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

TOMOYO Linux Cross Reference
Linux/mm/slab.c

Version: ~ [ linux-5.10-rc1 ] ~ [ linux-5.9.1 ] ~ [ linux-5.8.16 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.72 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.152 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.202 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.240 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.240 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  * linux/mm/slab.c
  4  * Written by Mark Hemment, 1996/97.
  5  * (markhe@nextd.demon.co.uk)
  6  *
  7  * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli
  8  *
  9  * Major cleanup, different bufctl logic, per-cpu arrays
 10  *      (c) 2000 Manfred Spraul
 11  *
 12  * Cleanup, make the head arrays unconditional, preparation for NUMA
 13  *      (c) 2002 Manfred Spraul
 14  *
 15  * An implementation of the Slab Allocator as described in outline in;
 16  *      UNIX Internals: The New Frontiers by Uresh Vahalia
 17  *      Pub: Prentice Hall      ISBN 0-13-101908-2
 18  * or with a little more detail in;
 19  *      The Slab Allocator: An Object-Caching Kernel Memory Allocator
 20  *      Jeff Bonwick (Sun Microsystems).
 21  *      Presented at: USENIX Summer 1994 Technical Conference
 22  *
 23  * The memory is organized in caches, one cache for each object type.
 24  * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct)
 25  * Each cache consists out of many slabs (they are small (usually one
 26  * page long) and always contiguous), and each slab contains multiple
 27  * initialized objects.
 28  *
 29  * This means, that your constructor is used only for newly allocated
 30  * slabs and you must pass objects with the same initializations to
 31  * kmem_cache_free.
 32  *
 33  * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM,
 34  * normal). If you need a special memory type, then must create a new
 35  * cache for that memory type.
 36  *
 37  * In order to reduce fragmentation, the slabs are sorted in 3 groups:
 38  *   full slabs with 0 free objects
 39  *   partial slabs
 40  *   empty slabs with no allocated objects
 41  *
 42  * If partial slabs exist, then new allocations come from these slabs,
 43  * otherwise from empty slabs or new slabs are allocated.
 44  *
 45  * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache
 46  * during kmem_cache_destroy(). The caller must prevent concurrent allocs.
 47  *
 48  * Each cache has a short per-cpu head array, most allocs
 49  * and frees go into that array, and if that array overflows, then 1/2
 50  * of the entries in the array are given back into the global cache.
 51  * The head array is strictly LIFO and should improve the cache hit rates.
 52  * On SMP, it additionally reduces the spinlock operations.
 53  *
 54  * The c_cpuarray may not be read with enabled local interrupts -
 55  * it's changed with a smp_call_function().
 56  *
 57  * SMP synchronization:
 58  *  constructors and destructors are called without any locking.
 59  *  Several members in struct kmem_cache and struct slab never change, they
 60  *      are accessed without any locking.
 61  *  The per-cpu arrays are never accessed from the wrong cpu, no locking,
 62  *      and local interrupts are disabled so slab code is preempt-safe.
 63  *  The non-constant members are protected with a per-cache irq spinlock.
 64  *
 65  * Many thanks to Mark Hemment, who wrote another per-cpu slab patch
 66  * in 2000 - many ideas in the current implementation are derived from
 67  * his patch.
 68  *
 69  * Further notes from the original documentation:
 70  *
 71  * 11 April '97.  Started multi-threading - markhe
 72  *      The global cache-chain is protected by the mutex 'slab_mutex'.
 73  *      The sem is only needed when accessing/extending the cache-chain, which
 74  *      can never happen inside an interrupt (kmem_cache_create(),
 75  *      kmem_cache_shrink() and kmem_cache_reap()).
 76  *
 77  *      At present, each engine can be growing a cache.  This should be blocked.
 78  *
 79  * 15 March 2005. NUMA slab allocator.
 80  *      Shai Fultheim <shai@scalex86.org>.
 81  *      Shobhit Dayal <shobhit@calsoftinc.com>
 82  *      Alok N Kataria <alokk@calsoftinc.com>
 83  *      Christoph Lameter <christoph@lameter.com>
 84  *
 85  *      Modified the slab allocator to be node aware on NUMA systems.
 86  *      Each node has its own list of partial, free and full slabs.
 87  *      All object allocations for a node occur from node specific slab lists.
 88  */
 89 
 90 #include        <linux/slab.h>
 91 #include        <linux/mm.h>
 92 #include        <linux/poison.h>
 93 #include        <linux/swap.h>
 94 #include        <linux/cache.h>
 95 #include        <linux/interrupt.h>
 96 #include        <linux/init.h>
 97 #include        <linux/compiler.h>
 98 #include        <linux/cpuset.h>
 99 #include        <linux/proc_fs.h>
100 #include        <linux/seq_file.h>
101 #include        <linux/notifier.h>
102 #include        <linux/kallsyms.h>
103 #include        <linux/cpu.h>
104 #include        <linux/sysctl.h>
105 #include        <linux/module.h>
106 #include        <linux/rcupdate.h>
107 #include        <linux/string.h>
108 #include        <linux/uaccess.h>
109 #include        <linux/nodemask.h>
110 #include        <linux/kmemleak.h>
111 #include        <linux/mempolicy.h>
112 #include        <linux/mutex.h>
113 #include        <linux/fault-inject.h>
114 #include        <linux/rtmutex.h>
115 #include        <linux/reciprocal_div.h>
116 #include        <linux/debugobjects.h>
117 #include        <linux/memory.h>
118 #include        <linux/prefetch.h>
119 #include        <linux/sched/task_stack.h>
120 
121 #include        <net/sock.h>
122 
123 #include        <asm/cacheflush.h>
124 #include        <asm/tlbflush.h>
125 #include        <asm/page.h>
126 
127 #include <trace/events/kmem.h>
128 
129 #include        "internal.h"
130 
131 #include        "slab.h"
132 
133 /*
134  * DEBUG        - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON.
135  *                0 for faster, smaller code (especially in the critical paths).
136  *
137  * STATS        - 1 to collect stats for /proc/slabinfo.
138  *                0 for faster, smaller code (especially in the critical paths).
139  *
140  * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible)
141  */
142 
143 #ifdef CONFIG_DEBUG_SLAB
144 #define DEBUG           1
145 #define STATS           1
146 #define FORCED_DEBUG    1
147 #else
148 #define DEBUG           0
149 #define STATS           0
150 #define FORCED_DEBUG    0
151 #endif
152 
153 /* Shouldn't this be in a header file somewhere? */
154 #define BYTES_PER_WORD          sizeof(void *)
155 #define REDZONE_ALIGN           max(BYTES_PER_WORD, __alignof__(unsigned long long))
156 
157 #ifndef ARCH_KMALLOC_FLAGS
158 #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN
159 #endif
160 
161 #define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \
162                                 <= SLAB_OBJ_MIN_SIZE) ? 1 : 0)
163 
164 #if FREELIST_BYTE_INDEX
165 typedef unsigned char freelist_idx_t;
166 #else
167 typedef unsigned short freelist_idx_t;
168 #endif
169 
170 #define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1)
171 
172 /*
173  * struct array_cache
174  *
175  * Purpose:
176  * - LIFO ordering, to hand out cache-warm objects from _alloc
177  * - reduce the number of linked list operations
178  * - reduce spinlock operations
179  *
180  * The limit is stored in the per-cpu structure to reduce the data cache
181  * footprint.
182  *
183  */
184 struct array_cache {
185         unsigned int avail;
186         unsigned int limit;
187         unsigned int batchcount;
188         unsigned int touched;
189         void *entry[];  /*
190                          * Must have this definition in here for the proper
191                          * alignment of array_cache. Also simplifies accessing
192                          * the entries.
193                          */
194 };
195 
196 struct alien_cache {
197         spinlock_t lock;
198         struct array_cache ac;
199 };
200 
201 /*
202  * Need this for bootstrapping a per node allocator.
203  */
204 #define NUM_INIT_LISTS (2 * MAX_NUMNODES)
205 static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS];
206 #define CACHE_CACHE 0
207 #define SIZE_NODE (MAX_NUMNODES)
208 
209 static int drain_freelist(struct kmem_cache *cache,
210                         struct kmem_cache_node *n, int tofree);
211 static void free_block(struct kmem_cache *cachep, void **objpp, int len,
212                         int node, struct list_head *list);
213 static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list);
214 static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp);
215 static void cache_reap(struct work_struct *unused);
216 
217 static inline void fixup_objfreelist_debug(struct kmem_cache *cachep,
218                                                 void **list);
219 static inline void fixup_slab_list(struct kmem_cache *cachep,
220                                 struct kmem_cache_node *n, struct page *page,
221                                 void **list);
222 static int slab_early_init = 1;
223 
224 #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node))
225 
226 static void kmem_cache_node_init(struct kmem_cache_node *parent)
227 {
228         INIT_LIST_HEAD(&parent->slabs_full);
229         INIT_LIST_HEAD(&parent->slabs_partial);
230         INIT_LIST_HEAD(&parent->slabs_free);
231         parent->total_slabs = 0;
232         parent->free_slabs = 0;
233         parent->shared = NULL;
234         parent->alien = NULL;
235         parent->colour_next = 0;
236         spin_lock_init(&parent->list_lock);
237         parent->free_objects = 0;
238         parent->free_touched = 0;
239 }
240 
241 #define MAKE_LIST(cachep, listp, slab, nodeid)                          \
242         do {                                                            \
243                 INIT_LIST_HEAD(listp);                                  \
244                 list_splice(&get_node(cachep, nodeid)->slab, listp);    \
245         } while (0)
246 
247 #define MAKE_ALL_LISTS(cachep, ptr, nodeid)                             \
248         do {                                                            \
249         MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid);  \
250         MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \
251         MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid);  \
252         } while (0)
253 
254 #define CFLGS_OBJFREELIST_SLAB  ((slab_flags_t __force)0x40000000U)
255 #define CFLGS_OFF_SLAB          ((slab_flags_t __force)0x80000000U)
256 #define OBJFREELIST_SLAB(x)     ((x)->flags & CFLGS_OBJFREELIST_SLAB)
257 #define OFF_SLAB(x)     ((x)->flags & CFLGS_OFF_SLAB)
258 
259 #define BATCHREFILL_LIMIT       16
260 /*
261  * Optimization question: fewer reaps means less probability for unnessary
262  * cpucache drain/refill cycles.
263  *
264  * OTOH the cpuarrays can contain lots of objects,
265  * which could lock up otherwise freeable slabs.
266  */
267 #define REAPTIMEOUT_AC          (2*HZ)
268 #define REAPTIMEOUT_NODE        (4*HZ)
269 
270 #if STATS
271 #define STATS_INC_ACTIVE(x)     ((x)->num_active++)
272 #define STATS_DEC_ACTIVE(x)     ((x)->num_active--)
273 #define STATS_INC_ALLOCED(x)    ((x)->num_allocations++)
274 #define STATS_INC_GROWN(x)      ((x)->grown++)
275 #define STATS_ADD_REAPED(x,y)   ((x)->reaped += (y))
276 #define STATS_SET_HIGH(x)                                               \
277         do {                                                            \
278                 if ((x)->num_active > (x)->high_mark)                   \
279                         (x)->high_mark = (x)->num_active;               \
280         } while (0)
281 #define STATS_INC_ERR(x)        ((x)->errors++)
282 #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++)
283 #define STATS_INC_NODEFREES(x)  ((x)->node_frees++)
284 #define STATS_INC_ACOVERFLOW(x)   ((x)->node_overflow++)
285 #define STATS_SET_FREEABLE(x, i)                                        \
286         do {                                                            \
287                 if ((x)->max_freeable < i)                              \
288                         (x)->max_freeable = i;                          \
289         } while (0)
290 #define STATS_INC_ALLOCHIT(x)   atomic_inc(&(x)->allochit)
291 #define STATS_INC_ALLOCMISS(x)  atomic_inc(&(x)->allocmiss)
292 #define STATS_INC_FREEHIT(x)    atomic_inc(&(x)->freehit)
293 #define STATS_INC_FREEMISS(x)   atomic_inc(&(x)->freemiss)
294 #else
295 #define STATS_INC_ACTIVE(x)     do { } while (0)
296 #define STATS_DEC_ACTIVE(x)     do { } while (0)
297 #define STATS_INC_ALLOCED(x)    do { } while (0)
298 #define STATS_INC_GROWN(x)      do { } while (0)
299 #define STATS_ADD_REAPED(x,y)   do { (void)(y); } while (0)
300 #define STATS_SET_HIGH(x)       do { } while (0)
301 #define STATS_INC_ERR(x)        do { } while (0)
302 #define STATS_INC_NODEALLOCS(x) do { } while (0)
303 #define STATS_INC_NODEFREES(x)  do { } while (0)
304 #define STATS_INC_ACOVERFLOW(x)   do { } while (0)
305 #define STATS_SET_FREEABLE(x, i) do { } while (0)
306 #define STATS_INC_ALLOCHIT(x)   do { } while (0)
307 #define STATS_INC_ALLOCMISS(x)  do { } while (0)
308 #define STATS_INC_FREEHIT(x)    do { } while (0)
309 #define STATS_INC_FREEMISS(x)   do { } while (0)
310 #endif
311 
312 #if DEBUG
313 
314 /*
315  * memory layout of objects:
316  * 0            : objp
317  * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that
318  *              the end of an object is aligned with the end of the real
319  *              allocation. Catches writes behind the end of the allocation.
320  * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1:
321  *              redzone word.
322  * cachep->obj_offset: The real object.
323  * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long]
324  * cachep->size - 1* BYTES_PER_WORD: last caller address
325  *                                      [BYTES_PER_WORD long]
326  */
327 static int obj_offset(struct kmem_cache *cachep)
328 {
329         return cachep->obj_offset;
330 }
331 
332 static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp)
333 {
334         BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
335         return (unsigned long long*) (objp + obj_offset(cachep) -
336                                       sizeof(unsigned long long));
337 }
338 
339 static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp)
340 {
341         BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
342         if (cachep->flags & SLAB_STORE_USER)
343                 return (unsigned long long *)(objp + cachep->size -
344                                               sizeof(unsigned long long) -
345                                               REDZONE_ALIGN);
346         return (unsigned long long *) (objp + cachep->size -
347                                        sizeof(unsigned long long));
348 }
349 
350 static void **dbg_userword(struct kmem_cache *cachep, void *objp)
351 {
352         BUG_ON(!(cachep->flags & SLAB_STORE_USER));
353         return (void **)(objp + cachep->size - BYTES_PER_WORD);
354 }
355 
356 #else
357 
358 #define obj_offset(x)                   0
359 #define dbg_redzone1(cachep, objp)      ({BUG(); (unsigned long long *)NULL;})
360 #define dbg_redzone2(cachep, objp)      ({BUG(); (unsigned long long *)NULL;})
361 #define dbg_userword(cachep, objp)      ({BUG(); (void **)NULL;})
362 
363 #endif
364 
365 #ifdef CONFIG_DEBUG_SLAB_LEAK
366 
367 static inline bool is_store_user_clean(struct kmem_cache *cachep)
368 {
369         return atomic_read(&cachep->store_user_clean) == 1;
370 }
371 
372 static inline void set_store_user_clean(struct kmem_cache *cachep)
373 {
374         atomic_set(&cachep->store_user_clean, 1);
375 }
376 
377 static inline void set_store_user_dirty(struct kmem_cache *cachep)
378 {
379         if (is_store_user_clean(cachep))
380                 atomic_set(&cachep->store_user_clean, 0);
381 }
382 
383 #else
384 static inline void set_store_user_dirty(struct kmem_cache *cachep) {}
385 
386 #endif
387 
388 /*
389  * Do not go above this order unless 0 objects fit into the slab or
390  * overridden on the command line.
391  */
392 #define SLAB_MAX_ORDER_HI       1
393 #define SLAB_MAX_ORDER_LO       0
394 static int slab_max_order = SLAB_MAX_ORDER_LO;
395 static bool slab_max_order_set __initdata;
396 
397 static inline struct kmem_cache *virt_to_cache(const void *obj)
398 {
399         struct page *page = virt_to_head_page(obj);
400         return page->slab_cache;
401 }
402 
403 static inline void *index_to_obj(struct kmem_cache *cache, struct page *page,
404                                  unsigned int idx)
405 {
406         return page->s_mem + cache->size * idx;
407 }
408 
409 /*
410  * We want to avoid an expensive divide : (offset / cache->size)
411  *   Using the fact that size is a constant for a particular cache,
412  *   we can replace (offset / cache->size) by
413  *   reciprocal_divide(offset, cache->reciprocal_buffer_size)
414  */
415 static inline unsigned int obj_to_index(const struct kmem_cache *cache,
416                                         const struct page *page, void *obj)
417 {
418         u32 offset = (obj - page->s_mem);
419         return reciprocal_divide(offset, cache->reciprocal_buffer_size);
420 }
421 
422 #define BOOT_CPUCACHE_ENTRIES   1
423 /* internal cache of cache description objs */
424 static struct kmem_cache kmem_cache_boot = {
425         .batchcount = 1,
426         .limit = BOOT_CPUCACHE_ENTRIES,
427         .shared = 1,
428         .size = sizeof(struct kmem_cache),
429         .name = "kmem_cache",
430 };
431 
432 static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
433 
434 static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
435 {
436         return this_cpu_ptr(cachep->cpu_cache);
437 }
438 
439 /*
440  * Calculate the number of objects and left-over bytes for a given buffer size.
441  */
442 static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size,
443                 slab_flags_t flags, size_t *left_over)
444 {
445         unsigned int num;
446         size_t slab_size = PAGE_SIZE << gfporder;
447 
448         /*
449          * The slab management structure can be either off the slab or
450          * on it. For the latter case, the memory allocated for a
451          * slab is used for:
452          *
453          * - @buffer_size bytes for each object
454          * - One freelist_idx_t for each object
455          *
456          * We don't need to consider alignment of freelist because
457          * freelist will be at the end of slab page. The objects will be
458          * at the correct alignment.
459          *
460          * If the slab management structure is off the slab, then the
461          * alignment will already be calculated into the size. Because
462          * the slabs are all pages aligned, the objects will be at the
463          * correct alignment when allocated.
