Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 1 | /* |
| 2 | * Slab allocator functions that are independent of the allocator strategy |
| 3 | * |
| 4 | * (C) 2012 Christoph Lameter <cl@linux.com> |
| 5 | */ |
| 6 | #include <linux/slab.h> |
| 7 | |
| 8 | #include <linux/mm.h> |
| 9 | #include <linux/poison.h> |
| 10 | #include <linux/interrupt.h> |
| 11 | #include <linux/memory.h> |
| 12 | #include <linux/compiler.h> |
| 13 | #include <linux/module.h> |
Christoph Lameter | 20cea96 | 2012-07-06 15:25:13 -0500 | [diff] [blame] | 14 | #include <linux/cpu.h> |
| 15 | #include <linux/uaccess.h> |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 16 | #include <linux/seq_file.h> |
| 17 | #include <linux/proc_fs.h> |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 18 | #include <asm/cacheflush.h> |
| 19 | #include <asm/tlbflush.h> |
| 20 | #include <asm/page.h> |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 21 | #include <linux/memcontrol.h> |
Andrey Ryabinin | 928cec9 | 2014-08-06 16:04:44 -0700 | [diff] [blame] | 22 | |
| 23 | #define CREATE_TRACE_POINTS |
Christoph Lameter | f1b6eb6 | 2013-09-04 16:35:34 +0000 | [diff] [blame] | 24 | #include <trace/events/kmem.h> |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 25 | |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 26 | #include "slab.h" |
| 27 | |
| 28 | enum slab_state slab_state; |
Christoph Lameter | 18004c5 | 2012-07-06 15:25:12 -0500 | [diff] [blame] | 29 | LIST_HEAD(slab_caches); |
| 30 | DEFINE_MUTEX(slab_mutex); |
Christoph Lameter | 9b030cb | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 31 | struct kmem_cache *kmem_cache; |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 32 | |
Joonsoo Kim | 07f361b | 2014-10-09 15:26:00 -0700 | [diff] [blame] | 33 | /* |
Joonsoo Kim | 423c929 | 2014-10-09 15:26:22 -0700 | [diff] [blame] | 34 | * Set of flags that will prevent slab merging |
| 35 | */ |
| 36 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
| 37 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
Alexander Potapenko | 7ed2f9e | 2016-03-25 14:21:59 -0700 | [diff] [blame^] | 38 | SLAB_FAILSLAB | SLAB_KASAN) |
Joonsoo Kim | 423c929 | 2014-10-09 15:26:22 -0700 | [diff] [blame] | 39 | |
Vladimir Davydov | 230e9fc | 2016-01-14 15:18:15 -0800 | [diff] [blame] | 40 | #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ |
| 41 | SLAB_NOTRACK | SLAB_ACCOUNT) |
Joonsoo Kim | 423c929 | 2014-10-09 15:26:22 -0700 | [diff] [blame] | 42 | |
| 43 | /* |
| 44 | * Merge control. If this is set then no merging of slab caches will occur. |
| 45 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
| 46 | */ |
| 47 | static int slab_nomerge; |
| 48 | |
| 49 | static int __init setup_slab_nomerge(char *str) |
| 50 | { |
| 51 | slab_nomerge = 1; |
| 52 | return 1; |
| 53 | } |
| 54 | |
| 55 | #ifdef CONFIG_SLUB |
| 56 | __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); |
| 57 | #endif |
| 58 | |
| 59 | __setup("slab_nomerge", setup_slab_nomerge); |
| 60 | |
| 61 | /* |
Joonsoo Kim | 07f361b | 2014-10-09 15:26:00 -0700 | [diff] [blame] | 62 | * Determine the size of a slab object |
| 63 | */ |
| 64 | unsigned int kmem_cache_size(struct kmem_cache *s) |
| 65 | { |
| 66 | return s->object_size; |
| 67 | } |
| 68 | EXPORT_SYMBOL(kmem_cache_size); |
| 69 | |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 70 | #ifdef CONFIG_DEBUG_VM |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 71 | static int kmem_cache_sanity_check(const char *name, size_t size) |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 72 | { |
| 73 | struct kmem_cache *s = NULL; |
| 74 | |
| 75 | if (!name || in_interrupt() || size < sizeof(void *) || |
| 76 | size > KMALLOC_MAX_SIZE) { |
| 77 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
| 78 | return -EINVAL; |
| 79 | } |
| 80 | |
| 81 | list_for_each_entry(s, &slab_caches, list) { |
| 82 | char tmp; |
| 83 | int res; |
| 84 | |
| 85 | /* |
| 86 | * This happens when the module gets unloaded and doesn't |
| 87 | * destroy its slab cache and no-one else reuses the vmalloc |
| 88 | * area of the module. Print a warning. |
| 89 | */ |
| 90 | res = probe_kernel_address(s->name, tmp); |
| 91 | if (res) { |
| 92 | pr_err("Slab cache with size %d has lost its name\n", |
| 93 | s->object_size); |
| 94 | continue; |
| 95 | } |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 96 | } |
| 97 | |
| 98 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
| 99 | return 0; |
| 100 | } |
| 101 | #else |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 102 | static inline int kmem_cache_sanity_check(const char *name, size_t size) |
Shuah Khan | 77be4b1 | 2012-08-16 00:09:46 -0700 | [diff] [blame] | 103 | { |
| 104 | return 0; |
| 105 | } |
| 106 | #endif |
| 107 | |
Christoph Lameter | 484748f | 2015-09-04 15:45:34 -0700 | [diff] [blame] | 108 | void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) |
| 109 | { |
| 110 | size_t i; |
| 111 | |
Jesper Dangaard Brouer | ca25719 | 2016-03-15 14:54:00 -0700 | [diff] [blame] | 112 | for (i = 0; i < nr; i++) { |
| 113 | if (s) |
| 114 | kmem_cache_free(s, p[i]); |
| 115 | else |
| 116 | kfree(p[i]); |
| 117 | } |
Christoph Lameter | 484748f | 2015-09-04 15:45:34 -0700 | [diff] [blame] | 118 | } |
| 119 | |
Jesper Dangaard Brouer | 865762a | 2015-11-20 15:57:58 -0800 | [diff] [blame] | 120 | int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, |
Christoph Lameter | 484748f | 2015-09-04 15:45:34 -0700 | [diff] [blame] | 121 | void **p) |
| 122 | { |
| 123 | size_t i; |
| 124 | |
| 125 | for (i = 0; i < nr; i++) { |
| 126 | void *x = p[i] = kmem_cache_alloc(s, flags); |
| 127 | if (!x) { |
| 128 | __kmem_cache_free_bulk(s, i, p); |
Jesper Dangaard Brouer | 865762a | 2015-11-20 15:57:58 -0800 | [diff] [blame] | 129 | return 0; |
Christoph Lameter | 484748f | 2015-09-04 15:45:34 -0700 | [diff] [blame] | 130 | } |
| 131 | } |
Jesper Dangaard Brouer | 865762a | 2015-11-20 15:57:58 -0800 | [diff] [blame] | 132 | return i; |
Christoph Lameter | 484748f | 2015-09-04 15:45:34 -0700 | [diff] [blame] | 133 | } |
| 134 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 135 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 136 | void slab_init_memcg_params(struct kmem_cache *s) |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 137 | { |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 138 | s->memcg_params.is_root_cache = true; |
Vladimir Davydov | 426589f | 2015-02-12 14:59:23 -0800 | [diff] [blame] | 139 | INIT_LIST_HEAD(&s->memcg_params.list); |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 140 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL); |
| 141 | } |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 142 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 143 | static int init_memcg_params(struct kmem_cache *s, |
| 144 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) |
| 145 | { |
| 146 | struct memcg_cache_array *arr; |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 147 | |
| 148 | if (memcg) { |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 149 | s->memcg_params.is_root_cache = false; |
| 150 | s->memcg_params.memcg = memcg; |
| 151 | s->memcg_params.root_cache = root_cache; |
| 152 | return 0; |
| 153 | } |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 154 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 155 | slab_init_memcg_params(s); |
| 156 | |
| 157 | if (!memcg_nr_cache_ids) |
| 158 | return 0; |
| 159 | |
| 160 | arr = kzalloc(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); |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 167 | return 0; |
| 168 | } |
