| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Slab allocator functions that are independent of the allocator strategy |
| 4 | * |
| 5 | * (C) 2012 Christoph Lameter <cl@linux.com> |
| 6 | */ |
| 7 | #include <linux/slab.h> |
| 8 | |
| 9 | #include <linux/mm.h> |
| 10 | #include <linux/poison.h> |
| 11 | #include <linux/interrupt.h> |
| 12 | #include <linux/memory.h> |
| 13 | #include <linux/cache.h> |
| 14 | #include <linux/compiler.h> |
| 15 | #include <linux/kfence.h> |
| 16 | #include <linux/module.h> |
| 17 | #include <linux/cpu.h> |
| 18 | #include <linux/uaccess.h> |
| 19 | #include <linux/seq_file.h> |
| 20 | #include <linux/dma-mapping.h> |
| 21 | #include <linux/swiotlb.h> |
| 22 | #include <linux/proc_fs.h> |
| 23 | #include <linux/debugfs.h> |
| 24 | #include <linux/kmemleak.h> |
| 25 | #include <linux/kasan.h> |
| 26 | #include <asm/cacheflush.h> |
| 27 | #include <asm/tlbflush.h> |
| 28 | #include <asm/page.h> |
| 29 | #include <linux/memcontrol.h> |
| 30 | #include <linux/stackdepot.h> |
| 31 | |
| 32 | #include "internal.h" |
| 33 | #include "slab.h" |
| 34 | |
| 35 | #define CREATE_TRACE_POINTS |
| 36 | #include <trace/events/kmem.h> |
| 37 | |
| 38 | enum slab_state slab_state; |
| 39 | LIST_HEAD(slab_caches); |
| 40 | DEFINE_MUTEX(slab_mutex); |
| 41 | struct kmem_cache *kmem_cache; |
| 42 | |
| 43 | static LIST_HEAD(slab_caches_to_rcu_destroy); |
| 44 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); |
| 45 | static DECLARE_WORK(slab_caches_to_rcu_destroy_work, |
| 46 | slab_caches_to_rcu_destroy_workfn); |
| 47 | |
| 48 | /* |
| 49 | * Set of flags that will prevent slab merging |
| 50 | */ |
| 51 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
| 52 | SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ |
| 53 | SLAB_FAILSLAB | SLAB_NO_MERGE) |
| 54 | |
| 55 | #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ |
| 56 | SLAB_CACHE_DMA32 | SLAB_ACCOUNT) |
| 57 | |
| 58 | /* |
| 59 | * Merge control. If this is set then no merging of slab caches will occur. |
| 60 | */ |
| 61 | static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); |
| 62 | |
| 63 | static int __init setup_slab_nomerge(char *str) |
| 64 | { |
| 65 | slab_nomerge = true; |
| 66 | return 1; |
| 67 | } |
| 68 | |
| 69 | static int __init setup_slab_merge(char *str) |
| 70 | { |
| 71 | slab_nomerge = false; |
| 72 | return 1; |
| 73 | } |
| 74 | |
| 75 | __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); |
| 76 | __setup_param("slub_merge", slub_merge, setup_slab_merge, 0); |
| 77 | |
| 78 | __setup("slab_nomerge", setup_slab_nomerge); |
| 79 | __setup("slab_merge", setup_slab_merge); |
| 80 | |
| 81 | /* |
| 82 | * Determine the size of a slab object |
| 83 | */ |
| 84 | unsigned int kmem_cache_size(struct kmem_cache *s) |
| 85 | { |
| 86 | return s->object_size; |
| 87 | } |
| 88 | EXPORT_SYMBOL(kmem_cache_size); |
| 89 | |
| 90 | #ifdef CONFIG_DEBUG_VM |
| 91 | static int kmem_cache_sanity_check(const char *name, unsigned int size) |
| 92 | { |
| 93 | if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) { |
| 94 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
| 95 | return -EINVAL; |
| 96 | } |
| 97 | |
| 98 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
| 99 | return 0; |
| 100 | } |
| 101 | #else |
| 102 | static inline int kmem_cache_sanity_check(const char *name, unsigned int size) |
| 103 | { |
| 104 | return 0; |
| 105 | } |
| 106 | #endif |
| 107 | |
| 108 | /* |
| 109 | * Figure out what the alignment of the objects will be given a set of |
| 110 | * flags, a user specified alignment and the size of the objects. |
| 111 | */ |
| 112 | static unsigned int calculate_alignment(slab_flags_t flags, |
| 113 | unsigned int align, unsigned int size) |
| 114 | { |
| 115 | /* |
| 116 | * If the user wants hardware cache aligned objects then follow that |
| 117 | * suggestion if the object is sufficiently large. |
| 118 | * |
| 119 | * The hardware cache alignment cannot override the specified |
| 120 | * alignment though. If that is greater then use it. |
| 121 | */ |
| 122 | if (flags & SLAB_HWCACHE_ALIGN) { |
| 123 | unsigned int ralign; |
| 124 | |
| 125 | ralign = cache_line_size(); |
| 126 | while (size <= ralign / 2) |
| 127 | ralign /= 2; |
| 128 | align = max(align, ralign); |
| 129 | } |
| 130 | |
| 131 | align = max(align, arch_slab_minalign()); |
| 132 | |
| 133 | return ALIGN(align, sizeof(void *)); |
| 134 | } |
| 135 | |
| 136 | /* |
| 137 | * Find a mergeable slab cache |
| 138 | */ |
| 139 | int slab_unmergeable(struct kmem_cache *s) |
| 140 | { |
| 141 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) |
| 142 | return 1; |
| 143 | |
| 144 | if (s->ctor) |
| 145 | return 1; |
| 146 | |
| 147 | #ifdef CONFIG_HARDENED_USERCOPY |
| 148 | if (s->usersize) |
| 149 | return 1; |
| 150 | #endif |
| 151 | |
| 152 | /* |
| 153 | * We may have set a slab to be unmergeable during bootstrap. |
| 154 | */ |
| 155 | if (s->refcount < 0) |
| 156 | return 1; |
| 157 | |
| 158 | return 0; |
| 159 | } |
| 160 | |
| 161 | struct kmem_cache *find_mergeable(unsigned int size, unsigned int align, |
| 162 | slab_flags_t flags, const char *name, void (*ctor)(void *)) |
| 163 | { |
| 164 | struct kmem_cache *s; |
| 165 | |
| 166 | if (slab_nomerge) |
| 167 | return NULL; |
| 168 | |
| 169 | if (ctor) |
| 170 | return NULL; |
| 171 | |
| 172 | size = ALIGN(size, sizeof(void *)); |
| 173 | align = calculate_alignment(flags, align, size); |
| 174 | size = ALIGN(size, align); |
| 175 | flags = kmem_cache_flags(flags, name); |
| 176 | |
| 177 | if (flags & SLAB_NEVER_MERGE) |
| 178 | return NULL; |
| 179 | |
| 180 | list_for_each_entry_reverse(s, &slab_caches, list) { |
| 181 | if (slab_unmergeable(s)) |
| 182 | continue; |
| 183 | |
| 184 | if (size > s->size) |
| 185 | continue; |
| 186 | |
| 187 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) |
| 188 | continue; |
| 189 | /* |
| 190 | * Check if alignment is compatible. |
| 191 | * Courtesy of Adrian Drzewiecki |
| 192 | */ |
| 193 | if ((s->size & ~(align - 1)) != s->size) |
| 194 | continue; |
| 195 | |
| 196 | if (s->size - size >= sizeof(void *)) |
