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 #ifdef SLAB_SUPPORTS_SYSFS 396 /* 397 * For a given kmem_cache, kmem_cache_destroy() should only be called 398 * once or there will be a use-after-free problem. The actual deletion 399 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock 400 * protection. So they are now done without holding those locks. 401 * 402 * Note that there will be a slight delay in the deletion of sysfs files 403 * if kmem_cache_release() is called indrectly from a work function. 404 */ 405 static void kmem_cache_release(struct kmem_cache *s) 406 { 407 if (slab_state >= FULL) { 408 sysfs_slab_unlink(s); 409 sysfs_slab_release(s); 410 } else { 411 slab_kmem_cache_release(s); 412 } 413 } 414 #else 415 static void kmem_cache_release(struct kmem_cache *s) 416 { 417 slab_kmem_cache_release(s); 418 } 419 #endif 420 421 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) 422 { 423 LIST_HEAD(to_destroy); 424 struct kmem_cache *s, *s2; 425 426 /* 427 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the 428 * @slab_caches_to_rcu_destroy list. The slab pages are freed 429 * through RCU and the associated kmem_cache are dereferenced 430 * while freeing the pages, so the kmem_caches should be freed only 431 * after the pending RCU operations are finished. As rcu_barrier() 432 * is a pretty slow operation, we batch all pending destructions 433 * asynchronously. 434 */ 435 mutex_lock(&slab_mutex); 436 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); 437 mutex_unlock(&slab_mutex); 438 439 if (list_empty(&to_destroy)) 440 return; 441 442 rcu_barrier(); 443 444 list_for_each_entry_safe(s, s2, &to_destroy, list) { 445 debugfs_slab_release(s); 446 kfence_shutdown_cache(s); 447 kmem_cache_release(s); 448 } 449 } 450 451 static int shutdown_cache(struct kmem_cache *s) 452 { 453 /* free asan quarantined objects */ 454 kasan_cache_shutdown(s); 455 456 if (__kmem_cache_shutdown(s) != 0) 457 return -EBUSY; 458 459 list_del(&s->list); 460 461 if (s->flags & SLAB_TYPESAFE_BY_RCU) { 462 list_add_tail(&s->list, &slab_caches_to_rcu_destroy); 463 schedule_work(&slab_caches_to_rcu_destroy_work); 464 } else { 465 kfence_shutdown_cache(s); 466 debugfs_slab_release(s); 467 } 468 469 return 0; 470 } 471 472 void slab_kmem_cache_release(struct kmem_cache *s) 473 { 474 __kmem_cache_release(s); 475 kfree_const(s->name); 476 kmem_cache_free(kmem_cache, s); 477 } 478 479 void kmem_cache_destroy(struct kmem_cache *s) 480 { 481 int err = -EBUSY; 482 bool rcu_set; 483 484 if (unlikely(!s) || !kasan_check_byte(s)) 485 return; 486 487 cpus_read_lock(); 488 mutex_lock(&slab_mutex); 489 490 rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU; 491 492 s->refcount--; 493 if (s->refcount) 494 goto out_unlock; 495 496 err = shutdown_cache(s); 497 WARN(err, "%s %s: Slab cache still has objects when called from %pS", 498 __func__, s->name, (void *)_RET_IP_); 499 out_unlock: 500 mutex_unlock(&slab_mutex); 501 cpus_read_unlock(); 502 if (!err && !rcu_set) 503 kmem_cache_release(s); 504 } 505 EXPORT_SYMBOL(kmem_cache_destroy); 506 507 /** 508 * kmem_cache_shrink - Shrink a cache. 509 * @cachep: The cache to shrink. 510 * 511 * Releases as many slabs as possible for a cache. 512 * To help debugging, a zero exit status indicates all slabs were released. 