| 169 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 170 | static void destroy_memcg_params(struct kmem_cache *s) |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 171 | { |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 172 | if (is_root_cache(s)) |
| 173 | kfree(rcu_access_pointer(s->memcg_params.memcg_caches)); |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 174 | } |
| 175 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 176 | static int update_memcg_params(struct kmem_cache *s, int new_array_size) |
Vladimir Davydov | 6f817f4 | 2014-10-09 15:28:47 -0700 | [diff] [blame] | 177 | { |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 178 | struct memcg_cache_array *old, *new; |
Vladimir Davydov | 6f817f4 | 2014-10-09 15:28:47 -0700 | [diff] [blame] | 179 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 180 | if (!is_root_cache(s)) |
| 181 | return 0; |
Vladimir Davydov | 6f817f4 | 2014-10-09 15:28:47 -0700 | [diff] [blame] | 182 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 183 | new = kzalloc(sizeof(struct memcg_cache_array) + |
| 184 | new_array_size * sizeof(void *), GFP_KERNEL); |
| 185 | if (!new) |
Vladimir Davydov | 6f817f4 | 2014-10-09 15:28:47 -0700 | [diff] [blame] | 186 | return -ENOMEM; |
| 187 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 188 | old = rcu_dereference_protected(s->memcg_params.memcg_caches, |
| 189 | lockdep_is_held(&slab_mutex)); |
| 190 | if (old) |
| 191 | memcpy(new->entries, old->entries, |
| 192 | memcg_nr_cache_ids * sizeof(void *)); |
Vladimir Davydov | 6f817f4 | 2014-10-09 15:28:47 -0700 | [diff] [blame] | 193 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 194 | rcu_assign_pointer(s->memcg_params.memcg_caches, new); |
| 195 | if (old) |
| 196 | kfree_rcu(old, rcu); |
Vladimir Davydov | 6f817f4 | 2014-10-09 15:28:47 -0700 | [diff] [blame] | 197 | return 0; |
| 198 | } |
| 199 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 200 | int memcg_update_all_caches(int num_memcgs) |
| 201 | { |
| 202 | struct kmem_cache *s; |
| 203 | int ret = 0; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 204 | |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 205 | mutex_lock(&slab_mutex); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 206 | list_for_each_entry(s, &slab_caches, list) { |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 207 | ret = update_memcg_params(s, num_memcgs); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 208 | /* |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 209 | * Instead of freeing the memory, we'll just leave the caches |
| 210 | * up to this point in an updated state. |
| 211 | */ |
| 212 | if (ret) |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 213 | break; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 214 | } |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 215 | mutex_unlock(&slab_mutex); |
| 216 | return ret; |
| 217 | } |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 218 | #else |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 219 | static inline int init_memcg_params(struct kmem_cache *s, |
| 220 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 221 | { |
| 222 | return 0; |
| 223 | } |
| 224 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 225 | static inline void destroy_memcg_params(struct kmem_cache *s) |
Vladimir Davydov | 33a690c | 2014-10-09 15:28:43 -0700 | [diff] [blame] | 226 | { |
| 227 | } |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 228 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 229 | |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 230 | /* |
Joonsoo Kim | 423c929 | 2014-10-09 15:26:22 -0700 | [diff] [blame] | 231 | * Find a mergeable slab cache |
| 232 | */ |
| 233 | int slab_unmergeable(struct kmem_cache *s) |
| 234 | { |
| 235 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) |
| 236 | return 1; |
| 237 | |
| 238 | if (!is_root_cache(s)) |
| 239 | return 1; |
| 240 | |
| 241 | if (s->ctor) |
| 242 | return 1; |
| 243 | |
| 244 | /* |
| 245 | * We may have set a slab to be unmergeable during bootstrap. |
| 246 | */ |
| 247 | if (s->refcount < 0) |
| 248 | return 1; |
| 249 | |
| 250 | return 0; |
| 251 | } |
| 252 | |
| 253 | struct kmem_cache *find_mergeable(size_t size, size_t align, |
| 254 | unsigned long flags, const char *name, void (*ctor)(void *)) |
| 255 | { |
| 256 | struct kmem_cache *s; |
| 257 | |
| 258 | if (slab_nomerge || (flags & SLAB_NEVER_MERGE)) |
| 259 | return NULL; |
| 260 | |
| 261 | if (ctor) |
| 262 | return NULL; |
| 263 | |
| 264 | size = ALIGN(size, sizeof(void *)); |
| 265 | align = calculate_alignment(flags, align, size); |
| 266 | size = ALIGN(size, align); |
| 267 | flags = kmem_cache_flags(size, flags, name, NULL); |
| 268 | |
Joonsoo Kim | 5436205 | 2014-12-10 15:42:18 -0800 | [diff] [blame] | 269 | list_for_each_entry_reverse(s, &slab_caches, list) { |
Joonsoo Kim | 423c929 | 2014-10-09 15:26:22 -0700 | [diff] [blame] | 270 | if (slab_unmergeable(s)) |
| 271 | continue; |
| 272 | |
| 273 | if (size > s->size) |
| 274 | continue; |
| 275 | |
| 276 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) |
| 277 | continue; |
| 278 | /* |
| 279 | * Check if alignment is compatible. |
| 280 | * Courtesy of Adrian Drzewiecki |
| 281 | */ |
| 282 | if ((s->size & ~(align - 1)) != s->size) |
| 283 | continue; |
| 284 | |
| 285 | if (s->size - size >= sizeof(void *)) |
| 286 | continue; |
| 287 | |
Joonsoo Kim | 95069ac8 | 2014-11-13 15:19:25 -0800 | [diff] [blame] | 288 | if (IS_ENABLED(CONFIG_SLAB) && align && |
| 289 | (align > s->align || s->align % align)) |
| 290 | continue; |
| 291 | |
Joonsoo Kim | 423c929 | 2014-10-09 15:26:22 -0700 | [diff] [blame] | 292 | return s; |
| 293 | } |
| 294 | return NULL; |
| 295 | } |
| 296 | |
| 297 | /* |
Christoph Lameter | 4590685 | 2012-11-28 16:23:16 +0000 | [diff] [blame] | 298 | * Figure out what the alignment of the objects will be given a set of |
| 299 | * flags, a user specified alignment and the size of the objects. |
| 300 | */ |
| 301 | unsigned long calculate_alignment(unsigned long flags, |
| 302 | unsigned long align, unsigned long size) |
| 303 | { |
| 304 | /* |
| 305 | * If the user wants hardware cache aligned objects then follow that |
| 306 | * suggestion if the object is sufficiently large. |
| 307 | * |
| 308 | * The hardware cache alignment cannot override the specified |
| 309 | * alignment though. If that is greater then use it. |
| 310 | */ |
| 311 | if (flags & SLAB_HWCACHE_ALIGN) { |
| 312 | unsigned long ralign = cache_line_size(); |
| 313 | while (size <= ralign / 2) |
| 314 | ralign /= 2; |
| 315 | align = max(align, ralign); |
| 316 | } |
| 317 | |
| 318 | if (align < ARCH_SLAB_MINALIGN) |
| 319 | align = ARCH_SLAB_MINALIGN; |
| 320 | |
| 321 | return ALIGN(align, sizeof(void *)); |
| 322 | } |
| 323 | |
Vladimir Davydov | c9a77a7 | 2015-11-05 18:45:08 -0800 | [diff] [blame] | 324 | static struct kmem_cache *create_cache(const char *name, |
| 325 | size_t object_size, size_t size, size_t align, |
| 326 | unsigned long flags, void (*ctor)(void *), |
| 327 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 328 | { |
| 329 | struct kmem_cache *s; |
| 330 | int err; |
| 331 | |
| 332 | err = -ENOMEM; |
| 333 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); |
| 334 | if (!s) |
| 335 | goto out; |
| 336 | |
| 337 | s->name = name; |
| 338 | s->object_size = object_size; |
| 339 | s->size = size; |
| 340 | s->align = align; |
| 341 | s->ctor = ctor; |
| 342 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 343 | err = init_memcg_params(s, memcg, root_cache); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 344 | if (err) |