| 197 | continue; |
| 198 | |
| 199 | return s; |
| 200 | } |
| 201 | return NULL; |
| 202 | } |
| 203 | |
| 204 | static struct kmem_cache *create_cache(const char *name, |
| 205 | unsigned int object_size, unsigned int align, |
| 206 | slab_flags_t flags, unsigned int useroffset, |
| 207 | unsigned int usersize, void (*ctor)(void *), |
| 208 | struct kmem_cache *root_cache) |
| 209 | { |
| 210 | struct kmem_cache *s; |
| 211 | int err; |
| 212 | |
| 213 | if (WARN_ON(useroffset + usersize > object_size)) |
| 214 | useroffset = usersize = 0; |
| 215 | |
| 216 | err = -ENOMEM; |
| 217 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); |
| 218 | if (!s) |
| 219 | goto out; |
| 220 | |
| 221 | s->name = name; |
| 222 | s->size = s->object_size = object_size; |
| 223 | s->align = align; |
| 224 | s->ctor = ctor; |
| 225 | #ifdef CONFIG_HARDENED_USERCOPY |
| 226 | s->useroffset = useroffset; |
| 227 | s->usersize = usersize; |
| 228 | #endif |
| 229 | |
| 230 | err = __kmem_cache_create(s, flags); |
| 231 | if (err) |
| 232 | goto out_free_cache; |
| 233 | |
| 234 | s->refcount = 1; |
| 235 | list_add(&s->list, &slab_caches); |
| 236 | return s; |
| 237 | |
| 238 | out_free_cache: |
| 239 | kmem_cache_free(kmem_cache, s); |
| 240 | out: |
| 241 | return ERR_PTR(err); |
| 242 | } |
| 243 | |
| 244 | /** |
| 245 | * kmem_cache_create_usercopy - Create a cache with a region suitable |
| 246 | * for copying to userspace |
| 247 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
| 248 | * @size: The size of objects to be created in this cache. |
| 249 | * @align: The required alignment for the objects. |
| 250 | * @flags: SLAB flags |
| 251 | * @useroffset: Usercopy region offset |
| 252 | * @usersize: Usercopy region size |
| 253 | * @ctor: A constructor for the objects. |
| 254 | * |
| 255 | * Cannot be called within a interrupt, but can be interrupted. |
| 256 | * The @ctor is run when new pages are allocated by the cache. |
| 257 | * |
| 258 | * The flags are |
| 259 | * |
| 260 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) |
| 261 | * to catch references to uninitialised memory. |
| 262 | * |
| 263 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check |
| 264 | * for buffer overruns. |
| 265 | * |
| 266 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
| 267 | * cacheline. This can be beneficial if you're counting cycles as closely |
| 268 | * as davem. |
| 269 | * |
| 270 | * Return: a pointer to the cache on success, NULL on failure. |
| 271 | */ |
| 272 | struct kmem_cache * |
| 273 | kmem_cache_create_usercopy(const char *name, |
| 274 | unsigned int size, unsigned int align, |
| 275 | slab_flags_t flags, |
| 276 | unsigned int useroffset, unsigned int usersize, |
| 277 | void (*ctor)(void *)) |
| 278 | { |
| 279 | struct kmem_cache *s = NULL; |
| 280 | const char *cache_name; |
| 281 | int err; |
| 282 | |
| 283 | #ifdef CONFIG_SLUB_DEBUG |
| 284 | /* |
| 285 | * If no slab_debug was enabled globally, the static key is not yet |
| 286 | * enabled by setup_slub_debug(). Enable it if the cache is being |
| 287 | * created with any of the debugging flags passed explicitly. |
| 288 | * It's also possible that this is the first cache created with |
| 289 | * SLAB_STORE_USER and we should init stack_depot for it. |
| 290 | */ |
| 291 | if (flags & SLAB_DEBUG_FLAGS) |
| 292 | static_branch_enable(&slub_debug_enabled); |
| 293 | if (flags & SLAB_STORE_USER) |
| 294 | stack_depot_init(); |
| 295 | #endif |
| 296 | |
| 297 | mutex_lock(&slab_mutex); |
| 298 | |
| 299 | err = kmem_cache_sanity_check(name, size); |
| 300 | if (err) { |
| 301 | goto out_unlock; |
| 302 | } |
| 303 | |
| 304 | /* Refuse requests with allocator specific flags */ |
| 305 | if (flags & ~SLAB_FLAGS_PERMITTED) { |
| 306 | err = -EINVAL; |
| 307 | goto out_unlock; |
| 308 | } |
| 309 | |
| 310 | /* |
| 311 | * Some allocators will constraint the set of valid flags to a subset |
| 312 | * of all flags. We expect them to define CACHE_CREATE_MASK in this |
| 313 | * case, and we'll just provide them with a sanitized version of the |
| 314 | * passed flags. |
| 315 | */ |
| 316 | flags &= CACHE_CREATE_MASK; |
| 317 | |
| 318 | /* Fail closed on bad usersize of useroffset values. */ |
| 319 | if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) || |
| 320 | WARN_ON(!usersize && useroffset) || |
| 321 | WARN_ON(size < usersize || size - usersize < useroffset)) |
| 322 | usersize = useroffset = 0; |
| 323 | |
| 324 | if (!usersize) |
| 325 | s = __kmem_cache_alias(name, size, align, flags, ctor); |
| 326 | if (s) |
| 327 | goto out_unlock; |
| 328 | |
| 329 | cache_name = kstrdup_const(name, GFP_KERNEL); |
| 330 | if (!cache_name) { |
| 331 | err = -ENOMEM; |
| 332 | goto out_unlock; |
| 333 | } |
| 334 | |
| 335 | s = create_cache(cache_name, size, |
| 336 | calculate_alignment(flags, align, size), |
| 337 | flags, useroffset, usersize, ctor, NULL); |
| 338 | if (IS_ERR(s)) { |
| 339 | err = PTR_ERR(s); |
| 340 | kfree_const(cache_name); |
| 341 | } |
| 342 | |
| 343 | out_unlock: |
| 344 | mutex_unlock(&slab_mutex); |
| 345 | |
| 346 | if (err) { |
| 347 | if (flags & SLAB_PANIC) |
| 348 | panic("%s: Failed to create slab '%s'. Error %d\n", |
| 349 | __func__, name, err); |
| 350 | else { |
| 351 | pr_warn("%s(%s) failed with error %d\n", |
| 352 | __func__, name, err); |
| 353 | dump_stack(); |
| 354 | } |
| 355 | return NULL; |
| 356 | } |
| 357 | return s; |
| 358 | } |
| 359 | EXPORT_SYMBOL(kmem_cache_create_usercopy); |
| 360 | |
| 361 | /** |
| 362 | * kmem_cache_create - Create a cache. |
| 363 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
| 364 | * @size: The size of objects to be created in this cache. |
| 365 | * @align: The required alignment for the objects. |
| 366 | * @flags: SLAB flags |
| 367 | * @ctor: A constructor for the objects. |
| 368 | * |
| 369 | * Cannot be called within a interrupt, but can be interrupted. |
| 370 | * The @ctor is run when new pages are allocated by the cache. |
| 371 | * |
| 372 | * The flags are |
| 373 | * |
| 374 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) |