464          */
465         if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) {
466                 num = slab_size / buffer_size;
467                 *left_over = slab_size % buffer_size;
468         } else {
469                 num = slab_size / (buffer_size + sizeof(freelist_idx_t));
470                 *left_over = slab_size %
471                         (buffer_size + sizeof(freelist_idx_t));
472         }
473 
474         return num;
475 }
476 
477 #if DEBUG
478 #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
479 
480 static void __slab_error(const char *function, struct kmem_cache *cachep,
481                         char *msg)
482 {
483         pr_err("slab error in %s(): cache `%s': %s\n",
484                function, cachep->name, msg);
485         dump_stack();
486         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
487 }
488 #endif
489 
490 /*
491  * By default on NUMA we use alien caches to stage the freeing of
492  * objects allocated from other nodes. This causes massive memory
493  * inefficiencies when using fake NUMA setup to split memory into a
494  * large number of small nodes, so it can be disabled on the command
495  * line
496   */
497 
498 static int use_alien_caches __read_mostly = 1;
499 static int __init noaliencache_setup(char *s)
500 {
501         use_alien_caches = 0;
502         return 1;
503 }
504 __setup("noaliencache", noaliencache_setup);
505 
506 static int __init slab_max_order_setup(char *str)
507 {
508         get_option(&str, &slab_max_order);
509         slab_max_order = slab_max_order < 0 ? 0 :
510                                 min(slab_max_order, MAX_ORDER - 1);
511         slab_max_order_set = true;
512 
513         return 1;
514 }
515 __setup("slab_max_order=", slab_max_order_setup);
516 
517 #ifdef CONFIG_NUMA
518 /*
519  * Special reaping functions for NUMA systems called from cache_reap().
520  * These take care of doing round robin flushing of alien caches (containing
521  * objects freed on different nodes from which they were allocated) and the
522  * flushing of remote pcps by calling drain_node_pages.
523  */
524 static DEFINE_PER_CPU(unsigned long, slab_reap_node);
525 
526 static void init_reap_node(int cpu)
527 {
528         per_cpu(slab_reap_node, cpu) = next_node_in(cpu_to_mem(cpu),
529                                                     node_online_map);
530 }
531 
532 static void next_reap_node(void)
533 {
534         int node = __this_cpu_read(slab_reap_node);
535 
536         node = next_node_in(node, node_online_map);
537         __this_cpu_write(slab_reap_node, node);
538 }
539 
540 #else
541 #define init_reap_node(cpu) do { } while (0)
542 #define next_reap_node(void) do { } while (0)
543 #endif
544 
545 /*
546  * Initiate the reap timer running on the target CPU.  We run at around 1 to 2Hz
547  * via the workqueue/eventd.
548  * Add the CPU number into the expiration time to minimize the possibility of
549  * the CPUs getting into lockstep and contending for the global cache chain
550  * lock.
551  */
552 static void start_cpu_timer(int cpu)
553 {
554         struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
555 
556         if (reap_work->work.func == NULL) {
557                 init_reap_node(cpu);
558                 INIT_DEFERRABLE_WORK(reap_work, cache_reap);
559                 schedule_delayed_work_on(cpu, reap_work,
560                                         __round_jiffies_relative(HZ, cpu));
561         }
562 }
563 
564 static void init_arraycache(struct array_cache *ac, int limit, int batch)
565 {
566         /*
567          * The array_cache structures contain pointers to free object.
568          * However, when such objects are allocated or transferred to another
569          * cache the pointers are not cleared and they could be counted as
570          * valid references during a kmemleak scan. Therefore, kmemleak must
571          * not scan such objects.
572          */
573         kmemleak_no_scan(ac);
574         if (ac) {
575                 ac->avail = 0;
576                 ac->limit = limit;
577                 ac->batchcount = batch;
578                 ac->touched = 0;
579         }
580 }
581 
582 static struct array_cache *alloc_arraycache(int node, int entries,
583                                             int batchcount, gfp_t gfp)
584 {
585         size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache);
586         struct array_cache *ac = NULL;
587 
588         ac = kmalloc_node(memsize, gfp, node);
589         init_arraycache(ac, entries, batchcount);
590         return ac;
591 }
592 
593 static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep,
594                                         struct page *page, void *objp)
595 {
596         struct kmem_cache_node *n;
597         int page_node;
598         LIST_HEAD(list);
599 
600         page_node = page_to_nid(page);
601         n = get_node(cachep, page_node);
602 
603         spin_lock(&n->list_lock);
604         free_block(cachep, &objp, 1, page_node, &list);
605         spin_unlock(&n->list_lock);
606 
607         slabs_destroy(cachep, &list);
608 }
609 
610 /*
611  * Transfer objects in one arraycache to another.
612  * Locking must be handled by the caller.
613  *
614  * Return the number of entries transferred.
615  */
616 static int transfer_objects(struct array_cache *to,
617                 struct array_cache *from, unsigned int max)
618 {
619         /* Figure out how many entries to transfer */
620         int nr = min3(from->avail, max, to->limit - to->avail);
621 
622         if (!nr)
623                 return 0;
624 
625         memcpy(to->entry + to->avail, from->entry + from->avail -nr,
626                         sizeof(void *) *nr);
627 
628         from->avail -= nr;
629         to->avail += nr;
630         return nr;
631 }
632 
633 #ifndef CONFIG_NUMA
634 
635 #define drain_alien_cache(cachep, alien) do { } while (0)
636 #define reap_alien(cachep, n) do { } while (0)
637 
638 static inline struct alien_cache **alloc_alien_cache(int node,
639                                                 int limit, gfp_t gfp)
640 {
641         return NULL;
642 }
643 
644 static inline void free_alien_cache(struct alien_cache **ac_ptr)
645 {
646 }
647 
648 static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
649 {
650         return 0;
651 }
652 
653 static inline void *alternate_node_alloc(struct kmem_cache *cachep,
654                 gfp_t flags)
655 {
656         return NULL;
657 }
658 
659 static inline void *____cache_alloc_node(struct kmem_cache *cachep,
660                  gfp_t flags, int nodeid)
661 {
662         return NULL;
663 }
664 
665 static inline gfp_t gfp_exact_node(gfp_t flags)
666 {
667         return flags & ~__GFP_NOFAIL;
668 }
669 
670 #else   /* CONFIG_NUMA */
671 
672 static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
673 static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
674 
675 static struct alien_cache *__alloc_alien_cache(int node, int entries,
676                                                 int batch, gfp_t gfp)
677 {
678         size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache);
679         struct alien_cache *alc = NULL;
680 
681         alc = kmalloc_node(memsize, gfp, node);
682         init_arraycache(&alc->ac, entries, batch);
683         spin_lock_init(&alc->lock);
684         return alc;
685 }
686 
687 static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
688 {
689         struct alien_cache **alc_ptr;
690         size_t memsize = sizeof(void *) * nr_node_ids;
691         int i;
692 
693         if (limit > 1)
694                 limit = 12;
695         alc_ptr = kzalloc_node(memsize, gfp, node);
696         if (!alc_ptr)
697                 return NULL;
698 
699         for_each_node(i) {
700                 if (i == node || !node_online(i))
701                         continue;
702                 alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp);
703                 if (!alc_ptr[i]) {
704                         for (i--; i >= 0; i--)
705                                 kfree(alc_ptr[i]);
706                         kfree(alc_ptr);
707                         return NULL;
708                 }
709         }
710         return alc_ptr;
711 }
712 
713 static void free_alien_cache(struct alien_cache **alc_ptr)
714 {
715         int i;
716 
717         if (!alc_ptr)
718                 return;
719         for_each_node(i)
720             kfree(alc_ptr[i]);
721         kfree(alc_ptr);
722 }
723 
724 static void __drain_alien_cache(struct kmem_cache *cachep,
725                                 struct array_cache *ac, int node,
726                                 struct list_head *list)
727 {
728         struct kmem_cache_node *n = get_node(cachep, node);
729 
730         if (ac->avail) {
731                 spin_lock(&n->list_lock);
732                 /*
733                  * Stuff objects into the remote nodes shared array first.
734                  * That way we could avoid the overhead of putting the objects
735                  * into the free lists and getting them back later.
736                  */
737                 if (n->shared)
738                         transfer_objects(n->shared, ac, ac->limit);
739 
740                 free_block(cachep, ac->entry, ac->avail, node, list);
741                 ac->avail = 0;
742                 spin_unlock(&n->list_lock);
743         }
744 }
745 
746 /*
747  * Called from cache_reap() to regularly drain alien caches round robin.
748  */
749 static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n)
750 {
751         int node = __this_cpu_read(slab_reap_node);
752 
753         if (n->alien) {
754                 struct alien_cache *alc = n->alien[node];
755                 struct array_cache *ac;
756 
757                 if (alc) {
758                         ac = &alc->ac;
759                         if (ac->avail && spin_trylock_irq(&alc->lock)) {
760                                 LIST_HEAD(list);
761 
762                                 __drain_alien_cache(cachep, ac, node, &list);
763                                 spin_unlock_irq(&alc->lock);
764                                 slabs_destroy(cachep, &list);
765                         }
766                 }
767         }
768 }
769 
770 static void drain_alien_cache(struct kmem_cache *cachep,
771                                 struct alien_cache **alien)
772 {
773         int i = 0;
774         struct alien_cache *alc;
775         struct array_cache *ac;
776         unsigned long flags;
777 
778         for_each_online_node(i) {
779                 alc = alien[i];
780                 if (alc) {
781                         LIST_HEAD(list);
782 
783                         ac = &alc->ac;
784                         spin_lock_irqsave(&alc->lock, flags);
785                         __drain_alien_cache(cachep, ac, i, &list);
786                         spin_unlock_irqrestore(&alc->lock, flags);
787                         slabs_destroy(cachep, &list);
788                 }
789         }
790 }
791 
792 static int __cache_free_alien(struct kmem_cache *cachep, void *objp,
793                                 int node, int page_node)
794 {
795         struct kmem_cache_node *n;
796         struct alien_cache *alien = NULL;
797         struct array_cache *ac;
798         LIST_HEAD(list);
799 
800         n = get_node(cachep, node);
801         STATS_INC_NODEFREES(cachep);
802         if (n->alien && n->alien[page_node]) {
803                 alien = n->alien[page_node];
804                 ac = &alien->ac;
805                 spin_lock(&alien->lock);
806                 if (unlikely(ac->avail == ac->limit)) {
807                         STATS_INC_ACOVERFLOW(cachep);
808                         __drain_alien_cache(cachep, ac, page_node, &list);
809                 }
810                 ac->entry[ac->avail++] = objp;
811                 spin_unlock(&alien->lock);
812                 slabs_destroy(cachep, &list);
813         } else {
814                 n = get_node(cachep, page_node);
815                 spin_lock(&n->list_lock);
816                 free_block(cachep, &objp, 1, page_node, &list);
817                 spin_unlock(&n->list_lock);
818                 slabs_destroy(cachep, &list);
819         }
820         return 1;
821 }
822 
823 static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
824 {
825         int page_node = page_to_nid(virt_to_page(objp));
826         int node = numa_mem_id();
827         /*
828          * Make sure we are not freeing a object from another node to the array
829          * cache on this cpu.
830          */
831         if (likely(node == page_node))
832                 return 0;
833 
834         return __cache_free_alien(cachep, objp, node, page_node);
835 }
836 
837 /*
838  * Construct gfp mask to allocate from a specific node but do not reclaim or
839  * warn about failures.
840  */
841 static inline gfp_t gfp_exact_node(gfp_t flags)
842 {
843         return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
844 }
845 #endif
846 
847 static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp)
848 {
849         struct kmem_cache_node *n;
850 
851         /*
852          * Set up the kmem_cache_node for cpu before we can
853          * begin anything. Make sure some other cpu on this
854          * node has not already allocated this
855          */
856         n = get_node(cachep, node);
857         if (n) {
858                 spin_lock_irq(&n->list_lock);
859                 n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount +
860                                 cachep->num;
861                 spin_unlock_irq(&n->list_lock);
862 
863                 return 0;
864         }
865 
866         n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node);
867         if (!n)
868                 return -ENOMEM;
869 
870         kmem_cache_node_init(n);
871         n->next_reap = jiffies + REAPTIMEOUT_NODE +
872                     ((unsigned long)cachep) % REAPTIMEOUT_NODE;
873 
874         n->free_limit =
875                 (1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num;
876 
877         /*
878          * The kmem_cache_nodes don't come and go as CPUs
879          * come and go.  slab_mutex is sufficient
880          * protection here.
881          */
882         cachep->node[node] = n;
883 
884         return 0;
885 }
886 
887 #if (defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)) || defined(CONFIG_SMP)
888 /*
889  * Allocates and initializes node for a node on each slab cache, used for
890  * either memory or cpu hotplug.  If memory is being hot-added, the kmem_cache_node
891  * will be allocated off-node since memory is not yet online for the new node.
892  * When hotplugging memory or a cpu, existing node are not replaced if
893  * already in use.
894  *
895  * Must hold slab_mutex.
896  */
897 static int init_cache_node_node(int node)
898 {
899         int ret;
900         struct kmem_cache *cachep;
901 
902         list_for_each_entry(cachep, &slab_caches, list) {
903                 ret = init_cache_node(cachep, node, GFP_KERNEL);
904                 if (ret)
905                         return ret;
906         }
907 
908         return 0;
909 }
910 #endif
911 
912 static int setup_kmem_cache_node(struct kmem_cache *cachep,
913                                 int node, gfp_t gfp, bool force_change)
914 {
915         int ret = -ENOMEM;
916         struct kmem_cache_node *n;
917         struct array_cache *old_shared = NULL;
918         struct array_cache *new_shared = NULL;
919         struct alien_cache **new_alien = NULL;
920         LIST_HEAD(list);
921 
922         if (use_alien_caches) {
923                 new_alien = alloc_alien_cache(node, cachep->limit, gfp);
924                 if (!new_alien)
925                         goto fail;
926         }
927 
928         if (cachep->shared) {
929                 new_shared = alloc_arraycache(node,
930                         cachep->shared * cachep->batchcount, 0xbaadf00d, gfp);
931                 if (!new_shared)
932                         goto fail;
933         }
934 
935         ret = init_cache_node(cachep, node, gfp);
936         if (ret)
937                 goto fail;
938 
939         n = get_node(cachep, node);
940         spin_lock_irq(&n->list_lock);
941         if (n->shared && force_change) {
942                 free_block(cachep, n->shared->entry,
943                                 n->shared->avail, node, &list);
944                 n->shared->avail = 0;
945         }
946 
947         if (!n->shared || force_change) {
948                 old_shared = n->shared;
949                 n->shared = new_shared;
950                 new_shared = NULL;
951         }
952 
953         if (!n->alien) {
954                 n->alien = new_alien;
955                 new_alien = NULL;
956         }
957 
958         spin_unlock_irq(&n->list_lock);
959         slabs_destroy(cachep, &list);
960 
961         /*
962          * To protect lockless access to n->shared during irq disabled context.
963          * If n->shared isn't NULL in irq disabled context, accessing to it is
964          * guaranteed to be valid until irq is re-enabled, because it will be
965          * freed after synchronize_sched().
966          */
967         if (old_shared && force_change)
968                 synchronize_sched();
969 
970 fail:
971         kfree(old_shared);
972         kfree(new_shared);
973         free_alien_cache(new_alien);
974 
975         return ret;
976 }
977 
978 #ifdef CONFIG_SMP
979 
980 static void cpuup_canceled(long cpu)
981 {
982         struct kmem_cache *cachep;
983         struct kmem_cache_node *n = NULL;
984         int node = cpu_to_mem(cpu);
985         const struct cpumask *mask = cpumask_of_node(node);
986 
987         list_for_each_entry(cachep, &slab_caches, list) {
988                 struct array_cache *nc;
989                 struct array_cache *shared;
990                 struct alien_cache **alien;
991                 LIST_HEAD(list);
992 
993                 n = get_node(cachep, node);
994                 if (!n)
995                         continue;
996 
997                 spin_lock_irq(&n->list_lock);
998 
999                 /* Free limit for this kmem_cache_node */
1000                 n->free_limit -= cachep->batchcount;
1001 
1002                 /* cpu is dead; no one can alloc from it. */
1003                 nc = per_cpu_ptr(cachep->cpu_cache, cpu);
1004                 if (nc) {
1005                         free_block(cachep, nc->entry, nc->avail, node, &list);
1006                         nc->avail = 0;
1007                 }
1008 
1009                 if (!cpumask_empty(mask)) {
1010                         spin_unlock_irq(&n->list_lock);
1011                         goto free_slab;
1012                 }
1013 
1014                 shared = n->shared;
1015                 if (shared) {
1016                         free_block(cachep, shared->entry,
1017                                    shared->avail, node, &list);
1018                         n->shared = NULL;
1019                 }
1020 
1021                 alien = n->alien;
1022                 n->alien = NULL;
1023 
1024                 spin_unlock_irq(&n->list_lock);
1025 
1026                 kfree(shared);
1027                 if (alien) {
1028                         drain_alien_cache(cachep, alien);
1029                         free_alien_cache(alien);
1030                 }
1031 
1032 free_slab:
1033                 slabs_destroy(cachep, &list);
1034         }
1035         /*
1036          * In the previous loop, all the objects were freed to
1037          * the respective cache's slabs,  now we can go ahead and
1038          * shrink each nodelist to its limit.
1039          */
1040         list_for_each_entry(cachep, &slab_caches, list) {
1041                 n = get_node(cachep, node);
1042                 if (!n)
1043                         continue;
1044                 drain_freelist(cachep, n, INT_MAX);
1045         }
1046 }
1047 
1048 static int cpuup_prepare(long cpu)
1049 {
1050         struct kmem_cache *cachep;
1051         int node = cpu_to_mem(cpu);
1052         int err;
1053 
1054         /*
1055          * We need to do this right in the beginning since
1056          * alloc_arraycache's are going to use this list.
1057          * kmalloc_node allows us to add the slab to the right
1058          * kmem_cache_node and not this cpu's kmem_cache_node
1059          */
1060         err = init_cache_node_node(node);
1061         if (err < 0)
1062                 goto bad;
1063 
1064         /*
1065          * Now we can go ahead with allocating the shared arrays and
1066          * array caches
1067          */
1068         list_for_each_entry(cachep, &slab_caches, list) {
1069                 err = setup_kmem_cache_node(cachep, node, GFP_KERNEL, false);
1070                 if (err)
1071                         goto bad;
1072         }
1073 
1074         return 0;
1075 bad:
1076         cpuup_canceled(cpu);
1077         return -ENOMEM;
1078 }
1079 
1080 int slab_prepare_cpu(unsigned int cpu)
1081 {
1082         int err;
1083 
1084         mutex_lock(&slab_mutex);
1085         err = cpuup_prepare(cpu);
1086         mutex_unlock(&slab_mutex);
1087         return err;
1088 }
1089 
1090 /*
1091  * This is called for a failed online attempt and for a successful
1092  * offline.
1093  *
1094  * Even if all the cpus of a node are down, we don't free the
1095  * kmem_list3 of any cache. This to avoid a race between cpu_down, and
1096  * a kmalloc allocation from another cpu for memory from the node of
1097  * the cpu going down.  The list3 structure is usually allocated from
1098  * kmem_cache_create() and gets destroyed at kmem_cache_destroy().
