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

TOMOYO Linux Cross Reference
Linux/mm/slab_common.c

Version: ~ [ linux-5.8 ] ~ [ linux-5.7.14 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.57 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.138 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.193 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.232 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.232 ] ~ [ 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  * Slab allocator functions that are independent of the allocator strategy
  4  *
  5  * (C) 2012 Christoph Lameter <cl@linux.com>
  6  */
  7 #include <linux/slab.h>
  8 
  9 #include <linux/mm.h>
 10 #include <linux/poison.h>
 11 #include <linux/interrupt.h>
 12 #include <linux/memory.h>
 13 #include <linux/cache.h>
 14 #include <linux/compiler.h>
 15 #include <linux/module.h>
 16 #include <linux/cpu.h>
 17 #include <linux/uaccess.h>
 18 #include <linux/seq_file.h>
 19 #include <linux/proc_fs.h>
 20 #include <asm/cacheflush.h>
 21 #include <asm/tlbflush.h>
 22 #include <asm/page.h>
 23 #include <linux/memcontrol.h>
 24 
 25 #define CREATE_TRACE_POINTS
 26 #include <trace/events/kmem.h>
 27 
 28 #include "slab.h"
 29 
 30 enum slab_state slab_state;
 31 LIST_HEAD(slab_caches);
 32 DEFINE_MUTEX(slab_mutex);
 33 struct kmem_cache *kmem_cache;
 34 
 35 #ifdef CONFIG_HARDENED_USERCOPY
 36 bool usercopy_fallback __ro_after_init =
 37                 IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK);
 38 module_param(usercopy_fallback, bool, 0400);
 39 MODULE_PARM_DESC(usercopy_fallback,
 40                 "WARN instead of reject usercopy whitelist violations");
 41 #endif
 42 
 43 static LIST_HEAD(slab_caches_to_rcu_destroy);
 44 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
 45 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
 46                     slab_caches_to_rcu_destroy_workfn);
 47 
 48 /*
 49  * Set of flags that will prevent slab merging
 50  */
 51 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
 52                 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
 53                 SLAB_FAILSLAB | SLAB_KASAN)
 54 
 55 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
 56                          SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
 57 
 58 /*
 59  * Merge control. If this is set then no merging of slab caches will occur.
 60  */
 61 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
 62 
 63 static int __init setup_slab_nomerge(char *str)
 64 {
 65         slab_nomerge = true;
 66         return 1;
 67 }
 68 
 69 #ifdef CONFIG_SLUB
 70 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
 71 #endif
 72 
 73 __setup("slab_nomerge", setup_slab_nomerge);
 74 
 75 /*
 76  * Determine the size of a slab object
 77  */
 78 unsigned int kmem_cache_size(struct kmem_cache *s)
 79 {
 80         return s->object_size;
 81 }
 82 EXPORT_SYMBOL(kmem_cache_size);
 83 
 84 #ifdef CONFIG_DEBUG_VM
 85 static int kmem_cache_sanity_check(const char *name, unsigned int size)
 86 {
 87         if (!name || in_interrupt() || size < sizeof(void *) ||
 88                 size > KMALLOC_MAX_SIZE) {
 89                 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
 90                 return -EINVAL;
 91         }
 92 
 93         WARN_ON(strchr(name, ' '));     /* It confuses parsers */
 94         return 0;
 95 }
 96 #else
 97 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
 98 {
 99         return 0;
100 }
101 #endif
102 
103 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
104 {
105         size_t i;
106 
107         for (i = 0; i < nr; i++) {
108                 if (s)
109                         kmem_cache_free(s, p[i]);
110                 else
111                         kfree(p[i]);
112         }
113 }
114 
115 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
116                                                                 void **p)
117 {
118         size_t i;
119 
120         for (i = 0; i < nr; i++) {
121                 void *x = p[i] = kmem_cache_alloc(s, flags);
122                 if (!x) {
123                         __kmem_cache_free_bulk(s, i, p);
124                         return 0;
125                 }
126         }
127         return i;
128 }
129 
130 #ifdef CONFIG_MEMCG_KMEM
131 
132 LIST_HEAD(slab_root_caches);
133 
134 void slab_init_memcg_params(struct kmem_cache *s)
135 {
136         s->memcg_params.root_cache = NULL;
137         RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
138         INIT_LIST_HEAD(&s->memcg_params.children);
139         s->memcg_params.dying = false;
140 }
141 
142 static int init_memcg_params(struct kmem_cache *s,
143                 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
144 {
145         struct memcg_cache_array *arr;
146 
147         if (root_cache) {
148                 s->memcg_params.root_cache = root_cache;
149                 s->memcg_params.memcg = memcg;
150                 INIT_LIST_HEAD(&s->memcg_params.children_node);
151                 INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node);
152                 return 0;
153         }
154 
155         slab_init_memcg_params(s);
156 
157         if (!memcg_nr_cache_ids)
158                 return 0;
159 
160         arr = kvzalloc(sizeof(struct memcg_cache_array) +
161                        memcg_nr_cache_ids * sizeof(void *),
162                        GFP_KERNEL);
163         if (!arr)
164                 return -ENOMEM;
165 
166         RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
167         return 0;
168 }
169 
170 static void destroy_memcg_params(struct kmem_cache *s)
171 {
172         if (is_root_cache(s))
173                 kvfree(rcu_access_pointer(s->memcg_params.memcg_caches));
174 }
175 
176 static void free_memcg_params(struct rcu_head *rcu)
177 {
178         struct memcg_cache_array *old;
179 
180         old = container_of(rcu, struct memcg_cache_array, rcu);
181         kvfree(old);
182 }
183 
184 static int update_memcg_params(struct kmem_cache *s, int new_array_size)
185 {
186         struct memcg_cache_array *old, *new;
187 
188         new = kvzalloc(sizeof(struct memcg_cache_array) +
189                        new_array_size * sizeof(void *), GFP_KERNEL);
190         if (!new)
191                 return -ENOMEM;
192 
193         old = rcu_dereference_protected(s->memcg_params.memcg_caches,
194                                         lockdep_is_held(&slab_mutex));
195         if (old)
196                 memcpy(new->entries, old->entries,
197                        memcg_nr_cache_ids * sizeof(void *));
198 
199         rcu_assign_pointer(s->memcg_params.memcg_caches, new);
200         if (old)
201                 call_rcu(&old->rcu, free_memcg_params);
202         return 0;
203 }
204 
205 int memcg_update_all_caches(int num_memcgs)
206 {
207         struct kmem_cache *s;
208         int ret = 0;
209 
210         mutex_lock(&slab_mutex);
211         list_for_each_entry(s, &slab_root_caches, root_caches_node) {
212                 ret = update_memcg_params(s, num_memcgs);
213                 /*
214                  * Instead of freeing the memory, we'll just leave the caches
215                  * up to this point in an updated state.