| 345 | goto out_free_cache; |
| 346 | |
| 347 | err = __kmem_cache_create(s, flags); |
| 348 | if (err) |
| 349 | goto out_free_cache; |
| 350 | |
| 351 | s->refcount = 1; |
| 352 | list_add(&s->list, &slab_caches); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 353 | out: |
| 354 | if (err) |
| 355 | return ERR_PTR(err); |
| 356 | return s; |
| 357 | |
| 358 | out_free_cache: |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 359 | destroy_memcg_params(s); |
Vaishali Thakkar | 7c4da06 | 2015-02-10 14:09:40 -0800 | [diff] [blame] | 360 | kmem_cache_free(kmem_cache, s); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 361 | goto out; |
| 362 | } |
Christoph Lameter | 4590685 | 2012-11-28 16:23:16 +0000 | [diff] [blame] | 363 | |
| 364 | /* |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 365 | * kmem_cache_create - Create a cache. |
| 366 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
| 367 | * @size: The size of objects to be created in this cache. |
| 368 | * @align: The required alignment for the objects. |
| 369 | * @flags: SLAB flags |
| 370 | * @ctor: A constructor for the objects. |
| 371 | * |
| 372 | * Returns a ptr to the cache on success, NULL on failure. |
| 373 | * Cannot be called within a interrupt, but can be interrupted. |
| 374 | * The @ctor is run when new pages are allocated by the cache. |
| 375 | * |
| 376 | * The flags are |
| 377 | * |
| 378 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) |
| 379 | * to catch references to uninitialised memory. |
| 380 | * |
| 381 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check |
| 382 | * for buffer overruns. |
| 383 | * |
| 384 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
| 385 | * cacheline. This can be beneficial if you're counting cycles as closely |
| 386 | * as davem. |
| 387 | */ |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 388 | struct kmem_cache * |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 389 | kmem_cache_create(const char *name, size_t size, size_t align, |
| 390 | unsigned long flags, void (*ctor)(void *)) |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 391 | { |
Alexandru Moise | 40911a7 | 2015-11-05 18:45:43 -0800 | [diff] [blame] | 392 | struct kmem_cache *s = NULL; |
Andrzej Hajda | 3dec16e | 2015-02-13 14:36:38 -0800 | [diff] [blame] | 393 | const char *cache_name; |
Vladimir Davydov | 3965fc3 | 2014-01-23 15:52:55 -0800 | [diff] [blame] | 394 | int err; |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 395 | |
Pekka Enberg | b920536 | 2012-08-16 10:12:18 +0300 | [diff] [blame] | 396 | get_online_cpus(); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 397 | get_online_mems(); |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 398 | memcg_get_cache_ids(); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 399 | |
Pekka Enberg | b920536 | 2012-08-16 10:12:18 +0300 | [diff] [blame] | 400 | mutex_lock(&slab_mutex); |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 401 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 402 | err = kmem_cache_sanity_check(name, size); |
Andrew Morton | 3aa24f5 | 2014-10-09 15:25:58 -0700 | [diff] [blame] | 403 | if (err) { |
Vladimir Davydov | 3965fc3 | 2014-01-23 15:52:55 -0800 | [diff] [blame] | 404 | goto out_unlock; |
Andrew Morton | 3aa24f5 | 2014-10-09 15:25:58 -0700 | [diff] [blame] | 405 | } |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 406 | |
Glauber Costa | d884392 | 2012-10-17 15:36:51 +0400 | [diff] [blame] | 407 | /* |
| 408 | * Some allocators will constraint the set of valid flags to a subset |
| 409 | * of all flags. We expect them to define CACHE_CREATE_MASK in this |
| 410 | * case, and we'll just provide them with a sanitized version of the |
| 411 | * passed flags. |
| 412 | */ |
| 413 | flags &= CACHE_CREATE_MASK; |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 414 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 415 | s = __kmem_cache_alias(name, size, align, flags, ctor); |
| 416 | if (s) |
Vladimir Davydov | 3965fc3 | 2014-01-23 15:52:55 -0800 | [diff] [blame] | 417 | goto out_unlock; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 418 | |
Andrzej Hajda | 3dec16e | 2015-02-13 14:36:38 -0800 | [diff] [blame] | 419 | cache_name = kstrdup_const(name, GFP_KERNEL); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 420 | if (!cache_name) { |
| 421 | err = -ENOMEM; |
| 422 | goto out_unlock; |
| 423 | } |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 424 | |
Vladimir Davydov | c9a77a7 | 2015-11-05 18:45:08 -0800 | [diff] [blame] | 425 | s = create_cache(cache_name, size, size, |
| 426 | calculate_alignment(flags, align, size), |
| 427 | flags, ctor, NULL, NULL); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 428 | if (IS_ERR(s)) { |
| 429 | err = PTR_ERR(s); |
Andrzej Hajda | 3dec16e | 2015-02-13 14:36:38 -0800 | [diff] [blame] | 430 | kfree_const(cache_name); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 431 | } |
Vladimir Davydov | 3965fc3 | 2014-01-23 15:52:55 -0800 | [diff] [blame] | 432 | |
| 433 | out_unlock: |
Christoph Lameter | 20cea96 | 2012-07-06 15:25:13 -0500 | [diff] [blame] | 434 | mutex_unlock(&slab_mutex); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 435 | |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 436 | memcg_put_cache_ids(); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 437 | put_online_mems(); |
Christoph Lameter | 20cea96 | 2012-07-06 15:25:13 -0500 | [diff] [blame] | 438 | put_online_cpus(); |
| 439 | |
Dave Jones | ba3253c7 | 2014-01-29 14:05:48 -0800 | [diff] [blame] | 440 | if (err) { |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 441 | if (flags & SLAB_PANIC) |
| 442 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", |
| 443 | name, err); |
| 444 | else { |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 445 | pr_warn("kmem_cache_create(%s) failed with error %d\n", |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 446 | name, err); |
| 447 | dump_stack(); |
| 448 | } |
Christoph Lameter | 686d550 | 2012-09-05 00:20:33 +0000 | [diff] [blame] | 449 | return NULL; |
| 450 | } |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 451 | return s; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 452 | } |
Christoph Lameter | 039363f | 2012-07-06 15:25:10 -0500 | [diff] [blame] | 453 | EXPORT_SYMBOL(kmem_cache_create); |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 454 | |
Vladimir Davydov | c9a77a7 | 2015-11-05 18:45:08 -0800 | [diff] [blame] | 455 | static int shutdown_cache(struct kmem_cache *s, |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 456 | struct list_head *release, bool *need_rcu_barrier) |
| 457 | { |
Vladimir Davydov | cd918c5 | 2015-11-05 18:45:14 -0800 | [diff] [blame] | 458 | if (__kmem_cache_shutdown(s) != 0) |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 459 | return -EBUSY; |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 460 | |
| 461 | if (s->flags & SLAB_DESTROY_BY_RCU) |
| 462 | *need_rcu_barrier = true; |
| 463 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 464 | list_move(&s->list, release); |
| 465 | return 0; |
| 466 | } |
| 467 | |
Vladimir Davydov | c9a77a7 | 2015-11-05 18:45:08 -0800 | [diff] [blame] | 468 | static void release_caches(struct list_head *release, bool need_rcu_barrier) |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 469 | { |
| 470 | struct kmem_cache *s, *s2; |
| 471 | |
| 472 | if (need_rcu_barrier) |
| 473 | rcu_barrier(); |
| 474 | |
| 475 | list_for_each_entry_safe(s, s2, release, list) { |
| 476 | #ifdef SLAB_SUPPORTS_SYSFS |
| 477 | sysfs_slab_remove(s); |
| 478 | #else |
