;
2241                 pobjects += page->objects - page->inuse;
2242 
2243                 page->pages = pages;
2244                 page->pobjects = pobjects;
2245                 page->next = oldpage;
2246 
2247         } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2248                                                                 != oldpage);
2249         if (unlikely(!s->cpu_partial)) {
2250                 unsigned long flags;
2251 
2252                 local_irq_save(flags);
2253                 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2254                 local_irq_restore(flags);
2255         }
2256         preempt_enable();
2257 #endif
2258 }
2259 
2260 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
2261 {
2262         stat(s, CPUSLAB_FLUSH);
2263         deactivate_slab(s, c->page, c->freelist);
2264 
2265         c->tid = next_tid(c->tid);
2266         c->page = NULL;
2267         c->freelist = NULL;
2268 }
2269 
2270 /*
2271  * Flush cpu slab.
2272  *
2273  * Called from IPI handler with interrupts disabled.
2274  */
2275 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
2276 {
2277         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2278 
2279         if (likely(c)) {
2280                 if (c->page)
2281                         flush_slab(s, c);
2282 
2283                 unfreeze_partials(s, c);
2284         }
2285 }
2286 
2287 static void flush_cpu_slab(void *d)
2288 {
2289         struct kmem_cache *s = d;
2290 
2291         __flush_cpu_slab(s, smp_processor_id());
2292 }
2293 
2294 static bool has_cpu_slab(int cpu, void *info)
2295 {
2296         struct kmem_cache *s = info;
2297         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2298 
2299         return c->page || c->partial;
2300 }
2301 
2302 static void flush_all(struct kmem_cache *s)
2303 {
2304         on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
2305 }
2306 
2307 /*
2308  * Check if the objects in a per cpu structure fit numa
2309  * locality expectations.
2310  */
2311 static inline int node_match(struct page *page, int node)
2312 {
2313 #ifdef CONFIG_NUMA
2314         if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node))
2315                 return 0;
2316 #endif
2317         return 1;
2318 }
2319 
2320 #ifdef CONFIG_SLUB_DEBUG
2321 static int count_free(struct page *page)
2322 {
2323         return page->objects - page->inuse;
2324 }
2325 
2326 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2327 {
2328         return atomic_long_read(&n->total_objects);
2329 }
2330 #endif /* CONFIG_SLUB_DEBUG */
2331 
2332 #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
2333 static unsigned long count_partial(struct kmem_cache_node *n,
2334                                         int (*get_count)(struct page *))
2335 {
2336         unsigned long flags;
2337         unsigned long x = 0;
2338         struct page *page;
2339 
2340         spin_lock_irqsave(&n->list_lock, flags);
2341         list_for_each_entry(page, &n->partial, lru)
2342                 x += get_count(page);
2343         spin_unlock_irqrestore(&n->list_lock, flags);
2344         return x;
2345 }
2346 #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
2347 
2348 static noinline void
2349 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2350 {
2351 #ifdef CONFIG_SLUB_DEBUG
2352         static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2353                                       DEFAULT_RATELIMIT_BURST);
2354         int node;
2355         struct kmem_cache_node *n;
2356 
2357         if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2358                 return;
2359 
2360         pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2361                 nid, gfpflags, &gfpflags);
2362         pr_warn("  cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n",
2363                 s->name, s->object_size, s->size, oo_order(s->oo),
2364                 oo_order(s->min));
2365 
2366         if (oo_order(s->min) > get_order(s->object_size))
2367                 pr_warn("  %s debugging increased min order, use slub_debug=O to disable.\n",
2368                         s->name);
2369 
2370         for_each_kmem_cache_node(s, node, n) {
2371                 unsigned long nr_slabs;
2372                 unsigned long nr_objs;
2373                 unsigned long nr_free;
2374 
2375                 nr_free  = count_partial(n, count_free);
2376                 nr_slabs = node_nr_slabs(n);
2377                 nr_objs  = node_nr_objs(n);
2378 
2379                 pr_warn("  node %d: slabs: %ld, objs: %ld, free: %ld\n",
2380                         node, nr_slabs, nr_objs, nr_free);
2381         }
2382 #endif
2383 }
2384 
2385 static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2386                         int node, struct kmem_cache_cpu **pc)
2387 {
2388         void *freelist;
2389         struct kmem_cache_cpu *c = *pc;
2390         struct page *page;
2391 
2392         freelist = get_partial(s, flags, node, c);
2393 
2394         if (freelist)
2395                 return freelist;
2396 
2397         page = new_slab(s, flags, node);
2398         if (page) {
2399                 c = raw_cpu_ptr(s->cpu_slab);
2400                 if (c->page)
2401                         flush_slab(s, c);
2402 
2403                 /*
2404                  * No other reference to the page yet so we can
2405                  * muck around with it freely without cmpxchg
2406                  */
2407                 freelist = page->freelist;
2408                 page->freelist = NULL;
2409 
2410                 stat(s, ALLOC_SLAB);
2411                 c->page = page;
2412                 *pc = c;
2413         } else
2414                 freelist = NULL;
2415 
2416         return freelist;
2417 }
2418 
2419 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2420 {
2421         if (unlikely(PageSlabPfmemalloc(page)))
2422                 return gfp_pfmemalloc_allowed(gfpflags);
2423 
2424         return true;
2425 }
2426 
2427 /*
2428  * Check the page->freelist of a page and either transfer the freelist to the
2429  * per cpu freelist or deactivate the page.
2430  *
2431  * The page is still frozen if the return value is not NULL.
2432  *
2433  * If this function returns NULL then the page has been unfrozen.
2434  *
2435  * This function must be called with interrupt disabled.
2436  */
2437 static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2438 {
2439         struct page new;
2440         unsigned long counters;
2441         void *freelist;
2442 
2443         do {
2444                 freelist = page->freelist;
2445                 counters = page->counters;
2446 
2447                 new.counters = counters;
2448                 VM_BUG_ON(!new.frozen);
2449 
2450                 new.inuse = page->objects;
2451                 new.frozen = freelist != NULL;
2452 
2453         } while (!__cmpxchg_double_slab(s, page,
2454                 freelist, counters,
2455                 NULL, new.counters,
2456                 "get_freelist"));
2457 
2458         return freelist;
2459 }
2460 
2461 /*
2462  * Slow path. The lockless freelist is empty or we need to perform
2463  * debugging duties.
2464  *
2465  * Processing is still very fast if new objects have been freed to the
2466  * regular freelist. In that case we simply take over the regular freelist
2467  * as the lockless freelist and zap the regular freelist.
2468  *
2469  * If that is not working then we fall back to the partial lists. We take the
2470  * first element of the freelist as the object to allocate now and move the
2471  * rest of the freelist to the lockless freelist.
2472  *
2473  * And if we were unable to get a new slab from the partial slab lists then
2474  * we need to allocate a new slab. This is the slowest path since it involves
2475  * a call to the page allocator and the setup of a new slab.
2476  *
2477  * Version of __slab_alloc to use when we know that interrupts are
2478  * already disabled (which is the case for bulk allocation).
2479  */
2480 static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2481                           unsigned long addr, struct kmem_cache_cpu *c)
2482 {
2483         void *freelist;
2484         struct page *page;
2485 
2486         page = c->page;
2487         if (!page)
2488                 goto new_slab;
2489 redo:
2490 
2491         if (unlikely(!node_match(page, node))) {
2492                 int searchnode = node;
2493 
2494                 if (node != NUMA_NO_NODE && !node_present_pages(node))
2495                         searchnode = node_to_mem_node(node);
2496 
2497                 if (unlikely(!node_match(page, searchnode))) {
2498                         stat(s, ALLOC_NODE_MISMATCH);
2499                         deactivate_slab(s, page, c->freelist);
2500                         c->page = NULL;
2501                         c->freelist = NULL;
2502                         goto new_slab;
2503                 }
2504         }
2505 
2506         /*
2507          * By rights, we should be searching for a slab page that was
2508          * PFMEMALLOC but right now, we are losing the pfmemalloc
2509          * information when the page leaves the per-cpu allocator
2510          */
2511         if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2512                 deactivate_slab(s, page, c->freelist);
2513                 c->page = NULL;
2514                 c->freelist = NULL;
2515                 goto new_slab;
2516         }
2517 
2518         /* must check again c->freelist in case of cpu migration or IRQ */
2519         freelist = c->freelist;
2520         if (freelist)
2521                 goto load_freelist;
2522 
2523         freelist = get_freelist(s, page);
2524 
2525         if (!freelist) {
2526                 c->page = NULL;
2527                 stat(s, DEACTIVATE_BYPASS);
2528                 goto new_slab;
2529         }
2530 
2531         stat(s, ALLOC_REFILL);
2532 
2533 load_freelist:
2534         /*
2535          * freelist is pointing to the list of objects to be used.
2536          * page is pointing to the page from which the objects are obtained.
2537          * That page must be frozen for per cpu allocations to work.
2538          */
2539         VM_BUG_ON(!c->page->frozen);
2540         c->freelist = get_freepointer(s, freelist);
2541         c->tid = next_tid(c->tid);
2542         return freelist;
2543 
2544 new_slab:
2545 
2546         if (c->partial) {
2547                 page = c->page = c->partial;
2548                 c->partial = page->next;
2549                 stat(s, CPU_PARTIAL_ALLOC);
2550                 c->freelist = NULL;
2551                 goto redo;
2552         }
2553 
2554         freelist = new_slab_objects(s, gfpflags, node, &c);
2555 
2556         if (unlikely(!freelist)) {
2557                 slab_out_of_memory(s, gfpflags, node);
2558                 return NULL;
2559         }
2560 
2561         page = c->page;
2562         if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
2563                 goto load_freelist;
2564 
2565         /* Only entered in the debug case */
2566         if (kmem_cache_debug(s) &&
2567                         !alloc_debug_processing(s, page, freelist, addr))
2568                 goto new_slab;  /* Slab failed checks. Next slab needed */
2569 
2570         deactivate_slab(s, page, get_freepointer(s, freelist));
2571         c->page = NULL;
2572         c->freelist = NULL;
2573         return freelist;
2574 }
2575 
2576 /*
2577  * Another one that disabled interrupt and compensates for possible
2578  * cpu changes by refetching the per cpu area pointer.
2579  */
2580 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2581                           unsigned long addr, struct kmem_cache_cpu *c)
2582 {
2583         void *p;
2584         unsigned long flags;
2585 
2586         local_irq_save(flags);
2587 #ifdef CONFIG_PREEMPT
2588         /*
2589          * We may have been preempted and rescheduled on a different
2590          * cpu before disabling interrupts. Need to reload cpu area
2591          * pointer.
2592          */
2593         c = this_cpu_ptr(s->cpu_slab);
2594 #endif
2595 
2596         p = ___slab_alloc(s, gfpflags, node, addr, c);
2597         local_irq_restore(flags);
2598         return p;
2599 }
2600 
2601 /*
2602  * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2603  * have the fastpath folded into their functions. So no function call
2604  * overhead for requests that can be satisfied on the fastpath.
2605  *
2606  * The fastpath works by first checking if the lockless freelist can be used.
2607  * If not then __slab_alloc is called for slow processing.
2608  *
2609  * Otherwise we can simply pick the next object from the lockless free list.
2610  */
2611 static __always_inline void *slab_alloc_node(struct kmem_cache *s,
2612                 gfp_t gfpflags, int node, unsigned long addr)
2613 {
2614         void *object;
2615         struct kmem_cache_cpu *c;
2616         struct page *page;
2617         unsigned long tid;
2618 
2619         s = slab_pre_alloc_hook(s, gfpflags);
2620         if (!s)
2621                 return NULL;
2622 redo:
2623         /*
2624          * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2625          * enabled. We may switch back and forth between cpus while
2626          * reading from one cpu area. That does not matter as long
2627          * as we end up on the original cpu again when doing the cmpxchg.
2628          *
2629          * We should guarantee that tid and kmem_cache are retrieved on
2630          * the same cpu. It could be different if CONFIG_PREEMPT so we need
2631          * to check if it is matched or not.
2632          */
2633         do {
2634                 tid = this_cpu_read(s->cpu_slab->tid);
2635                 c = raw_cpu_ptr(s->cpu_slab);
2636         } while (IS_ENABLED(CONFIG_PREEMPT) &&
2637                  unlikely(tid != READ_ONCE(c->tid)));
2638 
2639         /*
2640          * Irqless object alloc/free algorithm used here depends on sequence
2641          * of fetching cpu_slab's data. tid should be fetched before anything
2642          * on c to guarantee that object and page associated with previous tid
2643          * won't be used with current tid. If we fetch tid first, object and
2644          * page could be one associated with next tid and our alloc/free
2645          * request will be failed. In this case, we will retry. So, no problem.
2646          */
2647         barrier();
2648 
2649         /*
2650          * The transaction ids are globally unique per cpu and per operation on
2651          * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2652          * occurs on the right processor and that there was no operation on the
2653          * linked list in between.
2654          */
2655 
2656         object = c->freelist;
2657         page = c->page;
2658         if (unlikely(!object || !node_match(page, node))) {
2659                 object = __slab_alloc(s, gfpflags, node, addr, c);
2660                 stat(s, ALLOC_SLOWPATH);
2661         } else {
2662                 void *next_object = get_freepointer_safe(s, object);
2663 
2664                 /*
2665                  * The cmpxchg will only match if there was no additional
2666                  * operation and if we are on the right processor.
2667                  *
2668                  * The cmpxchg does the following atomically (without lock
2669                  * semantics!)
2670                  * 1. Relocate first pointer to the current per cpu area.
2671                  * 2. Verify that tid and freelist have not been changed
2672                  * 3. If they were not changed replace tid and freelist
2673                  *
2674                  * Since this is without lock semantics the protection is only
2675                  * against code executing on this cpu *not* from access by
2676                  * other cpus.