216                  */
217                 if (ret)
218                         break;
219         }
220         mutex_unlock(&slab_mutex);
221         return ret;
222 }
223 
224 void memcg_link_cache(struct kmem_cache *s)
225 {
226         if (is_root_cache(s)) {
227                 list_add(&s->root_caches_node, &slab_root_caches);
228         } else {
229                 list_add(&s->memcg_params.children_node,
230                          &s->memcg_params.root_cache->memcg_params.children);
231                 list_add(&s->memcg_params.kmem_caches_node,
232                          &s->memcg_params.memcg->kmem_caches);
233         }
234 }
235 
236 static void memcg_unlink_cache(struct kmem_cache *s)
237 {
238         if (is_root_cache(s)) {
239                 list_del(&s->root_caches_node);
240         } else {
241                 list_del(&s->memcg_params.children_node);
242                 list_del(&s->memcg_params.kmem_caches_node);
243         }
244 }
245 #else
246 static inline int init_memcg_params(struct kmem_cache *s,
247                 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
248 {
249         return 0;
250 }
251 
252 static inline void destroy_memcg_params(struct kmem_cache *s)
253 {
254 }
255 
256 static inline void memcg_unlink_cache(struct kmem_cache *s)
257 {
258 }
259 #endif /* CONFIG_MEMCG_KMEM */
260 
261 /*
262  * Figure out what the alignment of the objects will be given a set of
263  * flags, a user specified alignment and the size of the objects.
264  */
265 static unsigned int calculate_alignment(slab_flags_t flags,
266                 unsigned int align, unsigned int size)
267 {
268         /*
269          * If the user wants hardware cache aligned objects then follow that
270          * suggestion if the object is sufficiently large.
271          *
272          * The hardware cache alignment cannot override the specified
273          * alignment though. If that is greater then use it.
274          */
275         if (flags & SLAB_HWCACHE_ALIGN) {
276                 unsigned int ralign;
277 
278                 ralign = cache_line_size();
279                 while (size <= ralign / 2)
280                         ralign /= 2;
281                 align = max(align, ralign);
282         }
283 
284         if (align < ARCH_SLAB_MINALIGN)
285                 align = ARCH_SLAB_MINALIGN;
286 
287         return ALIGN(align, sizeof(void *));
288 }
289 
290 /*
291  * Find a mergeable slab cache
292  */
293 int slab_unmergeable(struct kmem_cache *s)
294 {
295         if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
296                 return 1;
297 
298         if (!is_root_cache(s))
299                 return 1;
300 
301         if (s->ctor)
302                 return 1;
303 
304         if (s->usersize)
305                 return 1;
306 
307         /*
308          * We may have set a slab to be unmergeable during bootstrap.
309          */
310         if (s->refcount < 0)
311                 return 1;
312 
313         return 0;
314 }
315 
316 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
317                 slab_flags_t flags, const char *name, void (*ctor)(void *))
318 {
319         struct kmem_cache *s;
320 
321         if (slab_nomerge)
322                 return NULL;
323 
324         if (ctor)
325                 return NULL;
326 
327         size = ALIGN(size, sizeof(void *));
328         align = calculate_alignment(flags, align, size);
329         size = ALIGN(size, align);
330         flags = kmem_cache_flags(size, flags, name, NULL);
331 
332         if (flags & SLAB_NEVER_MERGE)
333                 return NULL;
334 
335         list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) {
336                 if (slab_unmergeable(s))
337                         continue;
338 
339                 if (size > s->size)
340                         continue;
341 
342                 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
343                         continue;
344                 /*
345                  * Check if alignment is compatible.
346                  * Courtesy of Adrian Drzewiecki
347                  */
348                 if ((s->size & ~(align - 1)) != s->size)
349                         continue;
350 
351                 if (s->size - size >= sizeof(void *))
352                         continue;
353 
354                 if (IS_ENABLED(CONFIG_SLAB) && align &&
355                         (align > s->align || s->align % align))
356                         continue;
357 
358                 return s;
359         }
360         return NULL;
361 }
362 
363 static struct kmem_cache *create_cache(const char *name,
364                 unsigned int object_size, unsigned int align,
365                 slab_flags_t flags, unsigned int useroffset,
366                 unsigned int usersize, void (*ctor)(void *),
367                 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
368 {
369         struct kmem_cache *s;
370         int err;
371 
372         if (WARN_ON(useroffset + usersize > object_size))
373                 useroffset = usersize = 0;
374 
375         err = -ENOMEM;
376         s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
377         if (!s)
378                 goto out;
379 
380         s->name = name;
381         s->size = s->object_size = object_size;
382         s->align = align;
383         s->ctor = ctor;
384         s->useroffset = useroffset;
385         s->usersize = usersize;
386 
387         err = init_memcg_params(s, memcg, root_cache);
388         if (err)
389                 goto out_free_cache;
390 
391         err = __kmem_cache_create(s, flags);
392         if (err)
393                 goto out_free_cache;
394 
395         s->refcount = 1;
396         list_add(&s->list, &slab_caches);
397         memcg_link_cache(s);
398 out:
399         if (err)
400                 return ERR_PTR(err);
401         return s;
402 
403 out_free_cache:
404         destroy_memcg_params(s);
405         kmem_cache_free(kmem_cache, s);
406         goto out;
407 }
408 
409 /**
410  * kmem_cache_create_usercopy - Create a cache with a region suitable
411  * for copying to userspace
412  * @name: A string which is used in /proc/slabinfo to identify this cache.
413  * @size: The size of objects to be created in this cache.
414  * @align: The required alignment for the objects.
415  * @flags: SLAB flags
416  * @useroffset: Usercopy region offset
417  * @usersize: Usercopy region size
418  * @ctor: A constructor for the objects.
419  *
420  * Cannot be called within a interrupt, but can be interrupted.