| 479 | slab_kmem_cache_release(s); |
| 480 | #endif |
| 481 | } |
| 482 | } |
| 483 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 484 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 485 | /* |
Vladimir Davydov | 776ed0f | 2014-06-04 16:10:02 -0700 | [diff] [blame] | 486 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 487 | * @memcg: The memory cgroup the new cache is for. |
| 488 | * @root_cache: The parent of the new cache. |
| 489 | * |
| 490 | * This function attempts to create a kmem cache that will serve allocation |
| 491 | * requests going from @memcg to @root_cache. The new cache inherits properties |
| 492 | * from its parent. |
| 493 | */ |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 494 | void memcg_create_kmem_cache(struct mem_cgroup *memcg, |
| 495 | struct kmem_cache *root_cache) |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 496 | { |
Vladimir Davydov | 3e0350a | 2015-02-10 14:11:44 -0800 | [diff] [blame] | 497 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */ |
Michal Hocko | 33398cf | 2015-09-08 15:01:02 -0700 | [diff] [blame] | 498 | struct cgroup_subsys_state *css = &memcg->css; |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 499 | struct memcg_cache_array *arr; |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 500 | struct kmem_cache *s = NULL; |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 501 | char *cache_name; |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 502 | int idx; |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 503 | |
| 504 | get_online_cpus(); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 505 | get_online_mems(); |
| 506 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 507 | mutex_lock(&slab_mutex); |
| 508 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 509 | /* |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 510 | * The memory cgroup could have been offlined while the cache |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 511 | * creation work was pending. |
| 512 | */ |
Vladimir Davydov | b6ecd2d | 2016-03-17 14:18:33 -0700 | [diff] [blame] | 513 | if (memcg->kmem_state != KMEM_ONLINE) |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 514 | goto out_unlock; |
| 515 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 516 | idx = memcg_cache_id(memcg); |
| 517 | arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches, |
| 518 | lockdep_is_held(&slab_mutex)); |
| 519 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 520 | /* |
| 521 | * Since per-memcg caches are created asynchronously on first |
| 522 | * allocation (see memcg_kmem_get_cache()), several threads can try to |
| 523 | * create the same cache, but only one of them may succeed. |
| 524 | */ |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 525 | if (arr->entries[idx]) |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 526 | goto out_unlock; |
| 527 | |
Vladimir Davydov | f100836 | 2015-02-12 14:59:29 -0800 | [diff] [blame] | 528 | cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf)); |
Vladimir Davydov | 073ee1c | 2014-06-04 16:08:23 -0700 | [diff] [blame] | 529 | cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name, |
Vladimir Davydov | f100836 | 2015-02-12 14:59:29 -0800 | [diff] [blame] | 530 | css->id, memcg_name_buf); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 531 | if (!cache_name) |
| 532 | goto out_unlock; |
| 533 | |
Vladimir Davydov | c9a77a7 | 2015-11-05 18:45:08 -0800 | [diff] [blame] | 534 | s = create_cache(cache_name, root_cache->object_size, |
| 535 | root_cache->size, root_cache->align, |
| 536 | root_cache->flags, root_cache->ctor, |
| 537 | memcg, root_cache); |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 538 | /* |
| 539 | * If we could not create a memcg cache, do not complain, because |
| 540 | * that's not critical at all as we can always proceed with the root |
| 541 | * cache. |
| 542 | */ |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 543 | if (IS_ERR(s)) { |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 544 | kfree(cache_name); |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 545 | goto out_unlock; |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 546 | } |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 547 | |
Vladimir Davydov | 426589f | 2015-02-12 14:59:23 -0800 | [diff] [blame] | 548 | list_add(&s->memcg_params.list, &root_cache->memcg_params.list); |
| 549 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 550 | /* |
| 551 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
| 552 | * barrier here to ensure nobody will see the kmem_cache partially |
| 553 | * initialized. |
| 554 | */ |
| 555 | smp_wmb(); |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 556 | arr->entries[idx] = s; |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 557 | |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 558 | out_unlock: |
| 559 | mutex_unlock(&slab_mutex); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 560 | |
| 561 | put_online_mems(); |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 562 | put_online_cpus(); |
| 563 | } |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 564 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 565 | void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg) |
| 566 | { |
| 567 | int idx; |
| 568 | struct memcg_cache_array *arr; |
Vladimir Davydov | d6e0b7f | 2015-02-12 14:59:47 -0800 | [diff] [blame] | 569 | struct kmem_cache *s, *c; |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 570 | |
| 571 | idx = memcg_cache_id(memcg); |
| 572 | |
Vladimir Davydov | d6e0b7f | 2015-02-12 14:59:47 -0800 | [diff] [blame] | 573 | get_online_cpus(); |
| 574 | get_online_mems(); |
| 575 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 576 | mutex_lock(&slab_mutex); |
| 577 | list_for_each_entry(s, &slab_caches, list) { |
| 578 | if (!is_root_cache(s)) |
| 579 | continue; |
| 580 | |
| 581 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, |
| 582 | lockdep_is_held(&slab_mutex)); |
Vladimir Davydov | d6e0b7f | 2015-02-12 14:59:47 -0800 | [diff] [blame] | 583 | c = arr->entries[idx]; |
| 584 | if (!c) |
| 585 | continue; |
| 586 | |
| 587 | __kmem_cache_shrink(c, true); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 588 | arr->entries[idx] = NULL; |
| 589 | } |
| 590 | mutex_unlock(&slab_mutex); |
Vladimir Davydov | d6e0b7f | 2015-02-12 14:59:47 -0800 | [diff] [blame] | 591 | |
| 592 | put_online_mems(); |
| 593 | put_online_cpus(); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 594 | } |
| 595 | |
Vladimir Davydov | d60fdcc | 2015-11-05 18:45:11 -0800 | [diff] [blame] | 596 | static int __shutdown_memcg_cache(struct kmem_cache *s, |
| 597 | struct list_head *release, bool *need_rcu_barrier) |
| 598 | { |
| 599 | BUG_ON(is_root_cache(s)); |
| 600 | |
| 601 | if (shutdown_cache(s, release, need_rcu_barrier)) |
| 602 | return -EBUSY; |
| 603 | |
| 604 | list_del(&s->memcg_params.list); |
| 605 | return 0; |
| 606 | } |
| 607 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 608 | void memcg_destroy_kmem_caches(struct mem_cgroup *memcg) |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 609 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 610 | LIST_HEAD(release); |
| 611 | bool need_rcu_barrier = false; |
| 612 | struct kmem_cache *s, *s2; |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 613 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 614 | get_online_cpus(); |