| 375 | * to catch references to uninitialised memory. |
| 376 | * |
| 377 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check |
| 378 | * for buffer overruns. |
| 379 | * |
| 380 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
| 381 | * cacheline. This can be beneficial if you're counting cycles as closely |
| 382 | * as davem. |
| 383 | * |
| 384 | * Return: a pointer to the cache on success, NULL on failure. |
| 385 | */ |
| 386 | struct kmem_cache * |
| 387 | kmem_cache_create(const char *name, unsigned int size, unsigned int align, |
| 388 | slab_flags_t flags, void (*ctor)(void *)) |
| 389 | { |
| 390 | return kmem_cache_create_usercopy(name, size, align, flags, 0, 0, |
| 391 | ctor); |
| 392 | } |
| 393 | EXPORT_SYMBOL(kmem_cache_create); |
| 394 | |
| 395 | static struct kmem_cache *kmem_buckets_cache __ro_after_init; |
| 396 | |
| 397 | /** |
| 398 | * kmem_buckets_create - Create a set of caches that handle dynamic sized |
| 399 | * allocations via kmem_buckets_alloc() |
| 400 | * @name: A prefix string which is used in /proc/slabinfo to identify this |
| 401 | * cache. The individual caches with have their sizes as the suffix. |
| 402 | * @flags: SLAB flags (see kmem_cache_create() for details). |
| 403 | * @useroffset: Starting offset within an allocation that may be copied |
| 404 | * to/from userspace. |
| 405 | * @usersize: How many bytes, starting at @useroffset, may be copied |
| 406 | * to/from userspace. |
| 407 | * @ctor: A constructor for the objects, run when new allocations are made. |
| 408 | * |
| 409 | * Cannot be called within an interrupt, but can be interrupted. |
| 410 | * |
| 411 | * Return: a pointer to the cache on success, NULL on failure. When |
| 412 | * CONFIG_SLAB_BUCKETS is not enabled, ZERO_SIZE_PTR is returned, and |
| 413 | * subsequent calls to kmem_buckets_alloc() will fall back to kmalloc(). |
| 414 | * (i.e. callers only need to check for NULL on failure.) |
| 415 | */ |
| 416 | kmem_buckets *kmem_buckets_create(const char *name, slab_flags_t flags, |
| 417 | unsigned int useroffset, |
| 418 | unsigned int usersize, |
| 419 | void (*ctor)(void *)) |
| 420 | { |
| 421 | kmem_buckets *b; |
| 422 | int idx; |
| 423 | |
| 424 | /* |
| 425 | * When the separate buckets API is not built in, just return |
| 426 | * a non-NULL value for the kmem_buckets pointer, which will be |
| 427 | * unused when performing allocations. |
| 428 | */ |
| 429 | if (!IS_ENABLED(CONFIG_SLAB_BUCKETS)) |
| 430 | return ZERO_SIZE_PTR; |
| 431 | |
| 432 | if (WARN_ON(!kmem_buckets_cache)) |
| 433 | return NULL; |
| 434 | |
| 435 | b = kmem_cache_alloc(kmem_buckets_cache, GFP_KERNEL|__GFP_ZERO); |
| 436 | if (WARN_ON(!b)) |
| 437 | return NULL; |
| 438 | |
| 439 | flags |= SLAB_NO_MERGE; |
| 440 | |
| 441 | for (idx = 0; idx < ARRAY_SIZE(kmalloc_caches[KMALLOC_NORMAL]); idx++) { |
| 442 | char *short_size, *cache_name; |
| 443 | unsigned int cache_useroffset, cache_usersize; |
| 444 | unsigned int size; |
| 445 | |
| 446 | if (!kmalloc_caches[KMALLOC_NORMAL][idx]) |
| 447 | continue; |
| 448 | |
| 449 | size = kmalloc_caches[KMALLOC_NORMAL][idx]->object_size; |
| 450 | if (!size) |
| 451 | continue; |
| 452 | |
| 453 | short_size = strchr(kmalloc_caches[KMALLOC_NORMAL][idx]->name, '-'); |
| 454 | if (WARN_ON(!short_size)) |
| 455 | goto fail; |
| 456 | |
| 457 | cache_name = kasprintf(GFP_KERNEL, "%s-%s", name, short_size + 1); |
| 458 | if (WARN_ON(!cache_name)) |
| 459 | goto fail; |
| 460 | |
| 461 | if (useroffset >= size) { |
| 462 | cache_useroffset = 0; |
| 463 | cache_usersize = 0; |
| 464 | } else { |
| 465 | cache_useroffset = useroffset; |
| 466 | cache_usersize = min(size - cache_useroffset, usersize); |
| 467 | } |
| 468 | (*b)[idx] = kmem_cache_create_usercopy(cache_name, size, |
| 469 | 0, flags, cache_useroffset, |
| 470 | cache_usersize, ctor); |
| 471 | kfree(cache_name); |
| 472 | if (WARN_ON(!(*b)[idx])) |
| 473 | goto fail; |
| 474 | } |
| 475 | |
| 476 | return b; |
| 477 | |
| 478 | fail: |
| 479 | for (idx = 0; idx < ARRAY_SIZE(kmalloc_caches[KMALLOC_NORMAL]); idx++) |
| 480 | kmem_cache_destroy((*b)[idx]); |
| 481 | kfree(b); |
| 482 | |
| 483 | return NULL; |
| 484 | } |
| 485 | EXPORT_SYMBOL(kmem_buckets_create); |
| 486 | |
| 487 | #ifdef SLAB_SUPPORTS_SYSFS |
| 488 | /* |
| 489 | * For a given kmem_cache, kmem_cache_destroy() should only be called |
| 490 | * once or there will be a use-after-free problem. The actual deletion |
| 491 | * and release of the kobject does not need slab_mutex or cpu_hotplug_lock |
| 492 | * protection. So they are now done without holding those locks. |
| 493 | * |
| 494 | * Note that there will be a slight delay in the deletion of sysfs files |
| 495 | * if kmem_cache_release() is called indrectly from a work function. |
| 496 | */ |
| 497 | static void kmem_cache_release(struct kmem_cache *s) |
| 498 | { |
| 499 | if (slab_state >= FULL) { |
| 500 | sysfs_slab_unlink(s); |
| 501 | sysfs_slab_release(s); |
| 502 | } else { |
| 503 | slab_kmem_cache_release(s); |
| 504 | } |
| 505 | } |
| 506 | #else |
| 507 | static void kmem_cache_release(struct kmem_cache *s) |
| 508 | { |
| 509 | slab_kmem_cache_release(s); |
| 510 | } |
| 511 | #endif |
| 512 | |
| 513 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) |
| 514 | { |
| 515 | LIST_HEAD(to_destroy); |
| 516 | struct kmem_cache *s, *s2; |
| 517 | |
| 518 | /* |
| 519 | * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the |
| 520 | * @slab_caches_to_rcu_destroy list. The slab pages are freed |
| 521 | * through RCU and the associated kmem_cache are dereferenced |
| 522 | * while freeing the pages, so the kmem_caches should be freed only |
| 523 | * after the pending RCU operations are finished. As rcu_barrier() |
| 524 | * is a pretty slow operation, we batch all pending destructions |
| 525 | * asynchronously. |
| 526 | */ |
| 527 | mutex_lock(&slab_mutex); |
| 528 | list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); |
| 529 | mutex_unlock(&slab_mutex); |
| 530 | |
| 531 | if (list_empty(&to_destroy)) |
| 532 | return; |
| 533 | |
| 534 | rcu_barrier(); |
