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 /* 44 * Set of flags that will prevent slab merging 45 */ 46 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ 47 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ 48 SLAB_FAILSLAB | SLAB_NO_MERGE) 49 50 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ 51 SLAB_CACHE_DMA32 | SLAB_ACCOUNT) 52 53 /* 54 * Merge control. If this is set then no merging of slab caches will occur. 55 */ 56 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); 57 58 static int __init setup_slab_nomerge(char *str) 59 { 60 slab_nomerge = true; 61 return 1; 62 } 63 64 static int __init setup_slab_merge(char *str) 65 { 66 slab_nomerge = false; 67 return 1; 68 } 69 70 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); 71 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0); 72 73 __setup("slab_nomerge", setup_slab_nomerge); 74 __setup("slab_merge", setup_slab_merge); 75 76 /* 77 * Determine the size of a slab object 78 */ 79 unsigned int kmem_cache_size(struct kmem_cache *s) 80 { 81 return s->object_size; 82 } 83 EXPORT_SYMBOL(kmem_cache_size); 84 85 #ifdef CONFIG_DEBUG_VM 86 87 static bool kmem_cache_is_duplicate_name(const char *name) 88 { 89 struct kmem_cache *s; 90 91 list_for_each_entry(s, &slab_caches, list) { 92 if (!strcmp(s->name, name)) 93 return true; 94 } 95 96 return false; 97 } 98 99 static int kmem_cache_sanity_check(const char *name, unsigned int size) 100 { 101 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) { 102 pr_err("kmem_cache_create(%s) integrity check failed\n", name); 103 return -EINVAL; 104 } 105 106 /* Duplicate names will confuse slabtop, et al */ 107 WARN(kmem_cache_is_duplicate_name(name), 108 "kmem_cache of name '%s' already exists\n", name); 109 110 WARN_ON(strchr(name, ' ')); /* It confuses parsers */ 111 return 0; 112 } 113 #else 114 static inline int kmem_cache_sanity_check(const char *name, unsigned int size) 115 { 116 return 0; 117 } 118 #endif 119 120 /* 121 * Figure out what the alignment of the objects will be given a set of 122 * flags, a user specified alignment and the size of the objects. 123 */ 124 static unsigned int calculate_alignment(slab_flags_t flags, 125 unsigned int align, unsigned int size) 126 { 127 /* 128 * If the user wants hardware cache aligned objects then follow that 129 * suggestion if the object is sufficiently large. 130 * 131 * The hardware cache alignment cannot override the specified 132 * alignment though. If that is greater then use it. 133 */ 134 if (flags & SLAB_HWCACHE_ALIGN) { 135 unsigned int ralign; 136 137 ralign = cache_line_size(); 138 while (size <= ralign / 2) 139 ralign /= 2; 140 align = max(align, ralign); 141 } 142 143 align = max(align, arch_slab_minalign()); 144 145 return ALIGN(align, sizeof(void *)); 146 } 147 148 /* 149 * Find a mergeable slab cache 150 */ 151 int slab_unmergeable(struct kmem_cache *s) 152 { 153 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) 154 return 1; 155 156 if (s->ctor) 157 return 1; 158 159 #ifdef CONFIG_HARDENED_USERCOPY 160 if (s->usersize) 161 return 1; 162 #endif 163 164 /* 165 * We may have set a slab to be unmergeable during bootstrap. 166 */ 167 if (s->refcount < 0) 168 return 1; 169 170 return 0; 171 } 172 173 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align, 174 slab_flags_t flags, const char *name, void (*ctor)(void *)) 175 { 176 struct kmem_cache *s; 177 178 if (slab_nomerge) 179 return NULL; 180 181 if (ctor) 182 return NULL; 183 184 flags = kmem_cache_flags(flags, name); 185 186 if (flags & SLAB_NEVER_MERGE) 187 return NULL; 188 189 size = ALIGN(size, sizeof(void *)); 190 align = calculate_alignment(flags, align, size); 191 size = ALIGN(size, align); 192 193 list_for_each_entry_reverse(s, &slab_caches, list) { 194 if (slab_unmergeable(s)) 195 continue; 196 197 if (size > s->size) 198 continue; 199 200 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) 201 continue; 202 /* 203 * Check if alignment is compatible. 204 * Courtesy of Adrian Drzewiecki 205 */ 206 if ((s->size & ~(align - 1)) != s->size) 207 continue; 208 209 if (s->size - size >= sizeof(void *)) 210 continue; 211 212 return s; 213 } 214 return NULL; 215 } 216 217 static struct kmem_cache *create_cache(const char *name, 218 unsigned int object_size, 219 struct kmem_cache_args *args, 220 slab_flags_t flags) 221 { 222 struct kmem_cache *s; 223 int err; 224 225 /* If a custom freelist pointer is requested make sure it's sane. */ 226 err = -EINVAL; 227 if (args->use_freeptr_offset && 228 (args->freeptr_offset >= object_size || 229 !