| 615 | get_online_mems(); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 616 | |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 617 | mutex_lock(&slab_mutex); |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 618 | list_for_each_entry_safe(s, s2, &slab_caches, list) { |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 619 | if (is_root_cache(s) || s->memcg_params.memcg != memcg) |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 620 | continue; |
| 621 | /* |
| 622 | * The cgroup is about to be freed and therefore has no charges |
| 623 | * left. Hence, all its caches must be empty by now. |
| 624 | */ |
Vladimir Davydov | d60fdcc | 2015-11-05 18:45:11 -0800 | [diff] [blame] | 625 | BUG_ON(__shutdown_memcg_cache(s, &release, &need_rcu_barrier)); |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 626 | } |
| 627 | mutex_unlock(&slab_mutex); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 628 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 629 | put_online_mems(); |
| 630 | put_online_cpus(); |
| 631 | |
Vladimir Davydov | c9a77a7 | 2015-11-05 18:45:08 -0800 | [diff] [blame] | 632 | release_caches(&release, need_rcu_barrier); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 633 | } |
Vladimir Davydov | d60fdcc | 2015-11-05 18:45:11 -0800 | [diff] [blame] | 634 | |
| 635 | static int shutdown_memcg_caches(struct kmem_cache *s, |
| 636 | struct list_head *release, bool *need_rcu_barrier) |
| 637 | { |
| 638 | struct memcg_cache_array *arr; |
| 639 | struct kmem_cache *c, *c2; |
| 640 | LIST_HEAD(busy); |
| 641 | int i; |
| 642 | |
| 643 | BUG_ON(!is_root_cache(s)); |
| 644 | |
| 645 | /* |
| 646 | * First, shutdown active caches, i.e. caches that belong to online |
| 647 | * memory cgroups. |
| 648 | */ |
| 649 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, |
| 650 | lockdep_is_held(&slab_mutex)); |
| 651 | for_each_memcg_cache_index(i) { |
| 652 | c = arr->entries[i]; |
| 653 | if (!c) |
| 654 | continue; |
| 655 | if (__shutdown_memcg_cache(c, release, need_rcu_barrier)) |
| 656 | /* |
| 657 | * The cache still has objects. Move it to a temporary |
| 658 | * list so as not to try to destroy it for a second |
| 659 | * time while iterating over inactive caches below. |
| 660 | */ |
| 661 | list_move(&c->memcg_params.list, &busy); |
| 662 | else |
| 663 | /* |
| 664 | * The cache is empty and will be destroyed soon. Clear |
| 665 | * the pointer to it in the memcg_caches array so that |
| 666 | * it will never be accessed even if the root cache |
| 667 | * stays alive. |
| 668 | */ |
| 669 | arr->entries[i] = NULL; |
| 670 | } |
| 671 | |
| 672 | /* |
| 673 | * Second, shutdown all caches left from memory cgroups that are now |
| 674 | * offline. |
| 675 | */ |
| 676 | list_for_each_entry_safe(c, c2, &s->memcg_params.list, |
| 677 | memcg_params.list) |
| 678 | __shutdown_memcg_cache(c, release, need_rcu_barrier); |
| 679 | |
| 680 | list_splice(&busy, &s->memcg_params.list); |
| 681 | |
| 682 | /* |
| 683 | * A cache being destroyed must be empty. In particular, this means |
| 684 | * that all per memcg caches attached to it must be empty too. |
| 685 | */ |
| 686 | if (!list_empty(&s->memcg_params.list)) |
| 687 | return -EBUSY; |
| 688 | return 0; |
| 689 | } |
| 690 | #else |
| 691 | static inline int shutdown_memcg_caches(struct kmem_cache *s, |
| 692 | struct list_head *release, bool *need_rcu_barrier) |
| 693 | { |
| 694 | return 0; |
| 695 | } |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 696 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ |
Vladimir Davydov | 794b124 | 2014-04-07 15:39:26 -0700 | [diff] [blame] | 697 | |
Christoph Lameter | 41a2128 | 2014-05-06 12:50:08 -0700 | [diff] [blame] | 698 | void slab_kmem_cache_release(struct kmem_cache *s) |
| 699 | { |
Dmitry Safonov | 52b4b95 | 2016-02-17 13:11:37 -0800 | [diff] [blame] | 700 | __kmem_cache_release(s); |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 701 | destroy_memcg_params(s); |
Andrzej Hajda | 3dec16e | 2015-02-13 14:36:38 -0800 | [diff] [blame] | 702 | kfree_const(s->name); |
Christoph Lameter | 41a2128 | 2014-05-06 12:50:08 -0700 | [diff] [blame] | 703 | kmem_cache_free(kmem_cache, s); |
| 704 | } |
| 705 | |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 706 | void kmem_cache_destroy(struct kmem_cache *s) |
| 707 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 708 | LIST_HEAD(release); |
| 709 | bool need_rcu_barrier = false; |
Vladimir Davydov | d60fdcc | 2015-11-05 18:45:11 -0800 | [diff] [blame] | 710 | int err; |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 711 | |
Sergey Senozhatsky | 3942d29 | 2015-09-08 15:00:50 -0700 | [diff] [blame] | 712 | if (unlikely(!s)) |
| 713 | return; |
| 714 | |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 715 | get_online_cpus(); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 716 | get_online_mems(); |
| 717 | |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 718 | mutex_lock(&slab_mutex); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 719 | |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 720 | s->refcount--; |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 721 | if (s->refcount) |
| 722 | goto out_unlock; |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 723 | |
Vladimir Davydov | d60fdcc | 2015-11-05 18:45:11 -0800 | [diff] [blame] | 724 | err = shutdown_memcg_caches(s, &release, &need_rcu_barrier); |
| 725 | if (!err) |
Vladimir Davydov | cd918c5 | 2015-11-05 18:45:14 -0800 | [diff] [blame] | 726 | err = shutdown_cache(s, &release, &need_rcu_barrier); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 727 | |
Vladimir Davydov | cd918c5 | 2015-11-05 18:45:14 -0800 | [diff] [blame] | 728 | if (err) { |
Joe Perches | 756a025f0 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 729 | pr_err("kmem_cache_destroy %s: Slab cache still has objects\n", |
| 730 | s->name); |
Vladimir Davydov | cd918c5 | 2015-11-05 18:45:14 -0800 | [diff] [blame] | 731 | dump_stack(); |
| 732 | } |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 733 | out_unlock: |
| 734 | mutex_unlock(&slab_mutex); |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 735 | |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 736 | put_online_mems(); |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 737 | put_online_cpus(); |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 738 | |
Vladimir Davydov | c9a77a7 | 2015-11-05 18:45:08 -0800 | [diff] [blame] | 739 | release_caches(&release, need_rcu_barrier); |
Christoph Lameter | 945cf2b | 2012-09-04 23:18:33 +0000 | [diff] [blame] | 740 | } |
| 741 | EXPORT_SYMBOL(kmem_cache_destroy); |
| 742 | |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 743 | /** |
| 744 | * kmem_cache_shrink - Shrink a cache. |
| 745 | * @cachep: The cache to shrink. |
| 746 | * |
| 747 | * Releases as many slabs as possible for a cache. |
| 748 | * To help debugging, a zero exit status indicates all slabs were released. |
| 749 | */ |
| 750 | int kmem_cache_shrink(struct kmem_cache *cachep) |
| 751 | { |
| 752 | int ret; |
| 753 | |
| 754 | get_online_cpus(); |
| 755 | get_online_mems(); |
Vladimir Davydov | d6e0b7f | 2015-02-12 14:59:47 -0800 | [diff] [blame] | 756 | ret = __kmem_cache_shrink(cachep, false); |
Vladimir Davydov | 03afc0e | 2014-06-04 16:07:20 -0700 | [diff] [blame] | 757 | put_online_mems(); |
| 758 | put_online_cpus(); |
| 759 | return ret; |
| 760 | } |
| 761 | EXPORT_SYMBOL(kmem_cache_shrink); |
| 762 | |
Denis Kirjanov | fda9012 | 2015-11-05 18:44:59 -0800 | [diff] [blame] | 763 | bool slab_is_available(void) |
Christoph Lameter | 97d0660 | 2012-07-06 15:25:11 -0500 | [diff] [blame] | 764 | { |
| 765 | return slab_state >= UP; |
| 766 | } |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 767 | |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 768 | #ifndef CONFIG_SLOB |