513 * 514 * Return: %0 if all slabs were released, non-zero otherwise 515 */ 516 int kmem_cache_shrink(struct kmem_cache *cachep) 517 { 518 kasan_cache_shrink(cachep); 519 520 return __kmem_cache_shrink(cachep); 521 } 522 EXPORT_SYMBOL(kmem_cache_shrink); 523 524 bool slab_is_available(void) 525 { 526 return slab_state >= UP; 527 } 528 529 #ifdef CONFIG_PRINTK 530 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab) 531 { 532 if (__kfence_obj_info(kpp, object, slab)) 533 return; 534 __kmem_obj_info(kpp, object, slab); 535 } 536 537 /** 538 * kmem_dump_obj - Print available slab provenance information 539 * @object: slab object for which to find provenance information. 540 * 541 * This function uses pr_cont(), so that the caller is expected to have 542 * printed out whatever preamble is appropriate. The provenance information 543 * depends on the type of object and on how much debugging is enabled. 544 * For a slab-cache object, the fact that it is a slab object is printed, 545 * and, if available, the slab name, return address, and stack trace from 546 * the allocation and last free path of that object. 547 * 548 * Return: %true if the pointer is to a not-yet-freed object from 549 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer 550 * is to an already-freed object, and %false otherwise. 551 */ 552 bool kmem_dump_obj(void *object) 553 { 554 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc"; 555 int i; 556 struct slab *slab; 557 unsigned long ptroffset; 558 struct kmem_obj_info kp = { }; 559 560 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */ 561 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object)) 562 return false; 563 slab = virt_to_slab(object); 564 if (!slab) 565 return false; 566 567 kmem_obj_info(&kp, object, slab); 568 if (kp.kp_slab_cache) 569 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name); 570 else 571 pr_cont(" slab%s", cp); 572 if (is_kfence_address(object)) 573 pr_cont(" (kfence)"); 574 if (kp.kp_objp) 575 pr_cont(" start %px", kp.kp_objp); 576 if (kp.kp_data_offset) 577 pr_cont(" data offset %lu", kp.kp_data_offset); 578 if (kp.kp_objp) { 579 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset; 580 pr_cont(" pointer offset %lu", ptroffset); 581 } 582 if (kp.kp_slab_cache && kp.kp_slab_cache->object_size) 583 pr_cont(" size %u", kp.kp_slab_cache->object_size); 584 if (kp.kp_ret) 585 pr_cont(" allocated at %pS\n", kp.kp_ret); 586 else 587 pr_cont("\n"); 588 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) { 589 if (!kp.kp_stack[i]) 590 break; 591 pr_info(" %pS\n", kp.kp_stack[i]); 592 } 593 594 if (kp.kp_free_stack[0]) 595 pr_cont(" Free path:\n"); 596 597 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) { 598 if (!kp.kp_free_stack[i]) 599 break; 600 pr_info(" %pS\n", kp.kp_free_stack[i]); 601 } 602 603 return true; 604 } 605 EXPORT_SYMBOL_GPL(kmem_dump_obj); 606 #endif 607 608 /* Create a cache during boot when no slab services are available yet */ 609 void __init create_boot_cache(struct kmem_cache *s, const char *name, 610 unsigned int size, slab_flags_t flags, 611 unsigned int useroffset, unsigned int usersize) 612 { 613 int err; 614 unsigned int align = ARCH_KMALLOC_MINALIGN; 615 616 s->name = name; 617 s->size = s->object_size = size; 618 619 /* 620 * For power of two sizes, guarantee natural alignment for kmalloc 621 * caches, regardless of SL*B debugging options. 