| 535 | |
| 536 | list_for_each_entry_safe(s, s2, &to_destroy, list) { |
| 537 | debugfs_slab_release(s); |
| 538 | kfence_shutdown_cache(s); |
| 539 | kmem_cache_release(s); |
| 540 | } |
| 541 | } |
| 542 | |
| 543 | static int shutdown_cache(struct kmem_cache *s) |
| 544 | { |
| 545 | /* free asan quarantined objects */ |
| 546 | kasan_cache_shutdown(s); |
| 547 | |
| 548 | if (__kmem_cache_shutdown(s) != 0) |
| 549 | return -EBUSY; |
| 550 | |
| 551 | list_del(&s->list); |
| 552 | |
| 553 | if (s->flags & SLAB_TYPESAFE_BY_RCU) { |
| 554 | list_add_tail(&s->list, &slab_caches_to_rcu_destroy); |
| 555 | schedule_work(&slab_caches_to_rcu_destroy_work); |
| 556 | } else { |
| 557 | kfence_shutdown_cache(s); |
| 558 | debugfs_slab_release(s); |
| 559 | } |
| 560 | |
| 561 | return 0; |
| 562 | } |
| 563 | |
| 564 | void slab_kmem_cache_release(struct kmem_cache *s) |
| 565 | { |
| 566 | __kmem_cache_release(s); |
| 567 | kfree_const(s->name); |
| 568 | kmem_cache_free(kmem_cache, s); |
| 569 | } |
| 570 | |
| 571 | void kmem_cache_destroy(struct kmem_cache *s) |
| 572 | { |
| 573 | int err = -EBUSY; |
| 574 | bool rcu_set; |
| 575 | |
| 576 | if (unlikely(!s) || !kasan_check_byte(s)) |
| 577 | return; |
| 578 | |
| 579 | cpus_read_lock(); |
| 580 | mutex_lock(&slab_mutex); |
| 581 | |
| 582 | rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU; |
| 583 | |
| 584 | s->refcount--; |
| 585 | if (s->refcount) |
| 586 | goto out_unlock; |
| 587 | |
| 588 | err = shutdown_cache(s); |
| 589 | WARN(err, "%s %s: Slab cache still has objects when called from %pS", |
| 590 | __func__, s->name, (void *)_RET_IP_); |
| 591 | out_unlock: |
| 592 | mutex_unlock(&slab_mutex); |
| 593 | cpus_read_unlock(); |
| 594 | if (!err && !rcu_set) |
| 595 | kmem_cache_release(s); |
| 596 | } |
| 597 | EXPORT_SYMBOL(kmem_cache_destroy); |
| 598 | |
| 599 | /** |
| 600 | * kmem_cache_shrink - Shrink a cache. |
| 601 | * @cachep: The cache to shrink. |
| 602 | * |
| 603 | * Releases as many slabs as possible for a cache. |
| 604 | * To help debugging, a zero exit status indicates all slabs were released. |
| 605 | * |
| 606 | * Return: %0 if all slabs were released, non-zero otherwise |
| 607 | */ |
| 608 | int kmem_cache_shrink(struct kmem_cache *cachep) |
| 609 | { |
| 610 | kasan_cache_shrink(cachep); |
| 611 | |
| 612 | return __kmem_cache_shrink(cachep); |
| 613 | } |
| 614 | EXPORT_SYMBOL(kmem_cache_shrink); |
| 615 | |
| 616 | bool slab_is_available(void) |
| 617 | { |
| 618 | return slab_state >= UP; |
| 619 | } |
| 620 | |
| 621 | #ifdef CONFIG_PRINTK |
| 622 | static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab) |
| 623 | { |
| 624 | if (__kfence_obj_info(kpp, object, slab)) |
| 625 | return; |
| 626 | __kmem_obj_info(kpp, object, slab); |
| 627 | } |
| 628 | |
| 629 | /** |
| 630 | * kmem_dump_obj - Print available slab provenance information |
| 631 | * @object: slab object for which to find provenance information. |
| 632 | * |
| 633 | * This function uses pr_cont(), so that the caller is expected to have |
| 634 | * printed out whatever preamble is appropriate. The provenance information |
| 635 | * depends on the type of object and on how much debugging is enabled. |
| 636 | * For a slab-cache object, the fact that it is a slab object is printed, |
| 637 | * and, if available, the slab name, return address, and stack trace from |
| 638 | * the allocation and last free path of that object. |
| 639 | * |
| 640 | * Return: %true if the pointer is to a not-yet-freed object from |
| 641 | * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer |
| 642 | * is to an already-freed object, and %false otherwise. |
| 643 | */ |
| 644 | bool kmem_dump_obj(void *object) |
| 645 | { |
| 646 | char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc"; |
| 647 | int i; |
| 648 | struct slab *slab; |
| 649 | unsigned long ptroffset; |
| 650 | struct kmem_obj_info kp = { }; |
| 651 | |
| 652 | /* Some arches consider ZERO_SIZE_PTR to be a valid address. */ |
| 653 | if (object < (void *)PAGE_SIZE || !virt_addr_valid(object)) |
| 654 | return false; |
| 655 | slab = virt_to_slab(object); |
| 656 | if (!slab) |
| 657 | return false; |
| 658 | |
| 659 | kmem_obj_info(&kp, object, slab); |
| 660 | if (kp.kp_slab_cache) |
| 661 | pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name); |
| 662 | else |
| 663 | pr_cont(" slab%s", cp); |
| 664 | if (is_kfence_address(object)) |
| 665 | pr_cont(" (kfence)"); |
| 666 | if (kp.kp_objp) |
| 667 | pr_cont(" start %px", kp.kp_objp); |
| 668 | if (kp.kp_data_offset) |
| 669 | pr_cont(" data offset %lu", kp.kp_data_offset); |
| 670 | if (kp.kp_objp) { |
| 671 | ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset; |
| 672 | pr_cont(" pointer offset %lu", ptroffset); |
| 673 | } |
| 674 | if (kp.kp_slab_cache && kp.kp_slab_cache->object_size) |
| 675 | pr_cont(" size %u", kp.kp_slab_cache->object_size); |
| 676 | if (kp.kp_ret) |
| 677 | pr_cont(" allocated at %pS\n", kp.kp_ret); |
| 678 | else |
| 679 | pr_cont("\n"); |
| 680 | for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) { |
| 681 | if (!kp.kp_stack[i]) |
| 682 | break; |
| 683 | pr_info(" %pS\n", kp.kp_stack[i]); |
| 684 | } |
| 685 | |
| 686 | if (kp.kp_free_stack[0]) |
| 687 | pr_cont(" Free path:\n"); |
| 688 | |
| 689 | for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) { |
| 690 | if (!kp.kp_free_stack[i]) |
| 691 | break; |
| 692 | pr_info(" %pS\n", kp.kp_free_stack[i]); |
| 693 | } |
| 694 | |
| 695 | return true; |
| 696 | } |
| 697 | EXPORT_SYMBOL_GPL(kmem_dump_obj); |
| 698 | #endif |
| 699 | |
| 700 | /* Create a cache during boot when no slab services are available yet */ |
| 701 | void __init create_boot_cache(struct kmem_cache *s, const char *name, |
| 702 | unsigned int size, slab_flags_t flags, |
| 703 | unsigned int useroffset, unsigned int usersize) |
| 704 | { |
| 705 | int err; |
| 706 | unsigned int align = ARCH_KMALLOC_MINALIGN; |
| 707 | |
| 708 | s->name = name; |
| 709 | s->size = s->object_size = size; |
| 710 | |
| 711 | /* |
| 712 | * kmalloc caches guarantee alignment of at least the largest |