421  * The @ctor is run when new pages are allocated by the cache.
422  *
423  * The flags are
424  *
425  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
426  * to catch references to uninitialised memory.
427  *
428  * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
429  * for buffer overruns.
430  *
431  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
432  * cacheline.  This can be beneficial if you're counting cycles as closely
433  * as davem.
434  *
435  * Return: a pointer to the cache on success, NULL on failure.
436  */
437 struct kmem_cache *
438 kmem_cache_create_usercopy(const char *name,
439                   unsigned int size, unsigned int align,
440                   slab_flags_t flags,
441                   unsigned int useroffset, unsigned int usersize,
442                   void (*ctor)(void *))
443 {
444         struct kmem_cache *s = NULL;
445         const char *cache_name;
446         int err;
447 
448         get_online_cpus();
449         get_online_mems();
450         memcg_get_cache_ids();
451 
452         mutex_lock(&slab_mutex);
453 
454         err = kmem_cache_sanity_check(name, size);
455         if (err) {
456                 goto out_unlock;
457         }
458 
459         /* Refuse requests with allocator specific flags */
460         if (flags & ~SLAB_FLAGS_PERMITTED) {
461                 err = -EINVAL;
462                 goto out_unlock;
463         }
464 
465         /*
466          * Some allocators will constraint the set of valid flags to a subset
467          * of all flags. We expect them to define CACHE_CREATE_MASK in this
468          * case, and we'll just provide them with a sanitized version of the
469          * passed flags.
470          */
471         flags &= CACHE_CREATE_MASK;
472 
473         /* Fail closed on bad usersize of useroffset values. */
474         if (WARN_ON(!usersize && useroffset) ||
475             WARN_ON(size < usersize || size - usersize < useroffset))
476                 usersize = useroffset = 0;
477 
478         if (!usersize)
479                 s = __kmem_cache_alias(name, size, align, flags, ctor);
480         if (s)
481                 goto out_unlock;
482 
483         cache_name = kstrdup_const(name, GFP_KERNEL);
484         if (!cache_name) {
485                 err = -ENOMEM;
486                 goto out_unlock;
487         }
488 
489         s = create_cache(cache_name, size,
490                          calculate_alignment(flags, align, size),
491                          flags, useroffset, usersize, ctor, NULL, NULL);
492         if (IS_ERR(s)) {
493                 err = PTR_ERR(s);
494                 kfree_const(cache_name);
495         }
496 
497 out_unlock:
498         mutex_unlock(&slab_mutex);
499 
500         memcg_put_cache_ids();
501         put_online_mems();
502         put_online_cpus();
503 
504         if (err) {
505                 if (flags & SLAB_PANIC)
506                         panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
507                                 name, err);
508                 else {
509                         pr_warn("kmem_cache_create(%s) failed with error %d\n",
510                                 name, err);
511                         dump_stack();
512                 }
513                 return NULL;
514         }
515         return s;
516 }
517 EXPORT_SYMBOL(kmem_cache_create_usercopy);
518 
519 /**
520  * kmem_cache_create - Create a cache.
521  * @name: A string which is used in /proc/slabinfo to identify this cache.
522  * @size: The size of objects to be created in this cache.
523  * @align: The required alignment for the objects.
524  * @flags: SLAB flags
525  * @ctor: A constructor for the objects.
526  *
527  * Cannot be called within a interrupt, but can be interrupted.
528  * The @ctor is run when new pages are allocated by the cache.
529  *
530  * The flags are
531  *
532  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
533  * to catch references to uninitialised memory.
534  *
535  * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
536  * for buffer overruns.
537  *
538  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
539  * cacheline.  This can be beneficial if you're counting cycles as closely
540  * as davem.
541  *
542  * Return: a pointer to the cache on success, NULL on failure.
543  */
544 struct kmem_cache *
545 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
546                 slab_flags_t flags, void (*ctor)(void *))
547 {
548         return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
549                                           ctor);
550 }
551 EXPORT_SYMBOL(kmem_cache_create);
552 
553 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
554 {
555         LIST_HEAD(to_destroy);
556         struct kmem_cache *s, *s2;
557 
558         /*
559          * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
560          * @slab_caches_to_rcu_destroy list.  The slab pages are freed
561          * through RCU and and the associated kmem_cache are dereferenced
562          * while freeing the pages, so the kmem_caches should be freed only
563          * after the pending RCU operations are finished.  As rcu_barrier()
564          * is a pretty slow operation, we batch all pending destructions
565          * asynchronously.
566          */
567         mutex_lock(&slab_mutex);
568         list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
569         mutex_unlock(&slab_mutex);
570 
571         if (list_empty(&to_destroy))
572                 return;
573 
574         rcu_barrier();
575 
576         list_for_each_entry_safe(s, s2, &to_destroy, list) {
577 #ifdef SLAB_SUPPORTS_SYSFS
578                 sysfs_slab_release(s);
579 #else
580                 slab_kmem_cache_release(s);
581 #endif
582         }
583 }
584 
585 static int shutdown_cache(struct kmem_cache *s)
586 {
587         /* free asan quarantined objects */
588         kasan_cache_shutdown(s);
589 
590         if (__kmem_cache_shutdown(s) != 0)
591                 return -EBUSY;
592 
593         memcg_unlink_cache(s);
594         list_del(&s->list);
595 
596         if (s->flags & SLAB_TYPESAFE_BY_RCU) {
597 #ifdef SLAB_SUPPORTS_SYSFS
598                 sysfs_slab_unlink(s);
599 #endif
600                 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
601                 schedule_work(&slab_caches_to_rcu_destroy_work);
602         } else {
603 #ifdef SLAB_SUPPORTS_SYSFS
604                 sysfs_slab_unlink(s);
605                 sysfs_slab_release(s);
606 #else
607                 slab_kmem_cache_release(s);
608 #endif
609         }
610 
611         return 0;
612 }
613 
614 #ifdef CONFIG_MEMCG_KMEM
615 /*
616  * memcg_create_kmem_cache - Create a cache for a memory cgroup.
617  * @memcg: The memory cgroup the new cache is for.
618  * @root_cache: The parent of the new cache.
619  *
620  * This function attempts to create a kmem cache that will serve allocation
621  * requests going from @memcg to @root_cache. The new cache inherits properties
622  * from its parent.