1099  */
1100 int slab_dead_cpu(unsigned int cpu)
1101 {
1102         mutex_lock(&slab_mutex);
1103         cpuup_canceled(cpu);
1104         mutex_unlock(&slab_mutex);
1105         return 0;
1106 }
1107 #endif
1108 
1109 static int slab_online_cpu(unsigned int cpu)
1110 {
1111         start_cpu_timer(cpu);
1112         return 0;
1113 }
1114 
1115 static int slab_offline_cpu(unsigned int cpu)
1116 {
1117         /*
1118          * Shutdown cache reaper. Note that the slab_mutex is held so
1119          * that if cache_reap() is invoked it cannot do anything
1120          * expensive but will only modify reap_work and reschedule the
1121          * timer.
1122          */
1123         cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu));
1124         /* Now the cache_reaper is guaranteed to be not running. */
1125         per_cpu(slab_reap_work, cpu).work.func = NULL;
1126         return 0;
1127 }
1128 
1129 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
1130 /*
1131  * Drains freelist for a node on each slab cache, used for memory hot-remove.
1132  * Returns -EBUSY if all objects cannot be drained so that the node is not
1133  * removed.
1134  *
1135  * Must hold slab_mutex.
1136  */
1137 static int __meminit drain_cache_node_node(int node)
1138 {
1139         struct kmem_cache *cachep;
1140         int ret = 0;
1141 
1142         list_for_each_entry(cachep, &slab_caches, list) {
1143                 struct kmem_cache_node *n;
1144 
1145                 n = get_node(cachep, node);
1146                 if (!n)
1147                         continue;
1148 
1149                 drain_freelist(cachep, n, INT_MAX);
1150 
1151                 if (!list_empty(&n->slabs_full) ||
1152                     !list_empty(&n->slabs_partial)) {
1153                         ret = -EBUSY;
1154                         break;
1155                 }
1156         }
1157         return ret;
1158 }
1159 
1160 static int __meminit slab_memory_callback(struct notifier_block *self,
1161                                         unsigned long action, void *arg)
1162 {
1163         struct memory_notify *mnb = arg;
1164         int ret = 0;
1165         int nid;
1166 
1167         nid = mnb->status_change_nid;
1168         if (nid < 0)
1169                 goto out;
1170 
1171         switch (action) {
1172         case MEM_GOING_ONLINE:
1173                 mutex_lock(&slab_mutex);
1174                 ret = init_cache_node_node(nid);
1175                 mutex_unlock(&slab_mutex);
1176                 break;
1177         case MEM_GOING_OFFLINE:
1178                 mutex_lock(&slab_mutex);
1179                 ret = drain_cache_node_node(nid);
1180                 mutex_unlock(&slab_mutex);
1181                 break;
1182         case MEM_ONLINE:
1183         case MEM_OFFLINE:
1184         case MEM_CANCEL_ONLINE:
1185         case MEM_CANCEL_OFFLINE:
1186                 break;
1187         }
1188 out:
1189         return notifier_from_errno(ret);
1190 }
1191 #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */
1192 
1193 /*
1194  * swap the static kmem_cache_node with kmalloced memory
1195  */
1196 static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list,
1197                                 int nodeid)
1198 {
1199         struct kmem_cache_node *ptr;
1200 
1201         ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid);
1202         BUG_ON(!ptr);
1203 
1204         memcpy(ptr, list, sizeof(struct kmem_cache_node));
1205         /*
1206          * Do not assume that spinlocks can be initialized via memcpy:
1207          */
1208         spin_lock_init(&ptr->list_lock);
1209 
1210         MAKE_ALL_LISTS(cachep, ptr, nodeid);
1211         cachep->node[nodeid] = ptr;
1212 }
1213 
1214 /*
1215  * For setting up all the kmem_cache_node for cache whose buffer_size is same as
1216  * size of kmem_cache_node.
1217  */
1218 static void __init set_up_node(struct kmem_cache *cachep, int index)
1219 {
1220         int node;
1221 
1222         for_each_online_node(node) {
1223                 cachep->node[node] = &init_kmem_cache_node[index + node];
1224                 cachep->node[node]->next_reap = jiffies +
1225                     REAPTIMEOUT_NODE +
1226                     ((unsigned long)cachep) % REAPTIMEOUT_NODE;
1227         }
1228 }
1229 
1230 /*
1231  * Initialisation.  Called after the page allocator have been initialised and
1232  * before smp_init().
1233  */
1234 void __init kmem_cache_init(void)
1235 {
1236         int i;
1237 
1238         BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) <
1239                                         sizeof(struct rcu_head));
1240         kmem_cache = &kmem_cache_boot;
1241 
1242         if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1)
1243                 use_alien_caches = 0;
1244 
1245         for (i = 0; i < NUM_INIT_LISTS; i++)
1246                 kmem_cache_node_init(&init_kmem_cache_node[i]);
1247 
1248         /*
1249          * Fragmentation resistance on low memory - only use bigger
1250          * page orders on machines with more than 32MB of memory if
1251          * not overridden on the command line.
1252          */
1253         if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT)
1254                 slab_max_order = SLAB_MAX_ORDER_HI;
1255 
1256         /* Bootstrap is tricky, because several objects are allocated
1257          * from caches that do not exist yet:
1258          * 1) initialize the kmem_cache cache: it contains the struct
1259          *    kmem_cache structures of all caches, except kmem_cache itself:
1260          *    kmem_cache is statically allocated.
1261          *    Initially an __init data area is used for the head array and the
1262          *    kmem_cache_node structures, it's replaced with a kmalloc allocated
1263          *    array at the end of the bootstrap.
1264          * 2) Create the first kmalloc cache.
1265          *    The struct kmem_cache for the new cache is allocated normally.
1266          *    An __init data area is used for the head array.
1267          * 3) Create the remaining kmalloc caches, with minimally sized
1268          *    head arrays.
1269          * 4) Replace the __init data head arrays for kmem_cache and the first
1270          *    kmalloc cache with kmalloc allocated arrays.
1271          * 5) Replace the __init data for kmem_cache_node for kmem_cache and
1272          *    the other cache's with kmalloc allocated memory.
1273          * 6) Resize the head arrays of the kmalloc caches to their final sizes.
1274          */
1275 
1276         /* 1) create the kmem_cache */
1277 
1278         /*
1279          * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids
1280          */
1281         create_boot_cache(kmem_cache, "kmem_cache",
1282                 offsetof(struct kmem_cache, node) +
1283                                   nr_node_ids * sizeof(struct kmem_cache_node *),
1284                                   SLAB_HWCACHE_ALIGN);
1285         list_add(&kmem_cache->list, &slab_caches);
1286         slab_state = PARTIAL;
1287 
1288         /*
1289          * Initialize the caches that provide memory for the  kmem_cache_node
1290          * structures first.  Without this, further allocations will bug.
1291          */
1292         kmalloc_caches[INDEX_NODE] = create_kmalloc_cache(
1293                                 kmalloc_info[INDEX_NODE].name,
1294                                 kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS);
1295         slab_state = PARTIAL_NODE;
1296         setup_kmalloc_cache_index_table();
1297 
1298         slab_early_init = 0;
1299 
1300         /* 5) Replace the bootstrap kmem_cache_node */
1301         {
1302                 int nid;
1303 
1304                 for_each_online_node(nid) {
1305                         init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid);
1306 
1307                         init_list(kmalloc_caches[INDEX_NODE],
1308                                           &init_kmem_cache_node[SIZE_NODE + nid], nid);
1309                 }
1310         }
1311 
1312         create_kmalloc_caches(ARCH_KMALLOC_FLAGS);
1313 }
1314 
1315 void __init kmem_cache_init_late(void)
1316 {
1317         struct kmem_cache *cachep;
1318 
1319         slab_state = UP;
1320 
1321         /* 6) resize the head arrays to their final sizes */
1322         mutex_lock(&slab_mutex);
1323         list_for_each_entry(cachep, &slab_caches, list)
1324                 if (enable_cpucache(cachep, GFP_NOWAIT))
1325                         BUG();
1326         mutex_unlock(&slab_mutex);
1327 
1328         /* Done! */
1329         slab_state = FULL;
1330 
1331 #ifdef CONFIG_NUMA
1332         /*
1333          * Register a memory hotplug callback that initializes and frees
1334          * node.
1335          */
1336         hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
1337 #endif
1338 
1339         /*
1340          * The reap timers are started later, with a module init call: That part
1341          * of the kernel is not yet operational.
1342          */
1343 }
1344 
1345 static int __init cpucache_init(void)
1346 {
1347         int ret;
1348 
1349         /*
1350          * Register the timers that return unneeded pages to the page allocator
1351          */
1352         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "SLAB online",
1353                                 slab_online_cpu, slab_offline_cpu);
1354         WARN_ON(ret < 0);
1355 
1356         /* Done! */
1357         slab_state = FULL;
1358         return 0;
1359 }
1360 __initcall(cpucache_init);
1361 
1362 static noinline void
1363 slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid)
1364 {
1365 #if DEBUG
1366         struct kmem_cache_node *n;
1367         unsigned long flags;
1368         int node;
1369         static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
1370                                       DEFAULT_RATELIMIT_BURST);
1371 
1372         if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs))
1373                 return;
1374 
1375         pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
1376                 nodeid, gfpflags, &gfpflags);
1377         pr_warn("  cache: %s, object size: %d, order: %d\n",
1378                 cachep->name, cachep->size, cachep->gfporder);
1379 
1380         for_each_kmem_cache_node(cachep, node, n) {
1381                 unsigned long total_slabs, free_slabs, free_objs;
1382 
1383                 spin_lock_irqsave(&n->list_lock, flags);
1384                 total_slabs = n->total_slabs;
1385                 free_slabs = n->free_slabs;
1386                 free_objs = n->free_objects;
1387                 spin_unlock_irqrestore(&n->list_lock, flags);
1388 
1389                 pr_warn("  node %d: slabs: %ld/%ld, objs: %ld/%ld\n",
1390                         node, total_slabs - free_slabs, total_slabs,
1391                         (total_slabs * cachep->num) - free_objs,
1392                         total_slabs * cachep->num);
1393         }
1394 #endif
1395 }
1396 
1397 /*
1398  * Interface to system's page allocator. No need to hold the
1399  * kmem_cache_node ->list_lock.
1400  *
1401  * If we requested dmaable memory, we will get it. Even if we
1402  * did not request dmaable memory, we might get it, but that
1403  * would be relatively rare and ignorable.
1404  */
1405 static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags,
1406                                                                 int nodeid)
1407 {
1408         struct page *page;
1409         int nr_pages;
1410 
1411         flags |= cachep->allocflags;
1412 
1413         page = __alloc_pages_node(nodeid, flags, cachep->gfporder);
1414         if (!page) {
1415                 slab_out_of_memory(cachep, flags, nodeid);
1416                 return NULL;
1417         }
1418 
1419         if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) {
1420                 __free_pages(page, cachep->gfporder);
1421                 return NULL;
1422         }
1423 
1424         nr_pages = (1 << cachep->gfporder);
1425         if (cachep->flags & SLAB_RECLAIM_ACCOUNT)
1426                 mod_lruvec_page_state(page, NR_SLAB_RECLAIMABLE, nr_pages);
1427         else
1428                 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE, nr_pages);
1429 
1430         __SetPageSlab(page);
1431         /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */
1432         if (sk_memalloc_socks() && page_is_pfmemalloc(page))
1433                 SetPageSlabPfmemalloc(page);
1434 
1435         return page;
1436 }
1437 
1438 /*
1439  * Interface to system's page release.
1440  */
1441 static void kmem_freepages(struct kmem_cache *cachep, struct page *page)
1442 {
1443         int order = cachep->gfporder;
1444         unsigned long nr_freed = (1 << order);
1445 
1446         if (cachep->flags & SLAB_RECLAIM_ACCOUNT)
1447                 mod_lruvec_page_state(page, NR_SLAB_RECLAIMABLE, -nr_freed);
1448         else
1449                 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE, -nr_freed);
1450 
1451         BUG_ON(!PageSlab(page));
1452         __ClearPageSlabPfmemalloc(page);
1453         __ClearPageSlab(page);
1454         page_mapcount_reset(page);
1455         page->mapping = NULL;
1456 
1457         if (current->reclaim_state)
1458                 current->reclaim_state->reclaimed_slab += nr_freed;
1459         memcg_uncharge_slab(page, order, cachep);
1460         __free_pages(page, order);
1461 }
1462 
1463 static void kmem_rcu_free(struct rcu_head *head)
1464 {
1465         struct kmem_cache *cachep;
1466         struct page *page;
1467 
1468         page = container_of(head, struct page, rcu_head);
1469         cachep = page->slab_cache;
1470 
1471         kmem_freepages(cachep, page);
1472 }
1473 
1474 #if DEBUG
1475 static bool is_debug_pagealloc_cache(struct kmem_cache *cachep)
1476 {
1477         if (debug_pagealloc_enabled() && OFF_SLAB(cachep) &&
1478                 (cachep->size % PAGE_SIZE) == 0)
1479                 return true;
1480 
1481         return false;
1482 }
1483 
1484 #ifdef CONFIG_DEBUG_PAGEALLOC
1485 static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr,
1486                             unsigned long caller)
1487 {
1488         int size = cachep->object_size;
1489 
1490         addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)];
1491 
1492         if (size < 5 * sizeof(unsigned long))
1493                 return;
1494 
1495         *addr++ = 0x12345678;
1496         *addr++ = caller;
1497         *addr++ = smp_processor_id();
1498         size -= 3 * sizeof(unsigned long);
1499         {
1500                 unsigned long *sptr = &caller;
1501                 unsigned long svalue;
1502 
1503                 while (!kstack_end(sptr)) {
1504                         svalue = *sptr++;
1505                         if (kernel_text_address(svalue)) {
1506                                 *addr++ = svalue;
1507                                 size -= sizeof(unsigned long);
1508                                 if (size <= sizeof(unsigned long))
1509                                         break;
1510                         }
1511                 }
1512 
1513         }
1514         *addr++ = 0x87654321;
1515 }
1516 
1517 static void slab_kernel_map(struct kmem_cache *cachep, void *objp,
1518                                 int map, unsigned long caller)
1519 {
1520         if (!is_debug_pagealloc_cache(cachep))
1521                 return;
1522 
1523         if (caller)
1524                 store_stackinfo(cachep, objp, caller);
1525 
1526         kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map);
1527 }
1528 
1529 #else
1530 static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp,
1531                                 int map, unsigned long caller) {}
1532 
1533 #endif
1534 
1535 static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val)
1536 {
1537         int size = cachep->object_size;
1538         addr = &((char *)addr)[obj_offset(cachep)];
1539 
1540         memset(addr, val, size);
1541         *(unsigned char *)(addr + size - 1) = POISON_END;
1542 }
1543 
1544 static void dump_line(char *data, int offset, int limit)
1545 {
1546         int i;
1547         unsigned char error = 0;
1548         int bad_count = 0;
1549 
1550         pr_err("%03x: ", offset);
1551         for (i = 0; i < limit; i++) {
1552                 if (data[offset + i] != POISON_FREE) {
1553                         error = data[offset + i];
1554                         bad_count++;
1555                 }
1556         }
1557         print_hex_dump(KERN_CONT, "", 0, 16, 1,
1558                         &data[offset], limit, 1);
1559 
1560         if (bad_count == 1) {
1561                 error ^= POISON_FREE;
1562                 if (!(error & (error - 1))) {
1563                         pr_err("Single bit error detected. Probably bad RAM.\n");
1564 #ifdef CONFIG_X86
1565                         pr_err("Run memtest86+ or a similar memory test tool.\n");
1566 #else
1567                         pr_err("Run a memory test tool.\n");
1568 #endif
1569                 }
1570         }
1571 }
1572 #endif
1573 
1574 #if DEBUG
1575 
1576 static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines)
1577 {
1578         int i, size;
1579         char *realobj;
1580 
1581         if (cachep->flags & SLAB_RED_ZONE) {
1582                 pr_err("Redzone: 0x%llx/0x%llx\n",
1583                        *dbg_redzone1(cachep, objp),
1584                        *dbg_redzone2(cachep, objp));
1585         }
1586 
1587         if (cachep->flags & SLAB_STORE_USER)
1588                 pr_err("Last user: (%pSR)\n", *dbg_userword(cachep, objp));
1589         realobj = (char *)objp + obj_offset(cachep);
1590         size = cachep->object_size;
1591         for (i = 0; i < size && lines; i += 16, lines--) {
1592                 int limit;
1593                 limit = 16;
1594                 if (i + limit > size)
1595                         limit = size - i;
1596                 dump_line(realobj, i, limit);
1597         }
1598 }
1599 
1600 static void check_poison_obj(struct kmem_cache *cachep, void *objp)
1601 {
1602         char *realobj;
1603         int size, i;
1604         int lines = 0;
1605 
1606         if (is_debug_pagealloc_cache(cachep))
1607                 return;
1608 
1609         realobj = (char *)objp + obj_offset(cachep);
1610         size = cachep->object_size;
1611 
1612         for (i = 0; i < size; i++) {
1613                 char exp = POISON_FREE;
1614                 if (i == size - 1)
1615                         exp = POISON_END;
1616                 if (realobj[i] != exp) {
1617                         int limit;
1618                         /* Mismatch ! */
1619                         /* Print header */
1620                         if (lines == 0) {
1621                                 pr_err("Slab corruption (%s): %s start=%px, len=%d\n",
1622                                        print_tainted(), cachep->name,
1623                                        realobj, size);
1624                                 print_objinfo(cachep, objp, 0);
1625                         }
1626                         /* Hexdump the affected line */
1627                         i = (i / 16) * 16;
1628                         limit = 16;
1629                         if (i + limit > size)
1630                                 limit = size - i;
1631                         dump_line(realobj, i, limit);
1632                         i += 16;
1633                         lines++;
1634                         /* Limit to 5 lines */
1635                         if (lines > 5)
1636                                 break;
1637                 }
1638         }
1639         if (lines != 0) {
1640                 /* Print some data about the neighboring objects, if they
1641                  * exist:
1642                  */
1643                 struct page *page = virt_to_head_page(objp);
1644                 unsigned int objnr;
1645 
1646                 objnr = obj_to_index(cachep, page, objp);
1647                 if (objnr) {
1648                         objp = index_to_obj(cachep, page, objnr - 1);
1649                         realobj = (char *)objp + obj_offset(cachep);
1650                         pr_err("Prev obj: start=%px, len=%d\n", realobj, size);
1651                         print_objinfo(cachep, objp, 2);
1652                 }
1653                 if (objnr + 1 < cachep->num) {
1654                         objp = index_to_obj(cachep, page, objnr + 1);
1655                         realobj = (char *)objp + obj_offset(cachep);
1656                         pr_err("Next obj: start=%px, len=%d\n", realobj, size);
1657                         print_objinfo(cachep, objp, 2);
1658                 }
1659         }
1660 }
1661 #endif
1662 
1663 #if DEBUG
1664 static void slab_destroy_debugcheck(struct kmem_cache *cachep,
1665                                                 struct page *page)
1666 {
1667         int i;
1668 
1669         if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) {
1670                 poison_obj(cachep, page->freelist - obj_offset(cachep),
1671                         POISON_FREE);
1672         }
1673 
1674         for (i = 0; i < cachep->num; i++) {
1675                 void *objp = index_to_obj(cachep, page, i);
1676 
1677                 if (cachep->flags & SLAB_POISON) {
1678                         check_poison_obj(cachep, objp);
1679                         slab_kernel_map(cachep, objp, 1, 0);
1680                 }
1681                 if (cachep->flags & SLAB_RED_ZONE) {
1682                         if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
1683                                 slab_error(cachep, "start of a freed object was overwritten");
1684                         if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
1685                                 slab_error(cachep, "end of a freed object was overwritten");
1686                 }
1687         }
1688 }
1689 #else
1690 static void slab_destroy_debugcheck(struct kmem_cache *cachep,
1691                                                 struct page *page)
1692 {
1693 }
1694 #endif
1695 
1696 /**
1697  * slab_destroy - destroy and release all objects in a slab
1698  * @cachep: cache pointer being destroyed
1699  * @page: page pointer being destroyed
1700  *
1701  * Destroy all the objs in a slab page, and release the mem back to the system.