| 769 | /* Create a cache during boot when no slab services are available yet */ |
| 770 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, |
| 771 | unsigned long flags) |
| 772 | { |
| 773 | int err; |
| 774 | |
| 775 | s->name = name; |
| 776 | s->size = s->object_size = size; |
Christoph Lameter | 4590685 | 2012-11-28 16:23:16 +0000 | [diff] [blame] | 777 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 778 | |
| 779 | slab_init_memcg_params(s); |
| 780 | |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 781 | err = __kmem_cache_create(s, flags); |
| 782 | |
| 783 | if (err) |
Christoph Lameter | 31ba734 | 2013-01-10 19:00:53 +0000 | [diff] [blame] | 784 | panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n", |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 785 | name, size, err); |
| 786 | |
| 787 | s->refcount = -1; /* Exempt from merging for now */ |
| 788 | } |
| 789 | |
| 790 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, |
| 791 | unsigned long flags) |
| 792 | { |
| 793 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
| 794 | |
| 795 | if (!s) |
| 796 | panic("Out of memory when creating slab %s\n", name); |
| 797 | |
| 798 | create_boot_cache(s, name, size, flags); |
| 799 | list_add(&s->list, &slab_caches); |
| 800 | s->refcount = 1; |
| 801 | return s; |
| 802 | } |
| 803 | |
Christoph Lameter | 9425c58 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 804 | struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1]; |
| 805 | EXPORT_SYMBOL(kmalloc_caches); |
| 806 | |
| 807 | #ifdef CONFIG_ZONE_DMA |
| 808 | struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1]; |
| 809 | EXPORT_SYMBOL(kmalloc_dma_caches); |
| 810 | #endif |
| 811 | |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 812 | /* |
Christoph Lameter | 2c59dd6 | 2013-01-10 19:14:19 +0000 | [diff] [blame] | 813 | * Conversion table for small slabs sizes / 8 to the index in the |
| 814 | * kmalloc array. This is necessary for slabs < 192 since we have non power |
| 815 | * of two cache sizes there. The size of larger slabs can be determined using |
| 816 | * fls. |
| 817 | */ |
| 818 | static s8 size_index[24] = { |
| 819 | 3, /* 8 */ |
| 820 | 4, /* 16 */ |
| 821 | 5, /* 24 */ |
| 822 | 5, /* 32 */ |
| 823 | 6, /* 40 */ |
| 824 | 6, /* 48 */ |
| 825 | 6, /* 56 */ |
| 826 | 6, /* 64 */ |
| 827 | 1, /* 72 */ |
| 828 | 1, /* 80 */ |
| 829 | 1, /* 88 */ |
| 830 | 1, /* 96 */ |
| 831 | 7, /* 104 */ |
| 832 | 7, /* 112 */ |
| 833 | 7, /* 120 */ |
| 834 | 7, /* 128 */ |
| 835 | 2, /* 136 */ |
| 836 | 2, /* 144 */ |
| 837 | 2, /* 152 */ |
| 838 | 2, /* 160 */ |
| 839 | 2, /* 168 */ |
| 840 | 2, /* 176 */ |
| 841 | 2, /* 184 */ |
| 842 | 2 /* 192 */ |
| 843 | }; |
| 844 | |
| 845 | static inline int size_index_elem(size_t bytes) |
| 846 | { |
| 847 | return (bytes - 1) / 8; |
| 848 | } |
| 849 | |
| 850 | /* |
| 851 | * Find the kmem_cache structure that serves a given size of |
| 852 | * allocation |
| 853 | */ |
| 854 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) |
| 855 | { |
| 856 | int index; |
| 857 | |
Joonsoo Kim | 9de1bc8 | 2013-08-02 11:02:42 +0900 | [diff] [blame] | 858 | if (unlikely(size > KMALLOC_MAX_SIZE)) { |
Sasha Levin | 907985f | 2013-06-10 15:18:00 -0400 | [diff] [blame] | 859 | WARN_ON_ONCE(!(flags & __GFP_NOWARN)); |
Christoph Lameter | 6286ae9 | 2013-05-03 15:43:18 +0000 | [diff] [blame] | 860 | return NULL; |
Sasha Levin | 907985f | 2013-06-10 15:18:00 -0400 | [diff] [blame] | 861 | } |
Christoph Lameter | 6286ae9 | 2013-05-03 15:43:18 +0000 | [diff] [blame] | 862 | |
Christoph Lameter | 2c59dd6 | 2013-01-10 19:14:19 +0000 | [diff] [blame] | 863 | if (size <= 192) { |
| 864 | if (!size) |
| 865 | return ZERO_SIZE_PTR; |
| 866 | |
| 867 | index = size_index[size_index_elem(size)]; |
| 868 | } else |
| 869 | index = fls(size - 1); |
| 870 | |
| 871 | #ifdef CONFIG_ZONE_DMA |
Joonsoo Kim | b1e0541 | 2013-02-04 23:46:46 +0900 | [diff] [blame] | 872 | if (unlikely((flags & GFP_DMA))) |
Christoph Lameter | 2c59dd6 | 2013-01-10 19:14:19 +0000 | [diff] [blame] | 873 | return kmalloc_dma_caches[index]; |
| 874 | |
| 875 | #endif |
| 876 | return kmalloc_caches[index]; |
| 877 | } |
| 878 | |
| 879 | /* |
Gavin Guo | 4066c33 | 2015-06-24 16:55:54 -0700 | [diff] [blame] | 880 | * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. |
| 881 | * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is |
| 882 | * kmalloc-67108864. |
| 883 | */ |
| 884 | static struct { |
| 885 | const char *name; |
| 886 | unsigned long size; |
| 887 | } const kmalloc_info[] __initconst = { |
| 888 | {NULL, 0}, {"kmalloc-96", 96}, |
| 889 | {"kmalloc-192", 192}, {"kmalloc-8", 8}, |
| 890 | {"kmalloc-16", 16}, {"kmalloc-32", 32}, |
| 891 | {"kmalloc-64", 64}, {"kmalloc-128", 128}, |
| 892 | {"kmalloc-256", 256}, {"kmalloc-512", 512}, |
| 893 | {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048}, |
| 894 | {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192}, |
| 895 | {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768}, |
| 896 | {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072}, |
| 897 | {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288}, |
| 898 | {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152}, |
| 899 | {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608}, |
| 900 | {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432}, |
| 901 | {"kmalloc-67108864", 67108864} |
| 902 | }; |
| 903 | |
| 904 | /* |
Daniel Sanders | 34cc699 | 2015-06-24 16:55:57 -0700 | [diff] [blame] | 905 | * Patch up the size_index table if we have strange large alignment |
| 906 | * requirements for the kmalloc array. This is only the case for |
| 907 | * MIPS it seems. The standard arches will not generate any code here. |
| 908 | * |
| 909 | * Largest permitted alignment is 256 bytes due to the way we |
| 910 | * handle the index determination for the smaller caches. |
| 911 | * |
| 912 | * Make sure that nothing crazy happens if someone starts tinkering |
| 913 | * around with ARCH_KMALLOC_MINALIGN |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 914 | */ |
Daniel Sanders | 34cc699 | 2015-06-24 16:55:57 -0700 | [diff] [blame] | 915 | void __init setup_kmalloc_cache_index_table(void) |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 916 | { |
| 917 | int i; |
| 918 | |
Christoph Lameter | 2c59dd6 | 2013-01-10 19:14:19 +0000 | [diff] [blame] | 919 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
| 920 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); |
| 921 | |
| 922 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
| 923 | int elem = size_index_elem(i); |
| 924 | |
| 925 | if (elem >= ARRAY_SIZE(size_index)) |
| 926 | break; |
| 927 | size_index[elem] = KMALLOC_SHIFT_LOW; |
| 928 | } |
| 929 | |
| 930 | if (KMALLOC_MIN_SIZE >= 64) { |
| 931 | /* |
| 932 | * The 96 byte size cache is not used if the alignment |
| 933 | * is 64 byte. |
| 934 | */ |
| 935 | for (i = 64 + 8; i <= 96; i += 8) |
| 936 | size_index[size_index_elem(i)] = 7; |
| 937 | |
| 938 | } |
| 939 | |
| 940 | if (KMALLOC_MIN_SIZE >= 128) { |
| 941 | /* |
| 942 | * The 192 byte sized cache is not used if the alignment |
| 943 | * is 128 byte. Redirect kmalloc to use the 256 byte cache |
| 944 | * instead. |
| 945 | */ |
| 946 | for (i = 128 + 8; i <= 192; i += 8) |
| 947 | size_index[size_index_elem(i)] = 8; |
| 948 | } |
Daniel Sanders | 34cc699 | 2015-06-24 16:55:57 -0700 | [diff] [blame] | 949 | } |
| 950 | |
Christoph Lameter | ae6f246 | 2015-06-30 09:01:11 -0500 | [diff] [blame] | 951 | static void __init new_kmalloc_cache(int idx, unsigned long flags) |