(flags & SLAB_TYPESAFE_BY_RCU) || 230 !IS_ALIGNED(args->freeptr_offset, __alignof__(freeptr_t)))) 231 goto out; 232 233 err = -ENOMEM; 234 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); 235 if (!s) 236 goto out; 237 err = do_kmem_cache_create(s, name, object_size, args, flags); 238 if (err) 239 goto out_free_cache; 240 241 s->refcount = 1; 242 list_add(&s->list, &slab_caches); 243 return s; 244 245 out_free_cache: 246 kmem_cache_free(kmem_cache, s); 247 out: 248 return ERR_PTR(err); 249 } 250 251 /** 252 * __kmem_cache_create_args - Create a kmem cache. 253 * @name: A string which is used in /proc/slabinfo to identify this cache. 254 * @object_size: The size of objects to be created in this cache. 255 * @args: Additional arguments for the cache creation (see 256 * &struct kmem_cache_args). 257 * @flags: See the desriptions of individual flags. The common ones are listed 258 * in the description below. 259 * 260 * Not to be called directly, use the kmem_cache_create() wrapper with the same 261 * parameters. 262 * 263 * Commonly used @flags: 264 * 265 * &SLAB_ACCOUNT - Account allocations to memcg. 266 * 267 * &SLAB_HWCACHE_ALIGN - Align objects on cache line boundaries. 268 * 269 * &SLAB_RECLAIM_ACCOUNT - Objects are reclaimable. 270 * 271 * &SLAB_TYPESAFE_BY_RCU - Slab page (not individual objects) freeing delayed 272 * by a grace period - see the full description before using. 273 * 274 * Context: Cannot be called within a interrupt, but can be interrupted. 275 * 276 * Return: a pointer to the cache on success, NULL on failure. 277 */ 278 struct kmem_cache *__kmem_cache_create_args(const char *name, 279 unsigned int object_size, 280 struct kmem_cache_args *args, 281 slab_flags_t flags) 282 { 283 struct kmem_cache *s = NULL; 284 const char *cache_name; 285 int err; 286 287 #ifdef CONFIG_SLUB_DEBUG 288 /* 289 * If no slab_debug was enabled globally, the static key is not yet 290 * enabled by setup_slub_debug(). Enable it if the cache is being 291 * created with any of the debugging flags passed explicitly. 292 * It's also possible that this is the first cache created with 293 * SLAB_STORE_USER and we should init stack_depot for it. 294 */ 295 if (flags & SLAB_DEBUG_FLAGS) 296 static_branch_enable(&slub_debug_enabled); 297 if (flags & SLAB_STORE_USER) 298 stack_depot_init(); 299 #endif 300 301 mutex_lock(&slab_mutex); 302 303 err = kmem_cache_sanity_check(name, object_size); 304 if (err) { 305 goto out_unlock; 306 } 307 308 /* Refuse requests with allocator specific flags */ 309 if (flags & ~SLAB_FLAGS_PERMITTED) { 310 err = -EINVAL; 311 goto out_unlock; 312 } 313 314 /* 315 * Some allocators will constraint the set of valid flags to a subset 316 * of all flags. We expect them to define CACHE_CREATE_MASK in this 317 * case, and we'll just provide them with a sanitized version of the 318 * passed flags. 319 */ 320 flags &= CACHE_CREATE_MASK; 321 322 /* Fail closed on bad usersize of useroffset values. */ 323 if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) || 324 WARN_ON(!args->usersize && args->useroffset) || 325 WARN_ON(object_size < args->usersize || 326 object_size - args->usersize < args->useroffset)) 327 args->usersize = args->useroffset = 0; 328 329 if (!args->usersize) 330 s = __kmem_cache_alias(name, object_size, args->align, flags, 331 args->ctor); 332 if (s) 333 goto out_unlock; 334 335 cache_name = kstrdup_const(name, GFP_KERNEL); 336 if (!cache_name) { 337 err = -ENOMEM; 338 goto out_unlock; 339 } 340 341 args->align = calculate_alignment(flags, args->align, object_size); 342 s = create_cache(cache_name, object_size, args, flags); 343 if (IS_ERR(s)) { 344 err = PTR_ERR(s); 345 kfree_const(cache_name); 346 } 347 348 out_unlock: 349 mutex_unlock(&slab_mutex); 350 351 if (err) { 352 if (flags & SLAB_PANIC) 353 panic("%s: Failed to create slab '%s'. Error %d\n", 354 __func__, name, err); 355 else { 356 pr_warn("%s(%s) failed with error %d\n", 357 __func__, name, err); 358 dump_stack(); 359 } 360 return NULL; 361 } 362 return s; 363 } 364 EXPORT_SYMBOL(__kmem_cache_create_args); 365 366 static struct kmem_cache *kmem_buckets_cache __ro_after_init; 367 368 /** 369 * kmem_buckets_create - Create a set of caches that handle dynamic sized 370 * allocations via kmem_buckets_alloc() 371 * @name: A prefix string which is used in /proc/slabinfo to identify this 372 * cache. The individual caches with have their sizes as the suffix. 373 * @flags: SLAB flags (see kmem_cache_create() for details). 374 * @useroffset: Starting offset within an allocation that may be copied 375 * to/from userspace. 376 * @usersize: How many bytes, starting at @useroffset, may be copied 377 * to/from userspace. 