1702  * Before calling the slab page must have been unlinked from the cache. The
1703  * kmem_cache_node ->list_lock is not held/needed.
1704  */
1705 static void slab_destroy(struct kmem_cache *cachep, struct page *page)
1706 {
1707         void *freelist;
1708 
1709         freelist = page->freelist;
1710         slab_destroy_debugcheck(cachep, page);
1711         if (unlikely(cachep->flags & SLAB_TYPESAFE_BY_RCU))
1712                 call_rcu(&page->rcu_head, kmem_rcu_free);
1713         else
1714                 kmem_freepages(cachep, page);
1715 
1716         /*
1717          * From now on, we don't use freelist
1718          * although actual page can be freed in rcu context
1719          */
1720         if (OFF_SLAB(cachep))
1721                 kmem_cache_free(cachep->freelist_cache, freelist);
1722 }
1723 
1724 static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list)
1725 {
1726         struct page *page, *n;
1727 
1728         list_for_each_entry_safe(page, n, list, lru) {
1729                 list_del(&page->lru);
1730                 slab_destroy(cachep, page);
1731         }
1732 }
1733 
1734 /**
1735  * calculate_slab_order - calculate size (page order) of slabs
1736  * @cachep: pointer to the cache that is being created
1737  * @size: size of objects to be created in this cache.
1738  * @flags: slab allocation flags
1739  *
1740  * Also calculates the number of objects per slab.
1741  *
1742  * This could be made much more intelligent.  For now, try to avoid using
1743  * high order pages for slabs.  When the gfp() functions are more friendly
1744  * towards high-order requests, this should be changed.
1745  */
1746 static size_t calculate_slab_order(struct kmem_cache *cachep,
1747                                 size_t size, slab_flags_t flags)
1748 {
1749         size_t left_over = 0;
1750         int gfporder;
1751 
1752         for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) {
1753                 unsigned int num;
1754                 size_t remainder;
1755 
1756                 num = cache_estimate(gfporder, size, flags, &remainder);
1757                 if (!num)
1758                         continue;
1759 
1760                 /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */
1761                 if (num > SLAB_OBJ_MAX_NUM)
1762                         break;
1763 
1764                 if (flags & CFLGS_OFF_SLAB) {
1765                         struct kmem_cache *freelist_cache;
1766                         size_t freelist_size;
1767 
1768                         freelist_size = num * sizeof(freelist_idx_t);
1769                         freelist_cache = kmalloc_slab(freelist_size, 0u);
1770                         if (!freelist_cache)
1771                                 continue;
1772 
1773                         /*
1774                          * Needed to avoid possible looping condition
1775                          * in cache_grow_begin()
1776                          */
1777                         if (OFF_SLAB(freelist_cache))
1778                                 continue;
1779 
1780                         /* check if off slab has enough benefit */
1781                         if (freelist_cache->size > cachep->size / 2)
1782                                 continue;
1783                 }
1784 
1785                 /* Found something acceptable - save it away */
1786                 cachep->num = num;
1787                 cachep->gfporder = gfporder;
1788                 left_over = remainder;
1789 
1790                 /*
1791                  * A VFS-reclaimable slab tends to have most allocations
1792                  * as GFP_NOFS and we really don't want to have to be allocating
1793                  * higher-order pages when we are unable to shrink dcache.
1794                  */
1795                 if (flags & SLAB_RECLAIM_ACCOUNT)
1796                         break;
1797 
1798                 /*
1799                  * Large number of objects is good, but very large slabs are
1800                  * currently bad for the gfp()s.
1801                  */
1802                 if (gfporder >= slab_max_order)
1803                         break;
1804 
1805                 /*
1806                  * Acceptable internal fragmentation?
1807                  */
1808                 if (left_over * 8 <= (PAGE_SIZE << gfporder))
1809                         break;
1810         }
1811         return left_over;
1812 }
1813 
1814 static struct array_cache __percpu *alloc_kmem_cache_cpus(
1815                 struct kmem_cache *cachep, int entries, int batchcount)
1816 {
1817         int cpu;
1818         size_t size;
1819         struct array_cache __percpu *cpu_cache;
1820 
1821         size = sizeof(void *) * entries + sizeof(struct array_cache);
1822         cpu_cache = __alloc_percpu(size, sizeof(void *));
1823 
1824         if (!cpu_cache)
1825                 return NULL;
1826 
1827         for_each_possible_cpu(cpu) {
1828                 init_arraycache(per_cpu_ptr(cpu_cache, cpu),
1829                                 entries, batchcount);
1830         }
1831 
1832         return cpu_cache;
1833 }
1834 
1835 static int __ref setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp)
1836 {
1837         if (slab_state >= FULL)
1838                 return enable_cpucache(cachep, gfp);
1839 
1840         cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1);
1841         if (!cachep->cpu_cache)
1842                 return 1;
1843 
1844         if (slab_state == DOWN) {
1845                 /* Creation of first cache (kmem_cache). */
1846                 set_up_node(kmem_cache, CACHE_CACHE);
1847         } else if (slab_state == PARTIAL) {
1848                 /* For kmem_cache_node */
1849                 set_up_node(cachep, SIZE_NODE);
1850         } else {
1851                 int node;
1852 
1853                 for_each_online_node(node) {
1854                         cachep->node[node] = kmalloc_node(
1855                                 sizeof(struct kmem_cache_node), gfp, node);
1856                         BUG_ON(!cachep->node[node]);
1857                         kmem_cache_node_init(cachep->node[node]);
1858                 }
1859         }
1860 
1861         cachep->node[numa_mem_id()]->next_reap =
1862                         jiffies + REAPTIMEOUT_NODE +
1863                         ((unsigned long)cachep) % REAPTIMEOUT_NODE;
1864 
1865         cpu_cache_get(cachep)->avail = 0;
1866         cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES;
1867         cpu_cache_get(cachep)->batchcount = 1;
1868         cpu_cache_get(cachep)->touched = 0;
1869         cachep->batchcount = 1;
1870         cachep->limit = BOOT_CPUCACHE_ENTRIES;
1871         return 0;
1872 }
1873 
1874 slab_flags_t kmem_cache_flags(unsigned long object_size,
1875         slab_flags_t flags, const char *name,
1876         void (*ctor)(void *))
1877 {
1878         return flags;
1879 }
1880 
1881 struct kmem_cache *
1882 __kmem_cache_alias(const char *name, size_t size, size_t align,
1883                    slab_flags_t flags, void (*ctor)(void *))
1884 {
1885         struct kmem_cache *cachep;
1886 
1887         cachep = find_mergeable(size, align, flags, name, ctor);
1888         if (cachep) {
1889                 cachep->refcount++;
1890 
1891                 /*
1892                  * Adjust the object sizes so that we clear
1893                  * the complete object on kzalloc.
1894                  */
1895                 cachep->object_size = max_t(int, cachep->object_size, size);
1896         }
1897         return cachep;
1898 }
1899 
1900 static bool set_objfreelist_slab_cache(struct kmem_cache *cachep,
1901                         size_t size, slab_flags_t flags)
1902 {
1903         size_t left;
1904 
1905         cachep->num = 0;
1906 
1907         if (cachep->ctor || flags & SLAB_TYPESAFE_BY_RCU)
1908                 return false;
1909 
1910         left = calculate_slab_order(cachep, size,
1911                         flags | CFLGS_OBJFREELIST_SLAB);
1912         if (!cachep->num)
1913                 return false;
1914 
1915         if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size)
1916                 return false;
1917 
1918         cachep->colour = left / cachep->colour_off;
1919 
1920         return true;
1921 }
1922 
1923 static bool set_off_slab_cache(struct kmem_cache *cachep,
1924                         size_t size, slab_flags_t flags)
1925 {
1926         size_t left;
1927 
1928         cachep->num = 0;
1929 
1930         /*
1931          * Always use on-slab management when SLAB_NOLEAKTRACE
1932          * to avoid recursive calls into kmemleak.
1933          */
1934         if (flags & SLAB_NOLEAKTRACE)
1935                 return false;
1936 
1937         /*
1938          * Size is large, assume best to place the slab management obj
1939          * off-slab (should allow better packing of objs).
1940          */
1941         left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB);
1942         if (!cachep->num)
1943                 return false;
1944 
1945         /*
1946          * If the slab has been placed off-slab, and we have enough space then
1947          * move it on-slab. This is at the expense of any extra colouring.
1948          */
1949         if (left >= cachep->num * sizeof(freelist_idx_t))
1950                 return false;
1951 
1952         cachep->colour = left / cachep->colour_off;
1953 
1954         return true;
1955 }
1956 
1957 static bool set_on_slab_cache(struct kmem_cache *cachep,
1958                         size_t size, slab_flags_t flags)
1959 {
1960         size_t left;
1961 
1962         cachep->num = 0;
1963 
1964         left = calculate_slab_order(cachep, size, flags);
1965         if (!cachep->num)
1966                 return false;
1967 
1968         cachep->colour = left / cachep->colour_off;
1969 
1970         return true;
1971 }
1972 
1973 /**
1974  * __kmem_cache_create - Create a cache.
1975  * @cachep: cache management descriptor
1976  * @flags: SLAB flags
1977  *
1978  * Returns a ptr to the cache on success, NULL on failure.
1979  * Cannot be called within a int, but can be interrupted.
1980  * The @ctor is run when new pages are allocated by the cache.
1981  *
1982  * The flags are
1983  *
1984  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
1985  * to catch references to uninitialised memory.
1986  *
1987  * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
1988  * for buffer overruns.
1989  *
1990  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
1991  * cacheline.  This can be beneficial if you're counting cycles as closely
1992  * as davem.
1993  */
1994 int __kmem_cache_create(struct kmem_cache *cachep, slab_flags_t flags)
1995 {
1996         size_t ralign = BYTES_PER_WORD;
1997         gfp_t gfp;
1998         int err;
1999         size_t size = cachep->size;
2000 
2001 #if DEBUG
2002 #if FORCED_DEBUG
2003         /*
2004          * Enable redzoning and last user accounting, except for caches with
2005          * large objects, if the increased size would increase the object size
2006          * above the next power of two: caches with object sizes just above a
2007          * power of two have a significant amount of internal fragmentation.
2008          */
2009         if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN +
2010                                                 2 * sizeof(unsigned long long)))
2011                 flags |= SLAB_RED_ZONE | SLAB_STORE_USER;
2012         if (!(flags & SLAB_TYPESAFE_BY_RCU))
2013                 flags |= SLAB_POISON;
2014 #endif
2015 #endif
2016 
2017         /*
2018          * Check that size is in terms of words.  This is needed to avoid
2019          * unaligned accesses for some archs when redzoning is used, and makes
2020          * sure any on-slab bufctl's are also correctly aligned.
2021          */
2022         size = ALIGN(size, BYTES_PER_WORD);
2023 
2024         if (flags & SLAB_RED_ZONE) {
2025                 ralign = REDZONE_ALIGN;
2026                 /* If redzoning, ensure that the second redzone is suitably
2027                  * aligned, by adjusting the object size accordingly. */
2028                 size = ALIGN(size, REDZONE_ALIGN);
2029         }
2030 
2031         /* 3) caller mandated alignment */
2032         if (ralign < cachep->align) {
2033                 ralign = cachep->align;
2034         }
2035         /* disable debug if necessary */
2036         if (ralign > __alignof__(unsigned long long))
2037                 flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
2038         /*
2039          * 4) Store it.
2040          */
2041         cachep->align = ralign;
2042         cachep->colour_off = cache_line_size();
2043         /* Offset must be a multiple of the alignment. */
2044         if (cachep->colour_off < cachep->align)
2045                 cachep->colour_off = cachep->align;
2046 
2047         if (slab_is_available())
2048                 gfp = GFP_KERNEL;
2049         else
2050                 gfp = GFP_NOWAIT;
2051 
2052 #if DEBUG
2053 
2054         /*
2055          * Both debugging options require word-alignment which is calculated
2056          * into align above.
2057          */
2058         if (flags & SLAB_RED_ZONE) {
2059                 /* add space for red zone words */
2060                 cachep->obj_offset += sizeof(unsigned long long);
2061                 size += 2 * sizeof(unsigned long long);
2062         }
2063         if (flags & SLAB_STORE_USER) {
2064                 /* user store requires one word storage behind the end of
2065                  * the real object. But if the second red zone needs to be
2066                  * aligned to 64 bits, we must allow that much space.
2067                  */
2068                 if (flags & SLAB_RED_ZONE)
2069                         size += REDZONE_ALIGN;
2070                 else
2071                         size += BYTES_PER_WORD;
2072         }
2073 #endif
2074 
2075         kasan_cache_create(cachep, &size, &flags);
2076 
2077         size = ALIGN(size, cachep->align);
2078         /*
2079          * We should restrict the number of objects in a slab to implement
2080          * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition.
2081          */
2082         if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE)
2083                 size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align);
2084 
2085 #if DEBUG
2086         /*
2087          * To activate debug pagealloc, off-slab management is necessary
2088          * requirement. In early phase of initialization, small sized slab
2089          * doesn't get initialized so it would not be possible. So, we need
2090          * to check size >= 256. It guarantees that all necessary small
2091          * sized slab is initialized in current slab initialization sequence.
2092          */
2093         if (debug_pagealloc_enabled() && (flags & SLAB_POISON) &&
2094                 size >= 256 && cachep->object_size > cache_line_size()) {
2095                 if (size < PAGE_SIZE || size % PAGE_SIZE == 0) {
2096                         size_t tmp_size = ALIGN(size, PAGE_SIZE);
2097 
2098                         if (set_off_slab_cache(cachep, tmp_size, flags)) {
2099                                 flags |= CFLGS_OFF_SLAB;
2100                                 cachep->obj_offset += tmp_size - size;
2101                                 size = tmp_size;
2102                                 goto done;
2103                         }
2104                 }
2105         }
2106 #endif
2107 
2108         if (set_objfreelist_slab_cache(cachep, size, flags)) {
2109                 flags |= CFLGS_OBJFREELIST_SLAB;
2110                 goto done;
2111         }
2112 
2113         if (set_off_slab_cache(cachep, size, flags)) {
2114                 flags |= CFLGS_OFF_SLAB;
2115                 goto done;
2116         }
2117 
2118         if (set_on_slab_cache(cachep, size, flags))
2119                 goto done;
2120 
2121         return -E2BIG;
2122 
2123 done:
2124         cachep->freelist_size = cachep->num * sizeof(freelist_idx_t);
2125         cachep->flags = flags;
2126         cachep->allocflags = __GFP_COMP;
2127         if (flags & SLAB_CACHE_DMA)
2128                 cachep->allocflags |= GFP_DMA;
2129         if (flags & SLAB_RECLAIM_ACCOUNT)
2130                 cachep->allocflags |= __GFP_RECLAIMABLE;
2131         cachep->size = size;
2132         cachep->reciprocal_buffer_size = reciprocal_value(size);
2133 
2134 #if DEBUG
2135         /*
2136          * If we're going to use the generic kernel_map_pages()
2137          * poisoning, then it's going to smash the contents of
2138          * the redzone and userword anyhow, so switch them off.
2139          */
2140         if (IS_ENABLED(CONFIG_PAGE_POISONING) &&
2141                 (cachep->flags & SLAB_POISON) &&
2142                 is_debug_pagealloc_cache(cachep))
2143                 cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
2144 #endif
2145 
2146         if (OFF_SLAB(cachep)) {
2147                 cachep->freelist_cache =
2148                         kmalloc_slab(cachep->freelist_size, 0u);
2149         }
2150 
2151         err = setup_cpu_cache(cachep, gfp);
2152         if (err) {
2153                 __kmem_cache_release(cachep);
2154                 return err;
2155         }
2156 
2157         return 0;
2158 }
2159 
2160 #if DEBUG
2161 static void check_irq_off(void)
2162 {
2163         BUG_ON(!irqs_disabled());
2164 }
2165 
2166 static void check_irq_on(void)
2167 {
2168         BUG_ON(irqs_disabled());
2169 }
2170 
2171 static void check_mutex_acquired(void)
2172 {
2173         BUG_ON(!mutex_is_locked(&slab_mutex));
2174 }
2175 
2176 static void check_spinlock_acquired(struct kmem_cache *cachep)
2177 {
2178 #ifdef CONFIG_SMP
2179         check_irq_off();
2180         assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock);
2181 #endif
2182 }
2183 
2184 static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node)
2185 {
2186 #ifdef CONFIG_SMP
2187         check_irq_off();
2188         assert_spin_locked(&get_node(cachep, node)->list_lock);
2189 #endif
2190 }
2191 
2192 #else
2193 #define check_irq_off() do { } while(0)
2194 #define check_irq_on()  do { } while(0)
2195 #define check_mutex_acquired()  do { } while(0)
2196 #define check_spinlock_acquired(x) do { } while(0)
2197 #define check_spinlock_acquired_node(x, y) do { } while(0)
2198 #endif
2199 
2200 static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac,
2201                                 int node, bool free_all, struct list_head *list)
2202 {
2203         int tofree;
2204 
2205         if (!ac || !ac->avail)
2206                 return;
2207 
2208         tofree = free_all ? ac->avail : (ac->limit + 4) / 5;
2209         if (tofree > ac->avail)
2210                 tofree = (ac->avail + 1) / 2;
2211 
2212         free_block(cachep, ac->entry, tofree, node, list);
2213         ac->avail -= tofree;
2214         memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail);
2215 }
2216 
2217 static void do_drain(void *arg)
2218 {
2219         struct kmem_cache *cachep = arg;
2220         struct array_cache *ac;
2221         int node = numa_mem_id();
2222         struct kmem_cache_node *n;
2223         LIST_HEAD(list);
2224 
2225         check_irq_off();
2226         ac = cpu_cache_get(cachep);
2227         n = get_node(cachep, node);
2228         spin_lock(&n->list_lock);
2229         free_block(cachep, ac->entry, ac->avail, node, &list);
2230         spin_unlock(&n->list_lock);
2231         slabs_destroy(cachep, &list);
2232         ac->avail = 0;
2233 }
2234 
2235 static void drain_cpu_caches(struct kmem_cache *cachep)
2236 {
2237         struct kmem_cache_node *n;
2238         int node;
2239         LIST_HEAD(list);
2240 
2241         on_each_cpu(do_drain, cachep, 1);
2242         check_irq_on();
2243         for_each_kmem_cache_node(cachep, node, n)
2244                 if (n->alien)
2245                         drain_alien_cache(cachep, n->alien);
2246 
2247         for_each_kmem_cache_node(cachep, node, n) {
2248                 spin_lock_irq(&n->list_lock);
2249                 drain_array_locked(cachep, n->shared, node, true, &list);
2250                 spin_unlock_irq(&n->list_lock);
2251 
2252                 slabs_destroy(cachep, &list);
2253         }
2254 }
2255 
2256 /*
2257  * Remove slabs from the list of free slabs.