2677                  */
2678                 if (unlikely(!this_cpu_cmpxchg_double(
2679                                 s->cpu_slab->freelist, s->cpu_slab->tid,
2680                                 object, tid,
2681                                 next_object, next_tid(tid)))) {
2682 
2683                         note_cmpxchg_failure("slab_alloc", s, tid);
2684                         goto redo;
2685                 }
2686                 prefetch_freepointer(s, next_object);
2687                 stat(s, ALLOC_FASTPATH);
2688         }
2689 
2690         if (unlikely(gfpflags & __GFP_ZERO) && object)
2691                 memset(object, 0, s->object_size);
2692 
2693         slab_post_alloc_hook(s, gfpflags, 1, &object);
2694 
2695         return object;
2696 }
2697 
2698 static __always_inline void *slab_alloc(struct kmem_cache *s,
2699                 gfp_t gfpflags, unsigned long addr)
2700 {
2701         return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2702 }
2703 
2704 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2705 {
2706         void *ret = slab_alloc(s, gfpflags, _RET_IP_);
2707 
2708         trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2709                                 s->size, gfpflags);
2710 
2711         return ret;
2712 }
2713 EXPORT_SYMBOL(kmem_cache_alloc);
2714 
2715 #ifdef CONFIG_TRACING
2716 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
2717 {
2718         void *ret = slab_alloc(s, gfpflags, _RET_IP_);
2719         trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2720         kasan_kmalloc(s, ret, size, gfpflags);
2721         return ret;
2722 }
2723 EXPORT_SYMBOL(kmem_cache_alloc_trace);
2724 #endif
2725 
2726 #ifdef CONFIG_NUMA
2727 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2728 {
2729         void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
2730 
2731         trace_kmem_cache_alloc_node(_RET_IP_, ret,
2732                                     s->object_size, s->size, gfpflags, node);
2733 
2734         return ret;
2735 }
2736 EXPORT_SYMBOL(kmem_cache_alloc_node);
2737 
2738 #ifdef CONFIG_TRACING
2739 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
2740                                     gfp_t gfpflags,
2741                                     int node, size_t size)
2742 {
2743         void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
2744 
2745         trace_kmalloc_node(_RET_IP_, ret,
2746                            size, s->size, gfpflags, node);
2747 
2748         kasan_kmalloc(s, ret, size, gfpflags);
2749         return ret;
2750 }
2751 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
2752 #endif
2753 #endif
2754 
2755 /*
2756  * Slow path handling. This may still be called frequently since objects
2757  * have a longer lifetime than the cpu slabs in most processing loads.
2758  *
2759  * So we still attempt to reduce cache line usage. Just take the slab
2760  * lock and free the item. If there is no additional partial page
2761  * handling required then we can return immediately.
2762  */
2763 static void __slab_free(struct kmem_cache *s, struct page *page,
2764                         void *head, void *tail, int cnt,
2765                         unsigned long addr)
2766 
2767 {
2768         void *prior;
2769         int was_frozen;
2770         struct page new;
2771         unsigned long counters;
2772         struct kmem_cache_node *n = NULL;
2773         unsigned long uninitialized_var(flags);
2774 
2775         stat(s, FREE_SLOWPATH);
2776 
2777         if (kmem_cache_debug(s) &&
2778             !free_debug_processing(s, page, head, tail, cnt, addr))
2779                 return;
2780 
2781         do {
2782                 if (unlikely(n)) {
2783                         spin_unlock_irqrestore(&n->list_lock, flags);
2784                         n = NULL;
2785                 }
2786                 prior = page->freelist;
2787                 counters = page->counters;
2788                 set_freepointer(s, tail, prior);
2789                 new.counters = counters;
2790                 was_frozen = new.frozen;
2791                 new.inuse -= cnt;
2792                 if ((!new.inuse || !prior) && !was_frozen) {
2793 
2794                         if (kmem_cache_has_cpu_partial(s) && !prior) {
2795 
2796                                 /*
2797                                  * Slab was on no list before and will be
2798                                  * partially empty
2799                                  * We can defer the list move and instead
2800                                  * freeze it.
2801                                  */
2802                                 new.frozen = 1;
2803 
2804                         } else { /* Needs to be taken off a list */
2805 
2806                                 n = get_node(s, page_to_nid(page));
2807                                 /*
2808                                  * Speculatively acquire the list_lock.
2809                                  * If the cmpxchg does not succeed then we may
2810                                  * drop the list_lock without any processing.
2811                                  *
2812                                  * Otherwise the list_lock will synchronize with
2813                                  * other processors updating the list of slabs.
2814                                  */
2815                                 spin_lock_irqsave(&n->list_lock, flags);
2816 
2817                         }
2818                 }
2819 
2820         } while (!cmpxchg_double_slab(s, page,
2821                 prior, counters,
2822                 head, new.counters,
2823                 "__slab_free"));
2824 
2825         if (likely(!n)) {
2826 
2827                 /*
2828                  * If we just froze the page then put it onto the
2829                  * per cpu partial list.
2830                  */
2831                 if (new.frozen && !was_frozen) {
2832                         put_cpu_partial(s, page, 1);
2833                         stat(s, CPU_PARTIAL_FREE);
2834                 }
2835                 /*
2836                  * The list lock was not taken therefore no list
2837                  * activity can be necessary.
2838                  */
2839                 if (was_frozen)
2840                         stat(s, FREE_FROZEN);
2841                 return;
2842         }
2843 
2844         if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
2845                 goto slab_empty;
2846 
2847         /*
2848          * Objects left in the slab. If it was not on the partial list before
2849          * then add it.
2850          */
2851         if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2852                 if (kmem_cache_debug(s))
2853                         remove_full(s, n, page);
2854                 add_partial(n, page, DEACTIVATE_TO_TAIL);
2855                 stat(s, FREE_ADD_PARTIAL);
2856         }
2857         spin_unlock_irqrestore(&n->list_lock, flags);
2858         return;
2859 
2860 slab_empty:
2861         if (prior) {
2862                 /*
2863                  * Slab on the partial list.
2864                  */
2865                 remove_partial(n, page);
2866                 stat(s, FREE_REMOVE_PARTIAL);
2867         } else {
2868                 /* Slab must be on the full list */
2869                 remove_full(s, n, page);
2870         }
2871 
2872         spin_unlock_irqrestore(&n->list_lock, flags);
2873         stat(s, FREE_SLAB);
2874         discard_slab(s, page);
2875 }
2876 
2877 /*
2878  * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2879  * can perform fastpath freeing without additional function calls.
2880  *
2881  * The fastpath is only possible if we are freeing to the current cpu slab
2882  * of this processor. This typically the case if we have just allocated
2883  * the item before.
2884  *
2885  * If fastpath is not possible then fall back to __slab_free where we deal
2886  * with all sorts of special processing.
2887  *
2888  * Bulk free of a freelist with several objects (all pointing to the
2889  * same page) possible by specifying head and tail ptr, plus objects
2890  * count (cnt). Bulk free indicated by tail pointer being set.
2891  */
2892 static __always_inline void do_slab_free(struct kmem_cache *s,
2893                                 struct page *page, void *head, void *tail,
2894                                 int cnt, unsigned long addr)
2895 {
2896         void *tail_obj = tail ? : head;
2897         struct kmem_cache_cpu *c;
2898         unsigned long tid;
2899 redo:
2900         /*
2901          * Determine the currently cpus per cpu slab.
2902          * The cpu may change afterward. However that does not matter since
2903          * data is retrieved via this pointer. If we are on the same cpu
2904          * during the cmpxchg then the free will succeed.
2905          */
2906         do {
2907                 tid = this_cpu_read(s->cpu_slab->tid);
2908                 c = raw_cpu_ptr(s->cpu_slab);
2909         } while (IS_ENABLED(CONFIG_PREEMPT) &&
2910                  unlikely(tid != READ_ONCE(c->tid)));
2911 
2912         /* Same with comment on barrier() in slab_alloc_node() */
2913         barrier();
2914 
2915         if (likely(page == c->page)) {
2916                 set_freepointer(s, tail_obj, c->freelist);
2917 
2918                 if (unlikely(!this_cpu_cmpxchg_double(
2919                                 s->cpu_slab->freelist, s->cpu_slab->tid,
2920                                 c->freelist, tid,
2921                                 head, next_tid(tid)))) {
2922 
2923                         note_cmpxchg_failure("slab_free", s, tid);
2924                         goto redo;
2925                 }
2926                 stat(s, FREE_FASTPATH);
2927         } else
2928                 __slab_free(s, page, head, tail_obj, cnt, addr);
2929 
2930 }
2931 
2932 static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
2933                                       void *head, void *tail, int cnt,
2934                                       unsigned long addr)
2935 {
2936         slab_free_freelist_hook(s, head, tail);
2937         /*
2938          * slab_free_freelist_hook() could have put the items into quarantine.
2939          * If so, no need to free them.
2940          */
2941         if (s->flags & SLAB_KASAN && !(s->flags & SLAB_DESTROY_BY_RCU))
2942                 return;
2943         do_slab_free(s, page, head, tail, cnt, addr);
2944 }
2945 
2946 #ifdef CONFIG_KASAN
2947 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
2948 {
2949         do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
2950 }
2951 #endif
2952 
2953 void kmem_cache_free(struct kmem_cache *s, void *x)
2954 {
2955         s = cache_from_obj(s, x);
2956         if (!s)
2957                 return;
2958         slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
2959         trace_kmem_cache_free(_RET_IP_, x);
2960 }
2961 EXPORT_SYMBOL(kmem_cache_free);
2962 
2963 struct detached_freelist {
2964         struct page *page;
2965         void *tail;
2966         void *freelist;
2967         int cnt;
2968         struct kmem_cache *s;
2969 };
2970 
2971 /*
2972  * This function progressively scans the array with free objects (with
2973  * a limited look ahead) and extract objects belonging to the same
2974  * page.  It builds a detached freelist directly within the given
2975  * page/objects.  This can happen without any need for
2976  * synchronization, because the objects are owned by running process.
2977  * The freelist is build up as a single linked list in the objects.
2978  * The idea is, that this detached freelist can then be bulk
2979  * transferred to the real freelist(s), but only requiring a single
2980  * synchronization primitive.  Look ahead in the array is limited due
2981  * to performance reasons.
2982  */
2983 static inline
2984 int build_detached_freelist(struct kmem_cache *s, size_t size,
2985                             void **p, struct detached_freelist *df)
2986 {
2987         size_t first_skipped_index = 0;
2988         int lookahead = 3;
2989         void *object;
2990         struct page *page;
2991 
2992         /* Always re-init detached_freelist */
2993         df->page = NULL;
2994 
2995         do {
2996                 object = p[--size];
2997                 /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
2998         } while (!object && size);
2999 
3000         if (!object)
3001                 return 0;
3002 
3003         page = virt_to_head_page(object);
3004         if (!s) {
3005                 /* Handle kalloc'ed objects */
3006                 if (unlikely(!PageSlab(page))) {
3007                         BUG_ON(!PageCompound(page));
3008                         kfree_hook(object);
3009                         __free_pages(page, compound_order(page));
3010                         p[size] = NULL; /* mark object processed */
3011                         return size;
3012                 }
3013                 /* Derive kmem_cache from object */
3014                 df->s = page->slab_cache;
3015         } else {
3016                 df->s = cache_from_obj(s, object); /* Support for memcg */
3017         }
3018 
3019         /* Start new detached freelist */
3020         df->page = page;
3021         set_freepointer(df->s, object, NULL);
3022         df->tail = object;
3023         df->freelist = object;
3024         p[size] = NULL; /* mark object processed */
3025         df->cnt = 1;
3026 
3027         while (size) {
3028                 object = p[--size];
3029                 if (!object)
3030                         continue; /* Skip processed objects */
3031 
3032                 /* df->page is always set at this point */
3033                 if (df->page == virt_to_head_page(object)) {
3034                         /* Opportunity build freelist */
3035                         set_freepointer(df->s, object, df->freelist);
3036                         df->freelist = object;
3037                         df->cnt++;
3038                         p[size] = NULL; /* mark object processed */
3039 
3040                         continue;
3041                 }
3042 
3043                 /* Limit look ahead search */
3044                 if (!--lookahead)
3045                         break;
3046 
3047                 if (!first_skipped_index)
3048                         first_skipped_index = size + 1;
3049         }
3050 
3051         return first_skipped_index;
3052 }
3053 
3054 /* Note that interrupts must be enabled when calling this function. */
3055 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
3056 {
3057         if (WARN_ON(!size))
3058                 return;
3059 
3060         do {
3061                 struct detached_freelist df;
3062 
3063                 size = build_detached_freelist(s, size, p, &df);
3064                 if (unlikely(!df.page))
3065                         continue;
3066 
3067                 slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_);
3068         } while (likely(size));
3069 }
3070 EXPORT_SYMBOL(kmem_cache_free_bulk);
3071 
3072 /* Note that interrupts must be enabled when calling this function. */
3073 int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3074                           void **p)
3075 {
3076         struct kmem_cache_cpu *c;
3077         int i;
3078 
3079         /* memcg and kmem_cache debug support */
3080         s = slab_pre_alloc_hook(s, flags);
3081         if (unlikely(!s))
3082                 return false;
3083         /*
3084          * Drain objects in the per cpu slab, while disabling local
3085          * IRQs, which protects against PREEMPT and interrupts
3086          * handlers invoking normal fastpath.