378 * @ctor: A constructor for the objects, run when new allocations are made. 379 * 380 * Cannot be called within an interrupt, but can be interrupted. 381 * 382 * Return: a pointer to the cache on success, NULL on failure. When 383 * CONFIG_SLAB_BUCKETS is not enabled, ZERO_SIZE_PTR is returned, and 384 * subsequent calls to kmem_buckets_alloc() will fall back to kmalloc(). 385 * (i.e. callers only need to check for NULL on failure.) 386 */ 387 kmem_buckets *kmem_buckets_create(const char *name, slab_flags_t flags, 388 unsigned int useroffset, 389 unsigned int usersize, 390 void (*ctor)(void *)) 391 { 392 unsigned long mask = 0; 393 unsigned int idx; 394 kmem_buckets *b; 395 396 BUILD_BUG_ON(ARRAY_SIZE(kmalloc_caches[KMALLOC_NORMAL]) > BITS_PER_LONG); 397 398 /* 399 * When the separate buckets API is not built in, just return 400 * a non-NULL value for the kmem_buckets pointer, which will be 401 * unused when performing allocations. 402 */ 403 if (!IS_ENABLED(CONFIG_SLAB_BUCKETS)) 404 return ZERO_SIZE_PTR; 405 406 if (WARN_ON(!kmem_buckets_cache)) 407 return NULL; 408 409 b = kmem_cache_alloc(kmem_buckets_cache, GFP_KERNEL|__GFP_ZERO); 410 if (WARN_ON(!b)) 411 return NULL; 412 413 flags |= SLAB_NO_MERGE; 414 415 for (idx = 0; idx < ARRAY_SIZE(kmalloc_caches[KMALLOC_NORMAL]); idx++) { 416 char *short_size, *cache_name; 417 unsigned int cache_useroffset, cache_usersize; 418 unsigned int size, aligned_idx; 419 420 if (!kmalloc_caches[KMALLOC_NORMAL][idx]) 421 continue; 422 423 size = kmalloc_caches[KMALLOC_NORMAL][idx]->object_size; 424 if (!size) 425 continue; 426 427 short_size = strchr(kmalloc_caches[KMALLOC_NORMAL][idx]->name, '-'); 428 if (WARN_ON(!short_size)) 429 goto fail; 430 431 if (useroffset >= size) { 432 cache_useroffset = 0; 433 cache_usersize = 0; 434 } else { 435 cache_useroffset = useroffset; 436 cache_usersize = min(size - cache_useroffset, usersize); 437 } 438 439 aligned_idx = __kmalloc_index(size, false); 440 if (!(*b)[aligned_idx]) { 441 cache_name = kasprintf(GFP_KERNEL, "%s-%s", name, short_size + 1); 442 if (WARN_ON(!cache_name)) 443 goto fail; 444 (*b)[aligned_idx] = kmem_cache_create_usercopy(cache_name, size, 445 0, flags, cache_useroffset, 446 cache_usersize, ctor); 447 kfree(cache_name); 448 if (WARN_ON(!(*b)[aligned_idx])) 449 goto fail; 450 set_bit(aligned_idx, &mask); 451 } 452 if (idx != aligned_idx) 453 (*b)[idx] = (*b)[aligned_idx]; 454 } 455 456 return b; 457 458 fail: 459 for_each_set_bit(idx, &mask, ARRAY_SIZE(kmalloc_caches[KMALLOC_NORMAL])) 460 kmem_cache_destroy((*b)[idx]); 461 kmem_cache_free(kmem_buckets_cache, b); 462 463 return NULL; 464 } 465 EXPORT_SYMBOL(kmem_buckets_create); 466 467 /* 468 * For a given kmem_cache, kmem_cache_destroy() should only be called 469 * once or there will be a use-after-free problem. The actual deletion 470 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock 471 * protection. So they are now done without holding those locks. 472 */ 473 static void kmem_cache_release(struct kmem_cache *s) 474 { 475 kfence_shutdown_cache(s); 476 if (__is_defined(SLAB_SUPPORTS_SYSFS) && slab_state >= FULL) 477 sysfs_slab_release(s); 478 else 479 slab_kmem_cache_release(s); 480 } 481 482 void slab_kmem_cache_release(struct kmem_cache *s) 483 { 484 __kmem_cache_release(s); 485 kfree_const(s->name); 486 kmem_cache_free(kmem_cache, s); 487 } 488 489 void kmem_cache_destroy(struct kmem_cache *s) 490 { 491 int err; 492 493 if (unlikely(!s) || !kasan_check_byte(s)) 494 return; 495 496 /* in-flight kfree_rcu()'s may include objects from our cache */ 497 kvfree_rcu_barrier(); 498 499 if (IS_ENABLED(CONFIG_SLUB_RCU_DEBUG) && 500 (s->flags & SLAB_TYPESAFE_BY_RCU)) { 501 /* 502 * Under CONFIG_SLUB_RCU_DEBUG, when objects in a 503 * SLAB_TYPESAFE_BY_RCU slab are freed, SLUB will internally 504 * defer their freeing with call_rcu(). 505 * Wait for such call_rcu() invocations here before actually 506 * destroying the cache. 507 * 508 * It doesn't matter that we haven't looked at the slab refcount 509 * yet - slabs with SLAB_TYPESAFE_BY_RCU can't be merged, so 510 * the refcount should be 1 here. 511 */ 512 rcu_barrier(); 513 } 514 515 cpus_read_lock(); 516 mutex_lock(&slab_mutex); 517 518 s->refcount--; 519 if (s->refcount) { 520 mutex_unlock(&slab_mutex); 521 cpus_read_unlock(); 522 return; 523 } 524 525 /* free asan quarantined objects */ 526 kasan_cache_shutdown(s); 527 528 err = __kmem_cache_shutdown(s); 529 if (!slab_in_kunit_test()) 530 WARN(err, "%s %s: Slab cache still has objects when called from %pS", 531 __func__, s->name, (void *)_RET_IP_); 532 533 list_del(&s->list); 534 535 mutex_unlock(&slab_mutex); 536 cpus_read_unlock(); 537 538 if (slab_state >= FULL) 539 sysfs_slab_unlink(s); 540 debugfs_slab_release(s); 541 542 if (err) 543 return; 544 545 if (s->flags & SLAB_TYPESAFE_BY_RCU) 546 rcu_barrier(); 547 548 kmem_cache_release(s); 549 } 550 EXPORT_SYMBOL(kmem_cache_destroy); 551 552 /** 553 * kmem_cache_shrink - Shrink a cache. 