2258  * Specify the number of slabs to drain in tofree.
2259  *
2260  * Returns the actual number of slabs released.
2261  */
2262 static int drain_freelist(struct kmem_cache *cache,
2263                         struct kmem_cache_node *n, int tofree)
2264 {
2265         struct list_head *p;
2266         int nr_freed;
2267         struct page *page;
2268 
2269         nr_freed = 0;
2270         while (nr_freed < tofree && !list_empty(&n->slabs_free)) {
2271 
2272                 spin_lock_irq(&n->list_lock);
2273                 p = n->slabs_free.prev;
2274                 if (p == &n->slabs_free) {
2275                         spin_unlock_irq(&n->list_lock);
2276                         goto out;
2277                 }
2278 
2279                 page = list_entry(p, struct page, lru);
2280                 list_del(&page->lru);
2281                 n->free_slabs--;
2282                 n->total_slabs--;
2283                 /*
2284                  * Safe to drop the lock. The slab is no longer linked
2285                  * to the cache.
2286                  */
2287                 n->free_objects -= cache->num;
2288                 spin_unlock_irq(&n->list_lock);
2289                 slab_destroy(cache, page);
2290                 nr_freed++;
2291         }
2292 out:
2293         return nr_freed;
2294 }
2295 
2296 int __kmem_cache_shrink(struct kmem_cache *cachep)
2297 {
2298         int ret = 0;
2299         int node;
2300         struct kmem_cache_node *n;
2301 
2302         drain_cpu_caches(cachep);
2303 
2304         check_irq_on();
2305         for_each_kmem_cache_node(cachep, node, n) {
2306                 drain_freelist(cachep, n, INT_MAX);
2307 
2308                 ret += !list_empty(&n->slabs_full) ||
2309                         !list_empty(&n->slabs_partial);
2310         }
2311         return (ret ? 1 : 0);
2312 }
2313 
2314 #ifdef CONFIG_MEMCG
2315 void __kmemcg_cache_deactivate(struct kmem_cache *cachep)
2316 {
2317         __kmem_cache_shrink(cachep);
2318 }
2319 #endif
2320 
2321 int __kmem_cache_shutdown(struct kmem_cache *cachep)
2322 {
2323         return __kmem_cache_shrink(cachep);
2324 }
2325 
2326 void __kmem_cache_release(struct kmem_cache *cachep)
2327 {
2328         int i;
2329         struct kmem_cache_node *n;
2330 
2331         cache_random_seq_destroy(cachep);
2332 
2333         free_percpu(cachep->cpu_cache);
2334 
2335         /* NUMA: free the node structures */
2336         for_each_kmem_cache_node(cachep, i, n) {
2337                 kfree(n->shared);
2338                 free_alien_cache(n->alien);
2339                 kfree(n);
2340                 cachep->node[i] = NULL;
2341         }
2342 }
2343 
2344 /*
2345  * Get the memory for a slab management obj.
2346  *
2347  * For a slab cache when the slab descriptor is off-slab, the
2348  * slab descriptor can't come from the same cache which is being created,
2349  * Because if it is the case, that means we defer the creation of
2350  * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point.
2351  * And we eventually call down to __kmem_cache_create(), which
2352  * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one.
2353  * This is a "chicken-and-egg" problem.
2354  *
2355  * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches,
2356  * which are all initialized during kmem_cache_init().
2357  */
2358 static void *alloc_slabmgmt(struct kmem_cache *cachep,
2359                                    struct page *page, int colour_off,
2360                                    gfp_t local_flags, int nodeid)
2361 {
2362         void *freelist;
2363         void *addr = page_address(page);
2364 
2365         page->s_mem = addr + colour_off;
2366         page->active = 0;
2367 
2368         if (OBJFREELIST_SLAB(cachep))
2369                 freelist = NULL;
2370         else if (OFF_SLAB(cachep)) {
2371                 /* Slab management obj is off-slab. */
2372                 freelist = kmem_cache_alloc_node(cachep->freelist_cache,
2373                                               local_flags, nodeid);
2374                 if (!freelist)
2375                         return NULL;
2376         } else {
2377                 /* We will use last bytes at the slab for freelist */
2378                 freelist = addr + (PAGE_SIZE << cachep->gfporder) -
2379                                 cachep->freelist_size;
2380         }
2381 
2382         return freelist;
2383 }
2384 
2385 static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx)
2386 {
2387         return ((freelist_idx_t *)page->freelist)[idx];
2388 }
2389 
2390 static inline void set_free_obj(struct page *page,
2391                                         unsigned int idx, freelist_idx_t val)
2392 {
2393         ((freelist_idx_t *)(page->freelist))[idx] = val;
2394 }
2395 
2396 static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page)
2397 {
2398 #if DEBUG
2399         int i;
2400 
2401         for (i = 0; i < cachep->num; i++) {
2402                 void *objp = index_to_obj(cachep, page, i);
2403 
2404                 if (cachep->flags & SLAB_STORE_USER)
2405                         *dbg_userword(cachep, objp) = NULL;
2406 
2407                 if (cachep->flags & SLAB_RED_ZONE) {
2408                         *dbg_redzone1(cachep, objp) = RED_INACTIVE;
2409                         *dbg_redzone2(cachep, objp) = RED_INACTIVE;
2410                 }
2411                 /*
2412                  * Constructors are not allowed to allocate memory from the same
2413                  * cache which they are a constructor for.  Otherwise, deadlock.
2414                  * They must also be threaded.
2415                  */
2416                 if (cachep->ctor && !(cachep->flags & SLAB_POISON)) {
2417                         kasan_unpoison_object_data(cachep,
2418                                                    objp + obj_offset(cachep));
2419                         cachep->ctor(objp + obj_offset(cachep));
2420                         kasan_poison_object_data(
2421                                 cachep, objp + obj_offset(cachep));
2422                 }
2423 
2424                 if (cachep->flags & SLAB_RED_ZONE) {
2425                         if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
2426                                 slab_error(cachep, "constructor overwrote the end of an object");
2427                         if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
2428                                 slab_error(cachep, "constructor overwrote the start of an object");
2429                 }
2430                 /* need to poison the objs? */
2431                 if (cachep->flags & SLAB_POISON) {
2432                         poison_obj(cachep, objp, POISON_FREE);
2433                         slab_kernel_map(cachep, objp, 0, 0);
2434                 }
2435         }
2436 #endif
2437 }
2438 
2439 #ifdef CONFIG_SLAB_FREELIST_RANDOM
2440 /* Hold information during a freelist initialization */
2441 union freelist_init_state {
2442         struct {
2443                 unsigned int pos;
2444                 unsigned int *list;
2445                 unsigned int count;
2446         };
2447         struct rnd_state rnd_state;
2448 };
2449 
2450 /*
2451  * Initialize the state based on the randomization methode available.
2452  * return true if the pre-computed list is available, false otherwize.
2453  */
2454 static bool freelist_state_initialize(union freelist_init_state *state,
2455                                 struct kmem_cache *cachep,
2456                                 unsigned int count)
2457 {
2458         bool ret;
2459         unsigned int rand;
2460 
2461         /* Use best entropy available to define a random shift */
2462         rand = get_random_int();
2463 
2464         /* Use a random state if the pre-computed list is not available */
2465         if (!cachep->random_seq) {
2466                 prandom_seed_state(&state->rnd_state, rand);
2467                 ret = false;
2468         } else {
2469                 state->list = cachep->random_seq;
2470                 state->count = count;
2471                 state->pos = rand % count;
2472                 ret = true;
2473         }
2474         return ret;
2475 }
2476 
2477 /* Get the next entry on the list and randomize it using a random shift */
2478 static freelist_idx_t next_random_slot(union freelist_init_state *state)
2479 {
2480         if (state->pos >= state->count)
2481                 state->pos = 0;
2482         return state->list[state->pos++];
2483 }
2484 
2485 /* Swap two freelist entries */
2486 static void swap_free_obj(struct page *page, unsigned int a, unsigned int b)
2487 {
2488         swap(((freelist_idx_t *)page->freelist)[a],
2489                 ((freelist_idx_t *)page->freelist)[b]);
2490 }
2491 
2492 /*
2493  * Shuffle the freelist initialization state based on pre-computed lists.
2494  * return true if the list was successfully shuffled, false otherwise.
2495  */
2496 static bool shuffle_freelist(struct kmem_cache *cachep, struct page *page)
2497 {
2498         unsigned int objfreelist = 0, i, rand, count = cachep->num;
2499         union freelist_init_state state;
2500         bool precomputed;
2501 
2502         if (count < 2)
2503                 return false;
2504 
2505         precomputed = freelist_state_initialize(&state, cachep, count);
2506 
2507         /* Take a random entry as the objfreelist */
2508         if (OBJFREELIST_SLAB(cachep)) {
2509                 if (!precomputed)
2510                         objfreelist = count - 1;
2511                 else
2512                         objfreelist = next_random_slot(&state);
2513                 page->freelist = index_to_obj(cachep, page, objfreelist) +
2514                                                 obj_offset(cachep);
2515                 count--;
2516         }
2517 
2518         /*
2519          * On early boot, generate the list dynamically.
2520          * Later use a pre-computed list for speed.
2521          */
2522         if (!precomputed) {
2523                 for (i = 0; i < count; i++)
2524                         set_free_obj(page, i, i);
2525 
2526                 /* Fisher-Yates shuffle */
2527                 for (i = count - 1; i > 0; i--) {
2528                         rand = prandom_u32_state(&state.rnd_state);
2529                         rand %= (i + 1);
2530                         swap_free_obj(page, i, rand);
2531                 }
2532         } else {
2533                 for (i = 0; i < count; i++)
2534                         set_free_obj(page, i, next_random_slot(&state));
2535         }
2536 
2537         if (OBJFREELIST_SLAB(cachep))
2538                 set_free_obj(page, cachep->num - 1, objfreelist);
2539 
2540         return true;
2541 }
2542 #else
2543 static inline bool shuffle_freelist(struct kmem_cache *cachep,
2544                                 struct page *page)
2545 {
2546         return false;
2547 }
2548 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
2549 
2550 static void cache_init_objs(struct kmem_cache *cachep,
2551                             struct page *page)
2552 {
2553         int i;
2554         void *objp;
2555         bool shuffled;
2556 
2557         cache_init_objs_debug(cachep, page);
2558 
2559         /* Try to randomize the freelist if enabled */
2560         shuffled = shuffle_freelist(cachep, page);
2561 
2562         if (!shuffled && OBJFREELIST_SLAB(cachep)) {
2563                 page->freelist = index_to_obj(cachep, page, cachep->num - 1) +
2564                                                 obj_offset(cachep);
2565         }
2566 
2567         for (i = 0; i < cachep->num; i++) {
2568                 objp = index_to_obj(cachep, page, i);
2569                 kasan_init_slab_obj(cachep, objp);
2570 
2571                 /* constructor could break poison info */
2572                 if (DEBUG == 0 && cachep->ctor) {
2573                         kasan_unpoison_object_data(cachep, objp);
2574                         cachep->ctor(objp);
2575                         kasan_poison_object_data(cachep, objp);
2576                 }
2577 
2578                 if (!shuffled)
2579                         set_free_obj(page, i, i);
2580         }
2581 }
2582 
2583 static void *slab_get_obj(struct kmem_cache *cachep, struct page *page)
2584 {
2585         void *objp;
2586 
2587         objp = index_to_obj(cachep, page, get_free_obj(page, page->active));
2588         page->active++;
2589 
2590 #if DEBUG
2591         if (cachep->flags & SLAB_STORE_USER)
2592                 set_store_user_dirty(cachep);
2593 #endif
2594 
2595         return objp;
2596 }
2597 
2598 static void slab_put_obj(struct kmem_cache *cachep,
2599                         struct page *page, void *objp)
2600 {
2601         unsigned int objnr = obj_to_index(cachep, page, objp);
2602 #if DEBUG
2603         unsigned int i;
2604 
2605         /* Verify double free bug */
2606         for (i = page->active; i < cachep->num; i++) {
2607                 if (get_free_obj(page, i) == objnr) {
2608                         pr_err("slab: double free detected in cache '%s', objp %px\n",
2609                                cachep->name, objp);
2610                         BUG();
2611                 }
2612         }
2613 #endif
2614         page->active--;
2615         if (!page->freelist)
2616                 page->freelist = objp + obj_offset(cachep);
2617 
2618         set_free_obj(page, page->active, objnr);
2619 }
2620 
2621 /*
2622  * Map pages beginning at addr to the given cache and slab. This is required
2623  * for the slab allocator to be able to lookup the cache and slab of a
2624  * virtual address for kfree, ksize, and slab debugging.
2625  */
2626 static void slab_map_pages(struct kmem_cache *cache, struct page *page,
2627                            void *freelist)
2628 {
2629         page->slab_cache = cache;
2630         page->freelist = freelist;
2631 }
2632 
2633 /*
2634  * Grow (by 1) the number of slabs within a cache.  This is called by
2635  * kmem_cache_alloc() when there are no active objs left in a cache.
2636  */
2637 static struct page *cache_grow_begin(struct kmem_cache *cachep,
2638                                 gfp_t flags, int nodeid)
2639 {
2640         void *freelist;
2641         size_t offset;
2642         gfp_t local_flags;
2643         int page_node;
2644         struct kmem_cache_node *n;
2645         struct page *page;
2646 
2647         /*
2648          * Be lazy and only check for valid flags here,  keeping it out of the
2649          * critical path in kmem_cache_alloc().
2650          */
2651         if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
2652                 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
2653                 flags &= ~GFP_SLAB_BUG_MASK;
2654                 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
2655                                 invalid_mask, &invalid_mask, flags, &flags);
2656                 dump_stack();
2657         }
2658         local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK);
2659 
2660         check_irq_off();
2661         if (gfpflags_allow_blocking(local_flags))
2662                 local_irq_enable();
2663 
2664         /*
2665          * Get mem for the objs.  Attempt to allocate a physical page from
2666          * 'nodeid'.
2667          */
2668         page = kmem_getpages(cachep, local_flags, nodeid);
2669         if (!page)
2670                 goto failed;
2671 
2672         page_node = page_to_nid(page);
2673         n = get_node(cachep, page_node);
2674 
2675         /* Get colour for the slab, and cal the next value. */
2676         n->colour_next++;
2677         if (n->colour_next >= cachep->colour)
2678                 n->colour_next = 0;
2679 
2680         offset = n->colour_next;
2681         if (offset >= cachep->colour)
2682                 offset = 0;
2683 
2684         offset *= cachep->colour_off;
2685 
2686         /* Get slab management. */
2687         freelist = alloc_slabmgmt(cachep, page, offset,
2688                         local_flags & ~GFP_CONSTRAINT_MASK, page_node);
2689         if (OFF_SLAB(cachep) && !freelist)
2690                 goto opps1;
2691 
2692         slab_map_pages(cachep, page, freelist);
2693 
2694         kasan_poison_slab(page);
2695         cache_init_objs(cachep, page);
2696 
2697         if (gfpflags_allow_blocking(local_flags))
2698                 local_irq_disable();
2699 
2700         return page;
2701 
2702 opps1:
2703         kmem_freepages(cachep, page);
2704 failed:
2705         if (gfpflags_allow_blocking(local_flags))
2706                 local_irq_disable();
2707         return NULL;
2708 }
2709 
2710 static void cache_grow_end(struct kmem_cache *cachep, struct page *page)
2711 {
2712         struct kmem_cache_node *n;
2713         void *list = NULL;
2714 
2715         check_irq_off();
2716 
2717         if (!page)
2718                 return;
2719 
2720         INIT_LIST_HEAD(&page->lru);
2721         n = get_node(cachep, page_to_nid(page));
2722 
2723         spin_lock(&n->list_lock);
2724         n->total_slabs++;
2725         if (!page->active) {
2726                 list_add_tail(&page->lru, &(n->slabs_free));
2727                 n->free_slabs++;
2728         } else
2729                 fixup_slab_list(cachep, n, page, &list);
2730 
2731         STATS_INC_GROWN(cachep);
2732         n->free_objects += cachep->num - page->active;
2733         spin_unlock(&n->list_lock);
2734 
2735         fixup_objfreelist_debug(cachep, &list);
2736 }
2737 
2738 #if DEBUG
2739 
2740 /*
2741  * Perform extra freeing checks:
2742  * - detect bad pointers.
2743  * - POISON/RED_ZONE checking
2744  */
2745 static void kfree_debugcheck(const void *objp)
2746 {
2747         if (!virt_addr_valid(objp)) {
2748                 pr_err("kfree_debugcheck: out of range ptr %lxh\n",
2749                        (unsigned long)objp);
2750                 BUG();
2751         }
2752 }
2753 
2754 static inline void verify_redzone_free(struct kmem_cache *cache, void *obj)
2755 {
2756         unsigned long long redzone1, redzone2;
2757 
2758         redzone1 = *dbg_redzone1(cache, obj);
2759         redzone2 = *dbg_redzone2(cache, obj);
2760 
2761         /*
2762          * Redzone is ok.