| 713 | * power-of-two divisor of the size. For power-of-two sizes, |
| 714 | * it is the size itself. |
| 715 | */ |
| 716 | if (flags & SLAB_KMALLOC) |
| 717 | align = max(align, 1U << (ffs(size) - 1)); |
| 718 | s->align = calculate_alignment(flags, align, size); |
| 719 | |
| 720 | #ifdef CONFIG_HARDENED_USERCOPY |
| 721 | s->useroffset = useroffset; |
| 722 | s->usersize = usersize; |
| 723 | #endif |
| 724 | |
| 725 | err = __kmem_cache_create(s, flags); |
| 726 | |
| 727 | if (err) |
| 728 | panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", |
| 729 | name, size, err); |
| 730 | |
| 731 | s->refcount = -1; /* Exempt from merging for now */ |
| 732 | } |
| 733 | |
| 734 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
| 735 | unsigned int size, |
| 736 | slab_flags_t flags) |
| 737 | { |
| 738 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
| 739 | |
| 740 | if (!s) |
| 741 | panic("Out of memory when creating slab %s\n", name); |
| 742 | |
| 743 | create_boot_cache(s, name, size, flags | SLAB_KMALLOC, 0, size); |
| 744 | list_add(&s->list, &slab_caches); |
| 745 | s->refcount = 1; |
| 746 | return s; |
| 747 | } |
| 748 | |
| 749 | kmem_buckets kmalloc_caches[NR_KMALLOC_TYPES] __ro_after_init = |
| 750 | { /* initialization for https://llvm.org/pr42570 */ }; |
| 751 | EXPORT_SYMBOL(kmalloc_caches); |
| 752 | |
| 753 | #ifdef CONFIG_RANDOM_KMALLOC_CACHES |
| 754 | unsigned long random_kmalloc_seed __ro_after_init; |
| 755 | EXPORT_SYMBOL(random_kmalloc_seed); |
| 756 | #endif |
| 757 | |
| 758 | /* |
| 759 | * Conversion table for small slabs sizes / 8 to the index in the |
| 760 | * kmalloc array. This is necessary for slabs < 192 since we have non power |
| 761 | * of two cache sizes there. The size of larger slabs can be determined using |
| 762 | * fls. |
| 763 | */ |
| 764 | u8 kmalloc_size_index[24] __ro_after_init = { |
| 765 | 3, /* 8 */ |
| 766 | 4, /* 16 */ |
| 767 | 5, /* 24 */ |
| 768 | 5, /* 32 */ |
| 769 | 6, /* 40 */ |
| 770 | 6, /* 48 */ |
| 771 | 6, /* 56 */ |
| 772 | 6, /* 64 */ |
| 773 | 1, /* 72 */ |
| 774 | 1, /* 80 */ |
| 775 | 1, /* 88 */ |
| 776 | 1, /* 96 */ |
| 777 | 7, /* 104 */ |
| 778 | 7, /* 112 */ |
| 779 | 7, /* 120 */ |
| 780 | 7, /* 128 */ |
| 781 | 2, /* 136 */ |
| 782 | 2, /* 144 */ |
| 783 | 2, /* 152 */ |
| 784 | 2, /* 160 */ |
| 785 | 2, /* 168 */ |
| 786 | 2, /* 176 */ |
| 787 | 2, /* 184 */ |
| 788 | 2 /* 192 */ |
| 789 | }; |
| 790 | |
| 791 | size_t kmalloc_size_roundup(size_t size) |
| 792 | { |
| 793 | if (size && size <= KMALLOC_MAX_CACHE_SIZE) { |
| 794 | /* |
| 795 | * The flags don't matter since size_index is common to all. |
| 796 | * Neither does the caller for just getting ->object_size. |
| 797 | */ |
| 798 | return kmalloc_slab(size, NULL, GFP_KERNEL, 0)->object_size; |
| 799 | } |
| 800 | |
| 801 | /* Above the smaller buckets, size is a multiple of page size. */ |
| 802 | if (size && size <= KMALLOC_MAX_SIZE) |
| 803 | return PAGE_SIZE << get_order(size); |
| 804 | |
| 805 | /* |
| 806 | * Return 'size' for 0 - kmalloc() returns ZERO_SIZE_PTR |
| 807 | * and very large size - kmalloc() may fail. |
| 808 | */ |
| 809 | return size; |
| 810 | |
| 811 | } |
| 812 | EXPORT_SYMBOL(kmalloc_size_roundup); |
| 813 | |
| 814 | #ifdef CONFIG_ZONE_DMA |
| 815 | #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz, |
| 816 | #else |
| 817 | #define KMALLOC_DMA_NAME(sz) |
| 818 | #endif |
| 819 | |
| 820 | #ifdef CONFIG_MEMCG |
| 821 | #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz, |
| 822 | #else |
| 823 | #define KMALLOC_CGROUP_NAME(sz) |
| 824 | #endif |
| 825 | |
| 826 | #ifndef CONFIG_SLUB_TINY |
| 827 | #define KMALLOC_RCL_NAME(sz) .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz, |
| 828 | #else |
| 829 | #define KMALLOC_RCL_NAME(sz) |
| 830 | #endif |
| 831 | |
| 832 | #ifdef CONFIG_RANDOM_KMALLOC_CACHES |
| 833 | #define __KMALLOC_RANDOM_CONCAT(a, b) a ## b |
| 834 | #define KMALLOC_RANDOM_NAME(N, sz) __KMALLOC_RANDOM_CONCAT(KMA_RAND_, N)(sz) |
| 835 | #define KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 1] = "kmalloc-rnd-01-" #sz, |
| 836 | #define KMA_RAND_2(sz) KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 2] = "kmalloc-rnd-02-" #sz, |
| 837 | #define KMA_RAND_3(sz) KMA_RAND_2(sz) .name[KMALLOC_RANDOM_START + 3] = "kmalloc-rnd-03-" #sz, |
| 838 | #define KMA_RAND_4(sz) KMA_RAND_3(sz) .name[KMALLOC_RANDOM_START + 4] = "kmalloc-rnd-04-" #sz, |
| 839 | #define KMA_RAND_5(sz) KMA_RAND_4(sz) .name[KMALLOC_RANDOM_START + 5] = "kmalloc-rnd-05-" #sz, |
| 840 | #define KMA_RAND_6(sz) KMA_RAND_5(sz) .name[KMALLOC_RANDOM_START + 6] = "kmalloc-rnd-06-" #sz, |
| 841 | #define KMA_RAND_7(sz) KMA_RAND_6(sz) .name[KMALLOC_RANDOM_START + 7] = "kmalloc-rnd-07-" #sz, |
| 842 | #define KMA_RAND_8(sz) KMA_RAND_7(sz) .name[KMALLOC_RANDOM_START + 8] = "kmalloc-rnd-08-" #sz, |
| 843 | #define KMA_RAND_9(sz) KMA_RAND_8(sz) .name[KMALLOC_RANDOM_START + 9] = "kmalloc-rnd-09-" #sz, |
| 844 | #define KMA_RAND_10(sz) KMA_RAND_9(sz) .name[KMALLOC_RANDOM_START + 10] = "kmalloc-rnd-10-" #sz, |
| 845 | #define KMA_RAND_11(sz) KMA_RAND_10(sz) .name[KMALLOC_RANDOM_START + 11] = "kmalloc-rnd-11-" #sz, |
| 846 | #define KMA_RAND_12(sz) KMA_RAND_11(sz) .name[KMALLOC_RANDOM_START + 12] = "kmalloc-rnd-12-" #sz, |
| 847 | #define KMA_RAND_13(sz) KMA_RAND_12(sz) .name[KMALLOC_RANDOM_START + 13] = "kmalloc-rnd-13-" #sz, |
| 848 | #define KMA_RAND_14(sz) KMA_RAND_13(sz) .name[KMALLOC_RANDOM_START + 14] = "kmalloc-rnd-14-" #sz, |
| 849 | #define KMA_RAND_15(sz) KMA_RAND_14(sz) .name[KMALLOC_RANDOM_START + 15] = "kmalloc-rnd-15-" #sz, |
| 850 | #else // CONFIG_RANDOM_KMALLOC_CACHES |
| 851 | #define KMALLOC_RANDOM_NAME(N, sz) |
| 852 | #endif |
| 853 | |
| 854 | #define INIT_KMALLOC_INFO(__size, __short_size) \ |
| 855 | { \ |
| 856 | .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ |
| 857 | KMALLOC_RCL_NAME(__short_size) \ |
| 858 | KMALLOC_CGROUP_NAME(__short_size) \ |
| 859 | KMALLOC_DMA_NAME(__short_size) \ |
| 860 | KMALLOC_RANDOM_NAME(RANDOM_KMALLOC_CACHES_NR, __short_size) \ |
| 861 | .size = __size, \ |
| 862 | } |
| 863 | |
| 864 | /* |