554 * @cachep: The cache to shrink. 555 * 556 * Releases as many slabs as possible for a cache. 557 * To help debugging, a zero exit status indicates all slabs were released. 558 * 559 * Return: %0 if all slabs were released, non-zero otherwise 560 */ 561 int kmem_cache_shrink(struct kmem_cache *cachep) 562 { 563 kasan_cache_shrink(cachep); 564 565 return __kmem_cache_shrink(cachep); 566 } 567 EXPORT_SYMBOL(kmem_cache_shrink); 568 569 bool slab_is_available(void) 570 { 571 return slab_state >= UP; 572 } 573 574 #ifdef CONFIG_PRINTK 575 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab) 576 { 577 if (__kfence_obj_info(kpp, object, slab)) 578 return; 579 __kmem_obj_info(kpp, object, slab); 580 } 581 582 /** 583 * kmem_dump_obj - Print available slab provenance information 584 * @object: slab object for which to find provenance information. 585 * 586 * This function uses pr_cont(), so that the caller is expected to have 587 * printed out whatever preamble is appropriate. The provenance information 588 * depends on the type of object and on how much debugging is enabled. 589 * For a slab-cache object, the fact that it is a slab object is printed, 590 * and, if available, the slab name, return address, and stack trace from 591 * the allocation and last free path of that object. 592 * 593 * Return: %true if the pointer is to a not-yet-freed object from 594 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer 595 * is to an already-freed object, and %false otherwise. 596 */ 597 bool kmem_dump_obj(void *object) 598 { 599 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc"; 600 int i; 601 struct slab *slab; 602 unsigned long ptroffset; 603 struct kmem_obj_info kp = { }; 604 605 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */ 606 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object)) 607 return false; 608 slab = virt_to_slab(object); 609 if (!slab) 610 return false; 611 612 kmem_obj_info(&kp, object, slab); 613 if (kp.kp_slab_cache) 614 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name); 615 else 616 pr_cont(" slab%s", cp); 617 if (is_kfence_address(object)) 618 pr_cont(" (kfence)"); 619 if (kp.kp_objp) 620 pr_cont(" start %px", kp.kp_objp); 621 if (kp.kp_data_offset) 622 pr_cont(" data offset %lu", kp.kp_data_offset); 623 if (kp.kp_objp) { 624 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset; 625 pr_cont(" pointer offset %lu", ptroffset); 626 } 627 if (kp.kp_slab_cache && kp.kp_slab_cache->object_size) 628 pr_cont(" size %u", kp.kp_slab_cache->object_size); 629 if (kp.kp_ret) 630 pr_cont(" allocated at %pS\n", kp.kp_ret); 631 else 632 pr_cont("\n"); 633 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) { 634 if (!kp.kp_stack[i]) 635 break; 636 pr_info(" %pS\n", kp.kp_stack[i]); 637 } 638 639 if (kp.kp_free_stack[0]) 640 pr_cont(" Free path:\n"); 641 642 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) { 643 if (!kp.kp_free_stack[i]) 644 break; 645 pr_info(" %pS\n", kp.kp_free_stack[i]); 646 } 647 648 return true; 649 } 650 EXPORT_SYMBOL_GPL(kmem_dump_obj); 651 #endif 652 653 /* Create a cache during boot when no slab services are available yet */ 654 void __init create_boot_cache(struct kmem_cache *s, const char *name, 655 unsigned int size, slab_flags_t flags, 656 unsigned int useroffset, unsigned int usersize) 657 { 658 int err; 659 unsigned int align = ARCH_KMALLOC_MINALIGN; 660 struct kmem_cache_args kmem_args = {}; 661 662 /* 663 * kmalloc caches guarantee alignment of at least the largest 664 * power-of-two divisor of the size. For power-of-two sizes, 665 * it is the size itself. 