2763          */
2764         if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE)
2765                 return;
2766 
2767         if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE)
2768                 slab_error(cache, "double free detected");
2769         else
2770                 slab_error(cache, "memory outside object was overwritten");
2771 
2772         pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n",
2773                obj, redzone1, redzone2);
2774 }
2775 
2776 static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp,
2777                                    unsigned long caller)
2778 {
2779         unsigned int objnr;
2780         struct page *page;
2781 
2782         BUG_ON(virt_to_cache(objp) != cachep);
2783 
2784         objp -= obj_offset(cachep);
2785         kfree_debugcheck(objp);
2786         page = virt_to_head_page(objp);
2787 
2788         if (cachep->flags & SLAB_RED_ZONE) {
2789                 verify_redzone_free(cachep, objp);
2790                 *dbg_redzone1(cachep, objp) = RED_INACTIVE;
2791                 *dbg_redzone2(cachep, objp) = RED_INACTIVE;
2792         }
2793         if (cachep->flags & SLAB_STORE_USER) {
2794                 set_store_user_dirty(cachep);
2795                 *dbg_userword(cachep, objp) = (void *)caller;
2796         }
2797 
2798         objnr = obj_to_index(cachep, page, objp);
2799 
2800         BUG_ON(objnr >= cachep->num);
2801         BUG_ON(objp != index_to_obj(cachep, page, objnr));
2802 
2803         if (cachep->flags & SLAB_POISON) {
2804                 poison_obj(cachep, objp, POISON_FREE);
2805                 slab_kernel_map(cachep, objp, 0, caller);
2806         }
2807         return objp;
2808 }
2809 
2810 #else
2811 #define kfree_debugcheck(x) do { } while(0)
2812 #define cache_free_debugcheck(x,objp,z) (objp)
2813 #endif
2814 
2815 static inline void fixup_objfreelist_debug(struct kmem_cache *cachep,
2816                                                 void **list)
2817 {
2818 #if DEBUG
2819         void *next = *list;
2820         void *objp;
2821 
2822         while (next) {
2823                 objp = next - obj_offset(cachep);
2824                 next = *(void **)next;
2825                 poison_obj(cachep, objp, POISON_FREE);
2826         }
2827 #endif
2828 }
2829 
2830 static inline void fixup_slab_list(struct kmem_cache *cachep,
2831                                 struct kmem_cache_node *n, struct page *page,
2832                                 void **list)
2833 {
2834         /* move slabp to correct slabp list: */
2835         list_del(&page->lru);
2836         if (page->active == cachep->num) {
2837                 list_add(&page->lru, &n->slabs_full);
2838                 if (OBJFREELIST_SLAB(cachep)) {
2839 #if DEBUG
2840                         /* Poisoning will be done without holding the lock */
2841                         if (cachep->flags & SLAB_POISON) {
2842                                 void **objp = page->freelist;
2843 
2844                                 *objp = *list;
2845                                 *list = objp;
2846                         }
2847 #endif
2848                         page->freelist = NULL;
2849                 }
2850         } else
2851                 list_add(&page->lru, &n->slabs_partial);
2852 }
2853 
2854 /* Try to find non-pfmemalloc slab if needed */
2855 static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n,
2856                                         struct page *page, bool pfmemalloc)
2857 {
2858         if (!page)
2859                 return NULL;
2860 
2861         if (pfmemalloc)
2862                 return page;
2863 
2864         if (!PageSlabPfmemalloc(page))
2865                 return page;
2866 
2867         /* No need to keep pfmemalloc slab if we have enough free objects */
2868         if (n->free_objects > n->free_limit) {
2869                 ClearPageSlabPfmemalloc(page);
2870                 return page;
2871         }
2872 
2873         /* Move pfmemalloc slab to the end of list to speed up next search */
2874         list_del(&page->lru);
2875         if (!page->active) {
2876                 list_add_tail(&page->lru, &n->slabs_free);
2877                 n->free_slabs++;
2878         } else
2879                 list_add_tail(&page->lru, &n->slabs_partial);
2880 
2881         list_for_each_entry(page, &n->slabs_partial, lru) {
2882                 if (!PageSlabPfmemalloc(page))
2883                         return page;
2884         }
2885 
2886         n->free_touched = 1;
2887         list_for_each_entry(page, &n->slabs_free, lru) {
2888                 if (!PageSlabPfmemalloc(page)) {
2889                         n->free_slabs--;
2890                         return page;
2891                 }
2892         }
2893 
2894         return NULL;
2895 }
2896 
2897 static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc)
2898 {
2899         struct page *page;
2900 
2901         assert_spin_locked(&n->list_lock);
2902         page = list_first_entry_or_null(&n->slabs_partial, struct page, lru);
2903         if (!page) {
2904                 n->free_touched = 1;
2905                 page = list_first_entry_or_null(&n->slabs_free, struct page,
2906                                                 lru);
2907                 if (page)
2908                         n->free_slabs--;
2909         }
2910 
2911         if (sk_memalloc_socks())
2912                 page = get_valid_first_slab(n, page, pfmemalloc);
2913 
2914         return page;
2915 }
2916 
2917 static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep,
2918                                 struct kmem_cache_node *n, gfp_t flags)
2919 {
2920         struct page *page;
2921         void *obj;
2922         void *list = NULL;
2923 
2924         if (!gfp_pfmemalloc_allowed(flags))
2925                 return NULL;
2926 
2927         spin_lock(&n->list_lock);
2928         page = get_first_slab(n, true);
2929         if (!page) {
2930                 spin_unlock(&n->list_lock);
2931                 return NULL;
2932         }
2933 
2934         obj = slab_get_obj(cachep, page);
2935         n->free_objects--;
2936 
2937         fixup_slab_list(cachep, n, page, &list);
2938 
2939         spin_unlock(&n->list_lock);
2940         fixup_objfreelist_debug(cachep, &list);
2941 
2942         return obj;
2943 }
2944 
2945 /*
2946  * Slab list should be fixed up by fixup_slab_list() for existing slab
2947  * or cache_grow_end() for new slab
2948  */
2949 static __always_inline int alloc_block(struct kmem_cache *cachep,
2950                 struct array_cache *ac, struct page *page, int batchcount)
2951 {
2952         /*
2953          * There must be at least one object available for
2954          * allocation.
2955          */
2956         BUG_ON(page->active >= cachep->num);
2957 
2958         while (page->active < cachep->num && batchcount--) {
2959                 STATS_INC_ALLOCED(cachep);
2960                 STATS_INC_ACTIVE(cachep);
2961                 STATS_SET_HIGH(cachep);
2962 
2963                 ac->entry[ac->avail++] = slab_get_obj(cachep, page);
2964         }
2965 
2966         return batchcount;
2967 }
2968 
2969 static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags)
2970 {
2971         int batchcount;
2972         struct kmem_cache_node *n;
2973         struct array_cache *ac, *shared;
2974         int node;
2975         void *list = NULL;
2976         struct page *page;
2977 
2978         check_irq_off();
2979         node = numa_mem_id();
2980 
2981         ac = cpu_cache_get(cachep);
2982         batchcount = ac->batchcount;
2983         if (!ac->touched && batchcount > BATCHREFILL_LIMIT) {
2984                 /*
2985                  * If there was little recent activity on this cache, then
2986                  * perform only a partial refill.  Otherwise we could generate
2987                  * refill bouncing.
2988                  */
2989                 batchcount = BATCHREFILL_LIMIT;
2990         }
2991         n = get_node(cachep, node);
2992 
2993         BUG_ON(ac->avail > 0 || !n);
2994         shared = READ_ONCE(n->shared);
2995         if (!n->free_objects && (!shared || !shared->avail))
2996                 goto direct_grow;
2997 
2998         spin_lock(&n->list_lock);
2999         shared = READ_ONCE(n->shared);
3000 
3001         /* See if we can refill from the shared array */
3002         if (shared && transfer_objects(ac, shared, batchcount)) {
3003                 shared->touched = 1;
3004                 goto alloc_done;
3005         }
3006 
3007         while (batchcount > 0) {
3008                 /* Get slab alloc is to come from. */
3009                 page = get_first_slab(n, false);
3010                 if (!page)
3011                         goto must_grow;
3012 
3013                 check_spinlock_acquired(cachep);
3014 
3015                 batchcount = alloc_block(cachep, ac, page, batchcount);
3016                 fixup_slab_list(cachep, n, page, &list);
3017         }
3018 
3019 must_grow:
3020         n->free_objects -= ac->avail;
3021 alloc_done:
3022         spin_unlock(&n->list_lock);
3023         fixup_objfreelist_debug(cachep, &list);
3024 
3025 direct_grow:
3026         if (unlikely(!ac->avail)) {
3027                 /* Check if we can use obj in pfmemalloc slab */
3028                 if (sk_memalloc_socks()) {
3029                         void *obj = cache_alloc_pfmemalloc(cachep, n, flags);
3030 
3031                         if (obj)
3032                                 return obj;
3033                 }
3034 
3035                 page = cache_grow_begin(cachep, gfp_exact_node(flags), node);
3036 
3037                 /*
3038                  * cache_grow_begin() can reenable interrupts,
3039                  * then ac could change.
3040                  */
3041                 ac = cpu_cache_get(cachep);
3042                 if (!ac->avail && page)
3043                         alloc_block(cachep, ac, page, batchcount);
3044                 cache_grow_end(cachep, page);
3045 
3046                 if (!ac->avail)
3047                         return NULL;
3048         }
3049         ac->touched = 1;
3050 
3051         return ac->entry[--ac->avail];
3052 }
3053 
3054 static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep,
3055                                                 gfp_t flags)
3056 {
3057         might_sleep_if(gfpflags_allow_blocking(flags));
3058 }
3059 
3060 #if DEBUG
3061 static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep,
3062                                 gfp_t flags, void *objp, unsigned long caller)
3063 {
3064         if (!objp)
3065                 return objp;
3066         if (cachep->flags & SLAB_POISON) {
3067                 check_poison_obj(cachep, objp);
3068                 slab_kernel_map(cachep, objp, 1, 0);
3069                 poison_obj(cachep, objp, POISON_INUSE);
3070         }
3071         if (cachep->flags & SLAB_STORE_USER)
3072                 *dbg_userword(cachep, objp) = (void *)caller;
3073 
3074         if (cachep->flags & SLAB_RED_ZONE) {
3075                 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE ||
3076                                 *dbg_redzone2(cachep, objp) != RED_INACTIVE) {
3077                         slab_error(cachep, "double free, or memory outside object was overwritten");
3078                         pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n",
3079                                objp, *dbg_redzone1(cachep, objp),
3080                                *dbg_redzone2(cachep, objp));
3081                 }
3082                 *dbg_redzone1(cachep, objp) = RED_ACTIVE;
3083                 *dbg_redzone2(cachep, objp) = RED_ACTIVE;
3084         }
3085 
3086         objp += obj_offset(cachep);
3087         if (cachep->ctor && cachep->flags & SLAB_POISON)
3088                 cachep->ctor(objp);
3089         if (ARCH_SLAB_MINALIGN &&
3090             ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) {
3091                 pr_err("0x%px: not aligned to ARCH_SLAB_MINALIGN=%d\n",
3092                        objp, (int)ARCH_SLAB_MINALIGN);
3093         }
3094         return objp;
3095 }
3096 #else
3097 #define cache_alloc_debugcheck_after(a,b,objp,d) (objp)
3098 #endif
3099 
3100 static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags)
3101 {
3102         void *objp;
3103         struct array_cache *ac;
3104 
3105         check_irq_off();
3106 
3107         ac = cpu_cache_get(cachep);
3108         if (likely(ac->avail)) {
3109                 ac->touched = 1;
3110                 objp = ac->entry[--ac->avail];
3111 
3112                 STATS_INC_ALLOCHIT(cachep);
3113                 goto out;
3114         }
3115 
3116         STATS_INC_ALLOCMISS(cachep);
3117         objp = cache_alloc_refill(cachep, flags);
3118         /*
3119          * the 'ac' may be updated by cache_alloc_refill(),
3120          * and kmemleak_erase() requires its correct value.
3121          */
3122         ac = cpu_cache_get(cachep);
3123 
3124 out:
3125         /*
3126          * To avoid a false negative, if an object that is in one of the
3127          * per-CPU caches is leaked, we need to make sure kmemleak doesn't
3128          * treat the array pointers as a reference to the object.
3129          */
3130         if (objp)
3131                 kmemleak_erase(&ac->entry[ac->avail]);
3132         return objp;
3133 }
3134 
3135 #ifdef CONFIG_NUMA
3136 /*
3137  * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set.
3138  *
3139  * If we are in_interrupt, then process context, including cpusets and
3140  * mempolicy, may not apply and should not be used for allocation policy.
3141  */
3142 static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags)
3143 {
3144         int nid_alloc, nid_here;
3145 
3146         if (in_interrupt() || (flags & __GFP_THISNODE))
3147                 return NULL;
3148         nid_alloc = nid_here = numa_mem_id();
3149         if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD))
3150                 nid_alloc = cpuset_slab_spread_node();
3151         else if (current->mempolicy)
3152                 nid_alloc = mempolicy_slab_node();
3153         if (nid_alloc != nid_here)
3154                 return ____cache_alloc_node(cachep, flags, nid_alloc);
3155         return NULL;
3156 }
3157 
3158 /*
3159  * Fallback function if there was no memory available and no objects on a
3160  * certain node and fall back is permitted. First we scan all the
3161  * available node for available objects. If that fails then we
3162  * perform an allocation without specifying a node. This allows the page
3163  * allocator to do its reclaim / fallback magic. We then insert the
3164  * slab into the proper nodelist and then allocate from it.
3165  */
3166 static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags)
3167 {
3168         struct zonelist *zonelist;
3169         struct zoneref *z;
3170         struct zone *zone;
3171         enum zone_type high_zoneidx = gfp_zone(flags);
3172         void *obj = NULL;
3173         struct page *page;
3174         int nid;
3175         unsigned int cpuset_mems_cookie;
3176 
3177         if (flags & __GFP_THISNODE)
3178                 return NULL;
3179 
3180 retry_cpuset:
3181         cpuset_mems_cookie = read_mems_allowed_begin();
3182         zonelist = node_zonelist(mempolicy_slab_node(), flags);
3183 
3184 retry:
3185         /*
3186          * Look through allowed nodes for objects available
3187          * from existing per node queues.
3188          */
3189         for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
3190                 nid = zone_to_nid(zone);
3191 
3192                 if (cpuset_zone_allowed(zone, flags) &&
3193                         get_node(cache, nid) &&
3194                         get_node(cache, nid)->free_objects) {
3195                                 obj = ____cache_alloc_node(cache,
3196                                         gfp_exact_node(flags), nid);
3197                                 if (obj)
3198                                         break;
3199                 }
3200         }
3201 
3202         if (!obj) {
3203                 /*
3204                  * This allocation will be performed within the constraints
3205                  * of the current cpuset / memory policy requirements.
3206                  * We may trigger various forms of reclaim on the allowed
3207                  * set and go into memory reserves if necessary.
3208                  */
3209                 page = cache_grow_begin(cache, flags, numa_mem_id());
3210                 cache_grow_end(cache, page);
3211                 if (page) {
3212                         nid = page_to_nid(page);
3213                         obj = ____cache_alloc_node(cache,
3214                                 gfp_exact_node(flags), nid);
3215 
3216                         /*
3217                          * Another processor may allocate the objects in
3218                          * the slab since we are not holding any locks.
3219                          */
3220                         if (!obj)
3221                                 goto retry;
3222                 }
3223         }
3224 
3225         if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie)))
3226                 goto retry_cpuset;
3227         return obj;
3228 }
3229 
3230 /*
3231  * A interface to enable slab creation on nodeid
3232  */
3233 static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags,
3234                                 int nodeid)
3235 {
3236         struct page *page;
3237         struct kmem_cache_node *n;
3238         void *obj = NULL;
3239         void *list = NULL;
3240 
3241         VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES);
3242         n = get_node(cachep, nodeid);
3243         BUG_ON(!n);
3244 
3245         check_irq_off();
3246         spin_lock(&n->list_lock);
3247         page = get_first_slab(n, false);
3248         if (!page)
3249                 goto must_grow;
3250 
3251         check_spinlock_acquired_node(cachep, nodeid);
3252 
3253         STATS_INC_NODEALLOCS(cachep);
3254         STATS_INC_ACTIVE(cachep);
3255         STATS_SET_HIGH(cachep);
3256 
3257         BUG_ON(page->active == cachep->num);
3258 
3259         obj = slab_get_obj(cachep, page);
3260         n->free_objects--;
3261 
3262         fixup_slab_list(cachep, n, page, &list);
3263 
3264         spin_unlock(&n->list_lock);
3265         fixup_objfreelist_debug(cachep, &list);
3266         return obj;
3267 
3268 must_grow:
3269         spin_unlock(&n->list_lock);
3270         page = cache_grow_begin(cachep, gfp_exact_node(flags), nodeid);
3271         if (page) {
3272                 /* This slab isn't counted yet so don't update free_objects */
3273                 obj = slab_get_obj(cachep, page);
3274         }
3275         cache_grow_end(cachep, page);
3276 
3277         return obj ? obj : fallback_alloc(cachep, flags);
3278 }
3279 
3280 static __always_inline void *
3281 slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid,
3282                    unsigned long caller)
3283 {
3284         unsigned long save_flags;
3285         void *ptr;
3286         int slab_node = numa_mem_id();
3287 
3288         flags &= gfp_allowed_mask;
3289         cachep = slab_pre_alloc_hook(cachep, flags);
3290         if (unlikely(!cachep))
3291                 return NULL;
3292 
3293         cache_alloc_debugcheck_before(cachep, flags);
3294         local_irq_save(save_flags);
3295 
3296         if (nodeid == NUMA_NO_NODE)
3297                 nodeid = slab_node;
3298 
3299         if (unlikely(!get_node(cachep, nodeid))) {
3300                 /* Node not bootstrapped yet */
3301                 ptr = fallback_alloc(cachep, flags);
3302                 goto out;
3303         }
3304 
3305         if (nodeid == slab_node) {
3306                 /*
3307                  * Use the locally cached objects if possible.
3308                  * However ____cache_alloc does not allow fallback
3309                  * to other nodes. It may fail while we still have
3310                  * objects on other nodes available.
