blob: 1408ada9ff7295aae17c2d8bb9fdfdfb9c23ec86 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001/* SPDX-License-Identifier: GPL-2.0 */
Christoph Lameter97d06602012-07-06 15:25:11 -05002#ifndef MM_SLAB_H
3#define MM_SLAB_H
4/*
5 * Internal slab definitions
6 */
7
Matthew Wilcox (Oracle)d1220192021-10-04 14:45:51 +01008/* Reuses the bits in struct page */
9struct slab {
10 unsigned long __page_flags;
11 union {
12 struct list_head slab_list;
13 struct { /* Partial pages */
14 struct slab *next;
15#ifdef CONFIG_64BIT
16 int slabs; /* Nr of slabs left */
17#else
18 short int slabs;
19#endif
20 };
21 struct rcu_head rcu_head;
22 };
23 struct kmem_cache *slab_cache; /* not slob */
24 /* Double-word boundary */
25 void *freelist; /* first free object */
26 union {
27 void *s_mem; /* slab: first object */
28 unsigned long counters; /* SLUB */
29 struct { /* SLUB */
30 unsigned inuse:16;
31 unsigned objects:15;
32 unsigned frozen:1;
33 };
34 };
35
36 union {
37 unsigned int active; /* SLAB */
38 int units; /* SLOB */
39 };
40 atomic_t __page_refcount;
41#ifdef CONFIG_MEMCG
42 unsigned long memcg_data;
43#endif
44};
45
46#define SLAB_MATCH(pg, sl) \
47 static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl))
48SLAB_MATCH(flags, __page_flags);
49SLAB_MATCH(compound_head, slab_list); /* Ensure bit 0 is clear */
50SLAB_MATCH(slab_list, slab_list);
51SLAB_MATCH(rcu_head, rcu_head);
52SLAB_MATCH(slab_cache, slab_cache);
53SLAB_MATCH(s_mem, s_mem);
54SLAB_MATCH(active, active);
55SLAB_MATCH(_refcount, __page_refcount);
56#ifdef CONFIG_MEMCG
57SLAB_MATCH(memcg_data, memcg_data);
58#endif
59#undef SLAB_MATCH
60static_assert(sizeof(struct slab) <= sizeof(struct page));
61
62/**
63 * folio_slab - Converts from folio to slab.
64 * @folio: The folio.
65 *
66 * Currently struct slab is a different representation of a folio where
67 * folio_test_slab() is true.
68 *
69 * Return: The slab which contains this folio.
70 */
71#define folio_slab(folio) (_Generic((folio), \
72 const struct folio *: (const struct slab *)(folio), \
73 struct folio *: (struct slab *)(folio)))
74
75/**
76 * slab_folio - The folio allocated for a slab
77 * @slab: The slab.
78 *
79 * Slabs are allocated as folios that contain the individual objects and are
80 * using some fields in the first struct page of the folio - those fields are
81 * now accessed by struct slab. It is occasionally necessary to convert back to
82 * a folio in order to communicate with the rest of the mm. Please use this
83 * helper function instead of casting yourself, as the implementation may change
84 * in the future.
85 */
86#define slab_folio(s) (_Generic((s), \
87 const struct slab *: (const struct folio *)s, \
88 struct slab *: (struct folio *)s))
89
90/**
91 * page_slab - Converts from first struct page to slab.
92 * @p: The first (either head of compound or single) page of slab.
93 *
94 * A temporary wrapper to convert struct page to struct slab in situations where
95 * we know the page is the compound head, or single order-0 page.
96 *
97 * Long-term ideally everything would work with struct slab directly or go
98 * through folio to struct slab.
99 *
100 * Return: The slab which contains this page
101 */
102#define page_slab(p) (_Generic((p), \
103 const struct page *: (const struct slab *)(p), \
104 struct page *: (struct slab *)(p)))
105
106/**
107 * slab_page - The first struct page allocated for a slab
108 * @slab: The slab.
109 *
110 * A convenience wrapper for converting slab to the first struct page of the
111 * underlying folio, to communicate with code not yet converted to folio or
112 * struct slab.
113 */
114#define slab_page(s) folio_page(slab_folio(s), 0)
115
116/*
117 * If network-based swap is enabled, sl*b must keep track of whether pages
118 * were allocated from pfmemalloc reserves.