666 */ 667 if (flags & SLAB_KMALLOC) 668 align = max(align, 1U << (ffs(size) - 1)); 669 kmem_args.align = calculate_alignment(flags, align, size); 670 671 #ifdef CONFIG_HARDENED_USERCOPY 672 kmem_args.useroffset = useroffset; 673 kmem_args.usersize = usersize; 674 #endif 675 676 err = do_kmem_cache_create(s, name, size, &kmem_args, flags); 677 678 if (err) 679 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", 680 name, size, err); 681 682 s->refcount = -1; /* Exempt from merging for now */ 683 } 684 685 static struct kmem_cache *__init create_kmalloc_cache(const char *name, 686 unsigned int size, 687 slab_flags_t flags) 688 { 689 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); 690 691 if (!s) 692 panic("Out of memory when creating slab %s\n", name); 693 694 create_boot_cache(s, name, size, flags | SLAB_KMALLOC, 0, size); 695 list_add(&s->list, &slab_caches); 696 s->refcount = 1; 697 return s; 698 } 699 700 kmem_buckets kmalloc_caches[NR_KMALLOC_TYPES] __ro_after_init = 701 { /* initialization for https://llvm.org/pr42570 */ }; 702 EXPORT_SYMBOL(kmalloc_caches); 703 704 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 705 unsigned long random_kmalloc_seed __ro_after_init; 706 EXPORT_SYMBOL(random_kmalloc_seed); 707 #endif 708 709 /* 710 * Conversion table for small slabs sizes / 8 to the index in the 711 * kmalloc array. This is necessary for slabs < 192 since we have non power 712 * of two cache sizes there. The size of larger slabs can be determined using 713 * fls. 714 */ 715 u8 kmalloc_size_index[24] __ro_after_init = { 716 3, /* 8 */ 717 4, /* 16 */ 718 5, /* 24 */ 719 5, /* 32 */ 720 6, /* 40 */ 721 6, /* 48 */ 722 6, /* 56 */ 723 6, /* 64 */ 724 1, /* 72 */ 725 1, /* 80 */ 726 1, /* 88 */ 727 1, /* 96 */ 728 7, /* 104 */ 729 7, /* 112 */ 730 7, /* 120 */ 731 7, /* 128 */ 732 2, /* 136 */ 733 2, /* 144 */ 734 2, /* 152 */ 735 2, /* 160 */ 736 2, /* 168 */ 737 2, /* 176 */ 738 2, /* 184 */ 739 2 /* 192 */ 740 }; 741 742 size_t kmalloc_size_roundup(size_t size) 743 { 744 if (size && size <= KMALLOC_MAX_CACHE_SIZE) { 745 /* 746 * The flags don't matter since size_index is common to all. 747 * Neither does the caller for just getting ->object_size. 748 */ 749 return kmalloc_slab(size, NULL, GFP_KERNEL, 0)->object_size; 750 } 751 752 /* Above the smaller buckets, size is a multiple of page size. */ 753 if (size && size <= KMALLOC_MAX_SIZE) 754 return PAGE_SIZE << get_order(size); 755 756 /* 757 * Return 'size' for 0 - kmalloc() returns ZERO_SIZE_PTR 758 * and very large size - kmalloc() may fail. 759 */ 760 return size; 761 762 } 763 EXPORT_SYMBOL(kmalloc_size_roundup); 764 765 #ifdef CONFIG_ZONE_DMA 766 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz, 767 #else 768 #define KMALLOC_DMA_NAME(sz) 769 #endif 770 771 #ifdef CONFIG_MEMCG 772 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz, 773 #else 774 #define KMALLOC_CGROUP_NAME(sz) 775 #endif 776 777 #ifndef CONFIG_SLUB_TINY 778 #define KMALLOC_RCL_NAME(sz) .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz, 779 #else 780 #define KMALLOC_RCL_NAME(sz) 781 #endif 782 783 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 784 #define __KMALLOC_RANDOM_CONCAT(a, b) a ## b 785 #define KMALLOC_RANDOM_NAME(N, sz) __KMALLOC_RANDOM_CONCAT(KMA_RAND_, N)(sz) 786 #define KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 1] = "kmalloc-rnd-01-" #sz, 787 #define KMA_RAND_2(sz) KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 2] = "kmalloc-rnd-02-" #sz, 788 #define KMA_RAND_3(sz) KMA_RAND_2(sz) .name[KMALLOC_RANDOM_START + 3] = "kmalloc-rnd-03-" #sz, 789 #define KMA_RAND_4(sz) KMA_RAND_3(sz) .name[KMALLOC_RANDOM_START + 4] = "kmalloc-rnd-04-" #sz, 790 #define KMA_RAND_5(sz) KMA_RAND_4(sz) .name[KMALLOC_RANDOM_START + 5] = "kmalloc-rnd-05-" #sz, 791 #define KMA_RAND_6(sz) KMA_RAND_5(sz) .name[KMALLOC_RANDOM_START + 6] = "kmalloc-rnd-06-" #sz, 792 #define KMA_RAND_7(sz) KMA_RAND_6(sz) .name[KMALLOC_RANDOM_START + 7] = "kmalloc-rnd-07-" #sz, 793 #define KMA_RAND_8(sz) KMA_RAND_7(sz) .name[KMALLOC_RANDOM_START + 8] = "kmalloc-rnd-08-" #sz, 794 #define KMA_RAND_9(sz) KMA_RAND_8(sz) .name[KMALLOC_RANDOM_START + 9] = "kmalloc-rnd-09-" #sz, 795 #define KMA_RAND_10(sz) KMA_RAND_9(sz) .name[KMALLOC_RANDOM_START + 10] = "kmalloc-rnd-10-" #sz, 796 #define KMA_RAND_11(sz) KMA_RAND_10(sz) .name[KMALLOC_RANDOM_START + 11] = "kmalloc-rnd-11-" #sz, 797 #define KMA_RAND_12(sz) KMA_RAND_11(sz) .name[KMALLOC_RANDOM_START + 12] = "kmalloc-rnd-12-" #sz, 798 #define KMA_RAND_13(sz) KMA_RAND_12(sz) .name[KMALLOC_RANDOM_START + 13] = "kmalloc-rnd-13-" #sz, 799 #define KMA_RAND_14(sz) KMA_RAND_13(sz) .name[KMALLOC_RANDOM_START + 14] = "kmalloc-rnd-14-" #sz, 800 #define KMA_RAND_15(sz) KMA_RAND_14(sz) .name[KMALLOC_RANDOM_START + 15] = "kmalloc-rnd-15-" #sz, 801 #else // CONFIG_RANDOM_KMALLOC_CACHES 802 #define KMALLOC_RANDOM_NAME(N, sz) 803 #endif 804 805 #define INIT_KMALLOC_INFO(__size, __short_size) \ 806 { \ 807 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ 808 KMALLOC_RCL_NAME(__short_size) \ 809 KMALLOC_CGROUP_NAME(__short_size) \ 810 KMALLOC_DMA_NAME(__short_size) \ 811 KMALLOC_RANDOM_NAME(RANDOM_KMALLOC_CACHES_NR, __short_size) \ 812 .size = __size, \ 813 } 814 815 /* 816 * kmalloc_info[] is to make slab_debug=,kmalloc-xx option work