623  */
624 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
625                              struct kmem_cache *root_cache)
626 {
627         static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
628         struct cgroup_subsys_state *css = &memcg->css;
629         struct memcg_cache_array *arr;
630         struct kmem_cache *s = NULL;
631         char *cache_name;
632         int idx;
633 
634         get_online_cpus();
635         get_online_mems();
636 
637         mutex_lock(&slab_mutex);
638 
639         /*
640          * The memory cgroup could have been offlined while the cache
641          * creation work was pending.
642          */
643         if (memcg->kmem_state != KMEM_ONLINE || root_cache->memcg_params.dying)
644                 goto out_unlock;
645 
646         idx = memcg_cache_id(memcg);
647         arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
648                                         lockdep_is_held(&slab_mutex));
649 
650         /*
651          * Since per-memcg caches are created asynchronously on first
652          * allocation (see memcg_kmem_get_cache()), several threads can try to
653          * create the same cache, but only one of them may succeed.
654          */
655         if (arr->entries[idx])
656                 goto out_unlock;
657 
658         cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
659         cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
660                                css->serial_nr, memcg_name_buf);
661         if (!cache_name)
662                 goto out_unlock;
663 
664         s = create_cache(cache_name, root_cache->object_size,
665                          root_cache->align,
666                          root_cache->flags & CACHE_CREATE_MASK,
667                          root_cache->useroffset, root_cache->usersize,
668                          root_cache->ctor, memcg, root_cache);
669         /*
670          * If we could not create a memcg cache, do not complain, because
671          * that's not critical at all as we can always proceed with the root
672          * cache.
673          */
674         if (IS_ERR(s)) {
675                 kfree(cache_name);
676                 goto out_unlock;
677         }
678 
679         /*
680          * Since readers won't lock (see cache_from_memcg_idx()), we need a
681          * barrier here to ensure nobody will see the kmem_cache partially
682          * initialized.
683          */
684         smp_wmb();
685         arr->entries[idx] = s;
686 
687 out_unlock:
688         mutex_unlock(&slab_mutex);
689 
690         put_online_mems();
691         put_online_cpus();
692 }
693 
694 static void kmemcg_deactivate_workfn(struct work_struct *work)
695 {
696         struct kmem_cache *s = container_of(work, struct kmem_cache,
697                                             memcg_params.deact_work);
698 
699         get_online_cpus();
700         get_online_mems();
701 
702         mutex_lock(&slab_mutex);
703 
704         s->memcg_params.deact_fn(s);
705 
706         mutex_unlock(&slab_mutex);
707 
708         put_online_mems();
709         put_online_cpus();
710 
711         /* done, put the ref from slab_deactivate_memcg_cache_rcu_sched() */
712         css_put(&s->memcg_params.memcg->css);
713 }
714 
715 static void kmemcg_deactivate_rcufn(struct rcu_head *head)
716 {
717         struct kmem_cache *s = container_of(head, struct kmem_cache,
718                                             memcg_params.deact_rcu_head);
719 
720         /*
721          * We need to grab blocking locks.  Bounce to ->deact_work.  The
722          * work item shares the space with the RCU head and can't be
723          * initialized eariler.
724          */
725         INIT_WORK(&s->memcg_params.deact_work, kmemcg_deactivate_workfn);
726         queue_work(memcg_kmem_cache_wq, &s->memcg_params.deact_work);
727 }
728 
729 /**
730  * slab_deactivate_memcg_cache_rcu_sched - schedule deactivation after a
731  *                                         sched RCU grace period
732  * @s: target kmem_cache
733  * @deact_fn: deactivation function to call
734  *
735  * Schedule @deact_fn to be invoked with online cpus, mems and slab_mutex
736  * held after a sched RCU grace period.  The slab is guaranteed to stay
737  * alive until @deact_fn is finished.  This is to be used from
738  * __kmemcg_cache_deactivate().
739  */
740 void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s,
741                                            void (*deact_fn)(struct kmem_cache *))
742 {
743         if (WARN_ON_ONCE(is_root_cache(s)) ||
744             WARN_ON_ONCE(s->memcg_params.deact_fn))
745                 return;
746 
747         if (s->memcg_params.root_cache->memcg_params.dying)
748                 return;
749 
750         /* pin memcg so that @s doesn't get destroyed in the middle */
751         css_get(&s->memcg_params.memcg->css);
752 
753         s->memcg_params.deact_fn = deact_fn;
754         call_rcu(&s->memcg_params.deact_rcu_head, kmemcg_deactivate_rcufn);
755 }
756 
757 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
758 {
759         int idx;
760         struct memcg_cache_array *arr;
761         struct kmem_cache *s, *c;
762 
763         idx = memcg_cache_id(memcg);
764 
765         get_online_cpus();
766         get_online_mems();
767 
768         mutex_lock(&slab_mutex);
769         list_for_each_entry(s, &slab_root_caches, root_caches_node) {
770                 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
771                                                 lockdep_is_held(&slab_mutex));
772                 c = arr->entries[idx];
773                 if (!c)
774                         continue;
775 
776                 __kmemcg_cache_deactivate(c);
777                 arr->entries[idx] = NULL;
778         }
779         mutex_unlock(&slab_mutex);
780 
781         put_online_mems();
782         put_online_cpus();
783 }
784 
785 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
786 {
787         struct kmem_cache *s, *s2;
788 
789         get_online_cpus();
790         get_online_mems();
791 
792         mutex_lock(&slab_mutex);
793         list_for_each_entry_safe(s, s2, &memcg->kmem_caches,
794                                  memcg_params.kmem_caches_node) {
795                 /*
796                  * The cgroup is about to be freed and therefore has no charges
797                  * left. Hence, all its caches must be empty by now.
798                  */
799                 BUG_ON(shutdown_cache(s));
800         }
801         mutex_unlock(&slab_mutex);
802 
803         put_online_mems();
804         put_online_cpus();
805 }
806 
807 static int shutdown_memcg_caches(struct kmem_cache *s)
808 {
809         struct memcg_cache_array *arr;
810         struct kmem_cache *c, *c2;
811         LIST_HEAD(busy);
812         int i;
813 
814         BUG_ON(!is_root_cache(s));
815 
816         /*
817          * First, shutdown active caches, i.e. caches that belong to online
818          * memory cgroups.