3087          */
3088         local_irq_disable();
3089         c = this_cpu_ptr(s->cpu_slab);
3090 
3091         for (i = 0; i < size; i++) {
3092                 void *object = c->freelist;
3093 
3094                 if (unlikely(!object)) {
3095                         /*
3096                          * Invoking slow path likely have side-effect
3097                          * of re-populating per CPU c->freelist
3098                          */
3099                         p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
3100                                             _RET_IP_, c);
3101                         if (unlikely(!p[i]))
3102                                 goto error;
3103 
3104                         c = this_cpu_ptr(s->cpu_slab);
3105                         continue; /* goto for-loop */
3106                 }
3107                 c->freelist = get_freepointer(s, object);
3108                 p[i] = object;
3109         }
3110         c->tid = next_tid(c->tid);
3111         local_irq_enable();
3112 
3113         /* Clear memory outside IRQ disabled fastpath loop */
3114         if (unlikely(flags & __GFP_ZERO)) {
3115                 int j;
3116 
3117                 for (j = 0; j < i; j++)
3118                         memset(p[j], 0, s->object_size);
3119         }
3120 
3121         /* memcg and kmem_cache debug support */
3122         slab_post_alloc_hook(s, flags, size, p);
3123         return i;
3124 error:
3125         local_irq_enable();
3126         slab_post_alloc_hook(s, flags, i, p);
3127         __kmem_cache_free_bulk(s, i, p);
3128         return 0;
3129 }
3130 EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3131 
3132 
3133 /*
3134  * Object placement in a slab is made very easy because we always start at
3135  * offset 0. If we tune the size of the object to the alignment then we can
3136  * get the required alignment by putting one properly sized object after
3137  * another.
3138  *
3139  * Notice that the allocation order determines the sizes of the per cpu
3140  * caches. Each processor has always one slab available for allocations.
3141  * Increasing the allocation order reduces the number of times that slabs
3142  * must be moved on and off the partial lists and is therefore a factor in
3143  * locking overhead.
3144  */
3145 
3146 /*
3147  * Mininum / Maximum order of slab pages. This influences locking overhead
3148  * and slab fragmentation. A higher order reduces the number of partial slabs
3149  * and increases the number of allocations possible without having to
3150  * take the list_lock.
3151  */
3152 static int slub_min_order;
3153 static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
3154 static int slub_min_objects;
3155 
3156 /*
3157  * Calculate the order of allocation given an slab object size.
3158  *
3159  * The order of allocation has significant impact on performance and other
3160  * system components. Generally order 0 allocations should be preferred since
3161  * order 0 does not cause fragmentation in the page allocator. Larger objects
3162  * be problematic to put into order 0 slabs because there may be too much
3163  * unused space left. We go to a higher order if more than 1/16th of the slab
3164  * would be wasted.
3165  *
3166  * In order to reach satisfactory performance we must ensure that a minimum
3167  * number of objects is in one slab. Otherwise we may generate too much
3168  * activity on the partial lists which requires taking the list_lock. This is
3169  * less a concern for large slabs though which are rarely used.
3170  *
3171  * slub_max_order specifies the order where we begin to stop considering the
3172  * number of objects in a slab as critical. If we reach slub_max_order then
3173  * we try to keep the page order as low as possible. So we accept more waste
3174  * of space in favor of a small page order.
3175  *
3176  * Higher order allocations also allow the placement of more objects in a
3177  * slab and thereby reduce object handling overhead. If the user has
3178  * requested a higher mininum order then we start with that one instead of
3179  * the smallest order which will fit the object.
3180  */
3181 static inline int slab_order(int size, int min_objects,
3182                                 int max_order, int fract_leftover, int reserved)
3183 {
3184         int order;
3185         int rem;
3186         int min_order = slub_min_order;
3187 
3188         if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
3189                 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
3190 
3191         for (order = max(min_order, get_order(min_objects * size + reserved));
3192                         order <= max_order; order++) {
3193 
3194                 unsigned long slab_size = PAGE_SIZE << order;
3195 
3196                 rem = (slab_size - reserved) % size;
3197 
3198                 if (rem <= slab_size / fract_leftover)
3199                         break;
3200         }
3201 
3202         return order;
3203 }
3204 
3205 static inline int calculate_order(int size, int reserved)
3206 {
3207         int order;
3208         int min_objects;
3209         int fraction;
3210         int max_objects;
3211 
3212         /*
3213          * Attempt to find best configuration for a slab. This
3214          * works by first attempting to generate a layout with
3215          * the best configuration and backing off gradually.
3216          *
3217          * First we increase the acceptable waste in a slab. Then
3218          * we reduce the minimum objects required in a slab.
3219          */
3220         min_objects = slub_min_objects;
3221         if (!min_objects)
3222                 min_objects = 4 * (fls(nr_cpu_ids) + 1);
3223         max_objects = order_objects(slub_max_order, size, reserved);
3224         min_objects = min(min_objects, max_objects);
3225 
3226         while (min_objects > 1) {
3227                 fraction = 16;
3228                 while (fraction >= 4) {
3229                         order = slab_order(size, min_objects,
3230                                         slub_max_order, fraction, reserved);
3231                         if (order <= slub_max_order)
3232                                 return order;
3233                         fraction /= 2;
3234                 }
3235                 min_objects--;
3236         }
3237 
3238         /*
3239          * We were unable to place multiple objects in a slab. Now
3240          * lets see if we can place a single object there.
3241          */
3242         order = slab_order(size, 1, slub_max_order, 1, reserved);
3243         if (order <= slub_max_order)
3244                 return order;
3245 
3246         /*
3247          * Doh this slab cannot be placed using slub_max_order.
3248          */
3249         order = slab_order(size, 1, MAX_ORDER, 1, reserved);
3250         if (order < MAX_ORDER)
3251                 return order;
3252         return -ENOSYS;
3253 }
3254 
3255 static void
3256 init_kmem_cache_node(struct kmem_cache_node *n)
3257 {
3258         n->nr_partial = 0;
3259         spin_lock_init(&n->list_lock);
3260         INIT_LIST_HEAD(&n->partial);
3261 #ifdef CONFIG_SLUB_DEBUG
3262         atomic_long_set(&n->nr_slabs, 0);
3263         atomic_long_set(&n->total_objects, 0);
3264         INIT_LIST_HEAD(&n->full);
3265 #endif
3266 }
3267 
3268 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
3269 {
3270         BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
3271                         KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
3272 
3273         /*
3274          * Must align to double word boundary for the double cmpxchg
3275          * instructions to work; see __pcpu_double_call_return_bool().
3276          */
3277         s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3278                                      2 * sizeof(void *));
3279 
3280         if (!s->cpu_slab)
3281                 return 0;
3282 
3283         init_kmem_cache_cpus(s);
3284 
3285         return 1;
3286 }
3287 
3288 static struct kmem_cache *kmem_cache_node;
3289 
3290 /*
3291  * No kmalloc_node yet so do it by hand. We know that this is the first
3292  * slab on the node for this slabcache. There are no concurrent accesses
3293  * possible.
3294  *
3295  * Note that this function only works on the kmem_cache_node
3296  * when allocating for the kmem_cache_node. This is used for bootstrapping
3297  * memory on a fresh node that has no slab structures yet.
3298  */
3299 static void early_kmem_cache_node_alloc(int node)
3300 {
3301         struct page *page;
3302         struct kmem_cache_node *n;
3303 
3304         BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
3305 
3306         page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
3307 
3308         BUG_ON(!page);
3309         if (page_to_nid(page) != node) {
3310                 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3311                 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
3312         }
3313 
3314         n = page->freelist;
3315         BUG_ON(!n);
3316         page->freelist = get_freepointer(kmem_cache_node, n);
3317         page->inuse = 1;
3318         page->frozen = 0;
3319         kmem_cache_node->node[node] = n;
3320 #ifdef CONFIG_SLUB_DEBUG
3321         init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
3322         init_tracking(kmem_cache_node, n);
3323 #endif
3324         kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node),
3325                       GFP_KERNEL);
3326         init_kmem_cache_node(n);
3327         inc_slabs_node(kmem_cache_node, node, page->objects);
3328 
3329         /*
3330          * No locks need to be taken here as it has just been
3331          * initialized and there is no concurrent access.
3332          */
3333         __add_partial(n, page, DEACTIVATE_TO_HEAD);
3334 }
3335 
3336 static void free_kmem_cache_nodes(struct kmem_cache *s)
3337 {
3338         int node;
3339         struct kmem_cache_node *n;
3340 
3341         for_each_kmem_cache_node(s, node, n) {
3342                 kmem_cache_free(kmem_cache_node, n);
3343                 s->node[node] = NULL;
3344         }
3345 }
3346 
3347 void __kmem_cache_release(struct kmem_cache *s)
3348 {
3349         cache_random_seq_destroy(s);
3350         free_percpu(s->cpu_slab);
3351         free_kmem_cache_nodes(s);
3352 }
3353 
3354 static int init_kmem_cache_nodes(struct kmem_cache *s)
3355 {
3356         int node;
3357 
3358         for_each_node_state(node, N_NORMAL_MEMORY) {
3359                 struct kmem_cache_node *n;
3360 
3361                 if (slab_state == DOWN) {
3362                         early_kmem_cache_node_alloc(node);
3363                         continue;
3364                 }
3365                 n = kmem_cache_alloc_node(kmem_cache_node,
3366                                                 GFP_KERNEL, node);
3367 
3368                 if (!n) {
3369                         free_kmem_cache_nodes(s);
3370                         return 0;
3371                 }
3372 
3373                 s->node[node] = n;
3374                 init_kmem_cache_node(n);
3375         }
3376         return 1;
3377 }
3378 
3379 static void set_min_partial(struct kmem_cache *s, unsigned long min)
3380 {
3381         if (min < MIN_PARTIAL)
3382                 min = MIN_PARTIAL;
3383         else if (min > MAX_PARTIAL)
3384                 min = MAX_PARTIAL;
3385         s->min_partial = min;
3386 }
3387 
3388 /*
3389  * calculate_sizes() determines the order and the distribution of data within
3390  * a slab object.
3391  */
3392 static int calculate_sizes(struct kmem_cache *s, int forced_order)
3393 {
3394         unsigned long flags = s->flags;
3395         size_t size = s->object_size;
3396         int order;
3397 
3398         /*
3399          * Round up object size to the next word boundary. We can only
3400          * place the free pointer at word boundaries and this determines
3401          * the possible location of the free pointer.
3402          */
3403         size = ALIGN(size, sizeof(void *));
3404 
3405 #ifdef CONFIG_SLUB_DEBUG
3406         /*
3407          * Determine if we can poison the object itself. If the user of
3408          * the slab may touch the object after free or before allocation
3409          * then we should never poison the object itself.
3410          */
3411         if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
3412                         !s->ctor)
3413                 s->flags |= __OBJECT_POISON;
3414         else
3415                 s->flags &= ~__OBJECT_POISON;
3416 
3417 
3418         /*
3419          * If we are Redzoning then check if there is some space between the
3420          * end of the object and the free pointer. If not then add an
3421          * additional word to have some bytes to store Redzone information.
3422          */
3423         if ((flags & SLAB_RED_ZONE) && size == s->object_size)
3424                 size += sizeof(void *);
3425 #endif
3426 
3427         /*
3428          * With that we have determined the number of bytes in actual use
3429          * by the object. This is the potential offset to the free pointer.
3430          */
3431         s->inuse = size;
3432 
3433         if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
3434                 s->ctor)) {
3435                 /*
3436                  * Relocate free pointer after the object if it is not
3437                  * permitted to overwrite the first word of the object on
3438                  * kmem_cache_free.
3439                  *
3440                  * This is the case if we do RCU, have a constructor or
3441                  * destructor or are poisoning the objects.
3442                  */
3443                 s->offset = size;
3444                 size += sizeof(void *);
3445         }
3446 
3447 #ifdef CONFIG_SLUB_DEBUG
3448         if (flags & SLAB_STORE_USER)
3449                 /*
3450                  * Need to store information about allocs and frees after
3451                  * the object.
3452                  */
3453                 size += 2 * sizeof(struct track);
3454 #endif
3455 
3456         kasan_cache_create(s, &size, &s->flags);
3457 #ifdef CONFIG_SLUB_DEBUG
3458         if (flags & SLAB_RED_ZONE) {
3459                 /*
3460                  * Add some empty padding so that we can catch
3461                  * overwrites from earlier objects rather than let
3462                  * tracking information or the free pointer be
3463                  * corrupted if a user writes before the start
3464                  * of the object.
3465                  */
3466                 size += sizeof(void *);
3467 
3468                 s->red_left_pad = sizeof(void *);
3469                 s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3470                 size += s->red_left_pad;
3471         }
3472 #endif
3473 
3474         /*
3475          * SLUB stores one object immediately after another beginning from
3476          * offset 0. In order to align the objects we have to simply size
3477          * each object to conform to the alignment.
3478          */
3479         size = ALIGN(size, s->align);
3480         s->size = size;
3481         if (forced_order >= 0)
3482                 order = forced_order;
3483         else
3484                 order = calculate_order(size, s->reserved);
3485 
3486         if (order < 0)
3487                 return 0;
3488 
3489         s->allocflags = 0;
3490         if (order)
3491                 s->allocflags |= __GFP_COMP;
3492 
3493         if (s->flags & SLAB_CACHE_DMA)
3494                 s->allocflags |= GFP_DMA;
3495 
3496         if (s->flags & SLAB_RECLAIM_ACCOUNT)
3497                 s->allocflags |= __GFP_RECLAIMABLE;
3498 
3499         /*
3500          * Determine the number of objects per slab
3501          */
3502         s->oo = oo_make(order, size, s->reserved);
3503         s->min = oo_make(get_order(size), size, s->reserved);
3504         if (oo_objects(s->oo) > oo_objects(s->max))
3505                 s->max = s->oo;
3506 
3507         return !!oo_objects(s->oo);
3508 }
3509 
3510 static int kmem_cache_open(struct kmem_cache *s, unsigned long flags)
3511 {
3512         s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
3513         s->reserved = 0;
3514 
3515         if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
3516                 s->reserved = sizeof(struct rcu_head);
3517 
3518         if (!calculate_sizes(s, -1))
3519                 goto error;
3520         if (disable_higher_order_debug) {
3521                 /*
3522                  * Disable debugging flags that store metadata if the min slab
3523                  * order increased.
3524                  */
3525                 if (get_order(s->size) > get_order(s->object_size)) {
3526                         s->flags &= ~DEBUG_METADATA_FLAGS;
3527                         s->offset = 0;
3528                         if (!calculate_sizes(s, -1))
3529                                 goto error;
3530                 }
3531         }
3532 
3533 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3534     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3535         if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
3536                 /* Enable fast mode */
3537                 s->flags |= __CMPXCHG_DOUBLE;
3538 #endif
3539 
3540         /*
3541          * The larger the object size is, the more pages we want on the partial
3542          * list to avoid pounding the page allocator excessively.