119 */
120static inline bool slab_test_pfmemalloc(const struct slab *slab)
121{
122 return folio_test_active((struct folio *)slab_folio(slab));
123}
124
125static inline void slab_set_pfmemalloc(struct slab *slab)
126{
127 folio_set_active(slab_folio(slab));
128}
129
130static inline void slab_clear_pfmemalloc(struct slab *slab)
131{
132 folio_clear_active(slab_folio(slab));
133}
134
135static inline void __slab_clear_pfmemalloc(struct slab *slab)
136{
137 __folio_clear_active(slab_folio(slab));
138}
139
140static inline void *slab_address(const struct slab *slab)
141{
142 return folio_address(slab_folio(slab));
143}
144
145static inline int slab_nid(const struct slab *slab)
146{
147 return folio_nid(slab_folio(slab));
148}
149
150static inline pg_data_t *slab_pgdat(const struct slab *slab)
151{
152 return folio_pgdat(slab_folio(slab));
153}
154
155static inline struct slab *virt_to_slab(const void *addr)
156{
157 struct folio *folio = virt_to_folio(addr);
158
159 if (!folio_test_slab(folio))
160 return NULL;
161
162 return folio_slab(folio);
163}
164
165static inline int slab_order(const struct slab *slab)
166{
167 return folio_order((struct folio *)slab_folio(slab));
168}
169
170static inline size_t slab_size(const struct slab *slab)
171{
172 return PAGE_SIZE << slab_order(slab);
173}
174
Joonsoo Kim07f361b2014-10-09 15:26:00 -0700175#ifdef CONFIG_SLOB
176/*
177 * Common fields provided in kmem_cache by all slab allocators
178 * This struct is either used directly by the allocator (SLOB)
179 * or the allocator must include definitions for all fields
180 * provided in kmem_cache_common in their definition of kmem_cache.
181 *
182 * Once we can do anonymous structs (C11 standard) we could put a
183 * anonymous struct definition in these allocators so that the
184 * separate allocations in the kmem_cache structure of SLAB and
185 * SLUB is no longer needed.
186 */
187struct kmem_cache {
188 unsigned int object_size;/* The original size of the object */
189 unsigned int size; /* The aligned/padded/added on size */
190 unsigned int align; /* Alignment as calculated */
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800191 slab_flags_t flags; /* Active flags on the slab */
Alexey Dobriyan7bbdb812018-04-05 16:21:31 -0700192 unsigned int useroffset;/* Usercopy region offset */
193 unsigned int usersize; /* Usercopy region size */
Joonsoo Kim07f361b2014-10-09 15:26:00 -0700194 const char *name; /* Slab name for sysfs */
195 int refcount; /* Use counter */
196 void (*ctor)(void *); /* Called on object slot creation */
197 struct list_head list; /* List of all slab caches on the system */
198};
199
200#endif /* CONFIG_SLOB */
201
202#ifdef CONFIG_SLAB
203#include <linux/slab_def.h>
204#endif
205
206#ifdef CONFIG_SLUB
207#include <linux/slub_def.h>
208#endif
209
210#include <linux/memcontrol.h>
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700211#include <linux/fault-inject.h>
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700212#include <linux/kasan.h>
213#include <linux/kmemleak.h>
Thomas Garnier7c00fce2016-07-26 15:21:56 -0700214#include <linux/random.h>
Peter Zijlstrad92a8cf2017-03-03 10:13:38 +0100215#include <linux/sched/mm.h>
Joonsoo Kim07f361b2014-10-09 15:26:00 -0700216
Christoph Lameter97d06602012-07-06 15:25:11 -0500217/*
218 * State of the slab allocator.
219 *
220 * This is used to describe the states of the allocator during bootup.
221 * Allocators use this to gradually bootstrap themselves. Most allocators
222 * have the problem that the structures used for managing slab caches are
223 * allocated from slab caches themselves.
224 */
225enum slab_state {
226 DOWN, /* No slab functionality yet */
227 PARTIAL, /* SLUB: kmem_cache_node available */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000228 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
Christoph Lameter97d06602012-07-06 15:25:11 -0500229 UP, /* Slab caches usable but not all extras yet */
230 FULL /* Everything is working */
231};
232
233extern enum slab_state slab_state;
234
Christoph Lameter18004c52012-07-06 15:25:12 -0500235/* The slab cache mutex protects the management structures during changes */
236extern struct mutex slab_mutex;
Christoph Lameter9b030cb2012-09-05 00:20:33 +0000237
238/* The list of all slab caches on the system */
Christoph Lameter18004c52012-07-06 15:25:12 -0500239extern struct list_head slab_caches;
240
Christoph Lameter9b030cb2012-09-05 00:20:33 +0000241/* The slab cache that manages slab cache information */