819          */
820         arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
821                                         lockdep_is_held(&slab_mutex));
822         for_each_memcg_cache_index(i) {
823                 c = arr->entries[i];
824                 if (!c)
825                         continue;
826                 if (shutdown_cache(c))
827                         /*
828                          * The cache still has objects. Move it to a temporary
829                          * list so as not to try to destroy it for a second
830                          * time while iterating over inactive caches below.
831                          */
832                         list_move(&c->memcg_params.children_node, &busy);
833                 else
834                         /*
835                          * The cache is empty and will be destroyed soon. Clear
836                          * the pointer to it in the memcg_caches array so that
837                          * it will never be accessed even if the root cache
838                          * stays alive.
839                          */
840                         arr->entries[i] = NULL;
841         }
842 
843         /*
844          * Second, shutdown all caches left from memory cgroups that are now
845          * offline.
846          */
847         list_for_each_entry_safe(c, c2, &s->memcg_params.children,
848                                  memcg_params.children_node)
849                 shutdown_cache(c);
850 
851         list_splice(&busy, &s->memcg_params.children);
852 
853         /*
854          * A cache being destroyed must be empty. In particular, this means
855          * that all per memcg caches attached to it must be empty too.
856          */
857         if (!list_empty(&s->memcg_params.children))
858                 return -EBUSY;
859         return 0;
860 }
861 
862 static void flush_memcg_workqueue(struct kmem_cache *s)
863 {
864         mutex_lock(&slab_mutex);
865         s->memcg_params.dying = true;
866         mutex_unlock(&slab_mutex);
867 
868         /*
869          * SLUB deactivates the kmem_caches through call_rcu. Make
870          * sure all registered rcu callbacks have been invoked.
871          */
872         if (IS_ENABLED(CONFIG_SLUB))
873                 rcu_barrier();
874 
875         /*
876          * SLAB and SLUB create memcg kmem_caches through workqueue and SLUB
877          * deactivates the memcg kmem_caches through workqueue. Make sure all
878          * previous workitems on workqueue are processed.
879          */
880         flush_workqueue(memcg_kmem_cache_wq);
881 }
882 #else
883 static inline int shutdown_memcg_caches(struct kmem_cache *s)
884 {
885         return 0;
886 }
887 
888 static inline void flush_memcg_workqueue(struct kmem_cache *s)
889 {
890 }
891 #endif /* CONFIG_MEMCG_KMEM */
892 
893 void slab_kmem_cache_release(struct kmem_cache *s)
894 {
895         __kmem_cache_release(s);
896         destroy_memcg_params(s);
897         kfree_const(s->name);
898         kmem_cache_free(kmem_cache, s);
899 }
900 
901 void kmem_cache_destroy(struct kmem_cache *s)
902 {
903         int err;
904 
905         if (unlikely(!s))
906                 return;
907 
908         flush_memcg_workqueue(s);
909 
910         get_online_cpus();
911         get_online_mems();
912 
913         mutex_lock(&slab_mutex);
914 
915         s->refcount--;
916         if (s->refcount)
917                 goto out_unlock;
918 
919         err = shutdown_memcg_caches(s);
920         if (!err)
921                 err = shutdown_cache(s);
922 
923         if (err) {
924                 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
925                        s->name);
926                 dump_stack();
927         }
928 out_unlock:
929         mutex_unlock(&slab_mutex);
930 
931         put_online_mems();
932         put_online_cpus();
933 }
934 EXPORT_SYMBOL(kmem_cache_destroy);
935 
936 /**
937  * kmem_cache_shrink - Shrink a cache.
938  * @cachep: The cache to shrink.
939  *
940  * Releases as many slabs as possible for a cache.
941  * To help debugging, a zero exit status indicates all slabs were released.
942  *
943  * Return: %0 if all slabs were released, non-zero otherwise
944  */
945 int kmem_cache_shrink(struct kmem_cache *cachep)
946 {
947         int ret;
948 
949         get_online_cpus();
950         get_online_mems();
951         kasan_cache_shrink(cachep);
952         ret = __kmem_cache_shrink(cachep);
953         put_online_mems();
954         put_online_cpus();
955         return ret;
956 }
957 EXPORT_SYMBOL(kmem_cache_shrink);
958 
959 bool slab_is_available(void)
960 {
961         return slab_state >= UP;
962 }
963 
964 #ifndef CONFIG_SLOB
965 /* Create a cache during boot when no slab services are available yet */
966 void __init create_boot_cache(struct kmem_cache *s, const char *name,
967                 unsigned int size, slab_flags_t flags,
968                 unsigned int useroffset, unsigned int usersize)
969 {
970         int err;
971 
972         s->name = name;
973         s->size = s->object_size = size;
974         s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
975         s->useroffset = useroffset;
976         s->usersize = usersize;
977 
978         slab_init_memcg_params(s);
979 
980         err = __kmem_cache_create(s, flags);
981 
982         if (err)
983                 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
984                                         name, size, err);
985 
986         s->refcount = -1;       /* Exempt from merging for now */
987 }
988 
989 struct kmem_cache *__init create_kmalloc_cache(const char *name,
990                 unsigned int size, slab_flags_t flags,
991                 unsigned int useroffset, unsigned int usersize)
992 {
993         struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
994 
995         if (!s)
996                 panic("Out of memory when creating slab %s\n", name);
997 
998         create_boot_cache(s, name, size, flags, useroffset, usersize);
999         list_add(&s->list, &slab_caches);
1000         memcg_link_cache(s);
1001         s->refcount = 1;
1002         return s;
1003 }
1004 
1005 struct kmem_cache *
1006 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
1007 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
1008 EXPORT_SYMBOL(kmalloc_caches);
1009 
1010 /*
1011  * Conversion table for small slabs sizes / 8 to the index in the
1012  * kmalloc array. This is necessary for slabs < 192 since we have non power
1013  * of two cache sizes there. The size of larger slabs can be determined using
1014  * fls.