3543          */
3544         set_min_partial(s, ilog2(s->size) / 2);
3545 
3546         /*
3547          * cpu_partial determined the maximum number of objects kept in the
3548          * per cpu partial lists of a processor.
3549          *
3550          * Per cpu partial lists mainly contain slabs that just have one
3551          * object freed. If they are used for allocation then they can be
3552          * filled up again with minimal effort. The slab will never hit the
3553          * per node partial lists and therefore no locking will be required.
3554          *
3555          * This setting also determines
3556          *
3557          * A) The number of objects from per cpu partial slabs dumped to the
3558          *    per node list when we reach the limit.
3559          * B) The number of objects in cpu partial slabs to extract from the
3560          *    per node list when we run out of per cpu objects. We only fetch
3561          *    50% to keep some capacity around for frees.
3562          */
3563         if (!kmem_cache_has_cpu_partial(s))
3564                 s->cpu_partial = 0;
3565         else if (s->size >= PAGE_SIZE)
3566                 s->cpu_partial = 2;
3567         else if (s->size >= 1024)
3568                 s->cpu_partial = 6;
3569         else if (s->size >= 256)
3570                 s->cpu_partial = 13;
3571         else
3572                 s->cpu_partial = 30;
3573 
3574 #ifdef CONFIG_NUMA
3575         s->remote_node_defrag_ratio = 1000;
3576 #endif
3577 
3578         /* Initialize the pre-computed randomized freelist if slab is up */
3579         if (slab_state >= UP) {
3580                 if (init_cache_random_seq(s))
3581                         goto error;
3582         }
3583 
3584         if (!init_kmem_cache_nodes(s))
3585                 goto error;
3586 
3587         if (alloc_kmem_cache_cpus(s))
3588                 return 0;
3589 
3590         free_kmem_cache_nodes(s);
3591 error:
3592         if (flags & SLAB_PANIC)
3593                 panic("Cannot create slab %s size=%lu realsize=%u order=%u offset=%u flags=%lx\n",
3594                       s->name, (unsigned long)s->size, s->size,
3595                       oo_order(s->oo), s->offset, flags);
3596         return -EINVAL;
3597 }
3598 
3599 static void list_slab_objects(struct kmem_cache *s, struct page *page,
3600                                                         const char *text)
3601 {
3602 #ifdef CONFIG_SLUB_DEBUG
3603         void *addr = page_address(page);
3604         void *p;
3605         unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3606                                      sizeof(long), GFP_ATOMIC);
3607         if (!map)
3608                 return;
3609         slab_err(s, page, text, s->name);
3610         slab_lock(page);
3611 
3612         get_map(s, page, map);
3613         for_each_object(p, s, addr, page->objects) {
3614 
3615                 if (!test_bit(slab_index(p, s, addr), map)) {
3616                         pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
3617                         print_tracking(s, p);
3618                 }
3619         }
3620         slab_unlock(page);
3621         kfree(map);
3622 #endif
3623 }
3624 
3625 /*
3626  * Attempt to free all partial slabs on a node.
3627  * This is called from __kmem_cache_shutdown(). We must take list_lock
3628  * because sysfs file might still access partial list after the shutdowning.
3629  */
3630 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
3631 {
3632         LIST_HEAD(discard);
3633         struct page *page, *h;
3634 
3635         BUG_ON(irqs_disabled());
3636         spin_lock_irq(&n->list_lock);
3637         list_for_each_entry_safe(page, h, &n->partial, lru) {
3638                 if (!page->inuse) {
3639                         remove_partial(n, page);
3640                         list_add(&page->lru, &discard);
3641                 } else {
3642                         list_slab_objects(s, page,
3643                         "Objects remaining in %s on __kmem_cache_shutdown()");
3644                 }
3645         }
3646         spin_unlock_irq(&n->list_lock);
3647 
3648         list_for_each_entry_safe(page, h, &discard, lru)
3649                 discard_slab(s, page);
3650 }
3651 
3652 /*
3653  * Release all resources used by a slab cache.
3654  */
3655 int __kmem_cache_shutdown(struct kmem_cache *s)
3656 {
3657         int node;
3658         struct kmem_cache_node *n;
3659 
3660         flush_all(s);
3661         /* Attempt to free all objects */
3662         for_each_kmem_cache_node(s, node, n) {
3663                 free_partial(s, n);
3664                 if (n->nr_partial || slabs_node(s, node))
3665                         return 1;
3666         }
3667         return 0;
3668 }
3669 
3670 /********************************************************************
3671  *              Kmalloc subsystem
3672  *******************************************************************/
3673 
3674 static int __init setup_slub_min_order(char *str)
3675 {
3676         get_option(&str, &slub_min_order);
3677 
3678         return 1;
3679 }
3680 
3681 __setup("slub_min_order=", setup_slub_min_order);
3682 
3683 static int __init setup_slub_max_order(char *str)
3684 {
3685         get_option(&str, &slub_max_order);
3686         slub_max_order = min(slub_max_order, MAX_ORDER - 1);
3687 
3688         return 1;
3689 }
3690 
3691 __setup("slub_max_order=", setup_slub_max_order);
3692 
3693 static int __init setup_slub_min_objects(char *str)
3694 {
3695         get_option(&str, &slub_min_objects);
3696 
3697         return 1;
3698 }
3699 
3700 __setup("slub_min_objects=", setup_slub_min_objects);
3701 
3702 void *__kmalloc(size_t size, gfp_t flags)
3703 {
3704         struct kmem_cache *s;
3705         void *ret;
3706 
3707         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
3708                 return kmalloc_large(size, flags);
3709 
3710         s = kmalloc_slab(size, flags);
3711 
3712         if (unlikely(ZERO_OR_NULL_PTR(s)))
3713                 return s;
3714 
3715         ret = slab_alloc(s, flags, _RET_IP_);
3716 
3717         trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
3718 
3719         kasan_kmalloc(s, ret, size, flags);
3720 
3721         return ret;
3722 }
3723 EXPORT_SYMBOL(__kmalloc);
3724 
3725 #ifdef CONFIG_NUMA
3726 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3727 {
3728         struct page *page;
3729         void *ptr = NULL;
3730 
3731         flags |= __GFP_COMP | __GFP_NOTRACK;
3732         page = alloc_pages_node(node, flags, get_order(size));
3733         if (page)
3734                 ptr = page_address(page);
3735 
3736         kmalloc_large_node_hook(ptr, size, flags);
3737         return ptr;
3738 }
3739 
3740 void *__kmalloc_node(size_t size, gfp_t flags, int node)
3741 {
3742         struct kmem_cache *s;
3743         void *ret;
3744 
3745         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
3746                 ret = kmalloc_large_node(size, flags, node);
3747 
3748                 trace_kmalloc_node(_RET_IP_, ret,
3749                                    size, PAGE_SIZE << get_order(size),
3750                                    flags, node);
3751 
3752                 return ret;
3753         }
3754 
3755         s = kmalloc_slab(size, flags);
3756 
3757         if (unlikely(ZERO_OR_NULL_PTR(s)))
3758                 return s;
3759 
3760         ret = slab_alloc_node(s, flags, node, _RET_IP_);
3761 
3762         trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
3763 
3764         kasan_kmalloc(s, ret, size, flags);
3765 
3766         return ret;
3767 }
3768 EXPORT_SYMBOL(__kmalloc_node);
3769 #endif
3770 
3771 #ifdef CONFIG_HARDENED_USERCOPY
3772 /*
3773  * Rejects objects that are incorrectly sized.
3774  *
3775  * Returns NULL if check passes, otherwise const char * to name of cache
3776  * to indicate an error.
3777  */
3778 const char *__check_heap_object(const void *ptr, unsigned long n,
3779                                 struct page *page)
3780 {
3781         struct kmem_cache *s;
3782         unsigned long offset;
3783         size_t object_size;
3784 
3785         /* Find object and usable object size. */
3786         s = page->slab_cache;
3787         object_size = slab_ksize(s);
3788 
3789         /* Reject impossible pointers. */
3790         if (ptr < page_address(page))
3791                 return s->name;
3792 
3793         /* Find offset within object. */
3794         offset = (ptr - page_address(page)) % s->size;
3795 
3796         /* Adjust for redzone and reject if within the redzone. */
3797         if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) {
3798                 if (offset < s->red_left_pad)
3799                         return s->name;
3800                 offset -= s->red_left_pad;
3801         }
3802 
3803         /* Allow address range falling entirely within object size. */
3804         if (offset <= object_size && n <= object_size - offset)
3805                 return NULL;
3806 
3807         return s->name;
3808 }
3809 #endif /* CONFIG_HARDENED_USERCOPY */
3810 
3811 static size_t __ksize(const void *object)
3812 {
3813         struct page *page;
3814 
3815         if (unlikely(object == ZERO_SIZE_PTR))
3816                 return 0;
3817 
3818         page = virt_to_head_page(object);
3819 
3820         if (unlikely(!PageSlab(page))) {
3821                 WARN_ON(!PageCompound(page));
3822                 return PAGE_SIZE << compound_order(page);
3823         }
3824 
3825         return slab_ksize(page->slab_cache);
3826 }
3827 
3828 size_t ksize(const void *object)
3829 {
3830         size_t size = __ksize(object);
3831         /* We assume that ksize callers could use whole allocated area,
3832          * so we need to unpoison this area.
3833          */
3834         kasan_unpoison_shadow(object, size);
3835         return size;
3836 }
3837 EXPORT_SYMBOL(ksize);
3838 
3839 void kfree(const void *x)
3840 {
3841         struct page *page;
3842         void *object = (void *)x;
3843 
3844         trace_kfree(_RET_IP_, x);
3845 
3846         if (unlikely(ZERO_OR_NULL_PTR(x)))
3847                 return;
3848 
3849         page = virt_to_head_page(x);
3850         if (unlikely(!PageSlab(page))) {
3851                 BUG_ON(!PageCompound(page));
3852                 kfree_hook(x);
3853                 __free_pages(page, compound_order(page));
3854                 return;
3855         }
3856         slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
3857 }
3858 EXPORT_SYMBOL(kfree);
3859 
3860 #define SHRINK_PROMOTE_MAX 32
3861 
3862 /*
3863  * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3864  * up most to the head of the partial lists. New allocations will then
3865  * fill those up and thus they can be removed from the partial lists.
3866  *
3867  * The slabs with the least items are placed last. This results in them
3868  * being allocated from last increasing the chance that the last objects
3869  * are freed in them.
3870  */
3871 int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate)
3872 {
3873         int node;
3874         int i;
3875         struct kmem_cache_node *n;
3876         struct page *page;
3877         struct page *t;
3878         struct list_head discard;
3879         struct list_head promote[SHRINK_PROMOTE_MAX];
3880         unsigned long flags;
3881         int ret = 0;
3882 
3883         if (deactivate) {
3884                 /*
3885                  * Disable empty slabs caching. Used to avoid pinning offline
3886                  * memory cgroups by kmem pages that can be freed.
3887                  */
3888                 s->cpu_partial = 0;
3889                 s->min_partial = 0;
3890 
3891                 /*
3892                  * s->cpu_partial is checked locklessly (see put_cpu_partial),
3893                  * so we have to make sure the change is visible.
3894                  */
3895                 synchronize_sched();
3896         }
3897 
3898         flush_all(s);
3899         for_each_kmem_cache_node(s, node, n) {
3900                 INIT_LIST_HEAD(&discard);
3901                 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
3902                         INIT_LIST_HEAD(promote + i);
3903 
3904                 spin_lock_irqsave(&n->list_lock, flags);
3905 
3906                 /*
3907                  * Build lists of slabs to discard or promote.
3908                  *
3909                  * Note that concurrent frees may occur while we hold the
3910                  * list_lock. page->inuse here is the upper limit.
3911                  */
3912                 list_for_each_entry_safe(page, t, &n->partial, lru) {
3913                         int free = page->objects - page->inuse;
3914 
3915                         /* Do not reread page->inuse */
3916                         barrier();
3917 
3918                         /* We do not keep full slabs on the list */
3919                         BUG_ON(free <= 0);
3920 
3921                         if (free == page->objects) {
3922                                 list_move(&page->lru, &discard);
3923                                 n->nr_partial--;
3924                         } else if (free <= SHRINK_PROMOTE_MAX)
3925                                 list_move(&page->lru, promote + free - 1);
3926                 }
3927 
3928                 /*
3929                  * Promote the slabs filled up most to the head of the
3930                  * partial list.
3931                  */
3932                 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
3933                         list_splice(promote + i, &n->partial);
3934 
3935                 spin_unlock_irqrestore(&n->list_lock, flags);
3936 
3937                 /* Release empty slabs */
3938                 list_for_each_entry_safe(page, t, &discard, lru)
3939                         discard_slab(s, page);
3940 
3941                 if (slabs_node(s, node))
3942                         ret = 1;
3943         }
3944 
3945         return ret;
3946 }
3947 
3948 static int slab_mem_going_offline_callback(void *arg)
3949 {
3950         struct kmem_cache *s;
3951 
3952         mutex_lock(&slab_mutex);
3953         list_for_each_entry(s, &slab_caches, list)
3954                 __kmem_cache_shrink(s, false);
3955         mutex_unlock(&slab_mutex);
3956 
3957         return 0;
3958 }
3959 
3960 static void slab_mem_offline_callback(void *arg)
3961 {
3962         struct kmem_cache_node *n;
3963         struct kmem_cache *s;
3964         struct memory_notify *marg = arg;
3965         int offline_node;
3966 
3967         offline_node = marg->status_change_nid_normal;
3968 
3969         /*
3970          * If the node still has available memory. we need kmem_cache_node
3971          * for it yet.