242extern struct kmem_cache *kmem_cache;
243
Vlastimil Babkaaf3b5f82017-02-22 15:41:05 -0800244/* A table of kmalloc cache names and sizes */
245extern const struct kmalloc_info_struct {
Pengfei Licb5d9fb2019-11-30 17:49:21 -0800246 const char *name[NR_KMALLOC_TYPES];
Alexey Dobriyan55de8b92018-04-05 16:20:29 -0700247 unsigned int size;
Vlastimil Babkaaf3b5f82017-02-22 15:41:05 -0800248} kmalloc_info[];
249
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000250#ifndef CONFIG_SLOB
251/* Kmalloc array related functions */
Daniel Sanders34cc6992015-06-24 16:55:57 -0700252void setup_kmalloc_cache_index_table(void);
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800253void create_kmalloc_caches(slab_flags_t);
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000254
255/* Find the kmalloc slab corresponding for a certain size */
256struct kmem_cache *kmalloc_slab(size_t, gfp_t);
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000257#endif
258
Long Li44405092020-08-06 23:18:28 -0700259gfp_t kmalloc_fix_flags(gfp_t flags);
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000260
Christoph Lameter9b030cb2012-09-05 00:20:33 +0000261/* Functions provided by the slab allocators */
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800262int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
Christoph Lameter97d06602012-07-06 15:25:11 -0500263
Alexey Dobriyan55de8b92018-04-05 16:20:29 -0700264struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
265 slab_flags_t flags, unsigned int useroffset,
266 unsigned int usersize);
Christoph Lameter45530c42012-11-28 16:23:07 +0000267extern void create_boot_cache(struct kmem_cache *, const char *name,
Alexey Dobriyan361d5752018-04-05 16:20:33 -0700268 unsigned int size, slab_flags_t flags,
269 unsigned int useroffset, unsigned int usersize);
Christoph Lameter45530c42012-11-28 16:23:07 +0000270
Joonsoo Kim423c9292014-10-09 15:26:22 -0700271int slab_unmergeable(struct kmem_cache *s);
Alexey Dobriyanf4957d52018-04-05 16:20:37 -0700272struct kmem_cache *find_mergeable(unsigned size, unsigned align,
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800273 slab_flags_t flags, const char *name, void (*ctor)(void *));
Joonsoo Kim12220de2014-10-09 15:26:24 -0700274#ifndef CONFIG_SLOB
Glauber Costa2633d7a2012-12-18 14:22:34 -0800275struct kmem_cache *
Alexey Dobriyanf4957d52018-04-05 16:20:37 -0700276__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800277 slab_flags_t flags, void (*ctor)(void *));
Joonsoo Kim423c9292014-10-09 15:26:22 -0700278
Alexey Dobriyan0293d1f2018-04-05 16:21:24 -0700279slab_flags_t kmem_cache_flags(unsigned int object_size,
Nikolay Borisov37540002021-02-24 12:00:58 -0800280 slab_flags_t flags, const char *name);
Christoph Lametercbb79692012-09-05 00:18:32 +0000281#else
Glauber Costa2633d7a2012-12-18 14:22:34 -0800282static inline struct kmem_cache *
Alexey Dobriyanf4957d52018-04-05 16:20:37 -0700283__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800284 slab_flags_t flags, void (*ctor)(void *))
Christoph Lametercbb79692012-09-05 00:18:32 +0000285{ return NULL; }
Joonsoo Kim423c9292014-10-09 15:26:22 -0700286
Alexey Dobriyan0293d1f2018-04-05 16:21:24 -0700287static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
Nikolay Borisov37540002021-02-24 12:00:58 -0800288 slab_flags_t flags, const char *name)
Joonsoo Kim423c9292014-10-09 15:26:22 -0700289{
290 return flags;
291}
Christoph Lametercbb79692012-09-05 00:18:32 +0000292#endif
293
294
Glauber Costad8843922012-10-17 15:36:51 +0400295/* Legal flag mask for kmem_cache_create(), for various configurations */
Nicolas Boichat6d6ea1e2019-03-28 20:43:42 -0700296#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
297 SLAB_CACHE_DMA32 | SLAB_PANIC | \
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -0800298 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
Glauber Costad8843922012-10-17 15:36:51 +0400299
300#if defined(CONFIG_DEBUG_SLAB)
301#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
302#elif defined(CONFIG_SLUB_DEBUG)
303#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Laura Abbottbecfda62016-03-15 14:55:06 -0700304 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
Glauber Costad8843922012-10-17 15:36:51 +0400305#else
306#define SLAB_DEBUG_FLAGS (0)
307#endif
308
309#if defined(CONFIG_SLAB)
310#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
Vladimir Davydov230e9fc2016-01-14 15:18:15 -0800311 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
Levin, Alexander (Sasha Levin)75f296d2017-11-15 17:35:54 -0800312 SLAB_ACCOUNT)