at boot time. 817 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is 818 * kmalloc-2M. 819 */ 820 const struct kmalloc_info_struct kmalloc_info[] __initconst = { 821 INIT_KMALLOC_INFO(0, 0), 822 INIT_KMALLOC_INFO(96, 96), 823 INIT_KMALLOC_INFO(192, 192), 824 INIT_KMALLOC_INFO(8, 8), 825 INIT_KMALLOC_INFO(16, 16), 826 INIT_KMALLOC_INFO(32, 32), 827 INIT_KMALLOC_INFO(64, 64), 828 INIT_KMALLOC_INFO(128, 128), 829 INIT_KMALLOC_INFO(256, 256), 830 INIT_KMALLOC_INFO(512, 512), 831 INIT_KMALLOC_INFO(1024, 1k), 832 INIT_KMALLOC_INFO(2048, 2k), 833 INIT_KMALLOC_INFO(4096, 4k), 834 INIT_KMALLOC_INFO(8192, 8k), 835 INIT_KMALLOC_INFO(16384, 16k), 836 INIT_KMALLOC_INFO(32768, 32k), 837 INIT_KMALLOC_INFO(65536, 64k), 838 INIT_KMALLOC_INFO(131072, 128k), 839 INIT_KMALLOC_INFO(262144, 256k), 840 INIT_KMALLOC_INFO(524288, 512k), 841 INIT_KMALLOC_INFO(1048576, 1M), 842 INIT_KMALLOC_INFO(2097152, 2M) 843 }; 844 845 /* 846 * Patch up the size_index table if we have strange large alignment 847 * requirements for the kmalloc array. This is only the case for 848 * MIPS it seems. The standard arches will not generate any code here. 849 * 850 * Largest permitted alignment is 256 bytes due to the way we 851 * handle the index determination for the smaller caches. 852 * 853 * Make sure that nothing crazy happens if someone starts tinkering 854 * around with ARCH_KMALLOC_MINALIGN 855 */ 856 void __init setup_kmalloc_cache_index_table(void) 857 { 858 unsigned int i; 859 860 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || 861 !is_power_of_2(KMALLOC_MIN_SIZE)); 862 863 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { 864 unsigned int elem = size_index_elem(i); 865 866 if (elem >= ARRAY_SIZE(kmalloc_size_index)) 867 break; 868 kmalloc_size_index[elem] = KMALLOC_SHIFT_LOW; 869 } 870 871 if (KMALLOC_MIN_SIZE >= 64) { 872 /* 873 * The 96 byte sized cache is not used if the alignment 874 * is 64 byte. 875 */ 876 for (i = 64 + 8; i <= 96; i += 8) 877 kmalloc_size_index[size_index_elem(i)] = 7; 878 879 } 880 881 if (KMALLOC_MIN_SIZE >= 128) { 882 /* 883 * The 192 byte sized cache is not used if the alignment 884 * is 128 byte. Redirect kmalloc to use the 256 byte cache 885 * instead. 886 */ 887 for (i = 128 + 8; i <= 192; i += 8) 888 kmalloc_size_index[size_index_elem(i)] = 8; 889 } 890 } 891 892 static unsigned int __kmalloc_minalign(void) 893 { 894 unsigned int minalign = dma_get_cache_alignment(); 895 896 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && 897 is_swiotlb_allocated()) 898 minalign = ARCH_KMALLOC_MINALIGN; 899 900 return max(minalign, arch_slab_minalign()); 901 } 902 903 static void __init 904 new_kmalloc_cache(int idx, enum kmalloc_cache_type type) 905 { 906 slab_flags_t flags = 0; 907 unsigned int minalign = __kmalloc_minalign(); 908 unsigned int aligned_size = kmalloc_info[idx].size; 909 int aligned_idx = idx; 910 911 if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) { 912 flags |= SLAB_RECLAIM_ACCOUNT; 913 } else if (IS_ENABLED(CONFIG_MEMCG) && (type == KMALLOC_CGROUP)) { 914 if (mem_cgroup_kmem_disabled()) { 915 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx]; 916 return; 917 } 918 flags |= SLAB_ACCOUNT; 919 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) { 920 flags |= SLAB_CACHE_DMA; 921 } 922 923 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 924 if (type >= KMALLOC_RANDOM_START && type <= KMALLOC_RANDOM_END) 925 flags |= SLAB_NO_MERGE; 926 #endif 927 928 /* 929 * If CONFIG_MEMCG is enabled, disable cache merging for 930 * KMALLOC_NORMAL caches. 931 */ 932 if (IS_ENABLED(CONFIG_MEMCG) && (type == KMALLOC_NORMAL)) 933 flags |= SLAB_NO_MERGE; 934 935 if (minalign > ARCH_KMALLOC_MINALIGN) { 936 aligned_size = ALIGN(aligned_size, minalign); 937 aligned_idx = __kmalloc_index(aligned_size, false); 938 } 939 940 if (!kmalloc_caches[type][aligned_idx]) 941 kmalloc_caches[type][aligned_idx] = create_kmalloc_cache( 942 kmalloc_info[aligned_idx].name[type], 943 aligned_size, flags); 944 if (idx != aligned_idx) 945 kmalloc_caches[type][idx] = kmalloc_caches[type][aligned_idx]; 946 } 947 948 /* 949 * Create the kmalloc array. Some of the regular kmalloc arrays 950 * may already have been created because they were needed to 951 * enable allocations for slab creation. 