3972          */
3973         if (offline_node < 0)
3974                 return;
3975 
3976         mutex_lock(&slab_mutex);
3977         list_for_each_entry(s, &slab_caches, list) {
3978                 n = get_node(s, offline_node);
3979                 if (n) {
3980                         /*
3981                          * if n->nr_slabs > 0, slabs still exist on the node
3982                          * that is going down. We were unable to free them,
3983                          * and offline_pages() function shouldn't call this
3984                          * callback. So, we must fail.
3985                          */
3986                         BUG_ON(slabs_node(s, offline_node));
3987 
3988                         s->node[offline_node] = NULL;
3989                         kmem_cache_free(kmem_cache_node, n);
3990                 }
3991         }
3992         mutex_unlock(&slab_mutex);
3993 }
3994 
3995 static int slab_mem_going_online_callback(void *arg)
3996 {
3997         struct kmem_cache_node *n;
3998         struct kmem_cache *s;
3999         struct memory_notify *marg = arg;
4000         int nid = marg->status_change_nid_normal;
4001         int ret = 0;
4002 
4003         /*
4004          * If the node's memory is already available, then kmem_cache_node is
4005          * already created. Nothing to do.
4006          */
4007         if (nid < 0)
4008                 return 0;
4009 
4010         /*
4011          * We are bringing a node online. No memory is available yet. We must
4012          * allocate a kmem_cache_node structure in order to bring the node
4013          * online.
4014          */
4015         mutex_lock(&slab_mutex);
4016         list_for_each_entry(s, &slab_caches, list) {
4017                 /*
4018                  * XXX: kmem_cache_alloc_node will fallback to other nodes
4019                  *      since memory is not yet available from the node that
4020                  *      is brought up.
4021                  */
4022                 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
4023                 if (!n) {
4024                         ret = -ENOMEM;
4025                         goto out;
4026                 }
4027                 init_kmem_cache_node(n);
4028                 s->node[nid] = n;
4029         }
4030 out:
4031         mutex_unlock(&slab_mutex);
4032         return ret;
4033 }
4034 
4035 static int slab_memory_callback(struct notifier_block *self,
4036                                 unsigned long action, void *arg)
4037 {
4038         int ret = 0;
4039 
4040         switch (action) {
4041         case MEM_GOING_ONLINE:
4042                 ret = slab_mem_going_online_callback(arg);
4043                 break;
4044         case MEM_GOING_OFFLINE:
4045                 ret = slab_mem_going_offline_callback(arg);
4046                 break;
4047         case MEM_OFFLINE:
4048         case MEM_CANCEL_ONLINE:
4049                 slab_mem_offline_callback(arg);
4050                 break;
4051         case MEM_ONLINE:
4052         case MEM_CANCEL_OFFLINE:
4053                 break;
4054         }
4055         if (ret)
4056                 ret = notifier_from_errno(ret);
4057         else
4058                 ret = NOTIFY_OK;
4059         return ret;
4060 }
4061 
4062 static struct notifier_block slab_memory_callback_nb = {
4063         .notifier_call = slab_memory_callback,
4064         .priority = SLAB_CALLBACK_PRI,
4065 };
4066 
4067 /********************************************************************
4068  *                      Basic setup of slabs
4069  *******************************************************************/
4070 
4071 /*
4072  * Used for early kmem_cache structures that were allocated using
4073  * the page allocator. Allocate them properly then fix up the pointers
4074  * that may be pointing to the wrong kmem_cache structure.
4075  */
4076 
4077 static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
4078 {
4079         int node;
4080         struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
4081         struct kmem_cache_node *n;
4082 
4083         memcpy(s, static_cache, kmem_cache->object_size);
4084 
4085         /*
4086          * This runs very early, and only the boot processor is supposed to be
4087          * up.  Even if it weren't true, IRQs are not up so we couldn't fire
4088          * IPIs around.
4089          */
4090         __flush_cpu_slab(s, smp_processor_id());
4091         for_each_kmem_cache_node(s, node, n) {
4092                 struct page *p;
4093 
4094                 list_for_each_entry(p, &n->partial, lru)
4095                         p->slab_cache = s;
4096 
4097 #ifdef CONFIG_SLUB_DEBUG
4098                 list_for_each_entry(p, &n->full, lru)
4099                         p->slab_cache = s;
4100 #endif
4101         }
4102         slab_init_memcg_params(s);
4103         list_add(&s->list, &slab_caches);
4104         return s;
4105 }
4106 
4107 void __init kmem_cache_init(void)
4108 {
4109         static __initdata struct kmem_cache boot_kmem_cache,
4110                 boot_kmem_cache_node;
4111 
4112         if (debug_guardpage_minorder())
4113                 slub_max_order = 0;
4114 
4115         kmem_cache_node = &boot_kmem_cache_node;
4116         kmem_cache = &boot_kmem_cache;
4117 
4118         create_boot_cache(kmem_cache_node, "kmem_cache_node",
4119                 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN);
4120 
4121         register_hotmemory_notifier(&slab_memory_callback_nb);
4122 
4123         /* Able to allocate the per node structures */
4124         slab_state = PARTIAL;
4125 
4126         create_boot_cache(kmem_cache, "kmem_cache",
4127                         offsetof(struct kmem_cache, node) +
4128                                 nr_node_ids * sizeof(struct kmem_cache_node *),
4129                        SLAB_HWCACHE_ALIGN);
4130 
4131         kmem_cache = bootstrap(&boot_kmem_cache);
4132 
4133         /*
4134          * Allocate kmem_cache_node properly from the kmem_cache slab.
4135          * kmem_cache_node is separately allocated so no need to
4136          * update any list pointers.
4137          */
4138         kmem_cache_node = bootstrap(&boot_kmem_cache_node);
4139 
4140         /* Now we can use the kmem_cache to allocate kmalloc slabs */
4141         setup_kmalloc_cache_index_table();
4142         create_kmalloc_caches(0);
4143 
4144         /* Setup random freelists for each cache */
4145         init_freelist_randomization();
4146 
4147 #ifdef CONFIG_SMP
4148         register_cpu_notifier(&slab_notifier);
4149 #endif
4150 
4151         pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n",
4152                 cache_line_size(),
4153                 slub_min_order, slub_max_order, slub_min_objects,
4154                 nr_cpu_ids, nr_node_ids);
4155 }
4156 
4157 void __init kmem_cache_init_late(void)
4158 {
4159 }
4160 
4161 struct kmem_cache *
4162 __kmem_cache_alias(const char *name, size_t size, size_t align,
4163                    unsigned long flags, void (*ctor)(void *))
4164 {
4165         struct kmem_cache *s, *c;
4166 
4167         s = find_mergeable(size, align, flags, name, ctor);
4168         if (s) {
4169                 s->refcount++;
4170 
4171                 /*
4172                  * Adjust the object sizes so that we clear
4173                  * the complete object on kzalloc.
4174                  */
4175                 s->object_size = max(s->object_size, (int)size);
4176                 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
4177 
4178                 for_each_memcg_cache(c, s) {
4179                         c->object_size = s->object_size;
4180                         c->inuse = max_t(int, c->inuse,
4181                                          ALIGN(size, sizeof(void *)));
4182                 }
4183 
4184                 if (sysfs_slab_alias(s, name)) {
4185                         s->refcount--;
4186                         s = NULL;
4187                 }
4188         }
4189 
4190         return s;
4191 }
4192 
4193 int __kmem_cache_create(struct kmem_cache *s, unsigned long flags)
4194 {
4195         int err;
4196 
4197         err = kmem_cache_open(s, flags);
4198         if (err)
4199                 return err;
4200 
4201         /* Mutex is not taken during early boot */
4202         if (slab_state <= UP)
4203                 return 0;
4204 
4205         memcg_propagate_slab_attrs(s);
4206         err = sysfs_slab_add(s);
4207         if (err)
4208                 __kmem_cache_release(s);
4209 
4210         return err;
4211 }
4212 
4213 #ifdef CONFIG_SMP
4214 /*
4215  * Use the cpu notifier to insure that the cpu slabs are flushed when
4216  * necessary.
4217  */
4218 static int slab_cpuup_callback(struct notifier_block *nfb,
4219                 unsigned long action, void *hcpu)
4220 {
4221         long cpu = (long)hcpu;
4222         struct kmem_cache *s;
4223         unsigned long flags;
4224 
4225         switch (action) {
4226         case CPU_UP_CANCELED:
4227         case CPU_UP_CANCELED_FROZEN:
4228         case CPU_DEAD:
4229         case CPU_DEAD_FROZEN:
4230                 mutex_lock(&slab_mutex);
4231                 list_for_each_entry(s, &slab_caches, list) {
4232                         local_irq_save(flags);
4233                         __flush_cpu_slab(s, cpu);
4234                         local_irq_restore(flags);
4235                 }
4236                 mutex_unlock(&slab_mutex);
4237                 break;
4238         default:
4239                 break;
4240         }
4241         return NOTIFY_OK;
4242 }
4243 
4244 static struct notifier_block slab_notifier = {
4245         .notifier_call = slab_cpuup_callback
4246 };
4247 
4248 #endif
4249 
4250 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
4251 {
4252         struct kmem_cache *s;
4253         void *ret;
4254 
4255         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
4256                 return kmalloc_large(size, gfpflags);
4257 
4258         s = kmalloc_slab(size, gfpflags);
4259 
4260         if (unlikely(ZERO_OR_NULL_PTR(s)))
4261                 return s;
4262 
4263         ret = slab_alloc(s, gfpflags, caller);
4264 
4265         /* Honor the call site pointer we received. */
4266         trace_kmalloc(caller, ret, size, s->size, gfpflags);
4267 
4268         return ret;
4269 }
4270 
4271 #ifdef CONFIG_NUMA
4272 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
4273                                         int node, unsigned long caller)
4274 {
4275         struct kmem_cache *s;
4276         void *ret;
4277 
4278         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
4279                 ret = kmalloc_large_node(size, gfpflags, node);
4280 
4281                 trace_kmalloc_node(caller, ret,
4282                                    size, PAGE_SIZE << get_order(size),
4283                                    gfpflags, node);
4284 
4285                 return ret;
4286         }
4287 
4288         s = kmalloc_slab(size, gfpflags);
4289 
4290         if (unlikely(ZERO_OR_NULL_PTR(s)))
4291                 return s;
4292 
4293         ret = slab_alloc_node(s, gfpflags, node, caller);
4294 
4295         /* Honor the call site pointer we received. */
4296         trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
4297 
4298         return ret;
4299 }
4300 #endif
4301 
4302 #ifdef CONFIG_SYSFS
4303 static int count_inuse(struct page *page)
4304 {
4305         return page->inuse;
4306 }
4307 
4308 static int count_total(struct page *page)
4309 {
4310         return page->objects;
4311 }
4312 #endif
4313 
4314 #ifdef CONFIG_SLUB_DEBUG
4315 static int validate_slab(struct kmem_cache *s, struct page *page,
4316                                                 unsigned long *map)
4317 {
4318         void *p;
4319         void *addr = page_address(page);
4320 
4321         if (!check_slab(s, page) ||
4322                         !on_freelist(s, page, NULL))
4323                 return 0;
4324 
4325         /* Now we know that a valid freelist exists */
4326         bitmap_zero(map, page->objects);
4327 
4328         get_map(s, page, map);
4329         for_each_object(p, s, addr, page->objects) {
4330                 if (test_bit(slab_index(p, s, addr), map))
4331                         if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4332                                 return 0;
4333         }
4334 
4335         for_each_object(p, s, addr, page->objects)
4336                 if (!test_bit(slab_index(p, s, addr), map))
4337                         if (!check_object(s, page, p, SLUB_RED_ACTIVE))
4338                                 return 0;
4339         return 1;
4340 }
4341 
4342 static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4343                                                 unsigned long *map)
4344 {
4345         slab_lock(page);
4346         validate_slab(s, page, map);
4347         slab_unlock(page);
4348 }
4349 
4350 static int validate_slab_node(struct kmem_cache *s,
4351                 struct kmem_cache_node *n, unsigned long *map)
4352 {
4353         unsigned long count = 0;
4354         struct page *page;
4355         unsigned long flags;
4356 
4357         spin_lock_irqsave(&n->list_lock, flags);
4358 
4359         list_for_each_entry(page, &n->partial, lru) {
4360                 validate_slab_slab(s, page, map);
4361                 count++;
4362         }
4363         if (count != n->nr_partial)
4364                 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4365                        s->name, count, n->nr_partial);
4366 
4367         if (!(s->flags & SLAB_STORE_USER))
4368                 goto out;
4369 
4370         list_for_each_entry(page, &n->full, lru) {
4371                 validate_slab_slab(s, page, map);
4372                 count++;
4373         }
4374         if (count != atomic_long_read(&n->nr_slabs))
4375                 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4376                        s->name, count, atomic_long_read(&n->nr_slabs));
4377 
4378 out:
4379         spin_unlock_irqrestore(&n->list_lock, flags);
4380         return count;
4381 }
4382 
4383 static long validate_slab_cache(struct kmem_cache *s)
4384 {
4385         int node;
4386         unsigned long count = 0;
4387         unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4388                                 sizeof(unsigned long), GFP_KERNEL);
4389         struct kmem_cache_node *n;
4390 
4391         if (!map)
4392                 return -ENOMEM;
4393 
4394         flush_all(s);
4395         for_each_kmem_cache_node(s, node, n)
4396                 count += validate_slab_node(s, n, map);
4397         kfree(map);
4398         return count;
4399 }
4400 /*
4401  * Generate lists of code addresses where slabcache objects are allocated
4402  * and freed.