Christoph Lameter | a9730fc | 2015-06-29 09:28:08 -0500 | [diff] [blame] | 952 | { |
| 953 | kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name, |
| 954 | kmalloc_info[idx].size, flags); |
| 955 | } |
| 956 | |
Daniel Sanders | 34cc699 | 2015-06-24 16:55:57 -0700 | [diff] [blame] | 957 | /* |
| 958 | * Create the kmalloc array. Some of the regular kmalloc arrays |
| 959 | * may already have been created because they were needed to |
| 960 | * enable allocations for slab creation. |
| 961 | */ |
| 962 | void __init create_kmalloc_caches(unsigned long flags) |
| 963 | { |
| 964 | int i; |
| 965 | |
Christoph Lameter | a9730fc | 2015-06-29 09:28:08 -0500 | [diff] [blame] | 966 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { |
| 967 | if (!kmalloc_caches[i]) |
| 968 | new_kmalloc_cache(i, flags); |
Chris Mason | 956e46e | 2013-05-08 15:56:28 -0400 | [diff] [blame] | 969 | |
| 970 | /* |
Christoph Lameter | a9730fc | 2015-06-29 09:28:08 -0500 | [diff] [blame] | 971 | * Caches that are not of the two-to-the-power-of size. |
| 972 | * These have to be created immediately after the |
| 973 | * earlier power of two caches |
Chris Mason | 956e46e | 2013-05-08 15:56:28 -0400 | [diff] [blame] | 974 | */ |
Christoph Lameter | a9730fc | 2015-06-29 09:28:08 -0500 | [diff] [blame] | 975 | if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6) |
| 976 | new_kmalloc_cache(1, flags); |
| 977 | if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7) |
| 978 | new_kmalloc_cache(2, flags); |
Christoph Lameter | 8a965b3 | 2013-05-03 18:04:18 +0000 | [diff] [blame] | 979 | } |
| 980 | |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 981 | /* Kmalloc array is now usable */ |
| 982 | slab_state = UP; |
| 983 | |
Christoph Lameter | f97d5f6 | 2013-01-10 19:12:17 +0000 | [diff] [blame] | 984 | #ifdef CONFIG_ZONE_DMA |
| 985 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { |
| 986 | struct kmem_cache *s = kmalloc_caches[i]; |
| 987 | |
| 988 | if (s) { |
| 989 | int size = kmalloc_size(i); |
| 990 | char *n = kasprintf(GFP_NOWAIT, |
| 991 | "dma-kmalloc-%d", size); |
| 992 | |
| 993 | BUG_ON(!n); |
| 994 | kmalloc_dma_caches[i] = create_kmalloc_cache(n, |
| 995 | size, SLAB_CACHE_DMA | flags); |
| 996 | } |
| 997 | } |
| 998 | #endif |
| 999 | } |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 1000 | #endif /* !CONFIG_SLOB */ |
| 1001 | |
Vladimir Davydov | cea371f | 2014-06-04 16:07:04 -0700 | [diff] [blame] | 1002 | /* |
| 1003 | * To avoid unnecessary overhead, we pass through large allocation requests |
| 1004 | * directly to the page allocator. We use __GFP_COMP, because we will need to |
| 1005 | * know the allocation order to free the pages properly in kfree. |
| 1006 | */ |
Vladimir Davydov | 5238343 | 2014-06-04 16:06:39 -0700 | [diff] [blame] | 1007 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
| 1008 | { |
| 1009 | void *ret; |
| 1010 | struct page *page; |
| 1011 | |
| 1012 | flags |= __GFP_COMP; |
| 1013 | page = alloc_kmem_pages(flags, order); |
| 1014 | ret = page ? page_address(page) : NULL; |
| 1015 | kmemleak_alloc(ret, size, 1, flags); |
Andrey Ryabinin | 0316bec | 2015-02-13 14:39:42 -0800 | [diff] [blame] | 1016 | kasan_kmalloc_large(ret, size); |
Vladimir Davydov | 5238343 | 2014-06-04 16:06:39 -0700 | [diff] [blame] | 1017 | return ret; |
| 1018 | } |
| 1019 | EXPORT_SYMBOL(kmalloc_order); |
| 1020 | |
Christoph Lameter | f1b6eb6 | 2013-09-04 16:35:34 +0000 | [diff] [blame] | 1021 | #ifdef CONFIG_TRACING |
| 1022 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) |
| 1023 | { |
| 1024 | void *ret = kmalloc_order(size, flags, order); |
| 1025 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); |
| 1026 | return ret; |
| 1027 | } |
| 1028 | EXPORT_SYMBOL(kmalloc_order_trace); |
| 1029 | #endif |
Christoph Lameter | 45530c4 | 2012-11-28 16:23:07 +0000 | [diff] [blame] | 1030 | |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1031 | #ifdef CONFIG_SLABINFO |
Wanpeng Li | e9b4db2 | 2013-07-04 08:33:24 +0800 | [diff] [blame] | 1032 | |
| 1033 | #ifdef CONFIG_SLAB |
| 1034 | #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR) |
| 1035 | #else |
| 1036 | #define SLABINFO_RIGHTS S_IRUSR |
| 1037 | #endif |
| 1038 | |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 1039 | static void print_slabinfo_header(struct seq_file *m) |
Glauber Costa | bcee6e2 | 2012-10-19 18:20:26 +0400 | [diff] [blame] | 1040 | { |
| 1041 | /* |
| 1042 | * Output format version, so at least we can change it |
| 1043 | * without _too_ many complaints. |
| 1044 | */ |
| 1045 | #ifdef CONFIG_DEBUG_SLAB |
| 1046 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); |
| 1047 | #else |
| 1048 | seq_puts(m, "slabinfo - version: 2.1\n"); |
| 1049 | #endif |
Joe Perches | 756a025f0 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 1050 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); |
Glauber Costa | bcee6e2 | 2012-10-19 18:20:26 +0400 | [diff] [blame] | 1051 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
| 1052 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); |
| 1053 | #ifdef CONFIG_DEBUG_SLAB |
Joe Perches | 756a025f0 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 1054 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
Glauber Costa | bcee6e2 | 2012-10-19 18:20:26 +0400 | [diff] [blame] | 1055 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
| 1056 | #endif |
| 1057 | seq_putc(m, '\n'); |
| 1058 | } |
| 1059 | |
Vladimir Davydov | 1df3b26 | 2014-12-10 15:42:16 -0800 | [diff] [blame] | 1060 | void *slab_start(struct seq_file *m, loff_t *pos) |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1061 | { |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1062 | mutex_lock(&slab_mutex); |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1063 | return seq_list_start(&slab_caches, *pos); |
| 1064 | } |
| 1065 | |
Wanpeng Li | 276a243 | 2013-07-08 08:08:28 +0800 | [diff] [blame] | 1066 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1067 | { |
| 1068 | return seq_list_next(p, &slab_caches, pos); |
| 1069 | } |
| 1070 | |
Wanpeng Li | 276a243 | 2013-07-08 08:08:28 +0800 | [diff] [blame] | 1071 | void slab_stop(struct seq_file *m, void *p) |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1072 | { |
| 1073 | mutex_unlock(&slab_mutex); |
| 1074 | } |
| 1075 | |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 1076 | static void |
| 1077 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1078 | { |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 1079 | struct kmem_cache *c; |
| 1080 | struct slabinfo sinfo; |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 1081 | |
| 1082 | if (!is_root_cache(s)) |
| 1083 | return; |
| 1084 | |
Vladimir Davydov | 426589f | 2015-02-12 14:59:23 -0800 | [diff] [blame] | 1085 | for_each_memcg_cache(c, s) { |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 1086 | memset(&sinfo, 0, sizeof(sinfo)); |
| 1087 | get_slabinfo(c, &sinfo); |
| 1088 | |
| 1089 | info->active_slabs += sinfo.active_slabs; |
| 1090 | info->num_slabs += sinfo.num_slabs; |
| 1091 | info->shared_avail += sinfo.shared_avail; |
| 1092 | info->active_objs += sinfo.active_objs; |
| 1093 | info->num_objs += sinfo.num_objs; |
| 1094 | } |
| 1095 | } |
| 1096 | |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 1097 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 1098 | { |
Glauber Costa | 0d7561c | 2012-10-19 18:20:27 +0400 | [diff] [blame] | 1099 | struct slabinfo sinfo; |
| 1100 | |
| 1101 | memset(&sinfo, 0, sizeof(sinfo)); |
| 1102 | get_slabinfo(s, &sinfo); |
| 1103 | |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 1104 | memcg_accumulate_slabinfo(s, &sinfo); |