Glauber Costad8843922012-10-17 15:36:51 +0400313#elif defined(CONFIG_SLUB)
314#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
Levin, Alexander (Sasha Levin)75f296d2017-11-15 17:35:54 -0800315 SLAB_TEMPORARY | SLAB_ACCOUNT)
Glauber Costad8843922012-10-17 15:36:51 +0400316#else
Rustam Kovhaev34dbc3aa2021-11-19 16:43:37 -0800317#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE)
Glauber Costad8843922012-10-17 15:36:51 +0400318#endif
319
Thomas Garniere70954f2016-12-12 16:41:38 -0800320/* Common flags available with current configuration */
Glauber Costad8843922012-10-17 15:36:51 +0400321#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
322
Thomas Garniere70954f2016-12-12 16:41:38 -0800323/* Common flags permitted for kmem_cache_create */
324#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
325 SLAB_RED_ZONE | \
326 SLAB_POISON | \
327 SLAB_STORE_USER | \
328 SLAB_TRACE | \
329 SLAB_CONSISTENCY_CHECKS | \
330 SLAB_MEM_SPREAD | \
331 SLAB_NOLEAKTRACE | \
332 SLAB_RECLAIM_ACCOUNT | \
333 SLAB_TEMPORARY | \
Thomas Garniere70954f2016-12-12 16:41:38 -0800334 SLAB_ACCOUNT)
335
Shakeel Buttf9e13c02018-04-05 16:21:57 -0700336bool __kmem_cache_empty(struct kmem_cache *);
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000337int __kmem_cache_shutdown(struct kmem_cache *);
Dmitry Safonov52b4b952016-02-17 13:11:37 -0800338void __kmem_cache_release(struct kmem_cache *);
Tejun Heoc9fc5862017-02-22 15:41:27 -0800339int __kmem_cache_shrink(struct kmem_cache *);
Christoph Lameter41a21282014-05-06 12:50:08 -0700340void slab_kmem_cache_release(struct kmem_cache *);
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000341
Glauber Costab7454ad2012-10-19 18:20:25 +0400342struct seq_file;
343struct file;
Glauber Costab7454ad2012-10-19 18:20:25 +0400344
Glauber Costa0d7561c2012-10-19 18:20:27 +0400345struct slabinfo {
346 unsigned long active_objs;
347 unsigned long num_objs;
348 unsigned long active_slabs;
349 unsigned long num_slabs;
350 unsigned long shared_avail;
351 unsigned int limit;
352 unsigned int batchcount;
353 unsigned int shared;
354 unsigned int objects_per_slab;
355 unsigned int cache_order;
356};
357
358void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
359void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
Glauber Costab7454ad2012-10-19 18:20:25 +0400360ssize_t slabinfo_write(struct file *file, const char __user *buffer,
361 size_t count, loff_t *ppos);
Glauber Costaba6c4962012-12-18 14:22:27 -0800362
Christoph Lameter484748f2015-09-04 15:45:34 -0700363/*
364 * Generic implementation of bulk operations
365 * These are useful for situations in which the allocator cannot
Jesper Dangaard Brouer9f706d62016-03-15 14:54:03 -0700366 * perform optimizations. In that case segments of the object listed
Christoph Lameter484748f2015-09-04 15:45:34 -0700367 * may be allocated or freed using these operations.
368 */
369void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -0800370int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
Christoph Lameter484748f2015-09-04 15:45:34 -0700371
Muchun Song1a984c42020-12-14 19:06:24 -0800372static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
Roman Gushchin6cea1d52019-07-11 20:56:16 -0700373{
374 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
Roman Gushchind42f3242020-08-06 23:20:39 -0700375 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
Roman Gushchin6cea1d52019-07-11 20:56:16 -0700376}
377
Vlastimil Babkae42f1742020-08-06 23:19:05 -0700378#ifdef CONFIG_SLUB_DEBUG
379#ifdef CONFIG_SLUB_DEBUG_ON
380DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
381#else
382DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
383#endif
384extern void print_tracking(struct kmem_cache *s, void *object);
Oliver Glitta1f9f78b2021-06-28 19:34:33 -0700385long validate_slab_cache(struct kmem_cache *s);
Marco Elver0d4a0622021-07-14 21:26:34 -0700386static inline bool __slub_debug_enabled(void)
387{
388 return static_branch_unlikely(&slub_debug_enabled);
389}
Vlastimil Babkae42f1742020-08-06 23:19:05 -0700390#else
391static inline void print_tracking(struct kmem_cache *s, void *object)
392{
393}
Marco Elver0d4a0622021-07-14 21:26:34 -0700394static inline bool __slub_debug_enabled(void)
395{
396 return false;
397}
Vlastimil Babkae42f1742020-08-06 23:19:05 -0700398#endif
399
400/*
401 * Returns true if any of the specified slub_debug flags is enabled for the
402 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
403 * the static key.