4403  */
4404 
4405 struct location {
4406         unsigned long count;
4407         unsigned long addr;
4408         long long sum_time;
4409         long min_time;
4410         long max_time;
4411         long min_pid;
4412         long max_pid;
4413         DECLARE_BITMAP(cpus, NR_CPUS);
4414         nodemask_t nodes;
4415 };
4416 
4417 struct loc_track {
4418         unsigned long max;
4419         unsigned long count;
4420         struct location *loc;
4421 };
4422 
4423 static void free_loc_track(struct loc_track *t)
4424 {
4425         if (t->max)
4426                 free_pages((unsigned long)t->loc,
4427                         get_order(sizeof(struct location) * t->max));
4428 }
4429 
4430 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
4431 {
4432         struct location *l;
4433         int order;
4434 
4435         order = get_order(sizeof(struct location) * max);
4436 
4437         l = (void *)__get_free_pages(flags, order);
4438         if (!l)
4439                 return 0;
4440 
4441         if (t->count) {
4442                 memcpy(l, t->loc, sizeof(struct location) * t->count);
4443                 free_loc_track(t);
4444         }
4445         t->max = max;
4446         t->loc = l;
4447         return 1;
4448 }
4449 
4450 static int add_location(struct loc_track *t, struct kmem_cache *s,
4451                                 const struct track *track)
4452 {
4453         long start, end, pos;
4454         struct location *l;
4455         unsigned long caddr;
4456         unsigned long age = jiffies - track->when;
4457 
4458         start = -1;
4459         end = t->count;
4460 
4461         for ( ; ; ) {
4462                 pos = start + (end - start + 1) / 2;
4463 
4464                 /*
4465                  * There is nothing at "end". If we end up there
4466                  * we need to add something to before end.
4467                  */
4468                 if (pos == end)
4469                         break;
4470 
4471                 caddr = t->loc[pos].addr;
4472                 if (track->addr == caddr) {
4473 
4474                         l = &t->loc[pos];
4475                         l->count++;
4476                         if (track->when) {
4477                                 l->sum_time += age;
4478                                 if (age < l->min_time)
4479                                         l->min_time = age;
4480                                 if (age > l->max_time)
4481                                         l->max_time = age;
4482 
4483                                 if (track->pid < l->min_pid)
4484                                         l->min_pid = track->pid;
4485                                 if (track->pid > l->max_pid)
4486                                         l->max_pid = track->pid;
4487 
4488                                 cpumask_set_cpu(track->cpu,
4489                                                 to_cpumask(l->cpus));
4490                         }
4491                         node_set(page_to_nid(virt_to_page(track)), l->nodes);
4492                         return 1;
4493                 }
4494 
4495                 if (track->addr < caddr)
4496                         end = pos;
4497                 else
4498                         start = pos;
4499         }
4500 
4501         /*
4502          * Not found. Insert new tracking element.
4503          */
4504         if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
4505                 return 0;
4506 
4507         l = t->loc + pos;
4508         if (pos < t->count)
4509                 memmove(l + 1, l,
4510                         (t->count - pos) * sizeof(struct location));
4511         t->count++;
4512         l->count = 1;
4513         l->addr = track->addr;
4514         l->sum_time = age;
4515         l->min_time = age;
4516         l->max_time = age;
4517         l->min_pid = track->pid;
4518         l->max_pid = track->pid;
4519         cpumask_clear(to_cpumask(l->cpus));
4520         cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
4521         nodes_clear(l->nodes);
4522         node_set(page_to_nid(virt_to_page(track)), l->nodes);
4523         return 1;
4524 }
4525 
4526 static void process_slab(struct loc_track *t, struct kmem_cache *s,
4527                 struct page *page, enum track_item alloc,
4528                 unsigned long *map)
4529 {
4530         void *addr = page_address(page);
4531         void *p;
4532 
4533         bitmap_zero(map, page->objects);
4534         get_map(s, page, map);
4535 
4536         for_each_object(p, s, addr, page->objects)
4537                 if (!test_bit(slab_index(p, s, addr), map))
4538                         add_location(t, s, get_track(s, p, alloc));
4539 }
4540 
4541 static int list_locations(struct kmem_cache *s, char *buf,
4542                                         enum track_item alloc)
4543 {
4544         int len = 0;
4545         unsigned long i;
4546         struct loc_track t = { 0, 0, NULL };
4547         int node;
4548         unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4549                                      sizeof(unsigned long), GFP_KERNEL);
4550         struct kmem_cache_node *n;
4551 
4552         if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4553                                      GFP_TEMPORARY)) {
4554                 kfree(map);
4555                 return sprintf(buf, "Out of memory\n");
4556         }
4557         /* Push back cpu slabs */
4558         flush_all(s);
4559 
4560         for_each_kmem_cache_node(s, node, n) {
4561                 unsigned long flags;
4562                 struct page *page;
4563 
4564                 if (!atomic_long_read(&n->nr_slabs))
4565                         continue;
4566 
4567                 spin_lock_irqsave(&n->list_lock, flags);
4568                 list_for_each_entry(page, &n->partial, lru)
4569                         process_slab(&t, s, page, alloc, map);
4570                 list_for_each_entry(page, &n->full, lru)
4571                         process_slab(&t, s, page, alloc, map);
4572                 spin_unlock_irqrestore(&n->list_lock, flags);
4573         }
4574 
4575         for (i = 0; i < t.count; i++) {
4576                 struct location *l = &t.loc[i];
4577 
4578                 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
4579                         break;
4580                 len += sprintf(buf + len, "%7ld ", l->count);
4581 
4582                 if (l->addr)
4583                         len += sprintf(buf + len, "%pS", (void *)l->addr);
4584                 else
4585                         len += sprintf(buf + len, "<not-available>");
4586 
4587                 if (l->sum_time != l->min_time) {
4588                         len += sprintf(buf + len, " age=%ld/%ld/%ld",
4589                                 l->min_time,
4590                                 (long)div_u64(l->sum_time, l->count),
4591                                 l->max_time);
4592                 } else
4593                         len += sprintf(buf + len, " age=%ld",
4594                                 l->min_time);
4595 
4596                 if (l->min_pid != l->max_pid)
4597                         len += sprintf(buf + len, " pid=%ld-%ld",
4598                                 l->min_pid, l->max_pid);
4599                 else
4600                         len += sprintf(buf + len, " pid=%ld",
4601                                 l->min_pid);
4602 
4603                 if (num_online_cpus() > 1 &&
4604                                 !cpumask_empty(to_cpumask(l->cpus)) &&
4605                                 len < PAGE_SIZE - 60)
4606                         len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4607                                          " cpus=%*pbl",
4608                                          cpumask_pr_args(to_cpumask(l->cpus)));
4609 
4610                 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
4611                                 len < PAGE_SIZE - 60)
4612                         len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4613                                          " nodes=%*pbl",
4614                                          nodemask_pr_args(&l->nodes));
4615 
4616                 len += sprintf(buf + len, "\n");
4617         }
4618 
4619         free_loc_track(&t);
4620         kfree(map);
4621         if (!t.count)
4622                 len += sprintf(buf, "No data\n");
4623         return len;
4624 }
4625 #endif
4626 
4627 #ifdef SLUB_RESILIENCY_TEST
4628 static void __init resiliency_test(void)
4629 {
4630         u8 *p;
4631 
4632         BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
4633 
4634         pr_err("SLUB resiliency testing\n");
4635         pr_err("-----------------------\n");
4636         pr_err("A. Corruption after allocation\n");
4637 
4638         p = kzalloc(16, GFP_KERNEL);
4639         p[16] = 0x12;
4640         pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4641                p + 16);
4642 
4643         validate_slab_cache(kmalloc_caches[4]);
4644 
4645         /* Hmmm... The next two are dangerous */
4646         p = kzalloc(32, GFP_KERNEL);
4647         p[32 + sizeof(void *)] = 0x34;
4648         pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4649                p);
4650         pr_err("If allocated object is overwritten then not detectable\n\n");
4651 
4652         validate_slab_cache(kmalloc_caches[5]);
4653         p = kzalloc(64, GFP_KERNEL);
4654         p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4655         *p = 0x56;
4656         pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4657                p);
4658         pr_err("If allocated object is overwritten then not detectable\n\n");
4659         validate_slab_cache(kmalloc_caches[6]);
4660 
4661         pr_err("\nB. Corruption after free\n");
4662         p = kzalloc(128, GFP_KERNEL);
4663         kfree(p);
4664         *p = 0x78;
4665         pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
4666         validate_slab_cache(kmalloc_caches[7]);
4667 
4668         p = kzalloc(256, GFP_KERNEL);
4669         kfree(p);
4670         p[50] = 0x9a;
4671         pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
4672         validate_slab_cache(kmalloc_caches[8]);
4673 
4674         p = kzalloc(512, GFP_KERNEL);
4675         kfree(p);
4676         p[512] = 0xab;
4677         pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
4678         validate_slab_cache(kmalloc_caches[9]);
4679 }
4680 #else
4681 #ifdef CONFIG_SYSFS
4682 static void resiliency_test(void) {};
4683 #endif
4684 #endif
4685 
4686 #ifdef CONFIG_SYSFS
4687 enum slab_stat_type {
4688         SL_ALL,                 /* All slabs */
4689         SL_PARTIAL,             /* Only partially allocated slabs */
4690         SL_CPU,                 /* Only slabs used for cpu caches */
4691         SL_OBJECTS,             /* Determine allocated objects not slabs */
4692         SL_TOTAL                /* Determine object capacity not slabs */
4693 };
4694 
4695 #define SO_ALL          (1 << SL_ALL)
4696 #define SO_PARTIAL      (1 << SL_PARTIAL)
4697 #define SO_CPU          (1 << SL_CPU)
4698 #define SO_OBJECTS      (1 << SL_OBJECTS)
4699 #define SO_TOTAL        (1 << SL_TOTAL)
4700 
4701 static ssize_t show_slab_objects(struct kmem_cache *s,
4702                             char *buf, unsigned long flags)
4703 {
4704         unsigned long total = 0;
4705         int node;
4706         int x;
4707         unsigned long *nodes;
4708 
4709         nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
4710         if (!nodes)
4711                 return -ENOMEM;
4712 
4713         if (flags & SO_CPU) {
4714                 int cpu;
4715 
4716                 for_each_possible_cpu(cpu) {
4717                         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4718                                                                cpu);
4719                         int node;
4720                         struct page *page;
4721 
4722                         page = READ_ONCE(c->page);
4723                         if (!page)
4724                                 continue;
4725 
4726                         node = page_to_nid(page);
4727                         if (flags & SO_TOTAL)
4728                                 x = page->objects;
4729                         else if (flags & SO_OBJECTS)
4730                                 x = page->inuse;
4731                         else
4732                                 x = 1;
4733 
4734                         total += x;
4735                         nodes[node] += x;
4736 
4737                         page = READ_ONCE(c->partial);
4738                         if (page) {
4739                                 node = page_to_nid(page);
4740                                 if (flags & SO_TOTAL)
4741                                         WARN_ON_ONCE(1);
4742                                 else if (flags & SO_OBJECTS)
4743                                         WARN_ON_ONCE(1);
4744                                 else
4745                                         x = page->pages;
4746                                 total += x;
4747                                 nodes[node] += x;
4748                         }
4749                 }
4750         }
4751 
4752         get_online_mems();
4753 #ifdef CONFIG_SLUB_DEBUG
4754         if (flags & SO_ALL) {
4755                 struct kmem_cache_node *n;
4756 
4757                 for_each_kmem_cache_node(s, node, n) {
4758 
4759                         if (flags & SO_TOTAL)
4760                                 x = atomic_long_read(&n->total_objects);
4761                         else if (flags & SO_OBJECTS)
4762                                 x = atomic_long_read(&n->total_objects) -
4763                                         count_partial(n, count_free);
4764                         else
4765                                 x = atomic_long_read(&n->nr_slabs);
4766                         total += x;
4767                         nodes[node] += x;
4768                 }
4769 
4770         } else
4771 #endif
4772         if (flags & SO_PARTIAL) {
4773                 struct kmem_cache_node *n;
4774 
4775                 for_each_kmem_cache_node(s, node, n) {
4776                         if (flags & SO_TOTAL)
4777                                 x = count_partial(n, count_total);
4778                         else if (flags & SO_OBJECTS)
4779                                 x = count_partial(n, count_inuse);
4780                         else
4781                                 x = n->nr_partial;
4782                         total += x;
4783                         nodes[node] += x;
4784                 }
4785         }
4786         x = sprintf(buf, "%lu", total);
4787 #ifdef CONFIG_NUMA
4788         for (node = 0; node < nr_node_ids; node++)
4789                 if (nodes[node])
4790                         x += sprintf(buf + x, " N%d=%lu",
4791                                         node, nodes[node]);
4792 #endif
4793         put_online_mems();
4794         kfree(nodes);
4795         return x + sprintf(buf + x, "\n");
4796 }
4797 
4798 #ifdef CONFIG_SLUB_DEBUG
4799 static int any_slab_objects(struct kmem_cache *s)
4800 {
4801         int node;
4802         struct kmem_cache_node *n;
4803 
4804         for_each_kmem_cache_node(s, node, n)
4805                 if (atomic_long_read(&n->total_objects))
4806                         return 1;
4807 
4808         return 0;
4809 }
4810 #endif
4811 
4812 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4813 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4814 
4815 struct slab_attribute {
4816         struct attribute attr;
4817         ssize_t (*show)(struct kmem_cache *s, char *buf);
4818         ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4819 };
4820 
4821 #define SLAB_ATTR_RO(_name) \
4822         static struct slab_attribute _name##_attr = \
4823         __ATTR(_name, 0400, _name##_show, NULL)
4824 
4825 #define SLAB_ATTR(_name) \
4826         static struct slab_attribute _name##_attr =  \
4827         __ATTR(_name, 0600, _name##_show, _name##_store)
4828 
4829 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4830 {
4831         return sprintf(buf, "%d\n", s->size);
4832 }
4833 SLAB_ATTR_RO(slab_size);
4834 
4835 static ssize_t align_show(struct kmem_cache *s, char *buf)
4836 {
4837         return sprintf(buf, "%d\n", s->align);
4838 }
4839 SLAB_ATTR_RO(align);
4840 
4841 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4842 {
4843         return sprintf(buf, "%d\n", s->object_size);
4844 }
4845 SLAB_ATTR_RO(object_size);
4846 
4847 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4848 {
4849         return sprintf(buf, "%d\n", oo_objects(s->oo));
4850 }
4851 SLAB_ATTR_RO(objs_per_slab);
4852 
4853 static ssize_t order_store(struct kmem_cache *s,
4854                                 const char *buf, size_t length)
4855 {
4856         unsigned long order;
4857         int err;
4858 
4859         err = kstrtoul(buf, 10, &order);
4860         if (err)
4861                 return err;
4862 
4863         if (order > slub_max_order || order < slub_min_order)
4864                 return -EINVAL;
4865 
4866         calculate_sizes(s, order);
4867         return length;
4868 }
4869 
4870 static ssize_t order_show(struct kmem_cache *s, char *buf)
4871 {
4872         return sprintf(buf, "%d\n", oo_order(s->oo));
4873 }
4874 SLAB_ATTR(order);
4875 
4876 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4877 {
4878         return sprintf(buf, "%lu\n", s->min_partial);
4879 }
4880 
4881 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4882                                  size_t length)
4883 {
4884         unsigned long min;
4885         int err;
4886 
4887         err = kstrtoul(buf, 10, &min);
4888         if (err)
4889                 return err;
4890 
4891         set_min_partial(s, min);
4892         return length;
4893 }
4894 SLAB_ATTR(min_partial);
4895 
4896 static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4897 {
4898         return sprintf(buf, "%u\n", s->cpu_partial);
4899 }
4900 