3311                  */
3312                 ptr = ____cache_alloc(cachep, flags);
3313                 if (ptr)
3314                         goto out;
3315         }
3316         /* ___cache_alloc_node can fall back to other nodes */
3317         ptr = ____cache_alloc_node(cachep, flags, nodeid);
3318   out:
3319         local_irq_restore(save_flags);
3320         ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller);
3321 
3322         if (unlikely(flags & __GFP_ZERO) && ptr)
3323                 memset(ptr, 0, cachep->object_size);
3324 
3325         slab_post_alloc_hook(cachep, flags, 1, &ptr);
3326         return ptr;
3327 }
3328 
3329 static __always_inline void *
3330 __do_cache_alloc(struct kmem_cache *cache, gfp_t flags)
3331 {
3332         void *objp;
3333 
3334         if (current->mempolicy || cpuset_do_slab_mem_spread()) {
3335                 objp = alternate_node_alloc(cache, flags);
3336                 if (objp)
3337                         goto out;
3338         }
3339         objp = ____cache_alloc(cache, flags);
3340 
3341         /*
3342          * We may just have run out of memory on the local node.
3343          * ____cache_alloc_node() knows how to locate memory on other nodes
3344          */
3345         if (!objp)
3346                 objp = ____cache_alloc_node(cache, flags, numa_mem_id());
3347 
3348   out:
3349         return objp;
3350 }
3351 #else
3352 
3353 static __always_inline void *
3354 __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
3355 {
3356         return ____cache_alloc(cachep, flags);
3357 }
3358 
3359 #endif /* CONFIG_NUMA */
3360 
3361 static __always_inline void *
3362 slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller)
3363 {
3364         unsigned long save_flags;
3365         void *objp;
3366 
3367         flags &= gfp_allowed_mask;
3368         cachep = slab_pre_alloc_hook(cachep, flags);
3369         if (unlikely(!cachep))
3370                 return NULL;
3371 
3372         cache_alloc_debugcheck_before(cachep, flags);
3373         local_irq_save(save_flags);
3374         objp = __do_cache_alloc(cachep, flags);
3375         local_irq_restore(save_flags);
3376         objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller);
3377         prefetchw(objp);
3378 
3379         if (unlikely(flags & __GFP_ZERO) && objp)
3380                 memset(objp, 0, cachep->object_size);
3381 
3382         slab_post_alloc_hook(cachep, flags, 1, &objp);
3383         return objp;
3384 }
3385 
3386 /*
3387  * Caller needs to acquire correct kmem_cache_node's list_lock
3388  * @list: List of detached free slabs should be freed by caller
3389  */
3390 static void free_block(struct kmem_cache *cachep, void **objpp,
3391                         int nr_objects, int node, struct list_head *list)
3392 {
3393         int i;
3394         struct kmem_cache_node *n = get_node(cachep, node);
3395         struct page *page;
3396 
3397         n->free_objects += nr_objects;
3398 
3399         for (i = 0; i < nr_objects; i++) {
3400                 void *objp;
3401                 struct page *page;
3402 
3403                 objp = objpp[i];
3404 
3405                 page = virt_to_head_page(objp);
3406                 list_del(&page->lru);
3407                 check_spinlock_acquired_node(cachep, node);
3408                 slab_put_obj(cachep, page, objp);
3409                 STATS_DEC_ACTIVE(cachep);
3410 
3411                 /* fixup slab chains */
3412                 if (page->active == 0) {
3413                         list_add(&page->lru, &n->slabs_free);
3414                         n->free_slabs++;
3415                 } else {
3416                         /* Unconditionally move a slab to the end of the
3417                          * partial list on free - maximum time for the
3418                          * other objects to be freed, too.
3419                          */
3420                         list_add_tail(&page->lru, &n->slabs_partial);
3421                 }
3422         }
3423 
3424         while (n->free_objects > n->free_limit && !list_empty(&n->slabs_free)) {
3425                 n->free_objects -= cachep->num;
3426 
3427                 page = list_last_entry(&n->slabs_free, struct page, lru);
3428                 list_move(&page->lru, list);
3429                 n->free_slabs--;
3430                 n->total_slabs--;
3431         }
3432 }
3433 
3434 static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac)
3435 {
3436         int batchcount;
3437         struct kmem_cache_node *n;
3438         int node = numa_mem_id();
3439         LIST_HEAD(list);
3440 
3441         batchcount = ac->batchcount;
3442 
3443         check_irq_off();
3444         n = get_node(cachep, node);
3445         spin_lock(&n->list_lock);
3446         if (n->shared) {
3447                 struct array_cache *shared_array = n->shared;
3448                 int max = shared_array->limit - shared_array->avail;
3449                 if (max) {
3450                         if (batchcount > max)
3451                                 batchcount = max;
3452                         memcpy(&(shared_array->entry[shared_array->avail]),
3453                                ac->entry, sizeof(void *) * batchcount);
3454                         shared_array->avail += batchcount;
3455                         goto free_done;
3456                 }
3457         }
3458 
3459         free_block(cachep, ac->entry, batchcount, node, &list);
3460 free_done:
3461 #if STATS
3462         {
3463                 int i = 0;
3464                 struct page *page;
3465 
3466                 list_for_each_entry(page, &n->slabs_free, lru) {
3467                         BUG_ON(page->active);
3468 
3469                         i++;
3470                 }
3471                 STATS_SET_FREEABLE(cachep, i);
3472         }
3473 #endif
3474         spin_unlock(&n->list_lock);
3475         slabs_destroy(cachep, &list);
3476         ac->avail -= batchcount;
3477         memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail);
3478 }
3479 
3480 /*
3481  * Release an obj back to its cache. If the obj has a constructed state, it must
3482  * be in this state _before_ it is released.  Called with disabled ints.
3483  */
3484 static inline void __cache_free(struct kmem_cache *cachep, void *objp,
3485                                 unsigned long caller)
3486 {
3487         /* Put the object into the quarantine, don't touch it for now. */
3488         if (kasan_slab_free(cachep, objp))
3489                 return;
3490 
3491         ___cache_free(cachep, objp, caller);
3492 }
3493 
3494 void ___cache_free(struct kmem_cache *cachep, void *objp,
3495                 unsigned long caller)
3496 {
3497         struct array_cache *ac = cpu_cache_get(cachep);
3498 
3499         check_irq_off();
3500         kmemleak_free_recursive(objp, cachep->flags);
3501         objp = cache_free_debugcheck(cachep, objp, caller);
3502 
3503         /*
3504          * Skip calling cache_free_alien() when the platform is not numa.
3505          * This will avoid cache misses that happen while accessing slabp (which
3506          * is per page memory  reference) to get nodeid. Instead use a global
3507          * variable to skip the call, which is mostly likely to be present in
3508          * the cache.
3509          */
3510         if (nr_online_nodes > 1 && cache_free_alien(cachep, objp))
3511                 return;
3512 
3513         if (ac->avail < ac->limit) {
3514                 STATS_INC_FREEHIT(cachep);
3515         } else {
3516                 STATS_INC_FREEMISS(cachep);
3517                 cache_flusharray(cachep, ac);
3518         }
3519 
3520         if (sk_memalloc_socks()) {
3521                 struct page *page = virt_to_head_page(objp);
3522 
3523                 if (unlikely(PageSlabPfmemalloc(page))) {
3524                         cache_free_pfmemalloc(cachep, page, objp);
3525                         return;
3526                 }
3527         }
3528 
3529         ac->entry[ac->avail++] = objp;
3530 }
3531 
3532 /**
3533  * kmem_cache_alloc - Allocate an object
3534  * @cachep: The cache to allocate from.
3535  * @flags: See kmalloc().
3536  *
3537  * Allocate an object from this cache.  The flags are only relevant
3538  * if the cache has no available objects.
3539  */
3540 void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
3541 {
3542         void *ret = slab_alloc(cachep, flags, _RET_IP_);
3543 
3544         kasan_slab_alloc(cachep, ret, flags);
3545         trace_kmem_cache_alloc(_RET_IP_, ret,
3546                                cachep->object_size, cachep->size, flags);
3547 
3548         return ret;
3549 }
3550 EXPORT_SYMBOL(kmem_cache_alloc);
3551 
3552 static __always_inline void
3553 cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags,
3554                                   size_t size, void **p, unsigned long caller)
3555 {
3556         size_t i;
3557 
3558         for (i = 0; i < size; i++)
3559                 p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller);
3560 }
3561 
3562 int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3563                           void **p)
3564 {
3565         size_t i;
3566 
3567         s = slab_pre_alloc_hook(s, flags);
3568         if (!s)
3569                 return 0;
3570 
3571         cache_alloc_debugcheck_before(s, flags);
3572 
3573         local_irq_disable();
3574         for (i = 0; i < size; i++) {
3575                 void *objp = __do_cache_alloc(s, flags);
3576 
3577                 if (unlikely(!objp))
3578                         goto error;
3579                 p[i] = objp;
3580         }
3581         local_irq_enable();
3582 
3583         cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_);
3584 
3585         /* Clear memory outside IRQ disabled section */
3586         if (unlikely(flags & __GFP_ZERO))
3587                 for (i = 0; i < size; i++)
3588                         memset(p[i], 0, s->object_size);
3589 
3590         slab_post_alloc_hook(s, flags, size, p);
3591         /* FIXME: Trace call missing. Christoph would like a bulk variant */
3592         return size;
3593 error:
3594         local_irq_enable();
3595         cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_);
3596         slab_post_alloc_hook(s, flags, i, p);
3597         __kmem_cache_free_bulk(s, i, p);
3598         return 0;
3599 }
3600 EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3601 
3602 #ifdef CONFIG_TRACING
3603 void *
3604 kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size)
3605 {
3606         void *ret;
3607 
3608         ret = slab_alloc(cachep, flags, _RET_IP_);
3609 
3610         kasan_kmalloc(cachep, ret, size, flags);
3611         trace_kmalloc(_RET_IP_, ret,
3612                       size, cachep->size, flags);
3613         return ret;
3614 }
3615 EXPORT_SYMBOL(kmem_cache_alloc_trace);
3616 #endif
3617 
3618 #ifdef CONFIG_NUMA
3619 /**
3620  * kmem_cache_alloc_node - Allocate an object on the specified node
3621  * @cachep: The cache to allocate from.
3622  * @flags: See kmalloc().
3623  * @nodeid: node number of the target node.
3624  *
3625  * Identical to kmem_cache_alloc but it will allocate memory on the given
3626  * node, which can improve the performance for cpu bound structures.
3627  *
3628  * Fallback to other node is possible if __GFP_THISNODE is not set.
3629  */
3630 void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid)
3631 {
3632         void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_);
3633 
3634         kasan_slab_alloc(cachep, ret, flags);
3635         trace_kmem_cache_alloc_node(_RET_IP_, ret,
3636                                     cachep->object_size, cachep->size,
3637                                     flags, nodeid);
3638 
3639         return ret;
3640 }
3641 EXPORT_SYMBOL(kmem_cache_alloc_node);
3642 
3643 #ifdef CONFIG_TRACING
3644 void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep,
3645                                   gfp_t flags,
3646                                   int nodeid,
3647                                   size_t size)
3648 {
3649         void *ret;
3650 
3651         ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_);
3652 
3653         kasan_kmalloc(cachep, ret, size, flags);
3654         trace_kmalloc_node(_RET_IP_, ret,
3655                            size, cachep->size,
3656                            flags, nodeid);
3657         return ret;
3658 }
3659 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
3660 #endif
3661 
3662 static __always_inline void *
3663 __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
3664 {
3665         struct kmem_cache *cachep;
3666         void *ret;
3667 
3668         cachep = kmalloc_slab(size, flags);
3669         if (unlikely(ZERO_OR_NULL_PTR(cachep)))
3670                 return cachep;
3671         ret = kmem_cache_alloc_node_trace(cachep, flags, node, size);
3672         kasan_kmalloc(cachep, ret, size, flags);
3673 
3674         return ret;
3675 }
3676 
3677 void *__kmalloc_node(size_t size, gfp_t flags, int node)
3678 {
3679         return __do_kmalloc_node(size, flags, node, _RET_IP_);
3680 }
3681 EXPORT_SYMBOL(__kmalloc_node);
3682 
3683 void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
3684                 int node, unsigned long caller)
3685 {
3686         return __do_kmalloc_node(size, flags, node, caller);
3687 }
3688 EXPORT_SYMBOL(__kmalloc_node_track_caller);
3689 #endif /* CONFIG_NUMA */
3690 
3691 /**
3692  * __do_kmalloc - allocate memory
3693  * @size: how many bytes of memory are required.
3694  * @flags: the type of memory to allocate (see kmalloc).
3695  * @caller: function caller for debug tracking of the caller
3696  */
3697 static __always_inline void *__do_kmalloc(size_t size, gfp_t flags,
3698                                           unsigned long caller)
3699 {
3700         struct kmem_cache *cachep;
3701         void *ret;
3702 
3703         cachep = kmalloc_slab(size, flags);
3704         if (unlikely(ZERO_OR_NULL_PTR(cachep)))
3705                 return cachep;
3706         ret = slab_alloc(cachep, flags, caller);
3707 
3708         kasan_kmalloc(cachep, ret, size, flags);
3709         trace_kmalloc(caller, ret,
3710                       size, cachep->size, flags);
3711 
3712         return ret;
3713 }
3714 
3715 void *__kmalloc(size_t size, gfp_t flags)
3716 {
3717         return __do_kmalloc(size, flags, _RET_IP_);
3718 }
3719 EXPORT_SYMBOL(__kmalloc);
3720 
3721 void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller)
3722 {
3723         return __do_kmalloc(size, flags, caller);
3724 }
3725 EXPORT_SYMBOL(__kmalloc_track_caller);
3726 
3727 /**
3728  * kmem_cache_free - Deallocate an object
3729  * @cachep: The cache the allocation was from.
3730  * @objp: The previously allocated object.
3731  *
3732  * Free an object which was previously allocated from this
3733  * cache.
3734  */
3735 void kmem_cache_free(struct kmem_cache *cachep, void *objp)
3736 {
3737         unsigned long flags;
3738         cachep = cache_from_obj(cachep, objp);
3739         if (!cachep)
3740                 return;
3741 
3742         local_irq_save(flags);
3743         debug_check_no_locks_freed(objp, cachep->object_size);
3744         if (!(cachep->flags & SLAB_DEBUG_OBJECTS))
3745                 debug_check_no_obj_freed(objp, cachep->object_size);
3746         __cache_free(cachep, objp, _RET_IP_);
3747         local_irq_restore(flags);
3748 
3749         trace_kmem_cache_free(_RET_IP_, objp);
3750 }
3751 EXPORT_SYMBOL(kmem_cache_free);
3752 
3753 void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p)
3754 {
3755         struct kmem_cache *s;
3756         size_t i;
3757 
3758         local_irq_disable();
3759         for (i = 0; i < size; i++) {
3760                 void *objp = p[i];
3761 
3762                 if (!orig_s) /* called via kfree_bulk */
3763                         s = virt_to_cache(objp);
3764                 else
3765                         s = cache_from_obj(orig_s, objp);
3766 
3767                 debug_check_no_locks_freed(objp, s->object_size);
3768                 if (!(s->flags & SLAB_DEBUG_OBJECTS))
3769                         debug_check_no_obj_freed(objp, s->object_size);
3770 
3771                 __cache_free(s, objp, _RET_IP_);
3772         }
3773         local_irq_enable();
3774 
3775         /* FIXME: add tracing */
3776 }
3777 EXPORT_SYMBOL(kmem_cache_free_bulk);
3778 
3779 /**
3780  * kfree - free previously allocated memory
3781  * @objp: pointer returned by kmalloc.
3782  *
3783  * If @objp is NULL, no operation is performed.
3784  *
3785  * Don't free memory not originally allocated by kmalloc()
3786  * or you will run into trouble.
3787  */
3788 void kfree(const void *objp)
3789 {
3790         struct kmem_cache *c;
3791         unsigned long flags;
3792 
3793         trace_kfree(_RET_IP_, objp);
3794 
3795         if (unlikely(ZERO_OR_NULL_PTR(objp)))
3796                 return;
3797         local_irq_save(flags);
3798         kfree_debugcheck(objp);
3799         c = virt_to_cache(objp);
3800         debug_check_no_locks_freed(objp, c->object_size);
3801 
3802         debug_check_no_obj_freed(objp, c->object_size);
3803         __cache_free(c, (void *)objp, _RET_IP_);
3804         local_irq_restore(flags);
3805 }
3806 EXPORT_SYMBOL(kfree);
3807 
3808 /*
3809  * This initializes kmem_cache_node or resizes various caches for all nodes.
3810  */
3811 static int setup_kmem_cache_nodes(struct kmem_cache *cachep, gfp_t gfp)
3812 {
3813         int ret;
3814         int node;
3815         struct kmem_cache_node *n;
3816 
3817         for_each_online_node(node) {
3818                 ret = setup_kmem_cache_node(cachep, node, gfp, true);
3819                 if (ret)
3820                         goto fail;
3821 
3822         }
3823 
3824         return 0;
3825 
3826 fail:
3827         if (!cachep->list.next) {
3828                 /* Cache is not active yet. Roll back what we did */
3829                 node--;
3830                 while (node >= 0) {
3831                         n = get_node(cachep, node);
3832                         if (n) {
3833                                 kfree(n->shared);
3834                                 free_alien_cache(n->alien);
3835                                 kfree(n);
3836                                 cachep->node[node] = NULL;
3837                         }
3838                         node--;
3839                 }
3840         }
3841         return -ENOMEM;
3842 }
3843 
3844 /* Always called with the slab_mutex held */
3845 static int __do_tune_cpucache(struct kmem_cache *cachep, int limit,
3846                                 int batchcount, int shared, gfp_t gfp)
3847 {
3848         struct array_cache __percpu *cpu_cache, *prev;
3849         int cpu;
3850 
3851         cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount);
3852         if (!cpu_cache)
3853                 return -ENOMEM;
3854 
3855         prev = cachep->cpu_cache;
3856         cachep->cpu_cache = cpu_cache;
3857         /*
3858          * Without a previous cpu_cache there's no need to synchronize remote
3859          * cpus, so skip the IPIs.