952 */ 953 void __init create_kmalloc_caches(void) 954 { 955 int i; 956 enum kmalloc_cache_type type; 957 958 /* 959 * Including KMALLOC_CGROUP if CONFIG_MEMCG defined 960 */ 961 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) { 962 /* Caches that are NOT of the two-to-the-power-of size. */ 963 if (KMALLOC_MIN_SIZE <= 32) 964 new_kmalloc_cache(1, type); 965 if (KMALLOC_MIN_SIZE <= 64) 966 new_kmalloc_cache(2, type); 967 968 /* Caches that are of the two-to-the-power-of size. */ 969 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) 970 new_kmalloc_cache(i, type); 971 } 972 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 973 random_kmalloc_seed = get_random_u64(); 974 #endif 975 976 /* Kmalloc array is now usable */ 977 slab_state = UP; 978 979 if (IS_ENABLED(CONFIG_SLAB_BUCKETS)) 980 kmem_buckets_cache = kmem_cache_create("kmalloc_buckets", 981 sizeof(kmem_buckets), 982 0, SLAB_NO_MERGE, NULL); 983 } 984 985 /** 986 * __ksize -- Report full size of underlying allocation 987 * @object: pointer to the object 988 * 989 * This should only be used internally to query the true size of allocations. 990 * It is not meant to be a way to discover the usable size of an allocation 991 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond 992 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS, 993 * and/or FORTIFY_SOURCE. 994 * 995 * Return: size of the actual memory used by @object in bytes 996 */ 997 size_t __ksize(const void *object) 998 { 999 struct folio *folio; 1000 1001 if (unlikely(object == ZERO_SIZE_PTR)) 1002 return 0; 1003 1004 folio = virt_to_folio(object); 1005 1006 if (unlikely(!folio_test_slab(folio))) { 1007 if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE)) 1008 return 0; 1009 if (WARN_ON(object != folio_address(folio))) 1010 return 0; 1011 return folio_size(folio); 1012 } 1013 1014 #ifdef CONFIG_SLUB_DEBUG 1015 skip_orig_size_check(folio_slab(folio)->slab_cache, object); 1016 #endif 1017 1018 return slab_ksize(folio_slab(folio)->slab_cache); 1019 } 1020 1021 gfp_t kmalloc_fix_flags(gfp_t flags) 1022 { 1023 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; 1024 1025 flags &= ~GFP_SLAB_BUG_MASK; 1026 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", 1027 invalid_mask, &invalid_mask, flags, &flags); 1028 dump_stack(); 1029 1030 return flags; 1031 } 1032 1033 #ifdef CONFIG_SLAB_FREELIST_RANDOM 1034 /* Randomize a generic freelist */ 1035 static void freelist_randomize(unsigned int *list, 1036 unsigned int count) 1037 { 1038 unsigned int rand; 1039 unsigned int i; 1040 1041 for (i = 0; i < count; i++) 1042 list[i] = i; 1043 1044 /* Fisher-Yates shuffle */ 1045 for (i = count - 1; i > 0; i--) { 1046 rand = get_random_u32_below(i + 1); 1047 swap(list[i], list[rand]); 1048 } 1049 } 1050 1051 /* Create a random sequence per cache */ 1052 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, 1053 gfp_t gfp) 1054 { 1055 1056 if (count < 2 || cachep->random_seq) 1057 return 0; 1058 1059 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); 1060 if (!cachep->random_seq) 1061 return -ENOMEM; 1062 1063 freelist_randomize(cachep->random_seq, count); 1064 return 0; 1065 } 1066 1067 /* Destroy the per-cache random freelist sequence */ 1068 void cache_random_seq_destroy(struct kmem_cache *cachep) 1069 { 1070 kfree(cachep->random_seq); 1071 cachep->random_seq = NULL; 1072 } 1073 #endif /* CONFIG_SLAB_FREELIST_RANDOM */ 1074 1075 #ifdef CONFIG_SLUB_DEBUG 1076 #define SLABINFO_RIGHTS (0400) 1077 1078 static void print_slabinfo_header(struct seq_file *m) 1079 { 1080 /* 1081 * Output format version, so at least we can change it 1082 * without _too_ many complaints. 1083 */ 1084 seq_puts(m, "slabinfo - version: 2.1\n"); 1085 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); 1086 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); 1087 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); 1088 seq_putc(m, '\n'); 1089 } 1090 1091 static void *slab_start(struct seq_file *m, loff_t *pos) 1092 { 1093 mutex_lock(&slab_mutex); 1094 return seq_list_start(&slab_caches, *pos); 1095 } 1096 1097 static void *slab_next(struct seq_file *m, void *p, loff_t *pos) 1098 { 1099 return seq_list_next(p, &slab_caches, pos); 1100 } 1101 1102 static void slab_stop(struct seq_file *m, void *p) 1103 { 1104 mutex_unlock(&slab_mutex); 1105 } 1106 1107 static void cache_show(struct kmem_cache *s, struct seq_file *m) 1108 { 1109 struct slabinfo sinfo; 1110 1111 memset(&sinfo, 0, sizeof(sinfo)); 1112 get_slabinfo(s, &sinfo); 1113 1114 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", 1115 s->name, sinfo.active_objs, sinfo.num_objs, s->size, 1116 sinfo.objects_per_slab, (1 << sinfo.cache_order)); 1117 1118 seq_printf(m, " : tunables %4u %4u %4u", 1119 sinfo.limit, sinfo.batchcount, sinfo.shared); 1120 seq_printf(m, " : slabdata %6lu %6lu %6lu", 1121 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); 1122 seq_putc(m, '\n'); 1123 } 1124 1125 static int slab_show(struct seq_file *m, void *p) 1126 { 1127 struct kmem_cache *s = list_entry(p, struct kmem_cache, list); 1128 1129 if (p == slab_caches.next) 1130 print_slabinfo_header(m); 1131 cache_show(s, m); 1132 return 0; 1133 } 1134 1135 void dump_unreclaimable_slab(void) 1136 { 1137 struct kmem_cache *s; 1138 struct slabinfo sinfo; 1139 1140 /* 1141 * Here acquiring slab_mutex is risky since we don't prefer to get 1142 * sleep in oom path. But, without mutex hold, it may introduce a 1143 * risk of crash. 