4901 static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4902                                  size_t length)
4903 {
4904         unsigned long objects;
4905         int err;
4906 
4907         err = kstrtoul(buf, 10, &objects);
4908         if (err)
4909                 return err;
4910         if (objects && !kmem_cache_has_cpu_partial(s))
4911                 return -EINVAL;
4912 
4913         s->cpu_partial = objects;
4914         flush_all(s);
4915         return length;
4916 }
4917 SLAB_ATTR(cpu_partial);
4918 
4919 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4920 {
4921         if (!s->ctor)
4922                 return 0;
4923         return sprintf(buf, "%pS\n", s->ctor);
4924 }
4925 SLAB_ATTR_RO(ctor);
4926 
4927 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4928 {
4929         return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
4930 }
4931 SLAB_ATTR_RO(aliases);
4932 
4933 static ssize_t partial_show(struct kmem_cache *s, char *buf)
4934 {
4935         return show_slab_objects(s, buf, SO_PARTIAL);
4936 }
4937 SLAB_ATTR_RO(partial);
4938 
4939 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4940 {
4941         return show_slab_objects(s, buf, SO_CPU);
4942 }
4943 SLAB_ATTR_RO(cpu_slabs);
4944 
4945 static ssize_t objects_show(struct kmem_cache *s, char *buf)
4946 {
4947         return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
4948 }
4949 SLAB_ATTR_RO(objects);
4950 
4951 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4952 {
4953         return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4954 }
4955 SLAB_ATTR_RO(objects_partial);
4956 
4957 static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
4958 {
4959         int objects = 0;
4960         int pages = 0;
4961         int cpu;
4962         int len;
4963 
4964         for_each_online_cpu(cpu) {
4965                 struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
4966 
4967                 if (page) {
4968                         pages += page->pages;
4969                         objects += page->pobjects;
4970                 }
4971         }
4972 
4973         len = sprintf(buf, "%d(%d)", objects, pages);
4974 
4975 #ifdef CONFIG_SMP
4976         for_each_online_cpu(cpu) {
4977                 struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
4978 
4979                 if (page && len < PAGE_SIZE - 20)
4980                         len += sprintf(buf + len, " C%d=%d(%d)", cpu,
4981                                 page->pobjects, page->pages);
4982         }
4983 #endif
4984         return len + sprintf(buf + len, "\n");
4985 }
4986 SLAB_ATTR_RO(slabs_cpu_partial);
4987 
4988 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4989 {
4990         return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4991 }
4992 
4993 static ssize_t reclaim_account_store(struct kmem_cache *s,
4994                                 const char *buf, size_t length)
4995 {
4996         s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4997         if (buf[0] == '1')
4998                 s->flags |= SLAB_RECLAIM_ACCOUNT;
4999         return length;
5000 }
5001 SLAB_ATTR(reclaim_account);
5002 
5003 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5004 {
5005         return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
5006 }
5007 SLAB_ATTR_RO(hwcache_align);
5008 
5009 #ifdef CONFIG_ZONE_DMA
5010 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5011 {
5012         return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5013 }
5014 SLAB_ATTR_RO(cache_dma);
5015 #endif
5016 
5017 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5018 {
5019         return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
5020 }
5021 SLAB_ATTR_RO(destroy_by_rcu);
5022 
5023 static ssize_t reserved_show(struct kmem_cache *s, char *buf)
5024 {
5025         return sprintf(buf, "%d\n", s->reserved);
5026 }
5027 SLAB_ATTR_RO(reserved);
5028 
5029 #ifdef CONFIG_SLUB_DEBUG
5030 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5031 {
5032         return show_slab_objects(s, buf, SO_ALL);
5033 }
5034 SLAB_ATTR_RO(slabs);
5035 
5036 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5037 {
5038         return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5039 }
5040 SLAB_ATTR_RO(total_objects);
5041 
5042 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5043 {
5044         return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
5045 }
5046 
5047 static ssize_t sanity_checks_store(struct kmem_cache *s,
5048                                 const char *buf, size_t length)
5049 {
5050         s->flags &= ~SLAB_CONSISTENCY_CHECKS;
5051         if (buf[0] == '1') {
5052                 s->flags &= ~__CMPXCHG_DOUBLE;
5053                 s->flags |= SLAB_CONSISTENCY_CHECKS;
5054         }
5055         return length;
5056 }
5057 SLAB_ATTR(sanity_checks);
5058 
5059 static ssize_t trace_show(struct kmem_cache *s, char *buf)
5060 {
5061         return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5062 }
5063 
5064 static ssize_t trace_store(struct kmem_cache *s, const char *buf,
5065                                                         size_t length)
5066 {
5067         /*
5068          * Tracing a merged cache is going to give confusing results
5069          * as well as cause other issues like converting a mergeable
5070          * cache into an umergeable one.
5071          */
5072         if (s->refcount > 1)
5073                 return -EINVAL;
5074 
5075         s->flags &= ~SLAB_TRACE;
5076         if (buf[0] == '1') {
5077                 s->flags &= ~__CMPXCHG_DOUBLE;
5078                 s->flags |= SLAB_TRACE;
5079         }
5080         return length;
5081 }
5082 SLAB_ATTR(trace);
5083 
5084 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5085 {
5086         return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5087 }
5088 
5089 static ssize_t red_zone_store(struct kmem_cache *s,
5090                                 const char *buf, size_t length)
5091 {
5092         if (any_slab_objects(s))
5093                 return -EBUSY;
5094 
5095         s->flags &= ~SLAB_RED_ZONE;
5096         if (buf[0] == '1') {
5097                 s->flags |= SLAB_RED_ZONE;
5098         }
5099         calculate_sizes(s, -1);
5100         return length;
5101 }
5102 SLAB_ATTR(red_zone);
5103 
5104 static ssize_t poison_show(struct kmem_cache *s, char *buf)
5105 {
5106         return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
5107 }
5108 
5109 static ssize_t poison_store(struct kmem_cache *s,
5110                                 const char *buf, size_t length)
5111 {
5112         if (any_slab_objects(s))
5113                 return -EBUSY;
5114 
5115         s->flags &= ~SLAB_POISON;
5116         if (buf[0] == '1') {
5117                 s->flags |= SLAB_POISON;
5118         }
5119         calculate_sizes(s, -1);
5120         return length;
5121 }
5122 SLAB_ATTR(poison);
5123 
5124 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5125 {
5126         return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5127 }
5128 
5129 static ssize_t store_user_store(struct kmem_cache *s,
5130                                 const char *buf, size_t length)
5131 {
5132         if (any_slab_objects(s))
5133                 return -EBUSY;
5134 
5135         s->flags &= ~SLAB_STORE_USER;
5136         if (buf[0] == '1') {
5137                 s->flags &= ~__CMPXCHG_DOUBLE;
5138                 s->flags |= SLAB_STORE_USER;
5139         }
5140         calculate_sizes(s, -1);
5141         return length;
5142 }
5143 SLAB_ATTR(store_user);
5144 
5145 static ssize_t validate_show(struct kmem_cache *s, char *buf)
5146 {
5147         return 0;
5148 }
5149 
5150 static ssize_t validate_store(struct kmem_cache *s,
5151                         const char *buf, size_t length)
5152 {
5153         int ret = -EINVAL;
5154 
5155         if (buf[0] == '1') {
5156                 ret = validate_slab_cache(s);
5157                 if (ret >= 0)
5158                         ret = length;
5159         }
5160         return ret;
5161 }
5162 SLAB_ATTR(validate);
5163 
5164 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
5165 {
5166         if (!(s->flags & SLAB_STORE_USER))
5167                 return -ENOSYS;
5168         return list_locations(s, buf, TRACK_ALLOC);
5169 }
5170 SLAB_ATTR_RO(alloc_calls);
5171 
5172 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
5173 {
5174         if (!(s->flags & SLAB_STORE_USER))
5175                 return -ENOSYS;
5176         return list_locations(s, buf, TRACK_FREE);
5177 }
5178 SLAB_ATTR_RO(free_calls);
5179 #endif /* CONFIG_SLUB_DEBUG */
5180 
5181 #ifdef CONFIG_FAILSLAB
5182 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5183 {
5184         return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5185 }
5186 
5187 static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5188                                                         size_t length)
5189 {
5190         if (s->refcount > 1)
5191                 return -EINVAL;
5192 
5193         s->flags &= ~SLAB_FAILSLAB;
5194         if (buf[0] == '1')
5195                 s->flags |= SLAB_FAILSLAB;
5196         return length;
5197 }
5198 SLAB_ATTR(failslab);
5199 #endif
5200 
5201 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5202 {
5203         return 0;
5204 }
5205 
5206 static ssize_t shrink_store(struct kmem_cache *s,
5207                         const char *buf, size_t length)
5208 {
5209         if (buf[0] == '1')
5210                 kmem_cache_shrink(s);
5211         else
5212                 return -EINVAL;
5213         return length;
5214 }
5215 SLAB_ATTR(shrink);
5216 
5217 #ifdef CONFIG_NUMA
5218 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
5219 {
5220         return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
5221 }
5222 
5223 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
5224                                 const char *buf, size_t length)
5225 {
5226         unsigned long ratio;
5227         int err;
5228 
5229         err = kstrtoul(buf, 10, &ratio);
5230         if (err)
5231                 return err;
5232 
5233         if (ratio <= 100)
5234                 s->remote_node_defrag_ratio = ratio * 10;
5235 
5236         return length;
5237 }
5238 SLAB_ATTR(remote_node_defrag_ratio);
5239 #endif
5240 
5241 #ifdef CONFIG_SLUB_STATS
5242 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5243 {
5244         unsigned long sum  = 0;
5245         int cpu;
5246         int len;
5247         int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
5248 
5249         if (!data)
5250                 return -ENOMEM;
5251 
5252         for_each_online_cpu(cpu) {
5253                 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
5254 
5255                 data[cpu] = x;
5256                 sum += x;
5257         }
5258 
5259         len = sprintf(buf, "%lu", sum);
5260 
5261 #ifdef CONFIG_SMP
5262         for_each_online_cpu(cpu) {
5263                 if (data[cpu] && len < PAGE_SIZE - 20)
5264                         len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
5265         }
5266 #endif
5267         kfree(data);
5268         return len + sprintf(buf + len, "\n");
5269 }
5270 
5271 static void clear_stat(struct kmem_cache *s, enum stat_item si)
5272 {
5273         int cpu;
5274 
5275         for_each_online_cpu(cpu)
5276                 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
5277 }
5278 
5279 #define STAT_ATTR(si, text)                                     \
5280 static ssize_t text##_show(struct kmem_cache *s, char *buf)     \
5281 {                                                               \
5282         return show_stat(s, buf, si);                           \
5283 }                                                               \
5284 static ssize_t text##_store(struct kmem_cache *s,               \
5285                                 const char *buf, size_t length) \
5286 {                                                               \
5287         if (buf[0] != '')                                      \
5288                 return -EINVAL;                                 \
5289         clear_stat(s, si);                                      \
5290         return length;                                          \
5291 }                                                               \
5292 SLAB_ATTR(text);                                                \
5293 
5294 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5295 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5296 STAT_ATTR(FREE_FASTPATH, free_fastpath);
5297 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5298 STAT_ATTR(FREE_FROZEN, free_frozen);
5299 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5300 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5301 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5302 STAT_ATTR(ALLOC_SLAB, alloc_slab);
5303 STAT_ATTR(ALLOC_REFILL, alloc_refill);
5304 STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
5305 STAT_ATTR(FREE_SLAB, free_slab);
5306 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5307 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5308 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5309 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5310 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5311 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
5312 STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
5313 STAT_ATTR(ORDER_FALLBACK, order_fallback);
5314 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5315 STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
5316 STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5317 STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
5318 STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5319 STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
5320 #endif
5321 
5322 static struct attribute *slab_attrs[] = {
5323         &slab_size_attr.attr,
5324         &object_size_attr.attr,
5325         &objs_per_slab_attr.attr,
5326         &order_attr.attr,
5327         &min_partial_attr.attr,
5328         &cpu_partial_attr.attr,
5329         &objects_attr.attr,
5330         &objects_partial_attr.attr,
5331         &partial_attr.attr,
5332         &cpu_slabs_attr.attr,
5333         &ctor_attr.attr,
5334         &aliases_attr.attr,
5335         &align_attr.attr,
5336         &hwcache_align_attr.attr,
5337         &reclaim_account_attr.attr,
5338         &destroy_by_rcu_attr.attr,
5339         &shrink_attr.attr,
5340         &reserved_attr.attr,
5341         &slabs_cpu_partial_attr.attr,
5342 #ifdef CONFIG_SLUB_DEBUG
5343         &total_objects_attr.attr,
5344         &slabs_attr.attr,
5345         &sanity_checks_attr.attr,
5346         &trace_attr.attr,
5347         &red_zone_attr.attr,
5348         &poison_attr.attr,
5349         &store_user_attr.attr,
5350         &validate_attr.attr,
5351         &alloc_calls_attr.attr,
5352         &free_calls_attr.attr,
5353 #endif
5354 #ifdef CONFIG_ZONE_DMA
5355         &cache_dma_attr.attr,
5356 #endif
5357 #ifdef CONFIG_NUMA
5358         &remote_node_defrag_ratio_attr.attr,
5359 #endif
5360 #ifdef CONFIG_SLUB_STATS
5361         &alloc_fastpath_attr.attr,
5362         &alloc_slowpath_attr.attr,
5363         &free_fastpath_attr.attr,
5364         &free_slowpath_attr.attr,
5365         &free_frozen_attr.attr,
5366         &free_add_partial_attr.attr,
5367         &free_remove_partial_attr.attr,
5368         &alloc_from_partial_attr.attr,
5369         &alloc_slab_attr.attr,
5370         &alloc_refill_attr.attr,
5371         &alloc_node_mismatch_attr.attr,
5372         &free_slab_attr.attr,
5373         &cpuslab_flush_attr.attr,
5374         &deactivate_full_attr.attr,
5375         &deactivate_empty_attr.attr,
5376         &deactivate_to_head_attr.attr,
5377         &deactivate_to_tail_attr.attr,
5378         &deactivate_remote_frees_attr.attr,
5379         &deactivate_bypass_attr.attr,
5380         &order_fallback_attr.attr,
5381         &cmpxchg_double_fail_attr.attr,
5382         &cmpxchg_double_cpu_fail_attr.attr,
5383         &cpu_partial_alloc_attr.attr,
5384         &cpu_partial_free_attr.attr,
5385         &cpu_partial_node_attr.attr,
5386         &cpu_partial_drain_attr.attr,
5387 #endif
5388 #ifdef CONFIG_FAILSLAB
5389         &failslab_attr.attr,
5390 #endif
5391 
5392         NULL
5393 };
5394 
5395 static struct attribute_group slab_attr_group = {
5396         .attrs = slab_attrs,
5397 };
5398 
5399 static ssize_t slab_attr_show(struct kobject *kobj,
5400                                 struct attribute *attr,
5401                                 char *buf)
5402 {
5403         struct slab_attribute *attribute;
5404         struct kmem_cache *s;
5405         int err;
5406 
5407         attribute = to_slab_attr(attr);
5408         s = to_slab(kobj);
5409 
5410         if (!attribute->show)
5411                 return -EIO;
5412 
5413         err = attribute->show(s, buf);
5414 
5415         return err;
5416 }
5417 
5418 static ssize_t slab_attr_store(struct kobject *kobj,
5419                                 struct attribute *attr,
5420                                 const char *buf, size_t len)
5421 {
5422         struct slab_attribute *attribute;
5423         struct kmem_cache *s;
5424         int err;
5425 
5426         attribute = to_slab_attr(attr);
5427         s = to_slab(kobj);
5428 
5429         if (!attribute->store)
5430                 return -EIO;
5431 
5432         err = attribute->store(s, buf, len);
5433 #ifdef CONFIG_MEMCG
5434         if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
5435                 struct kmem_cache *c;
5436 
5437                 mutex_lock(&slab_mutex);
5438                 if (s->max_attr_size < len)
5439                         s->max_attr_size = len;
5440 
5441                 /*
5442                  * This is a best effort propagation, so this function's return
5443                  * value will be determined by the parent cache only. This is
5444                  * basically because not all attributes will have a well
5445                  * defined semantics for rollbacks - most of the actions will
5446                  * have permanent effects.