3860          */
3861         if (prev)
3862                 kick_all_cpus_sync();
3863 
3864         check_irq_on();
3865         cachep->batchcount = batchcount;
3866         cachep->limit = limit;
3867         cachep->shared = shared;
3868 
3869         if (!prev)
3870                 goto setup_node;
3871 
3872         for_each_online_cpu(cpu) {
3873                 LIST_HEAD(list);
3874                 int node;
3875                 struct kmem_cache_node *n;
3876                 struct array_cache *ac = per_cpu_ptr(prev, cpu);
3877 
3878                 node = cpu_to_mem(cpu);
3879                 n = get_node(cachep, node);
3880                 spin_lock_irq(&n->list_lock);
3881                 free_block(cachep, ac->entry, ac->avail, node, &list);
3882                 spin_unlock_irq(&n->list_lock);
3883                 slabs_destroy(cachep, &list);
3884         }
3885         free_percpu(prev);
3886 
3887 setup_node:
3888         return setup_kmem_cache_nodes(cachep, gfp);
3889 }
3890 
3891 static int do_tune_cpucache(struct kmem_cache *cachep, int limit,
3892                                 int batchcount, int shared, gfp_t gfp)
3893 {
3894         int ret;
3895         struct kmem_cache *c;
3896 
3897         ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp);
3898 
3899         if (slab_state < FULL)
3900                 return ret;
3901 
3902         if ((ret < 0) || !is_root_cache(cachep))
3903                 return ret;
3904 
3905         lockdep_assert_held(&slab_mutex);
3906         for_each_memcg_cache(c, cachep) {
3907                 /* return value determined by the root cache only */
3908                 __do_tune_cpucache(c, limit, batchcount, shared, gfp);
3909         }
3910 
3911         return ret;
3912 }
3913 
3914 /* Called with slab_mutex held always */
3915 static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp)
3916 {
3917         int err;
3918         int limit = 0;
3919         int shared = 0;
3920         int batchcount = 0;
3921 
3922         err = cache_random_seq_create(cachep, cachep->num, gfp);
3923         if (err)
3924                 goto end;
3925 
3926         if (!is_root_cache(cachep)) {
3927                 struct kmem_cache *root = memcg_root_cache(cachep);
3928                 limit = root->limit;
3929                 shared = root->shared;
3930                 batchcount = root->batchcount;
3931         }
3932 
3933         if (limit && shared && batchcount)
3934                 goto skip_setup;
3935         /*
3936          * The head array serves three purposes:
3937          * - create a LIFO ordering, i.e. return objects that are cache-warm
3938          * - reduce the number of spinlock operations.
3939          * - reduce the number of linked list operations on the slab and
3940          *   bufctl chains: array operations are cheaper.
3941          * The numbers are guessed, we should auto-tune as described by
3942          * Bonwick.
3943          */
3944         if (cachep->size > 131072)
3945                 limit = 1;
3946         else if (cachep->size > PAGE_SIZE)
3947                 limit = 8;
3948         else if (cachep->size > 1024)
3949                 limit = 24;
3950         else if (cachep->size > 256)
3951                 limit = 54;
3952         else
3953                 limit = 120;
3954 
3955         /*
3956          * CPU bound tasks (e.g. network routing) can exhibit cpu bound
3957          * allocation behaviour: Most allocs on one cpu, most free operations
3958          * on another cpu. For these cases, an efficient object passing between
3959          * cpus is necessary. This is provided by a shared array. The array
3960          * replaces Bonwick's magazine layer.
3961          * On uniprocessor, it's functionally equivalent (but less efficient)
3962          * to a larger limit. Thus disabled by default.
3963          */
3964         shared = 0;
3965         if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1)
3966                 shared = 8;
3967 
3968 #if DEBUG
3969         /*
3970          * With debugging enabled, large batchcount lead to excessively long
3971          * periods with disabled local interrupts. Limit the batchcount
3972          */
3973         if (limit > 32)
3974                 limit = 32;
3975 #endif
3976         batchcount = (limit + 1) / 2;
3977 skip_setup:
3978         err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp);
3979 end:
3980         if (err)
3981                 pr_err("enable_cpucache failed for %s, error %d\n",
3982                        cachep->name, -err);
3983         return err;
3984 }
3985 
3986 /*
3987  * Drain an array if it contains any elements taking the node lock only if
3988  * necessary. Note that the node listlock also protects the array_cache
3989  * if drain_array() is used on the shared array.
3990  */
3991 static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n,
3992                          struct array_cache *ac, int node)
3993 {
3994         LIST_HEAD(list);
3995 
3996         /* ac from n->shared can be freed if we don't hold the slab_mutex. */
3997         check_mutex_acquired();
3998 
3999         if (!ac || !ac->avail)
4000                 return;
4001 
4002         if (ac->touched) {
4003                 ac->touched = 0;
4004                 return;
4005         }
4006 
4007         spin_lock_irq(&n->list_lock);
4008         drain_array_locked(cachep, ac, node, false, &list);
4009         spin_unlock_irq(&n->list_lock);
4010 
4011         slabs_destroy(cachep, &list);
4012 }
4013 
4014 /**
4015  * cache_reap - Reclaim memory from caches.
4016  * @w: work descriptor
4017  *
4018  * Called from workqueue/eventd every few seconds.
4019  * Purpose:
4020  * - clear the per-cpu caches for this CPU.
4021  * - return freeable pages to the main free memory pool.
4022  *
4023  * If we cannot acquire the cache chain mutex then just give up - we'll try
4024  * again on the next iteration.
4025  */
4026 static void cache_reap(struct work_struct *w)
4027 {
4028         struct kmem_cache *searchp;
4029         struct kmem_cache_node *n;
4030         int node = numa_mem_id();
4031         struct delayed_work *work = to_delayed_work(w);
4032 
4033         if (!mutex_trylock(&slab_mutex))
4034                 /* Give up. Setup the next iteration. */
4035                 goto out;
4036 
4037         list_for_each_entry(searchp, &slab_caches, list) {
4038                 check_irq_on();
4039 
4040                 /*
4041                  * We only take the node lock if absolutely necessary and we
4042                  * have established with reasonable certainty that
4043                  * we can do some work if the lock was obtained.
4044                  */
4045                 n = get_node(searchp, node);
4046 
4047                 reap_alien(searchp, n);
4048 
4049                 drain_array(searchp, n, cpu_cache_get(searchp), node);
4050 
4051                 /*
4052                  * These are racy checks but it does not matter
4053                  * if we skip one check or scan twice.
4054                  */
4055                 if (time_after(n->next_reap, jiffies))
4056                         goto next;
4057 
4058                 n->next_reap = jiffies + REAPTIMEOUT_NODE;
4059 
4060                 drain_array(searchp, n, n->shared, node);
4061 
4062                 if (n->free_touched)
4063                         n->free_touched = 0;
4064                 else {
4065                         int freed;
4066 
4067                         freed = drain_freelist(searchp, n, (n->free_limit +
4068                                 5 * searchp->num - 1) / (5 * searchp->num));
4069                         STATS_ADD_REAPED(searchp, freed);
4070                 }
4071 next:
4072                 cond_resched();
4073         }
4074         check_irq_on();
4075         mutex_unlock(&slab_mutex);
4076         next_reap_node();
4077 out:
4078         /* Set up the next iteration */
4079         schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC));
4080 }
4081 
4082 void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo)
4083 {
4084         unsigned long active_objs, num_objs, active_slabs;
4085         unsigned long total_slabs = 0, free_objs = 0, shared_avail = 0;
4086         unsigned long free_slabs = 0;
4087         int node;
4088         struct kmem_cache_node *n;
4089 
4090         for_each_kmem_cache_node(cachep, node, n) {
4091                 check_irq_on();
4092                 spin_lock_irq(&n->list_lock);
4093 
4094                 total_slabs += n->total_slabs;
4095                 free_slabs += n->free_slabs;
4096                 free_objs += n->free_objects;
4097 
4098                 if (n->shared)
4099                         shared_avail += n->shared->avail;
4100 
4101                 spin_unlock_irq(&n->list_lock);
4102         }
4103         num_objs = total_slabs * cachep->num;
4104         active_slabs = total_slabs - free_slabs;
4105         active_objs = num_objs - free_objs;
4106 
4107         sinfo->active_objs = active_objs;
4108         sinfo->num_objs = num_objs;
4109         sinfo->active_slabs = active_slabs;
4110         sinfo->num_slabs = total_slabs;
4111         sinfo->shared_avail = shared_avail;
4112         sinfo->limit = cachep->limit;
4113         sinfo->batchcount = cachep->batchcount;
4114         sinfo->shared = cachep->shared;
4115         sinfo->objects_per_slab = cachep->num;
4116         sinfo->cache_order = cachep->gfporder;
4117 }
4118 
4119 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep)
4120 {
4121 #if STATS
4122         {                       /* node stats */
4123                 unsigned long high = cachep->high_mark;
4124                 unsigned long allocs = cachep->num_allocations;
4125                 unsigned long grown = cachep->grown;
4126                 unsigned long reaped = cachep->reaped;
4127                 unsigned long errors = cachep->errors;
4128                 unsigned long max_freeable = cachep->max_freeable;
4129                 unsigned long node_allocs = cachep->node_allocs;
4130                 unsigned long node_frees = cachep->node_frees;
4131                 unsigned long overflows = cachep->node_overflow;
4132 
4133                 seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu",
4134                            allocs, high, grown,
4135                            reaped, errors, max_freeable, node_allocs,
4136                            node_frees, overflows);
4137         }
4138         /* cpu stats */
4139         {
4140                 unsigned long allochit = atomic_read(&cachep->allochit);
4141                 unsigned long allocmiss = atomic_read(&cachep->allocmiss);
4142                 unsigned long freehit = atomic_read(&cachep->freehit);
4143                 unsigned long freemiss = atomic_read(&cachep->freemiss);
4144 
4145                 seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu",
4146                            allochit, allocmiss, freehit, freemiss);
4147         }
4148 #endif
4149 }
4150 
4151 #define MAX_SLABINFO_WRITE 128
4152 /**
4153  * slabinfo_write - Tuning for the slab allocator
4154  * @file: unused
4155  * @buffer: user buffer
4156  * @count: data length
4157  * @ppos: unused
4158  */
4159 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
4160                        size_t count, loff_t *ppos)
4161 {
4162         char kbuf[MAX_SLABINFO_WRITE + 1], *tmp;
4163         int limit, batchcount, shared, res;
4164         struct kmem_cache *cachep;
4165 
4166         if (count > MAX_SLABINFO_WRITE)
4167                 return -EINVAL;
4168         if (copy_from_user(&kbuf, buffer, count))
4169                 return -EFAULT;
4170         kbuf[MAX_SLABINFO_WRITE] = '\0';
4171 
4172         tmp = strchr(kbuf, ' ');
4173         if (!tmp)
4174                 return -EINVAL;
4175         *tmp = '\0';
4176         tmp++;
4177         if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3)
4178                 return -EINVAL;
4179 
4180         /* Find the cache in the chain of caches. */
4181         mutex_lock(&slab_mutex);
4182         res = -EINVAL;
4183         list_for_each_entry(cachep, &slab_caches, list) {
4184                 if (!strcmp(cachep->name, kbuf)) {
4185                         if (limit < 1 || batchcount < 1 ||
4186                                         batchcount > limit || shared < 0) {
4187                                 res = 0;
4188                         } else {
4189                                 res = do_tune_cpucache(cachep, limit,
4190                                                        batchcount, shared,
4191                                                        GFP_KERNEL);
4192                         }
4193                         break;
4194                 }
4195         }
4196         mutex_unlock(&slab_mutex);
4197         if (res >= 0)
4198                 res = count;
4199         return res;
4200 }
4201 
4202 #ifdef CONFIG_DEBUG_SLAB_LEAK
4203 
4204 static inline int add_caller(unsigned long *n, unsigned long v)
4205 {
4206         unsigned long *p;
4207         int l;
4208         if (!v)
4209                 return 1;
4210         l = n[1];
4211         p = n + 2;
4212         while (l) {
4213                 int i = l/2;
4214                 unsigned long *q = p + 2 * i;
4215                 if (*q == v) {
4216                         q[1]++;
4217                         return 1;
4218                 }
4219                 if (*q > v) {
4220                         l = i;
4221                 } else {
4222                         p = q + 2;
4223                         l -= i + 1;
4224                 }
4225         }
4226         if (++n[1] == n[0])
4227                 return 0;
4228         memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n));
4229         p[0] = v;
4230         p[1] = 1;
4231         return 1;
4232 }
4233 
4234 static void handle_slab(unsigned long *n, struct kmem_cache *c,
4235                                                 struct page *page)
4236 {
4237         void *p;
4238         int i, j;
4239         unsigned long v;
4240 
4241         if (n[0] == n[1])
4242                 return;
4243         for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) {
4244                 bool active = true;
4245 
4246                 for (j = page->active; j < c->num; j++) {
4247                         if (get_free_obj(page, j) == i) {
4248                                 active = false;
4249                                 break;
4250                         }
4251                 }
4252 
4253                 if (!active)
4254                         continue;
4255 
4256                 /*
4257                  * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table
4258                  * mapping is established when actual object allocation and
4259                  * we could mistakenly access the unmapped object in the cpu
4260                  * cache.
4261                  */
4262                 if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v)))
4263                         continue;
4264 
4265                 if (!add_caller(n, v))
4266                         return;
4267         }
4268 }
4269 
4270 static void show_symbol(struct seq_file *m, unsigned long address)
4271 {
4272 #ifdef CONFIG_KALLSYMS
4273         unsigned long offset, size;
4274         char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN];
4275 
4276         if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) {
4277                 seq_printf(m, "%s+%#lx/%#lx", name, offset, size);
4278                 if (modname[0])
4279                         seq_printf(m, " [%s]", modname);
4280                 return;
4281         }
4282 #endif
4283         seq_printf(m, "%px", (void *)address);
4284 }
4285 
4286 static int leaks_show(struct seq_file *m, void *p)
4287 {
4288         struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list);
4289         struct page *page;
4290         struct kmem_cache_node *n;
4291         const char *name;
4292         unsigned long *x = m->private;
4293         int node;
4294         int i;
4295 
4296         if (!(cachep->flags & SLAB_STORE_USER))
4297                 return 0;
4298         if (!(cachep->flags & SLAB_RED_ZONE))
4299                 return 0;
4300 
4301         /*
4302          * Set store_user_clean and start to grab stored user information
4303          * for all objects on this cache. If some alloc/free requests comes
4304          * during the processing, information would be wrong so restart
4305          * whole processing.
4306          */
4307         do {
4308                 set_store_user_clean(cachep);
4309                 drain_cpu_caches(cachep);
4310 
4311                 x[1] = 0;
4312 
4313                 for_each_kmem_cache_node(cachep, node, n) {
4314 
4315                         check_irq_on();
4316                         spin_lock_irq(&n->list_lock);
4317 
4318                         list_for_each_entry(page, &n->slabs_full, lru)
4319                                 handle_slab(x, cachep, page);
4320                         list_for_each_entry(page, &n->slabs_partial, lru)
4321                                 handle_slab(x, cachep, page);
4322                         spin_unlock_irq(&n->list_lock);
4323                 }
4324         } while (!is_store_user_clean(cachep));
4325 
4326         name = cachep->name;
4327         if (x[0] == x[1]) {
4328                 /* Increase the buffer size */
4329                 mutex_unlock(&slab_mutex);
4330                 m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL);
4331                 if (!m->private) {
4332                         /* Too bad, we are really out */
4333                         m->private = x;
4334                         mutex_lock(&slab_mutex);
4335                         return -ENOMEM;
4336                 }
4337                 *(unsigned long *)m->private = x[0] * 2;
4338                 kfree(x);
4339                 mutex_lock(&slab_mutex);
4340                 /* Now make sure this entry will be retried */
4341                 m->count = m->size;
4342                 return 0;
4343         }
4344         for (i = 0; i < x[1]; i++) {
4345                 seq_printf(m, "%s: %lu ", name, x[2*i+3]);
4346                 show_symbol(m, x[2*i+2]);
4347                 seq_putc(m, '\n');
4348         }
4349 
4350         return 0;
4351 }
4352 
4353 static const struct seq_operations slabstats_op = {
4354         .start = slab_start,
4355         .next = slab_next,
4356         .stop = slab_stop,
4357         .show = leaks_show,
4358 };
4359 
4360 static int slabstats_open(struct inode *inode, struct file *file)
4361 {
4362         unsigned long *n;
4363 
4364         n = __seq_open_private(file, &slabstats_op, PAGE_SIZE);
4365         if (!n)
4366                 return -ENOMEM;
4367 
4368         *n = PAGE_SIZE / (2 * sizeof(unsigned long));
4369 
4370         return 0;
4371 }
4372 
4373 static const struct file_operations proc_slabstats_operations = {
4374         .open           = slabstats_open,
4375         .read           = seq_read,
4376         .llseek         = seq_lseek,
4377         .release        = seq_release_private,
4378 };
4379 #endif
4380 
4381 static int __init slab_proc_init(void)
4382 {
4383 #ifdef CONFIG_DEBUG_SLAB_LEAK
4384         proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations);
4385 #endif
4386         return 0;
4387 }
4388 module_init(slab_proc_init);
4389 
4390 #ifdef CONFIG_HARDENED_USERCOPY
4391 /*
4392  * Rejects objects that are incorrectly sized.
4393  *
4394  * Returns NULL if check passes, otherwise const char * to name of cache
4395  * to indicate an error.
4396  */
4397 const char *__check_heap_object(const void *ptr, unsigned long n,
4398                                 struct page *page)
4399 {
4400         struct kmem_cache *cachep;
4401         unsigned int objnr;
4402         unsigned long offset;
4403 
4404         /* Find and validate object. */
4405         cachep = page->slab_cache;
4406         objnr = obj_to_index(cachep, page, (void *)ptr);
4407         BUG_ON(objnr >= cachep->num);
4408 
4409         /* Find offset within object. */
4410         offset = ptr - index_to_obj(cachep, page, objnr) - obj_offset(cachep);
4411 
4412         /* Allow address range falling entirely within object size. */
4413         if (offset <= cachep->object_size && n <= cachep->object_size - offset)
4414                 return NULL;
4415 
4416         return cachep->name;
4417 }
4418 #endif /* CONFIG_HARDENED_USERCOPY */
4419 
4420 /**
4421  * ksize - get the actual amount of memory allocated for a given object
4422  * @objp: Pointer to the object
4423  *
4424  * kmalloc may internally round up allocations and return more memory
4425  * than requested. ksize() can be used to determine the actual amount of
4426  * memory allocated. The caller may use this additional memory, even though
4427  * a smaller amount of memory was initially specified with the kmalloc call.
4428  * The caller must guarantee that objp points to a valid object previously
4429  * allocated with either kmalloc() or kmem_cache_alloc(). The object
4430  * must not be freed during the duration of the call.
4431  */
4432 size_t ksize(const void *objp)
4433 {
4434         size_t size;
4435 
4436         BUG_ON(!objp);
4437         if (unlikely(objp == ZERO_SIZE_PTR))
4438                 return 0;
4439 
4440         size = virt_to_cache(objp)->object_size;
4441         /* We assume that ksize callers could use the whole allocated area,
4442          * so we need to unpoison this area.
4443          */
4444         kasan_unpoison_shadow(objp, size);
4445 
4446         return size;
4447 }
4448 EXPORT_SYMBOL(ksize);
4449 

~ [ 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