| 1105 | |
Glauber Costa | 0d7561c | 2012-10-19 18:20:27 +0400 | [diff] [blame] | 1106 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 1107 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
Glauber Costa | 0d7561c | 2012-10-19 18:20:27 +0400 | [diff] [blame] | 1108 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
| 1109 | |
| 1110 | seq_printf(m, " : tunables %4u %4u %4u", |
| 1111 | sinfo.limit, sinfo.batchcount, sinfo.shared); |
| 1112 | seq_printf(m, " : slabdata %6lu %6lu %6lu", |
| 1113 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); |
| 1114 | slabinfo_show_stats(m, s); |
| 1115 | seq_putc(m, '\n'); |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1116 | } |
| 1117 | |
Vladimir Davydov | 1df3b26 | 2014-12-10 15:42:16 -0800 | [diff] [blame] | 1118 | static int slab_show(struct seq_file *m, void *p) |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 1119 | { |
| 1120 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); |
| 1121 | |
Vladimir Davydov | 1df3b26 | 2014-12-10 15:42:16 -0800 | [diff] [blame] | 1122 | if (p == slab_caches.next) |
| 1123 | print_slabinfo_header(m); |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 1124 | if (is_root_cache(s)) |
| 1125 | cache_show(s, m); |
| 1126 | return 0; |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 1127 | } |
| 1128 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 1129 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 1130 | int memcg_slab_show(struct seq_file *m, void *p) |
| 1131 | { |
| 1132 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); |
| 1133 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 1134 | |
| 1135 | if (p == slab_caches.next) |
| 1136 | print_slabinfo_header(m); |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 1137 | if (!is_root_cache(s) && s->memcg_params.memcg == memcg) |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 1138 | cache_show(s, m); |
| 1139 | return 0; |
| 1140 | } |
| 1141 | #endif |
| 1142 | |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1143 | /* |
| 1144 | * slabinfo_op - iterator that generates /proc/slabinfo |
| 1145 | * |
| 1146 | * Output layout: |
| 1147 | * cache-name |
| 1148 | * num-active-objs |
| 1149 | * total-objs |
| 1150 | * object size |
| 1151 | * num-active-slabs |
| 1152 | * total-slabs |
| 1153 | * num-pages-per-slab |
| 1154 | * + further values on SMP and with statistics enabled |
| 1155 | */ |
| 1156 | static const struct seq_operations slabinfo_op = { |
Vladimir Davydov | 1df3b26 | 2014-12-10 15:42:16 -0800 | [diff] [blame] | 1157 | .start = slab_start, |
Wanpeng Li | 276a243 | 2013-07-08 08:08:28 +0800 | [diff] [blame] | 1158 | .next = slab_next, |
| 1159 | .stop = slab_stop, |
Vladimir Davydov | 1df3b26 | 2014-12-10 15:42:16 -0800 | [diff] [blame] | 1160 | .show = slab_show, |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1161 | }; |
| 1162 | |
| 1163 | static int slabinfo_open(struct inode *inode, struct file *file) |
| 1164 | { |
| 1165 | return seq_open(file, &slabinfo_op); |
| 1166 | } |
| 1167 | |
| 1168 | static const struct file_operations proc_slabinfo_operations = { |
| 1169 | .open = slabinfo_open, |
| 1170 | .read = seq_read, |
| 1171 | .write = slabinfo_write, |
| 1172 | .llseek = seq_lseek, |
| 1173 | .release = seq_release, |
| 1174 | }; |
| 1175 | |
| 1176 | static int __init slab_proc_init(void) |
| 1177 | { |
Wanpeng Li | e9b4db2 | 2013-07-04 08:33:24 +0800 | [diff] [blame] | 1178 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, |
| 1179 | &proc_slabinfo_operations); |
Glauber Costa | b7454ad | 2012-10-19 18:20:25 +0400 | [diff] [blame] | 1180 | return 0; |
| 1181 | } |
| 1182 | module_init(slab_proc_init); |
| 1183 | #endif /* CONFIG_SLABINFO */ |
Andrey Ryabinin | 928cec9 | 2014-08-06 16:04:44 -0700 | [diff] [blame] | 1184 | |
| 1185 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, |
| 1186 | gfp_t flags) |
| 1187 | { |
| 1188 | void *ret; |
| 1189 | size_t ks = 0; |
| 1190 | |
| 1191 | if (p) |
| 1192 | ks = ksize(p); |
| 1193 | |
Andrey Ryabinin | 0316bec | 2015-02-13 14:39:42 -0800 | [diff] [blame] | 1194 | if (ks >= new_size) { |
| 1195 | kasan_krealloc((void *)p, new_size); |
Andrey Ryabinin | 928cec9 | 2014-08-06 16:04:44 -0700 | [diff] [blame] | 1196 | return (void *)p; |
Andrey Ryabinin | 0316bec | 2015-02-13 14:39:42 -0800 | [diff] [blame] | 1197 | } |
Andrey Ryabinin | 928cec9 | 2014-08-06 16:04:44 -0700 | [diff] [blame] | 1198 | |
| 1199 | ret = kmalloc_track_caller(new_size, flags); |
| 1200 | if (ret && p) |
| 1201 | memcpy(ret, p, ks); |
| 1202 | |
| 1203 | return ret; |
| 1204 | } |
| 1205 | |
| 1206 | /** |
| 1207 | * __krealloc - like krealloc() but don't free @p. |
| 1208 | * @p: object to reallocate memory for. |
| 1209 | * @new_size: how many bytes of memory are required. |
| 1210 | * @flags: the type of memory to allocate. |
| 1211 | * |
| 1212 | * This function is like krealloc() except it never frees the originally |
| 1213 | * allocated buffer. Use this if you don't want to free the buffer immediately |
| 1214 | * like, for example, with RCU. |
| 1215 | */ |
| 1216 | void *__krealloc(const void *p, size_t new_size, gfp_t flags) |
| 1217 | { |
| 1218 | if (unlikely(!new_size)) |
| 1219 | return ZERO_SIZE_PTR; |
| 1220 | |
| 1221 | return __do_krealloc(p, new_size, flags); |
| 1222 | |
| 1223 | } |
| 1224 | EXPORT_SYMBOL(__krealloc); |
| 1225 | |
| 1226 | /** |
| 1227 | * krealloc - reallocate memory. The contents will remain unchanged. |
| 1228 | * @p: object to reallocate memory for. |
| 1229 | * @new_size: how many bytes of memory are required. |
| 1230 | * @flags: the type of memory to allocate. |
| 1231 | * |
| 1232 | * The contents of the object pointed to are preserved up to the |
| 1233 | * lesser of the new and old sizes. If @p is %NULL, krealloc() |
| 1234 | * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a |
| 1235 | * %NULL pointer, the object pointed to is freed. |
| 1236 | */ |
| 1237 | void *krealloc(const void *p, size_t new_size, gfp_t flags) |
| 1238 | { |
| 1239 | void *ret; |
| 1240 | |
| 1241 | if (unlikely(!new_size)) { |
| 1242 | kfree(p); |
| 1243 | return ZERO_SIZE_PTR; |
| 1244 | } |
| 1245 | |
| 1246 | ret = __do_krealloc(p, new_size, flags); |
| 1247 | if (ret && p != ret) |
| 1248 | kfree(p); |
| 1249 | |
| 1250 | return ret; |
| 1251 | } |
| 1252 | EXPORT_SYMBOL(krealloc); |
| 1253 | |
| 1254 | /** |
| 1255 | * kzfree - like kfree but zero memory |
| 1256 | * @p: object to free memory of |
| 1257 | * |
| 1258 | * The memory of the object @p points to is zeroed before freed. |
| 1259 | * If @p is %NULL, kzfree() does nothing. |
| 1260 | * |
| 1261 | * Note: this function zeroes the whole allocated buffer which can be a good |
| 1262 | * deal bigger than the requested buffer size passed to kmalloc(). So be |
| 1263 | * careful when using this function in performance sensitive code. |
| 1264 | */ |
| 1265 | void kzfree(const void *p) |
| 1266 | { |
| 1267 | size_t ks; |
| 1268 | void *mem = (void *)p; |
| 1269 | |
| 1270 | if (unlikely(ZERO_OR_NULL_PTR(mem))) |
| 1271 | return; |
| 1272 | ks = ksize(mem); |
| 1273 | memset(mem, 0, ks); |
| 1274 | kfree(mem); |
| 1275 | } |
| 1276 | EXPORT_SYMBOL(kzfree); |
| 1277 | |
| 1278 | /* Tracepoints definitions. */ |
| 1279 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); |
| 1280 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); |
| 1281 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); |
| 1282 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); |
| 1283 | EXPORT_TRACEPOINT_SYMBOL(kfree); |
| 1284 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); |