1144 * Use mutex_trylock to protect the list traverse, dump nothing 1145 * without acquiring the mutex. 1146 */ 1147 if (!mutex_trylock(&slab_mutex)) { 1148 pr_warn("excessive unreclaimable slab but cannot dump stats\n"); 1149 return; 1150 } 1151 1152 pr_info("Unreclaimable slab info:\n"); 1153 pr_info("Name Used Total\n"); 1154 1155 list_for_each_entry(s, &slab_caches, list) { 1156 if (s->flags & SLAB_RECLAIM_ACCOUNT) 1157 continue; 1158 1159 get_slabinfo(s, &sinfo); 1160 1161 if (sinfo.num_objs > 0) 1162 pr_info("%-17s %10luKB %10luKB\n", s->name, 1163 (sinfo.active_objs * s->size) / 1024, 1164 (sinfo.num_objs * s->size) / 1024); 1165 } 1166 mutex_unlock(&slab_mutex); 1167 } 1168 1169 /* 1170 * slabinfo_op - iterator that generates /proc/slabinfo 1171 * 1172 * Output layout: 1173 * cache-name 1174 * num-active-objs 1175 * total-objs 1176 * object size 1177 * num-active-slabs 1178 * total-slabs 1179 * num-pages-per-slab 1180 * + further values on SMP and with statistics enabled 1181 */ 1182 static const struct seq_operations slabinfo_op = { 1183 .start = slab_start, 1184 .next = slab_next, 1185 .stop = slab_stop, 1186 .show = slab_show, 1187 }; 1188 1189 static int slabinfo_open(struct inode *inode, struct file *file) 1190 { 1191 return seq_open(file, &slabinfo_op); 1192 } 1193 1194 static const struct proc_ops slabinfo_proc_ops = { 1195 .proc_flags = PROC_ENTRY_PERMANENT, 1196 .proc_open = slabinfo_open, 1197 .proc_read = seq_read, 1198 .proc_lseek = seq_lseek, 1199 .proc_release = seq_release, 1200 }; 1201 1202 static int __init slab_proc_init(void) 1203 { 1204 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); 1205 return 0; 1206 } 1207 module_init(slab_proc_init); 1208 1209 #endif /* CONFIG_SLUB_DEBUG */ 1210 1211 /** 1212 * kfree_sensitive - Clear sensitive information in memory before freeing 1213 * @p: object to free memory of 1214 * 1215 * The memory of the object @p points to is zeroed before freed. 1216 * If @p is %NULL, kfree_sensitive() does nothing. 1217 * 1218 * Note: this function zeroes the whole allocated buffer which can be a good 1219 * deal bigger than the requested buffer size passed to kmalloc(). So be 1220 * careful when using this function in performance sensitive code. 1221 */ 1222 void kfree_sensitive(const void *p) 1223 { 1224 size_t ks; 1225 void *mem = (void *)p; 1226 1227 ks = ksize(mem); 1228 if (ks) { 1229 kasan_unpoison_range(mem, ks); 1230 memzero_explicit(mem, ks); 1231 } 1232 kfree(mem); 1233 } 1234 EXPORT_SYMBOL(kfree_sensitive); 1235 1236 size_t ksize(const void *objp) 1237 { 1238 /* 1239 * We need to first check that the pointer to the object is valid. 1240 * The KASAN report printed from ksize() is more useful, then when 1241 * it's printed later when the behaviour could be undefined due to 1242 * a potential use-after-free or double-free. 1243 * 1244 * We use kasan_check_byte(), which is supported for the hardware 1245 * tag-based KASAN mode, unlike kasan_check_read/write(). 1246 * 1247 * If the pointed to memory is invalid, we return 0 to avoid users of 1248 * ksize() writing to and potentially corrupting the memory region. 1249 * 1250 * We want to perform the check before __ksize(), to avoid potentially 1251 * crashing in __ksize() due to accessing invalid metadata. 1252 */ 1253 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp)) 1254 return 0; 1255 1256 return kfence_ksize(objp) ?: __ksize(objp); 1257 } 1258 EXPORT_SYMBOL(ksize); 1259 1260 #ifdef CONFIG_BPF_SYSCALL 1261 #include <linux/btf.h> 1262 1263 __bpf_kfunc_start_defs(); 1264 1265 __bpf_kfunc struct kmem_cache *bpf_get_kmem_cache(u64 addr) 1266 { 1267 struct slab *slab; 1268 1269 if (!virt_addr_valid((void *)(long)addr)) 1270 return NULL; 1271 1272 slab = virt_to_slab((void *)(long)addr); 1273 return slab ? slab->slab_cache : NULL; 1274 } 1275 1276 __bpf_kfunc_end_defs(); 1277 #endif /* CONFIG_BPF_SYSCALL */ 1278 1279 /* Tracepoints definitions. */ 1280 EXPORT_TRACEPOINT_SYMBOL(kmalloc); 1281 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); 1282 EXPORT_TRACEPOINT_SYMBOL(kfree); 1283 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); 1284 1285