| 865 | * kmalloc_info[] is to make slab_debug=,kmalloc-xx option work at boot time. |
| 866 | * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is |
| 867 | * kmalloc-2M. |
| 868 | */ |
| 869 | const struct kmalloc_info_struct kmalloc_info[] __initconst = { |
| 870 | INIT_KMALLOC_INFO(0, 0), |
| 871 | INIT_KMALLOC_INFO(96, 96), |
| 872 | INIT_KMALLOC_INFO(192, 192), |
| 873 | INIT_KMALLOC_INFO(8, 8), |
| 874 | INIT_KMALLOC_INFO(16, 16), |
| 875 | INIT_KMALLOC_INFO(32, 32), |
| 876 | INIT_KMALLOC_INFO(64, 64), |
| 877 | INIT_KMALLOC_INFO(128, 128), |
| 878 | INIT_KMALLOC_INFO(256, 256), |
| 879 | INIT_KMALLOC_INFO(512, 512), |
| 880 | INIT_KMALLOC_INFO(1024, 1k), |
| 881 | INIT_KMALLOC_INFO(2048, 2k), |
| 882 | INIT_KMALLOC_INFO(4096, 4k), |
| 883 | INIT_KMALLOC_INFO(8192, 8k), |
| 884 | INIT_KMALLOC_INFO(16384, 16k), |
| 885 | INIT_KMALLOC_INFO(32768, 32k), |
| 886 | INIT_KMALLOC_INFO(65536, 64k), |
| 887 | INIT_KMALLOC_INFO(131072, 128k), |
| 888 | INIT_KMALLOC_INFO(262144, 256k), |
| 889 | INIT_KMALLOC_INFO(524288, 512k), |
| 890 | INIT_KMALLOC_INFO(1048576, 1M), |
| 891 | INIT_KMALLOC_INFO(2097152, 2M) |
| 892 | }; |
| 893 | |
| 894 | /* |
| 895 | * Patch up the size_index table if we have strange large alignment |
| 896 | * requirements for the kmalloc array. This is only the case for |
| 897 | * MIPS it seems. The standard arches will not generate any code here. |
| 898 | * |
| 899 | * Largest permitted alignment is 256 bytes due to the way we |
| 900 | * handle the index determination for the smaller caches. |
| 901 | * |
| 902 | * Make sure that nothing crazy happens if someone starts tinkering |
| 903 | * around with ARCH_KMALLOC_MINALIGN |
| 904 | */ |
| 905 | void __init setup_kmalloc_cache_index_table(void) |
| 906 | { |
| 907 | unsigned int i; |
| 908 | |
| 909 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
| 910 | !is_power_of_2(KMALLOC_MIN_SIZE)); |
| 911 | |
| 912 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
| 913 | unsigned int elem = size_index_elem(i); |
| 914 | |
| 915 | if (elem >= ARRAY_SIZE(kmalloc_size_index)) |
| 916 | break; |
| 917 | kmalloc_size_index[elem] = KMALLOC_SHIFT_LOW; |
| 918 | } |
| 919 | |
| 920 | if (KMALLOC_MIN_SIZE >= 64) { |
| 921 | /* |
| 922 | * The 96 byte sized cache is not used if the alignment |
| 923 | * is 64 byte. |
| 924 | */ |
| 925 | for (i = 64 + 8; i <= 96; i += 8) |
| 926 | kmalloc_size_index[size_index_elem(i)] = 7; |
| 927 | |
| 928 | } |
| 929 | |
| 930 | if (KMALLOC_MIN_SIZE >= 128) { |
| 931 | /* |
| 932 | * The 192 byte sized cache is not used if the alignment |
| 933 | * is 128 byte. Redirect kmalloc to use the 256 byte cache |
| 934 | * instead. |
| 935 | */ |
| 936 | for (i = 128 + 8; i <= 192; i += 8) |
| 937 | kmalloc_size_index[size_index_elem(i)] = 8; |
| 938 | } |
| 939 | } |
| 940 | |
| 941 | static unsigned int __kmalloc_minalign(void) |
| 942 | { |
| 943 | unsigned int minalign = dma_get_cache_alignment(); |
| 944 | |
| 945 | if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && |
| 946 | is_swiotlb_allocated()) |
| 947 | minalign = ARCH_KMALLOC_MINALIGN; |
| 948 | |
| 949 | return max(minalign, arch_slab_minalign()); |
| 950 | } |
| 951 | |
| 952 | static void __init |
| 953 | new_kmalloc_cache(int idx, enum kmalloc_cache_type type) |
| 954 | { |
| 955 | slab_flags_t flags = 0; |
| 956 | unsigned int minalign = __kmalloc_minalign(); |
| 957 | unsigned int aligned_size = kmalloc_info[idx].size; |
| 958 | int aligned_idx = idx; |
| 959 | |
| 960 | if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) { |
| 961 | flags |= SLAB_RECLAIM_ACCOUNT; |
| 962 | } else if (IS_ENABLED(CONFIG_MEMCG) && (type == KMALLOC_CGROUP)) { |
| 963 | if (mem_cgroup_kmem_disabled()) { |
| 964 | kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx]; |
| 965 | return; |
| 966 | } |
| 967 | flags |= SLAB_ACCOUNT; |
| 968 | } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) { |
| 969 | flags |= SLAB_CACHE_DMA; |
| 970 | } |
| 971 | |
| 972 | #ifdef CONFIG_RANDOM_KMALLOC_CACHES |
| 973 | if (type >= KMALLOC_RANDOM_START && type <= KMALLOC_RANDOM_END) |
| 974 | flags |= SLAB_NO_MERGE; |
| 975 | #endif |
| 976 | |
| 977 | /* |
| 978 | * If CONFIG_MEMCG is enabled, disable cache merging for |
| 979 | * KMALLOC_NORMAL caches. |
| 980 | */ |
| 981 | if (IS_ENABLED(CONFIG_MEMCG) && (type == KMALLOC_NORMAL)) |
| 982 | flags |= SLAB_NO_MERGE; |
| 983 | |
| 984 | if (minalign > ARCH_KMALLOC_MINALIGN) { |
| 985 | aligned_size = ALIGN(aligned_size, minalign); |
| 986 | aligned_idx = __kmalloc_index(aligned_size, false); |
| 987 | } |
| 988 | |
| 989 | if (!kmalloc_caches[type][aligned_idx]) |
| 990 | kmalloc_caches[type][aligned_idx] = create_kmalloc_cache( |
| 991 | kmalloc_info[aligned_idx].name[type], |
| 992 | aligned_size, flags); |
| 993 | if (idx != aligned_idx) |
| 994 | kmalloc_caches[type][idx] = kmalloc_caches[type][aligned_idx]; |
| 995 | } |
| 996 | |
| 997 | /* |
| 998 | * Create the kmalloc array. Some of the regular kmalloc arrays |
| 999 | * may already have been created because they were needed to |
| 1000 | * enable allocations for slab creation. |
| 1001 | */ |
| 1002 | void __init create_kmalloc_caches(void) |
| 1003 | { |
| 1004 | int i; |
| 1005 | enum kmalloc_cache_type type; |
| 1006 | |
| 1007 | /* |
| 1008 | * Including KMALLOC_CGROUP if CONFIG_MEMCG defined |
| 1009 | */ |
| 1010 | for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) { |
| 1011 | /* Caches that are NOT of the two-to-the-power-of size. */ |
| 1012 | if (KMALLOC_MIN_SIZE <= 32) |
| 1013 | new_kmalloc_cache(1, type); |
| 1014 | if (KMALLOC_MIN_SIZE <= 64) |
| 1015 | new_kmalloc_cache(2, type); |
| 1016 | |
| 1017 | /* Caches that are of the two-to-the-power-of size. */ |
| 1018 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) |
| 1019 | new_kmalloc_cache(i, type); |
| 1020 | } |
| 1021 | #ifdef CONFIG_RANDOM_KMALLOC_CACHES |
| 1022 | random_kmalloc_seed = get_random_u64(); |
| 1023 | #endif |
| 1024 | |
| 1025 | /* Kmalloc array is now usable */ |
| 1026 | slab_state = UP; |
| 1027 | |
| 1028 | if (IS_ENABLED(CONFIG_SLAB_BUCKETS)) |