1015  */
1016 static u8 size_index[24] __ro_after_init = {
1017         3,      /* 8 */
1018         4,      /* 16 */
1019         5,      /* 24 */
1020         5,      /* 32 */
1021         6,      /* 40 */
1022         6,      /* 48 */
1023         6,      /* 56 */
1024         6,      /* 64 */
1025         1,      /* 72 */
1026         1,      /* 80 */
1027         1,      /* 88 */
1028         1,      /* 96 */
1029         7,      /* 104 */
1030         7,      /* 112 */
1031         7,      /* 120 */
1032         7,      /* 128 */
1033         2,      /* 136 */
1034         2,      /* 144 */
1035         2,      /* 152 */
1036         2,      /* 160 */
1037         2,      /* 168 */
1038         2,      /* 176 */
1039         2,      /* 184 */
1040         2       /* 192 */
1041 };
1042 
1043 static inline unsigned int size_index_elem(unsigned int bytes)
1044 {
1045         return (bytes - 1) / 8;
1046 }
1047 
1048 /*
1049  * Find the kmem_cache structure that serves a given size of
1050  * allocation
1051  */
1052 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
1053 {
1054         unsigned int index;
1055 
1056         if (size <= 192) {
1057                 if (!size)
1058                         return ZERO_SIZE_PTR;
1059 
1060                 index = size_index[size_index_elem(size)];
1061         } else {
1062                 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
1063                         return NULL;
1064                 index = fls(size - 1);
1065         }
1066 
1067         return kmalloc_caches[kmalloc_type(flags)][index];
1068 }
1069 
1070 /*
1071  * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
1072  * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
1073  * kmalloc-67108864.
1074  */
1075 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
1076         {NULL,                      0},         {"kmalloc-96",             96},
1077         {"kmalloc-192",           192},         {"kmalloc-8",               8},
1078         {"kmalloc-16",             16},         {"kmalloc-32",             32},
1079         {"kmalloc-64",             64},         {"kmalloc-128",           128},
1080         {"kmalloc-256",           256},         {"kmalloc-512",           512},
1081         {"kmalloc-1k",           1024},         {"kmalloc-2k",           2048},
1082         {"kmalloc-4k",           4096},         {"kmalloc-8k",           8192},
1083         {"kmalloc-16k",         16384},         {"kmalloc-32k",         32768},
1084         {"kmalloc-64k",         65536},         {"kmalloc-128k",       131072},
1085         {"kmalloc-256k",       262144},         {"kmalloc-512k",       524288},
1086         {"kmalloc-1M",        1048576},         {"kmalloc-2M",        2097152},
1087         {"kmalloc-4M",        4194304},         {"kmalloc-8M",        8388608},
1088         {"kmalloc-16M",      16777216},         {"kmalloc-32M",      33554432},
1089         {"kmalloc-64M",      67108864}
1090 };
1091 
1092 /*
1093  * Patch up the size_index table if we have strange large alignment
1094  * requirements for the kmalloc array. This is only the case for
1095  * MIPS it seems. The standard arches will not generate any code here.
1096  *
1097  * Largest permitted alignment is 256 bytes due to the way we
1098  * handle the index determination for the smaller caches.
1099  *
1100  * Make sure that nothing crazy happens if someone starts tinkering
1101  * around with ARCH_KMALLOC_MINALIGN
1102  */
1103 void __init setup_kmalloc_cache_index_table(void)
1104 {
1105         unsigned int i;
1106 
1107         BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
1108                 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
1109 
1110         for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
1111                 unsigned int elem = size_index_elem(i);
1112 
1113                 if (elem >= ARRAY_SIZE(size_index))
1114                         break;
1115                 size_index[elem] = KMALLOC_SHIFT_LOW;
1116         }
1117 
1118         if (KMALLOC_MIN_SIZE >= 64) {
1119                 /*
1120                  * The 96 byte size cache is not used if the alignment
1121                  * is 64 byte.
1122                  */
1123                 for (i = 64 + 8; i <= 96; i += 8)
1124                         size_index[size_index_elem(i)] = 7;
1125 
1126         }
1127 
1128         if (KMALLOC_MIN_SIZE >= 128) {
1129                 /*
1130                  * The 192 byte sized cache is not used if the alignment
1131                  * is 128 byte. Redirect kmalloc to use the 256 byte cache
1132                  * instead.
1133                  */
1134                 for (i = 128 + 8; i <= 192; i += 8)
1135                         size_index[size_index_elem(i)] = 8;
1136         }
1137 }
1138 
1139 static const char *
1140 kmalloc_cache_name(const char *prefix, unsigned int size)
1141 {
1142 
1143         static const char units[3] = "\0kM";
1144         int idx = 0;
1145 
1146         while (size >= 1024 && (size % 1024 == 0)) {
1147                 size /= 1024;
1148                 idx++;
1149         }
1150 
1151         return kasprintf(GFP_NOWAIT, "%s-%u%c", prefix, size, units[idx]);
1152 }
1153 
1154 static void __init
1155 new_kmalloc_cache(int idx, int type, slab_flags_t flags)
1156 {
1157         const char *name;
1158 
1159         if (type == KMALLOC_RECLAIM) {
1160                 flags |= SLAB_RECLAIM_ACCOUNT;
1161                 name = kmalloc_cache_name("kmalloc-rcl",
1162                                                 kmalloc_info[idx].size);
1163                 BUG_ON(!name);
1164         } else {
1165                 name = kmalloc_info[idx].name;
1166         }
1167 
1168         kmalloc_caches[type][idx] = create_kmalloc_cache(name,
1169                                         kmalloc_info[idx].size, flags, 0,
1170                                         kmalloc_info[idx].size);
1171 }
1172 
1173 /*
1174  * Create the kmalloc array. Some of the regular kmalloc arrays
1175  * may already have been created because they were needed to
1176  * enable allocations for slab creation.
1177  */
1178 void __init create_kmalloc_caches(slab_flags_t flags)
1179 {
1180         int i, type;
1181 
1182         for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) {
1183                 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
1184                         if (!kmalloc_caches[type][i])
1185                                 new_kmalloc_cache(i, type, flags);
1186 
1187                         /*
1188                          * Caches that are not of the two-to-the-power-of size.
1189                          * These have to be created immediately after the
1190                          * earlier power of two caches
1191                          */
1192                         if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
1193                                         !kmalloc_caches[type][1])
1194                                 new_kmalloc_cache(1, type, flags);
1195                         if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
1196                                         !kmalloc_caches[type][2])
1197                                 new_kmalloc_cache(2, type, flags);
1198                 }
1199         }
1200 
1201         /* Kmalloc array is now usable */
1202         slab_state = UP;
1203 
1204 #ifdef CONFIG_ZONE_DMA
1205         for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
1206                 struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
1207 
1208                 if (s) {
1209                         unsigned int size = kmalloc_size(i);
1210                         const char *n = kmalloc_cache_name("dma-kmalloc", size);
1211 
1212                         BUG_ON(!n);
1213                         kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
1214                                 n, size, SLAB_CACHE_DMA | flags, 0, 0);
1215                 }
1216         }
1217 #endif
1218 }
1219 #endif /* !CONFIG_SLOB */
1220 
1221 /*
1222  * To avoid unnecessary overhead, we pass through large allocation requests
1223  * directly to the page allocator. We use __GFP_COMP, because we will need to
1224  * know the allocation order to free the pages properly in kfree.