404 */
405static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
406{
Marco Elver0d4a0622021-07-14 21:26:34 -0700407 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
408 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
409 if (__slub_debug_enabled())
Vlastimil Babkae42f1742020-08-06 23:19:05 -0700410 return s->flags & flags;
Vlastimil Babkae42f1742020-08-06 23:19:05 -0700411 return false;
412}
413
Kirill Tkhai84c07d12018-08-17 15:47:25 -0700414#ifdef CONFIG_MEMCG_KMEM
Roman Gushchin10befea2020-08-06 23:21:27 -0700415int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
Roman Gushchin2e9bd482021-02-24 12:03:11 -0800416 gfp_t gfp, bool new_page);
Waiman Longfdbcb2a2021-06-28 19:37:19 -0700417void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
418 enum node_stat_item idx, int nr);
Roman Gushchin286e04b2020-08-06 23:20:52 -0700419
420static inline void memcg_free_page_obj_cgroups(struct page *page)
421{
Roman Gushchin270c6a72020-12-01 13:58:28 -0800422 kfree(page_objcgs(page));
Roman Gushchinbcfe06b2020-12-01 13:58:27 -0800423 page->memcg_data = 0;
Roman Gushchin286e04b2020-08-06 23:20:52 -0700424}
425
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700426static inline size_t obj_full_size(struct kmem_cache *s)
427{
428 /*
429 * For each accounted object there is an extra space which is used
430 * to store obj_cgroup membership. Charge it too.
431 */
432 return s->size + sizeof(struct obj_cgroup *);
433}
434
Roman Gushchinbecaba62020-12-05 22:14:45 -0800435/*
436 * Returns false if the allocation should fail.
437 */
438static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
439 struct obj_cgroup **objcgp,
440 size_t objects, gfp_t flags)
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700441{
Roman Gushchin98556092020-08-06 23:21:10 -0700442 struct obj_cgroup *objcg;
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700443
Roman Gushchinbecaba62020-12-05 22:14:45 -0800444 if (!memcg_kmem_enabled())
445 return true;
446
447 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
448 return true;
449
Roman Gushchin98556092020-08-06 23:21:10 -0700450 objcg = get_obj_cgroup_from_current();
451 if (!objcg)
Roman Gushchinbecaba62020-12-05 22:14:45 -0800452 return true;
Roman Gushchin98556092020-08-06 23:21:10 -0700453
454 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
455 obj_cgroup_put(objcg);
Roman Gushchinbecaba62020-12-05 22:14:45 -0800456 return false;
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700457 }
458
Roman Gushchinbecaba62020-12-05 22:14:45 -0800459 *objcgp = objcg;
460 return true;
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700461}
462
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700463static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
464 struct obj_cgroup *objcg,
Roman Gushchin10befea2020-08-06 23:21:27 -0700465 gfp_t flags, size_t size,
466 void **p)
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700467{
468 struct page *page;
469 unsigned long off;
470 size_t i;
471
Roman Gushchinbecaba62020-12-05 22:14:45 -0800472 if (!memcg_kmem_enabled() || !objcg)
Roman Gushchin10befea2020-08-06 23:21:27 -0700473 return;
474
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700475 for (i = 0; i < size; i++) {
476 if (likely(p[i])) {
477 page = virt_to_head_page(p[i]);
Roman Gushchin10befea2020-08-06 23:21:27 -0700478
Roman Gushchin270c6a72020-12-01 13:58:28 -0800479 if (!page_objcgs(page) &&
Roman Gushchin2e9bd482021-02-24 12:03:11 -0800480 memcg_alloc_page_obj_cgroups(page, s, flags,
481 false)) {
Roman Gushchin10befea2020-08-06 23:21:27 -0700482 obj_cgroup_uncharge(objcg, obj_full_size(s));
483 continue;
484 }
485
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700486 off = obj_to_index(s, page, p[i]);
487 obj_cgroup_get(objcg);
Roman Gushchin270c6a72020-12-01 13:58:28 -0800488 page_objcgs(page)[off] = objcg;
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700489 mod_objcg_state(objcg, page_pgdat(page),
490 cache_vmstat_idx(s), obj_full_size(s));
491 } else {
492 obj_cgroup_uncharge(objcg, obj_full_size(s));
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700493 }
494 }
495 obj_cgroup_put(objcg);
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700496}
497
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700498static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
499 void **p, int objects)
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700500{
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700501 struct kmem_cache *s;
Roman Gushchin270c6a72020-12-01 13:58:28 -0800502 struct obj_cgroup **objcgs;
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700503 struct obj_cgroup *objcg;
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700504 struct page *page;
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700505 unsigned int off;
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700506 int i;
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700507
Roman Gushchin10befea2020-08-06 23:21:27 -0700508 if (!memcg_kmem_enabled())
509 return;
510
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700511 for (i = 0; i < objects; i++) {
512 if (unlikely(!p[i]))
513 continue;
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700514
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700515 page = virt_to_head_page(p[i]);
Wang Hai121dffe2021-07-29 14:53:54 -0700516 objcgs = page_objcgs_check(page);
Roman Gushchin270c6a72020-12-01 13:58:28 -0800517 if (!objcgs)