| 1029 | kmem_buckets_cache = kmem_cache_create("kmalloc_buckets", |
| 1030 | sizeof(kmem_buckets), |
| 1031 | 0, SLAB_NO_MERGE, NULL); |
| 1032 | } |
| 1033 | |
| 1034 | /** |
| 1035 | * __ksize -- Report full size of underlying allocation |
| 1036 | * @object: pointer to the object |
| 1037 | * |
| 1038 | * This should only be used internally to query the true size of allocations. |
| 1039 | * It is not meant to be a way to discover the usable size of an allocation |
| 1040 | * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond |
| 1041 | * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS, |
| 1042 | * and/or FORTIFY_SOURCE. |
| 1043 | * |
| 1044 | * Return: size of the actual memory used by @object in bytes |
| 1045 | */ |
| 1046 | size_t __ksize(const void *object) |
| 1047 | { |
| 1048 | struct folio *folio; |
| 1049 | |
| 1050 | if (unlikely(object == ZERO_SIZE_PTR)) |
| 1051 | return 0; |
| 1052 | |
| 1053 | folio = virt_to_folio(object); |
| 1054 | |
| 1055 | if (unlikely(!folio_test_slab(folio))) { |
| 1056 | if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE)) |
| 1057 | return 0; |
| 1058 | if (WARN_ON(object != folio_address(folio))) |
| 1059 | return 0; |
| 1060 | return folio_size(folio); |
| 1061 | } |
| 1062 | |
| 1063 | #ifdef CONFIG_SLUB_DEBUG |
| 1064 | skip_orig_size_check(folio_slab(folio)->slab_cache, object); |
| 1065 | #endif |
| 1066 | |
| 1067 | return slab_ksize(folio_slab(folio)->slab_cache); |
| 1068 | } |
| 1069 | |
| 1070 | gfp_t kmalloc_fix_flags(gfp_t flags) |
| 1071 | { |
| 1072 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
| 1073 | |
| 1074 | flags &= ~GFP_SLAB_BUG_MASK; |
| 1075 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", |
| 1076 | invalid_mask, &invalid_mask, flags, &flags); |
| 1077 | dump_stack(); |
| 1078 | |
| 1079 | return flags; |
| 1080 | } |
| 1081 | |
| 1082 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
| 1083 | /* Randomize a generic freelist */ |
| 1084 | static void freelist_randomize(unsigned int *list, |
| 1085 | unsigned int count) |
| 1086 | { |
| 1087 | unsigned int rand; |
| 1088 | unsigned int i; |
| 1089 | |
| 1090 | for (i = 0; i < count; i++) |
| 1091 | list[i] = i; |
| 1092 | |
| 1093 | /* Fisher-Yates shuffle */ |
| 1094 | for (i = count - 1; i > 0; i--) { |
| 1095 | rand = get_random_u32_below(i + 1); |
| 1096 | swap(list[i], list[rand]); |
| 1097 | } |
| 1098 | } |
| 1099 | |
| 1100 | /* Create a random sequence per cache */ |
| 1101 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, |
| 1102 | gfp_t gfp) |
| 1103 | { |
| 1104 | |
| 1105 | if (count < 2 || cachep->random_seq) |
| 1106 | return 0; |
| 1107 | |
| 1108 | cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); |
| 1109 | if (!cachep->random_seq) |
| 1110 | return -ENOMEM; |
| 1111 | |
| 1112 | freelist_randomize(cachep->random_seq, count); |
| 1113 | return 0; |
| 1114 | } |
| 1115 | |
| 1116 | /* Destroy the per-cache random freelist sequence */ |
| 1117 | void cache_random_seq_destroy(struct kmem_cache *cachep) |
| 1118 | { |
| 1119 | kfree(cachep->random_seq); |
| 1120 | cachep->random_seq = NULL; |
| 1121 | } |
| 1122 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ |
| 1123 | |
| 1124 | #ifdef CONFIG_SLUB_DEBUG |
| 1125 | #define SLABINFO_RIGHTS (0400) |
| 1126 | |
| 1127 | static void print_slabinfo_header(struct seq_file *m) |
| 1128 | { |
| 1129 | /* |
| 1130 | * Output format version, so at least we can change it |
| 1131 | * without _too_ many complaints. |
| 1132 | */ |
| 1133 | seq_puts(m, "slabinfo - version: 2.1\n"); |
| 1134 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); |
| 1135 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
| 1136 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); |
| 1137 | seq_putc(m, '\n'); |
| 1138 | } |
| 1139 | |
| 1140 | static void *slab_start(struct seq_file *m, loff_t *pos) |
| 1141 | { |
| 1142 | mutex_lock(&slab_mutex); |
| 1143 | return seq_list_start(&slab_caches, *pos); |
| 1144 | } |
| 1145 | |
| 1146 | static void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
| 1147 | { |
| 1148 | return seq_list_next(p, &slab_caches, pos); |
| 1149 | } |
| 1150 | |
| 1151 | static void slab_stop(struct seq_file *m, void *p) |
| 1152 | { |
| 1153 | mutex_unlock(&slab_mutex); |
| 1154 | } |
| 1155 | |
| 1156 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
| 1157 | { |
| 1158 | struct slabinfo sinfo; |
| 1159 | |
| 1160 | memset(&sinfo, 0, sizeof(sinfo)); |
| 1161 | get_slabinfo(s, &sinfo); |
| 1162 | |
| 1163 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
| 1164 | s->name, sinfo.active_objs, sinfo.num_objs, s->size, |
| 1165 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
| 1166 | |
| 1167 | seq_printf(m, " : tunables %4u %4u %4u", |
| 1168 | sinfo.limit, sinfo.batchcount, sinfo.shared); |
| 1169 | seq_printf(m, " : slabdata %6lu %6lu %6lu", |
| 1170 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); |
| 1171 | seq_putc(m, '\n'); |
| 1172 | } |
| 1173 | |
| 1174 | static int slab_show(struct seq_file *m, void *p) |
| 1175 | { |
| 1176 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); |
| 1177 | |
| 1178 | if (p == slab_caches.next) |
| 1179 | print_slabinfo_header(m); |
| 1180 | cache_show(s, m); |
| 1181 | return 0; |
| 1182 | } |
| 1183 | |
| 1184 | void dump_unreclaimable_slab(void) |
| 1185 | { |
| 1186 | struct kmem_cache *s; |
| 1187 | struct slabinfo sinfo; |
| 1188 | |
| 1189 | /* |
| 1190 | * Here acquiring slab_mutex is risky since we don't prefer to get |
| 1191 | * sleep in oom path. But, without mutex hold, it may introduce a |
| 1192 | * risk of crash. |
| 1193 | * Use mutex_trylock to protect the list traverse, dump nothing |
| 1194 | * without acquiring the mutex. |
| 1195 | */ |
| 1196 | if (!mutex_trylock(&slab_mutex)) { |
| 1197 | pr_warn("excessive unreclaimable slab but cannot dump stats\n"); |
| 1198 | return; |
| 1199 | } |
| 1200 | |
| 1201 | pr_info("Unreclaimable slab info:\n"); |
| 1202 | pr_info("Name Used Total\n"); |
| 1203 | |
| 1204 | list_for_each_entry(s, &slab_caches, list) { |
| 1205 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