1225  */
1226 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1227 {
1228         void *ret;
1229         struct page *page;
1230 
1231         flags |= __GFP_COMP;
1232         page = alloc_pages(flags, order);
1233         ret = page ? page_address(page) : NULL;
1234         ret = kasan_kmalloc_large(ret, size, flags);
1235         /* As ret might get tagged, call kmemleak hook after KASAN. */
1236         kmemleak_alloc(ret, size, 1, flags);
1237         return ret;
1238 }
1239 EXPORT_SYMBOL(kmalloc_order);
1240 
1241 #ifdef CONFIG_TRACING
1242 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1243 {
1244         void *ret = kmalloc_order(size, flags, order);
1245         trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1246         return ret;
1247 }
1248 EXPORT_SYMBOL(kmalloc_order_trace);
1249 #endif
1250 
1251 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1252 /* Randomize a generic freelist */
1253 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1254                                unsigned int count)
1255 {
1256         unsigned int rand;
1257         unsigned int i;
1258 
1259         for (i = 0; i < count; i++)
1260                 list[i] = i;
1261 
1262         /* Fisher-Yates shuffle */
1263         for (i = count - 1; i > 0; i--) {
1264                 rand = prandom_u32_state(state);
1265                 rand %= (i + 1);
1266                 swap(list[i], list[rand]);
1267         }
1268 }
1269 
1270 /* Create a random sequence per cache */
1271 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1272                                     gfp_t gfp)
1273 {
1274         struct rnd_state state;
1275 
1276         if (count < 2 || cachep->random_seq)
1277                 return 0;
1278 
1279         cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1280         if (!cachep->random_seq)
1281                 return -ENOMEM;
1282 
1283         /* Get best entropy at this stage of boot */
1284         prandom_seed_state(&state, get_random_long());
1285 
1286         freelist_randomize(&state, cachep->random_seq, count);
1287         return 0;
1288 }
1289 
1290 /* Destroy the per-cache random freelist sequence */
1291 void cache_random_seq_destroy(struct kmem_cache *cachep)
1292 {
1293         kfree(cachep->random_seq);
1294         cachep->random_seq = NULL;
1295 }
1296 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1297 
1298 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1299 #ifdef CONFIG_SLAB
1300 #define SLABINFO_RIGHTS (0600)
1301 #else
1302 #define SLABINFO_RIGHTS (0400)
1303 #endif
1304 
1305 static void print_slabinfo_header(struct seq_file *m)
1306 {
1307         /*
1308          * Output format version, so at least we can change it
1309          * without _too_ many complaints.
1310          */
1311 #ifdef CONFIG_DEBUG_SLAB
1312         seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1313 #else
1314         seq_puts(m, "slabinfo - version: 2.1\n");
1315 #endif
1316         seq_puts(m, "# name            <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1317         seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1318         seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1319 #ifdef CONFIG_DEBUG_SLAB
1320         seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1321         seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1322 #endif
1323         seq_putc(m, '\n');
1324 }
1325 
1326 void *slab_start(struct seq_file *m, loff_t *pos)
1327 {
1328         mutex_lock(&slab_mutex);
1329         return seq_list_start(&slab_root_caches, *pos);
1330 }
1331 
1332 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1333 {
1334         return seq_list_next(p, &slab_root_caches, pos);
1335 }
1336 
1337 void slab_stop(struct seq_file *m, void *p)
1338 {
1339         mutex_unlock(&slab_mutex);
1340 }
1341 
1342 static void
1343 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1344 {
1345         struct kmem_cache *c;
1346         struct slabinfo sinfo;
1347 
1348         if (!is_root_cache(s))
1349                 return;
1350 
1351         for_each_memcg_cache(c, s) {
1352                 memset(&sinfo, 0, sizeof(sinfo));
1353                 get_slabinfo(c, &sinfo);
1354 
1355                 info->active_slabs += sinfo.active_slabs;
1356                 info->num_slabs += sinfo.num_slabs;
1357                 info->shared_avail += sinfo.shared_avail;
1358                 info->active_objs += sinfo.active_objs;
1359                 info->num_objs += sinfo.num_objs;
1360         }
1361 }
1362 
1363 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1364 {
1365         struct slabinfo sinfo;
1366 
1367         memset(&sinfo, 0, sizeof(sinfo));
1368         get_slabinfo(s, &sinfo);
1369 
1370         memcg_accumulate_slabinfo(s, &sinfo);
1371 
1372         seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1373                    cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1374                    sinfo.objects_per_slab, (1 << sinfo.cache_order));
1375 
1376         seq_printf(m, " : tunables %4u %4u %4u",
1377                    sinfo.limit, sinfo.batchcount, sinfo.shared);
1378         seq_printf(m, " : slabdata %6lu %6lu %6lu",
1379                    sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1380         slabinfo_show_stats(m, s);
1381         seq_putc(m, '\n');
1382 }
1383 
1384 static int slab_show(struct seq_file *m, void *p)
1385 {
1386         struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node);
1387 
1388         if (p == slab_root_caches.next)
1389                 print_slabinfo_header(m);
1390         cache_show(s, m);
1391         return 0;
1392 }
1393 
1394 void dump_unreclaimable_slab(void)
1395 {
1396         struct kmem_cache *s, *s2;
1397         struct slabinfo sinfo;
1398 
1399         /*
1400          * Here acquiring slab_mutex is risky since we don't prefer to get
1401          * sleep in oom path. But, without mutex hold, it may introduce a
1402          * risk of crash.
1403          * Use mutex_trylock to protect the list traverse, dump nothing
1404          * without acquiring the mutex.