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700518 continue;
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700519
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700520 if (!s_orig)
521 s = page->slab_cache;
522 else
523 s = s_orig;
Roman Gushchin10befea2020-08-06 23:21:27 -0700524
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700525 off = obj_to_index(s, page, p[i]);
Roman Gushchin270c6a72020-12-01 13:58:28 -0800526 objcg = objcgs[off];
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700527 if (!objcg)
528 continue;
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700529
Roman Gushchin270c6a72020-12-01 13:58:28 -0800530 objcgs[off] = NULL;
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700531 obj_cgroup_uncharge(objcg, obj_full_size(s));
532 mod_objcg_state(objcg, page_pgdat(page), cache_vmstat_idx(s),
533 -obj_full_size(s));
534 obj_cgroup_put(objcg);
535 }
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700536}
537
Kirill Tkhai84c07d12018-08-17 15:47:25 -0700538#else /* CONFIG_MEMCG_KMEM */
Roman Gushchin98556092020-08-06 23:21:10 -0700539static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
Roman Gushchin4d96ba32019-07-11 20:56:31 -0700540{
541 return NULL;
542}
543
Roman Gushchin286e04b2020-08-06 23:20:52 -0700544static inline int memcg_alloc_page_obj_cgroups(struct page *page,
Roman Gushchin2e9bd482021-02-24 12:03:11 -0800545 struct kmem_cache *s, gfp_t gfp,
546 bool new_page)
Roman Gushchin286e04b2020-08-06 23:20:52 -0700547{
548 return 0;
549}
550
551static inline void memcg_free_page_obj_cgroups(struct page *page)
552{
553}
554
Roman Gushchinbecaba62020-12-05 22:14:45 -0800555static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
556 struct obj_cgroup **objcgp,
557 size_t objects, gfp_t flags)
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700558{
Roman Gushchinbecaba62020-12-05 22:14:45 -0800559 return true;
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700560}
561
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700562static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
563 struct obj_cgroup *objcg,
Roman Gushchin10befea2020-08-06 23:21:27 -0700564 gfp_t flags, size_t size,
565 void **p)
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700566{
567}
568
Bharata B Raod1b2cf62020-10-13 16:53:09 -0700569static inline void memcg_slab_free_hook(struct kmem_cache *s,
570 void **p, int objects)
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700571{
572}
Kirill Tkhai84c07d12018-08-17 15:47:25 -0700573#endif /* CONFIG_MEMCG_KMEM */
Glauber Costab9ce5ef2012-12-18 14:22:46 -0800574
Kees Cooka64b5372019-07-11 20:53:26 -0700575static inline struct kmem_cache *virt_to_cache(const void *obj)
576{
Matthew Wilcox (Oracle)82c17752021-10-04 14:45:53 +0100577 struct slab *slab;
Kees Cooka64b5372019-07-11 20:53:26 -0700578
Matthew Wilcox (Oracle)82c17752021-10-04 14:45:53 +0100579 slab = virt_to_slab(obj);
580 if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n",
Kees Cooka64b5372019-07-11 20:53:26 -0700581 __func__))
582 return NULL;
Matthew Wilcox (Oracle)82c17752021-10-04 14:45:53 +0100583 return slab->slab_cache;
Kees Cooka64b5372019-07-11 20:53:26 -0700584}
585
Matthew Wilcox (Oracle)b9186532021-10-04 14:45:52 +0100586static __always_inline void account_slab(struct slab *slab, int order,
587 struct kmem_cache *s, gfp_t gfp)
Roman Gushchin6cea1d52019-07-11 20:56:16 -0700588{
Roman Gushchin2e9bd482021-02-24 12:03:11 -0800589 if (memcg_kmem_enabled() && (s->flags & SLAB_ACCOUNT))
Matthew Wilcox (Oracle)b9186532021-10-04 14:45:52 +0100590 memcg_alloc_page_obj_cgroups(slab_page(slab), s, gfp, true);
Roman Gushchin2e9bd482021-02-24 12:03:11 -0800591
Matthew Wilcox (Oracle)b9186532021-10-04 14:45:52 +0100592 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700593 PAGE_SIZE << order);
Roman Gushchin6cea1d52019-07-11 20:56:16 -0700594}
595
Matthew Wilcox (Oracle)b9186532021-10-04 14:45:52 +0100596static __always_inline void unaccount_slab(struct slab *slab, int order,
597 struct kmem_cache *s)
Roman Gushchin6cea1d52019-07-11 20:56:16 -0700598{
Roman Gushchin10befea2020-08-06 23:21:27 -0700599 if (memcg_kmem_enabled())
Matthew Wilcox (Oracle)b9186532021-10-04 14:45:52 +0100600 memcg_free_page_obj_cgroups(slab_page(slab));
Roman Gushchin98556092020-08-06 23:21:10 -0700601
Matthew Wilcox (Oracle)b9186532021-10-04 14:45:52 +0100602 mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
Roman Gushchinf2fe7b02020-08-06 23:20:59 -0700603 -(PAGE_SIZE << order));
Roman Gushchin6cea1d52019-07-11 20:56:16 -0700604}
605
Vlastimil Babkae42f1742020-08-06 23:19:05 -0700606static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
607{
608 struct kmem_cache *cachep;
609
610 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
Vlastimil Babkae42f1742020-08-06 23:19:05 -0700611 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
612 return s;
613
614 cachep = virt_to_cache(x);
Roman Gushchin10befea2020-08-06 23:21:27 -0700615 if (WARN(cachep && cachep != s,
Vlastimil Babkae42f1742020-08-06 23:19:05 -0700616 "%s: Wrong slab cache. %s but object is from %s\n",
617 __func__, s->name, cachep->name))
618 print_tracking(cachep, x);
619 return cachep;
620}
621
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700622static inline size_t slab_ksize(const struct kmem_cache *s)
623{
624#ifndef CONFIG_SLUB
625 return s->object_size;
626
627#else /* CONFIG_SLUB */
628# ifdef CONFIG_SLUB_DEBUG
629 /*
630 * Debugging requires use of the padding between object
631 * and whatever may come after it.