| 1206 | continue; |
| 1207 | |
| 1208 | get_slabinfo(s, &sinfo); |
| 1209 | |
| 1210 | if (sinfo.num_objs > 0) |
| 1211 | pr_info("%-17s %10luKB %10luKB\n", s->name, |
| 1212 | (sinfo.active_objs * s->size) / 1024, |
| 1213 | (sinfo.num_objs * s->size) / 1024); |
| 1214 | } |
| 1215 | mutex_unlock(&slab_mutex); |
| 1216 | } |
| 1217 | |
| 1218 | /* |
| 1219 | * slabinfo_op - iterator that generates /proc/slabinfo |
| 1220 | * |
| 1221 | * Output layout: |
| 1222 | * cache-name |
| 1223 | * num-active-objs |
| 1224 | * total-objs |
| 1225 | * object size |
| 1226 | * num-active-slabs |
| 1227 | * total-slabs |
| 1228 | * num-pages-per-slab |
| 1229 | * + further values on SMP and with statistics enabled |
| 1230 | */ |
| 1231 | static const struct seq_operations slabinfo_op = { |
| 1232 | .start = slab_start, |
| 1233 | .next = slab_next, |
| 1234 | .stop = slab_stop, |
| 1235 | .show = slab_show, |
| 1236 | }; |
| 1237 | |
| 1238 | static int slabinfo_open(struct inode *inode, struct file *file) |
| 1239 | { |
| 1240 | return seq_open(file, &slabinfo_op); |
| 1241 | } |
| 1242 | |
| 1243 | static const struct proc_ops slabinfo_proc_ops = { |
| 1244 | .proc_flags = PROC_ENTRY_PERMANENT, |
| 1245 | .proc_open = slabinfo_open, |
| 1246 | .proc_read = seq_read, |
| 1247 | .proc_lseek = seq_lseek, |
| 1248 | .proc_release = seq_release, |
| 1249 | }; |
| 1250 | |
| 1251 | static int __init slab_proc_init(void) |
| 1252 | { |
| 1253 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); |
| 1254 | return 0; |
| 1255 | } |
| 1256 | module_init(slab_proc_init); |
| 1257 | |
| 1258 | #endif /* CONFIG_SLUB_DEBUG */ |
| 1259 | |
| 1260 | static __always_inline __realloc_size(2) void * |
| 1261 | __do_krealloc(const void *p, size_t new_size, gfp_t flags) |
| 1262 | { |
| 1263 | void *ret; |
| 1264 | size_t ks; |
| 1265 | |
| 1266 | /* Check for double-free before calling ksize. */ |
| 1267 | if (likely(!ZERO_OR_NULL_PTR(p))) { |
| 1268 | if (!kasan_check_byte(p)) |
| 1269 | return NULL; |
| 1270 | ks = ksize(p); |
| 1271 | } else |
| 1272 | ks = 0; |
| 1273 | |
| 1274 | /* If the object still fits, repoison it precisely. */ |
| 1275 | if (ks >= new_size) { |
| 1276 | p = kasan_krealloc((void *)p, new_size, flags); |
| 1277 | return (void *)p; |
| 1278 | } |
| 1279 | |
| 1280 | ret = kmalloc_node_track_caller_noprof(new_size, flags, NUMA_NO_NODE, _RET_IP_); |
| 1281 | if (ret && p) { |
| 1282 | /* Disable KASAN checks as the object's redzone is accessed. */ |
| 1283 | kasan_disable_current(); |
| 1284 | memcpy(ret, kasan_reset_tag(p), ks); |
| 1285 | kasan_enable_current(); |
| 1286 | } |
| 1287 | |
| 1288 | return ret; |
| 1289 | } |
| 1290 | |
| 1291 | /** |
| 1292 | * krealloc - reallocate memory. The contents will remain unchanged. |
| 1293 | * @p: object to reallocate memory for. |
| 1294 | * @new_size: how many bytes of memory are required. |
| 1295 | * @flags: the type of memory to allocate. |
| 1296 | * |
| 1297 | * The contents of the object pointed to are preserved up to the |
| 1298 | * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored). |
| 1299 | * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size |
| 1300 | * is 0 and @p is not a %NULL pointer, the object pointed to is freed. |
| 1301 | * |
| 1302 | * Return: pointer to the allocated memory or %NULL in case of error |
| 1303 | */ |
| 1304 | void *krealloc_noprof(const void *p, size_t new_size, gfp_t flags) |
| 1305 | { |
| 1306 | void *ret; |
| 1307 | |
| 1308 | if (unlikely(!new_size)) { |
| 1309 | kfree(p); |
| 1310 | return ZERO_SIZE_PTR; |
| 1311 | } |
| 1312 | |
| 1313 | ret = __do_krealloc(p, new_size, flags); |
| 1314 | if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) |
| 1315 | kfree(p); |
| 1316 | |
| 1317 | return ret; |
| 1318 | } |
| 1319 | EXPORT_SYMBOL(krealloc_noprof); |
| 1320 | |
| 1321 | /** |
| 1322 | * kfree_sensitive - Clear sensitive information in memory before freeing |
| 1323 | * @p: object to free memory of |
| 1324 | * |
| 1325 | * The memory of the object @p points to is zeroed before freed. |
| 1326 | * If @p is %NULL, kfree_sensitive() does nothing. |
| 1327 | * |
| 1328 | * Note: this function zeroes the whole allocated buffer which can be a good |
| 1329 | * deal bigger than the requested buffer size passed to kmalloc(). So be |
| 1330 | * careful when using this function in performance sensitive code. |
| 1331 | */ |
| 1332 | void kfree_sensitive(const void *p) |
| 1333 | { |
| 1334 | size_t ks; |
| 1335 | void *mem = (void *)p; |
| 1336 | |
| 1337 | ks = ksize(mem); |
| 1338 | if (ks) { |
| 1339 | kasan_unpoison_range(mem, ks); |
| 1340 | memzero_explicit(mem, ks); |
| 1341 | } |
| 1342 | kfree(mem); |
| 1343 | } |
| 1344 | EXPORT_SYMBOL(kfree_sensitive); |
| 1345 | |
| 1346 | size_t ksize(const void *objp) |
| 1347 | { |
| 1348 | /* |
| 1349 | * We need to first check that the pointer to the object is valid. |
| 1350 | * The KASAN report printed from ksize() is more useful, then when |
| 1351 | * it's printed later when the behaviour could be undefined due to |
| 1352 | * a potential use-after-free or double-free. |
| 1353 | * |
| 1354 | * We use kasan_check_byte(), which is supported for the hardware |
| 1355 | * tag-based KASAN mode, unlike kasan_check_read/write(). |
| 1356 | * |
| 1357 | * If the pointed to memory is invalid, we return 0 to avoid users of |
| 1358 | * ksize() writing to and potentially corrupting the memory region. |
| 1359 | * |
| 1360 | * We want to perform the check before __ksize(), to avoid potentially |
| 1361 | * crashing in __ksize() due to accessing invalid metadata. |
| 1362 | */ |
| 1363 | if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp)) |
| 1364 | return 0; |
| 1365 | |
| 1366 | return kfence_ksize(objp) ?: __ksize(objp); |
| 1367 | } |
| 1368 | EXPORT_SYMBOL(ksize); |
| 1369 | |
| 1370 | /* Tracepoints definitions. */ |
| 1371 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); |
| 1372 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); |
| 1373 | EXPORT_TRACEPOINT_SYMBOL(kfree); |
| 1374 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); |
| 1375 | |