1405          */
1406         if (!mutex_trylock(&slab_mutex)) {
1407                 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1408                 return;
1409         }
1410 
1411         pr_info("Unreclaimable slab info:\n");
1412         pr_info("Name                      Used          Total\n");
1413 
1414         list_for_each_entry_safe(s, s2, &slab_caches, list) {
1415                 if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT))
1416                         continue;
1417 
1418                 get_slabinfo(s, &sinfo);
1419 
1420                 if (sinfo.num_objs > 0)
1421                         pr_info("%-17s %10luKB %10luKB\n", cache_name(s),
1422                                 (sinfo.active_objs * s->size) / 1024,
1423                                 (sinfo.num_objs * s->size) / 1024);
1424         }
1425         mutex_unlock(&slab_mutex);
1426 }
1427 
1428 #if defined(CONFIG_MEMCG)
1429 void *memcg_slab_start(struct seq_file *m, loff_t *pos)
1430 {
1431         struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
1432 
1433         mutex_lock(&slab_mutex);
1434         return seq_list_start(&memcg->kmem_caches, *pos);
1435 }
1436 
1437 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos)
1438 {
1439         struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
1440 
1441         return seq_list_next(p, &memcg->kmem_caches, pos);
1442 }
1443 
1444 void memcg_slab_stop(struct seq_file *m, void *p)
1445 {
1446         mutex_unlock(&slab_mutex);
1447 }
1448 
1449 int memcg_slab_show(struct seq_file *m, void *p)
1450 {
1451         struct kmem_cache *s = list_entry(p, struct kmem_cache,
1452                                           memcg_params.kmem_caches_node);
1453         struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
1454 
1455         if (p == memcg->kmem_caches.next)
1456                 print_slabinfo_header(m);
1457         cache_show(s, m);
1458         return 0;
1459 }
1460 #endif
1461 
1462 /*
1463  * slabinfo_op - iterator that generates /proc/slabinfo
1464  *
1465  * Output layout:
1466  * cache-name
1467  * num-active-objs
1468  * total-objs
1469  * object size
1470  * num-active-slabs
1471  * total-slabs
1472  * num-pages-per-slab
1473  * + further values on SMP and with statistics enabled
1474  */
1475 static const struct seq_operations slabinfo_op = {
1476         .start = slab_start,
1477         .next = slab_next,
1478         .stop = slab_stop,
1479         .show = slab_show,
1480 };
1481 
1482 static int slabinfo_open(struct inode *inode, struct file *file)
1483 {
1484         return seq_open(file, &slabinfo_op);
1485 }
1486 
1487 static const struct file_operations proc_slabinfo_operations = {
1488         .open           = slabinfo_open,
1489         .read           = seq_read,
1490         .write          = slabinfo_write,
1491         .llseek         = seq_lseek,
1492         .release        = seq_release,
1493 };
1494 
1495 static int __init slab_proc_init(void)
1496 {
1497         proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1498                                                 &proc_slabinfo_operations);
1499         return 0;
1500 }
1501 module_init(slab_proc_init);
1502 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1503 
1504 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1505                                            gfp_t flags)
1506 {
1507         void *ret;
1508         size_t ks = 0;
1509 
1510         if (p)
1511                 ks = ksize(p);
1512 
1513         if (ks >= new_size) {
1514                 p = kasan_krealloc((void *)p, new_size, flags);
1515                 return (void *)p;
1516         }
1517 
1518         ret = kmalloc_track_caller(new_size, flags);
1519         if (ret && p)
1520                 memcpy(ret, p, ks);
1521 
1522         return ret;
1523 }
1524 
1525 /**
1526  * __krealloc - like krealloc() but don't free @p.
1527  * @p: object to reallocate memory for.
1528  * @new_size: how many bytes of memory are required.
1529  * @flags: the type of memory to allocate.
1530  *
1531  * This function is like krealloc() except it never frees the originally
1532  * allocated buffer. Use this if you don't want to free the buffer immediately
1533  * like, for example, with RCU.
1534  *
1535  * Return: pointer to the allocated memory or %NULL in case of error
1536  */
1537 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1538 {
1539         if (unlikely(!new_size))
1540                 return ZERO_SIZE_PTR;
1541 
1542         return __do_krealloc(p, new_size, flags);
1543 
1544 }
1545 EXPORT_SYMBOL(__krealloc);
1546 
1547 /**
1548  * krealloc - reallocate memory. The contents will remain unchanged.
1549  * @p: object to reallocate memory for.
1550  * @new_size: how many bytes of memory are required.
1551  * @flags: the type of memory to allocate.
1552  *
1553  * The contents of the object pointed to are preserved up to the
1554  * lesser of the new and old sizes.  If @p is %NULL, krealloc()
1555  * behaves exactly like kmalloc().  If @new_size is 0 and @p is not a
1556  * %NULL pointer, the object pointed to is freed.
1557  *
1558  * Return: pointer to the allocated memory or %NULL in case of error
1559  */
1560 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1561 {
1562         void *ret;
1563 
1564         if (unlikely(!new_size)) {
1565                 kfree(p);
1566                 return ZERO_SIZE_PTR;
1567         }
1568 
1569         ret = __do_krealloc(p, new_size, flags);
1570         if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1571                 kfree(p);
1572 
1573         return ret;
1574 }
1575 EXPORT_SYMBOL(krealloc);
1576 
1577 /**
1578  * kzfree - like kfree but zero memory
1579  * @p: object to free memory of
1580  *
1581  * The memory of the object @p points to is zeroed before freed.
1582  * If @p is %NULL, kzfree() does nothing.
1583  *
1584  * Note: this function zeroes the whole allocated buffer which can be a good
1585  * deal bigger than the requested buffer size passed to kmalloc(). So be
1586  * careful when using this function in performance sensitive code.
1587  */
1588 void kzfree(const void *p)
1589 {
1590         size_t ks;
1591         void *mem = (void *)p;
1592 
1593         if (unlikely(ZERO_OR_NULL_PTR(mem)))
1594                 return;
1595         ks = ksize(mem);
1596         memset(mem, 0, ks);
1597         kfree(mem);
1598 }
1599 EXPORT_SYMBOL(kzfree);
1600 
1601 /* Tracepoints definitions. */
1602 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1603 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1604 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1605 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1606 EXPORT_TRACEPOINT_SYMBOL(kfree);
1607 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1608 
1609 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1610 {
1611         if (__should_failslab(s, gfpflags))
1612                 return -ENOMEM;
1613         return 0;
1614 }
1615 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);
1616 

~ [ 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