632 */
633 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
634 return s->object_size;
635# endif
Alexander Potapenko80a92012016-07-28 15:49:07 -0700636 if (s->flags & SLAB_KASAN)
637 return s->object_size;
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700638 /*
639 * If we have the need to store the freelist pointer
640 * back there or track user information then we can
641 * only use the space before that information.
642 */
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -0800643 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700644 return s->inuse;
645 /*
646 * Else we can use all the padding etc for the allocation
647 */
648 return s->size;
649#endif
650}
651
652static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700653 struct obj_cgroup **objcgp,
654 size_t size, gfp_t flags)
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700655{
656 flags &= gfp_allowed_mask;
Peter Zijlstrad92a8cf2017-03-03 10:13:38 +0100657
Daniel Vetter95d6c702020-12-14 19:08:34 -0800658 might_alloc(flags);
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700659
Jesper Dangaard Brouerfab99632016-03-15 14:53:38 -0700660 if (should_failslab(s, flags))
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700661 return NULL;
662
Roman Gushchinbecaba62020-12-05 22:14:45 -0800663 if (!memcg_slab_pre_alloc_hook(s, objcgp, size, flags))
664 return NULL;
Vladimir Davydov45264772016-07-26 15:24:21 -0700665
666 return s;
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700667}
668
Roman Gushchin964d4bd2020-08-06 23:20:56 -0700669static inline void slab_post_alloc_hook(struct kmem_cache *s,
Andrey Konovalovda844b72021-04-29 23:00:06 -0700670 struct obj_cgroup *objcg, gfp_t flags,
671 size_t size, void **p, bool init)
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700672{
673 size_t i;
674
675 flags &= gfp_allowed_mask;
Andrey Konovalovda844b72021-04-29 23:00:06 -0700676
677 /*
678 * As memory initialization might be integrated into KASAN,
679 * kasan_slab_alloc and initialization memset must be
680 * kept together to avoid discrepancies in behavior.
681 *
682 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
683 */
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700684 for (i = 0; i < size; i++) {
Andrey Konovalovda844b72021-04-29 23:00:06 -0700685 p[i] = kasan_slab_alloc(s, p[i], flags, init);
686 if (p[i] && init && !kasan_has_integrated_init())
687 memset(p[i], 0, s->object_size);
Andrey Konovalov53128242019-02-20 22:19:11 -0800688 kmemleak_alloc_recursive(p[i], s->object_size, 1,
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700689 s->flags, flags);
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700690 }
Vladimir Davydov45264772016-07-26 15:24:21 -0700691
Roman Gushchinbecaba62020-12-05 22:14:45 -0800692 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
Jesper Dangaard Brouer11c7aec2016-03-15 14:53:35 -0700693}
694
Christoph Lameter44c53562014-08-06 16:04:07 -0700695#ifndef CONFIG_SLOB
Christoph Lameterca349562013-01-10 19:14:19 +0000696/*
697 * The slab lists for all objects.