5447                  *
5448                  * Returning the error value of any of the children that fail
5449                  * is not 100 % defined, in the sense that users seeing the
5450                  * error code won't be able to know anything about the state of
5451                  * the cache.
5452                  *
5453                  * Only returning the error code for the parent cache at least
5454                  * has well defined semantics. The cache being written to
5455                  * directly either failed or succeeded, in which case we loop
5456                  * through the descendants with best-effort propagation.
5457                  */
5458                 for_each_memcg_cache(c, s)
5459                         attribute->store(c, buf, len);
5460                 mutex_unlock(&slab_mutex);
5461         }
5462 #endif
5463         return err;
5464 }
5465 
5466 static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5467 {
5468 #ifdef CONFIG_MEMCG
5469         int i;
5470         char *buffer = NULL;
5471         struct kmem_cache *root_cache;
5472 
5473         if (is_root_cache(s))
5474                 return;
5475 
5476         root_cache = s->memcg_params.root_cache;
5477 
5478         /*
5479          * This mean this cache had no attribute written. Therefore, no point
5480          * in copying default values around
5481          */
5482         if (!root_cache->max_attr_size)
5483                 return;
5484 
5485         for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5486                 char mbuf[64];
5487                 char *buf;
5488                 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
5489 
5490                 if (!attr || !attr->store || !attr->show)
5491                         continue;
5492 
5493                 /*
5494                  * It is really bad that we have to allocate here, so we will
5495                  * do it only as a fallback. If we actually allocate, though,
5496                  * we can just use the allocated buffer until the end.
5497                  *
5498                  * Most of the slub attributes will tend to be very small in
5499                  * size, but sysfs allows buffers up to a page, so they can
5500                  * theoretically happen.
5501                  */
5502                 if (buffer)
5503                         buf = buffer;
5504                 else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf))
5505                         buf = mbuf;
5506                 else {
5507                         buffer = (char *) get_zeroed_page(GFP_KERNEL);
5508                         if (WARN_ON(!buffer))
5509                                 continue;
5510                         buf = buffer;
5511                 }
5512 
5513                 attr->show(root_cache, buf);
5514                 attr->store(s, buf, strlen(buf));
5515         }
5516 
5517         if (buffer)
5518                 free_page((unsigned long)buffer);
5519 #endif
5520 }
5521 
5522 static void kmem_cache_release(struct kobject *k)
5523 {
5524         slab_kmem_cache_release(to_slab(k));
5525 }
5526 
5527 static const struct sysfs_ops slab_sysfs_ops = {
5528         .show = slab_attr_show,
5529         .store = slab_attr_store,
5530 };
5531 
5532 static struct kobj_type slab_ktype = {
5533         .sysfs_ops = &slab_sysfs_ops,
5534         .release = kmem_cache_release,
5535 };
5536 
5537 static int uevent_filter(struct kset *kset, struct kobject *kobj)
5538 {
5539         struct kobj_type *ktype = get_ktype(kobj);
5540 
5541         if (ktype == &slab_ktype)
5542                 return 1;
5543         return 0;
5544 }
5545 
5546 static const struct kset_uevent_ops slab_uevent_ops = {
5547         .filter = uevent_filter,
5548 };
5549 
5550 static struct kset *slab_kset;
5551 
5552 static inline struct kset *cache_kset(struct kmem_cache *s)
5553 {
5554 #ifdef CONFIG_MEMCG
5555         if (!is_root_cache(s))
5556                 return s->memcg_params.root_cache->memcg_kset;
5557 #endif
5558         return slab_kset;
5559 }
5560 
5561 #define ID_STR_LENGTH 64
5562 
5563 /* Create a unique string id for a slab cache:
5564  *
5565  * Format       :[flags-]size
5566  */
5567 static char *create_unique_id(struct kmem_cache *s)
5568 {
5569         char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5570         char *p = name;
5571 
5572         BUG_ON(!name);
5573 
5574         *p++ = ':';
5575         /*
5576          * First flags affecting slabcache operations. We will only
5577          * get here for aliasable slabs so we do not need to support
5578          * too many flags. The flags here must cover all flags that
5579          * are matched during merging to guarantee that the id is
5580          * unique.
5581          */
5582         if (s->flags & SLAB_CACHE_DMA)
5583                 *p++ = 'd';
5584         if (s->flags & SLAB_RECLAIM_ACCOUNT)
5585                 *p++ = 'a';
5586         if (s->flags & SLAB_CONSISTENCY_CHECKS)
5587                 *p++ = 'F';
5588         if (!(s->flags & SLAB_NOTRACK))
5589                 *p++ = 't';
5590         if (s->flags & SLAB_ACCOUNT)
5591                 *p++ = 'A';
5592         if (p != name + 1)
5593                 *p++ = '-';
5594         p += sprintf(p, "%07d", s->size);
5595 
5596         BUG_ON(p > name + ID_STR_LENGTH - 1);
5597         return name;
5598 }
5599 
5600 static int sysfs_slab_add(struct kmem_cache *s)
5601 {
5602         int err;
5603         const char *name;
5604         int unmergeable = slab_unmergeable(s);
5605 
5606         if (unmergeable) {
5607                 /*
5608                  * Slabcache can never be merged so we can use the name proper.
5609                  * This is typically the case for debug situations. In that
5610                  * case we can catch duplicate names easily.
5611                  */
5612                 sysfs_remove_link(&slab_kset->kobj, s->name);
5613                 name = s->name;
5614         } else {
5615                 /*
5616                  * Create a unique name for the slab as a target
5617                  * for the symlinks.
5618                  */
5619                 name = create_unique_id(s);
5620         }
5621 
5622         s->kobj.kset = cache_kset(s);
5623         err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
5624         if (err)
5625                 goto out;
5626 
5627         err = sysfs_create_group(&s->kobj, &slab_attr_group);
5628         if (err)
5629                 goto out_del_kobj;
5630 
5631 #ifdef CONFIG_MEMCG
5632         if (is_root_cache(s)) {
5633                 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5634                 if (!s->memcg_kset) {
5635                         err = -ENOMEM;
5636                         goto out_del_kobj;
5637                 }
5638         }
5639 #endif
5640 
5641         kobject_uevent(&s->kobj, KOBJ_ADD);
5642         if (!unmergeable) {
5643                 /* Setup first alias */
5644                 sysfs_slab_alias(s, s->name);
5645         }
5646 out:
5647         if (!unmergeable)
5648                 kfree(name);
5649         return err;
5650 out_del_kobj:
5651         kobject_del(&s->kobj);
5652         goto out;
5653 }
5654 
5655 void sysfs_slab_remove(struct kmem_cache *s)
5656 {
5657         if (slab_state < FULL)
5658                 /*
5659                  * Sysfs has not been setup yet so no need to remove the
5660                  * cache from sysfs.
5661                  */
5662                 return;
5663 
5664 #ifdef CONFIG_MEMCG
5665         kset_unregister(s->memcg_kset);
5666 #endif
5667         kobject_uevent(&s->kobj, KOBJ_REMOVE);
5668         kobject_del(&s->kobj);
5669         kobject_put(&s->kobj);
5670 }
5671 
5672 /*
5673  * Need to buffer aliases during bootup until sysfs becomes
5674  * available lest we lose that information.
5675  */
5676 struct saved_alias {
5677         struct kmem_cache *s;
5678         const char *name;
5679         struct saved_alias *next;
5680 };
5681 
5682 static struct saved_alias *alias_list;
5683 
5684 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5685 {
5686         struct saved_alias *al;
5687 
5688         if (slab_state == FULL) {
5689                 /*
5690                  * If we have a leftover link then remove it.
5691                  */
5692                 sysfs_remove_link(&slab_kset->kobj, name);
5693                 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
5694         }
5695 
5696         al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5697         if (!al)
5698                 return -ENOMEM;
5699 
5700         al->s = s;
5701         al->name = name;
5702         al->next = alias_list;
5703         alias_list = al;
5704         return 0;
5705 }
5706 
5707 static int __init slab_sysfs_init(void)
5708 {
5709         struct kmem_cache *s;
5710         int err;
5711 
5712         mutex_lock(&slab_mutex);
5713 
5714         slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
5715         if (!slab_kset) {
5716                 mutex_unlock(&slab_mutex);
5717                 pr_err("Cannot register slab subsystem.\n");
5718                 return -ENOSYS;
5719         }
5720 
5721         slab_state = FULL;
5722 
5723         list_for_each_entry(s, &slab_caches, list) {
5724                 err = sysfs_slab_add(s);
5725                 if (err)
5726                         pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5727                                s->name);
5728         }
5729 
5730         while (alias_list) {
5731                 struct saved_alias *al = alias_list;
5732 
5733                 alias_list = alias_list->next;
5734                 err = sysfs_slab_alias(al->s, al->name);
5735                 if (err)
5736                         pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5737                                al->name);
5738                 kfree(al);
5739         }
5740 
5741         mutex_unlock(&slab_mutex);
5742         resiliency_test();
5743         return 0;
5744 }
5745 
5746 __initcall(slab_sysfs_init);
5747 #endif /* CONFIG_SYSFS */
5748 
5749 /*
5750  * The /proc/slabinfo ABI
5751  */
5752 #ifdef CONFIG_SLABINFO
5753 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
5754 {
5755         unsigned long nr_slabs = 0;
5756         unsigned long nr_objs = 0;
5757         unsigned long nr_free = 0;
5758         int node;
5759         struct kmem_cache_node *n;
5760 
5761         for_each_kmem_cache_node(s, node, n) {
5762                 nr_slabs += node_nr_slabs(n);
5763                 nr_objs += node_nr_objs(n);
5764                 nr_free += count_partial(n, count_free);
5765         }
5766 
5767         sinfo->active_objs = nr_objs - nr_free;
5768         sinfo->num_objs = nr_objs;
5769         sinfo->active_slabs = nr_slabs;
5770         sinfo->num_slabs = nr_slabs;
5771         sinfo->objects_per_slab = oo_objects(s->oo);
5772         sinfo->cache_order = oo_order(s->oo);
5773 }
5774 
5775 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
5776 {
5777 }
5778 
5779 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5780                        size_t count, loff_t *ppos)
5781 {
5782         return -EIO;
5783 }
5784 #endif /* CONFIG_SLABINFO */
5785 

~ [ 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