698 */
699struct kmem_cache_node {
700 spinlock_t list_lock;
701
702#ifdef CONFIG_SLAB
703 struct list_head slabs_partial; /* partial list first, better asm code */
704 struct list_head slabs_full;
705 struct list_head slabs_free;
David Rientjesbf00bd32016-12-12 16:41:44 -0800706 unsigned long total_slabs; /* length of all slab lists */
707 unsigned long free_slabs; /* length of free slab list only */
Christoph Lameterca349562013-01-10 19:14:19 +0000708 unsigned long free_objects;
709 unsigned int free_limit;
710 unsigned int colour_next; /* Per-node cache coloring */
711 struct array_cache *shared; /* shared per node */
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700712 struct alien_cache **alien; /* on other nodes */
Christoph Lameterca349562013-01-10 19:14:19 +0000713 unsigned long next_reap; /* updated without locking */
714 int free_touched; /* updated without locking */
715#endif
716
717#ifdef CONFIG_SLUB
718 unsigned long nr_partial;
719 struct list_head partial;
720#ifdef CONFIG_SLUB_DEBUG
721 atomic_long_t nr_slabs;
722 atomic_long_t total_objects;
723 struct list_head full;
724#endif
725#endif
726
727};
Wanpeng Lie25839f2013-07-04 08:33:23 +0800728
Christoph Lameter44c53562014-08-06 16:04:07 -0700729static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
730{
731 return s->node[node];
732}
733
734/*
735 * Iterator over all nodes. The body will be executed for each node that has
736 * a kmem_cache_node structure allocated (which is true for all online nodes)
737 */
738#define for_each_kmem_cache_node(__s, __node, __n) \
Mikulas Patocka91635822014-10-09 15:26:20 -0700739 for (__node = 0; __node < nr_node_ids; __node++) \
740 if ((__n = get_node(__s, __node)))
Christoph Lameter44c53562014-08-06 16:04:07 -0700741
742#endif
743
Vladimir Davydov1df3b262014-12-10 15:42:16 -0800744void *slab_start(struct seq_file *m, loff_t *pos);
Wanpeng Li276a2432013-07-08 08:08:28 +0800745void *slab_next(struct seq_file *m, void *p, loff_t *pos);
746void slab_stop(struct seq_file *m, void *p);
Vladimir Davydovb0475012014-12-10 15:44:19 -0800747int memcg_slab_show(struct seq_file *m, void *p);
Andrey Ryabinin5240ab42014-08-06 16:04:14 -0700748
Yang Shi852d8be2017-11-15 17:32:07 -0800749#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
750void dump_unreclaimable_slab(void);
751#else
752static inline void dump_unreclaimable_slab(void)
753{
754}
755#endif
756
Alexander Potapenko55834c52016-05-20 16:59:11 -0700757void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
758
Thomas Garnier7c00fce2016-07-26 15:21:56 -0700759#ifdef CONFIG_SLAB_FREELIST_RANDOM
760int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
761 gfp_t gfp);
762void cache_random_seq_destroy(struct kmem_cache *cachep);
763#else
764static inline int cache_random_seq_create(struct kmem_cache *cachep,
765 unsigned int count, gfp_t gfp)
766{
767 return 0;
768}
769static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
770#endif /* CONFIG_SLAB_FREELIST_RANDOM */
771
Alexander Potapenko64713842019-07-11 20:59:19 -0700772static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
773{
Kees Cook51cba1e2021-04-01 16:23:43 -0700774 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
775 &init_on_alloc)) {
Alexander Potapenko64713842019-07-11 20:59:19 -0700776 if (c->ctor)
777 return false;
778 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
779 return flags & __GFP_ZERO;
780 return true;
781 }
782 return flags & __GFP_ZERO;
783}
784
785static inline bool slab_want_init_on_free(struct kmem_cache *c)
786{
Kees Cook51cba1e2021-04-01 16:23:43 -0700787 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
788 &init_on_free))
Alexander Potapenko64713842019-07-11 20:59:19 -0700789 return !(c->ctor ||
790 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
791 return false;
792}
793
Faiyaz Mohammed64dd6842021-06-28 19:34:55 -0700794#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
795void debugfs_slab_release(struct kmem_cache *);
796#else
797static inline void debugfs_slab_release(struct kmem_cache *s) { }
798#endif
799
Paul E. McKenney5bb1bb32021-01-07 13:46:11 -0800800#ifdef CONFIG_PRINTK
Paul E. McKenney8e7f37f2020-12-07 17:41:02 -0800801#define KS_ADDRS_COUNT 16
802struct kmem_obj_info {
803 void *kp_ptr;
804 struct page *kp_page;
805 void *kp_objp;
806 unsigned long kp_data_offset;
807 struct kmem_cache *kp_slab_cache;
808 void *kp_ret;
809 void *kp_stack[KS_ADDRS_COUNT];
Maninder Singhe548eaa2021-03-16 16:07:11 +0530810 void *kp_free_stack[KS_ADDRS_COUNT];
Paul E. McKenney8e7f37f2020-12-07 17:41:02 -0800811};
812void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct page *page);
Paul E. McKenney5bb1bb32021-01-07 13:46:11 -0800813#endif
Paul E. McKenney8e7f37f2020-12-07 17:41:02 -0800814
Andrey Ryabinin5240ab42014-08-06 16:04:14 -0700815#endif /* MM_SLAB_H */