622 */ 623 if (is_power_of_2(size)) 624 align = max(align, size); 625 s->align = calculate_alignment(flags, align, size); 626 627 #ifdef CONFIG_HARDENED_USERCOPY 628 s->useroffset = useroffset; 629 s->usersize = usersize; 630 #endif 631 632 err = __kmem_cache_create(s, flags); 633 634 if (err) 635 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", 636 name, size, err); 637 638 s->refcount = -1; /* Exempt from merging for now */ 639 } 640 641 static struct kmem_cache *__init create_kmalloc_cache(const char *name, 642 unsigned int size, 643 slab_flags_t flags) 644 { 645 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); 646 647 if (!s) 648 panic("Out of memory when creating slab %s\n", name); 649 650 create_boot_cache(s, name, size, flags | SLAB_KMALLOC, 0, size); 651 list_add(&s->list, &slab_caches); 652 s->refcount = 1; 653 return s; 654 } 655 656 struct kmem_cache * 657 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init = 658 { /* initialization for https://llvm.org/pr42570 */ }; 659 EXPORT_SYMBOL(kmalloc_caches); 660 661 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 662 unsigned long random_kmalloc_seed __ro_after_init; 663 EXPORT_SYMBOL(random_kmalloc_seed); 664 #endif 665 666 /* 667 * Conversion table for small slabs sizes / 8 to the index in the 668 * kmalloc array. This is necessary for slabs < 192 since we have non power 669 * of two cache sizes there. The size of larger slabs can be determined using 670 * fls. 671 */ 672 u8 kmalloc_size_index[24] __ro_after_init = { 673 3, /* 8 */ 674 4, /* 16 */ 675 5, /* 24 */ 676 5, /* 32 */ 677 6, /* 40 */ 678 6, /* 48 */ 679 6, /* 56 */ 680 6, /* 64 */ 681 1, /* 72 */ 682 1, /* 80 */ 683 1, /* 88 */ 684 1, /* 96 */ 685 7, /* 104 */ 686 7, /* 112 */ 687 7, /* 120 */ 688 7, /* 128 */ 689 2, /* 136 */ 690 2, /* 144 */ 691 2, /* 152 */ 692 2, /* 160 */ 693 2, /* 168 */ 694 2, /* 176 */ 695 2, /* 184 */ 696 2 /* 192 */ 697 }; 698 699 size_t kmalloc_size_roundup(size_t size) 700 { 701 if (size && size <= KMALLOC_MAX_CACHE_SIZE) { 702 /* 703 * The flags don't matter since size_index is common to all. 704 * Neither does the caller for just getting ->object_size. 705 */ 706 return kmalloc_slab(size, GFP_KERNEL, 0)->object_size; 707 } 708 709 /* Above the smaller buckets, size is a multiple of page size. */ 710 if (size && size <= KMALLOC_MAX_SIZE) 711 return PAGE_SIZE << get_order(size); 712 713 /* 714 * Return 'size' for 0 - kmalloc() returns ZERO_SIZE_PTR 715 * and very large size - kmalloc() may fail. 716 */ 717 return size; 718 719 } 720 EXPORT_SYMBOL(kmalloc_size_roundup); 721 722 #ifdef CONFIG_ZONE_DMA 723 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz, 724 #else 725 #define KMALLOC_DMA_NAME(sz) 726 #endif 727 728 #ifdef CONFIG_MEMCG_KMEM 729 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz, 730 #else 731 #define KMALLOC_CGROUP_NAME(sz) 732 #endif 733 734 #ifndef CONFIG_SLUB_TINY 735 #define KMALLOC_RCL_NAME(sz) .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz, 736 #else 737 #define KMALLOC_RCL_NAME(sz) 738 #endif 739 740 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 741 #define __KMALLOC_RANDOM_CONCAT(a, b) a ## b 742 #define KMALLOC_RANDOM_NAME(N, sz) __KMALLOC_RANDOM_CONCAT(KMA_RAND_, N)(sz) 743 #define KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 1] = "kmalloc-rnd-01-" #sz, 744 #define KMA_RAND_2(sz) KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 2] = "kmalloc-rnd-02-" #sz, 745 #define KMA_RAND_3(sz) KMA_RAND_2(sz) .name[KMALLOC_RANDOM_START + 3] = "kmalloc-rnd-03-" #sz, 746 #define KMA_RAND_4(sz) KMA_RAND_3(sz) .name[KMALLOC_RANDOM_START + 4] = "kmalloc-rnd-04-" #sz, 747 #define KMA_RAND_5(sz) KMA_RAND_4(sz) .name[KMALLOC_RANDOM_START + 5] = "kmalloc-rnd-05-" #sz, 748 #define KMA_RAND_6(sz) KMA_RAND_5(sz) .name[KMALLOC_RANDOM_START + 6] = "kmalloc-rnd-06-" #sz, 749 #define KMA_RAND_7(sz) KMA_RAND_6(sz) .name[KMALLOC_RANDOM_START + 7] = "kmalloc-rnd-07-" #sz, 750 #define KMA_RAND_8(sz) KMA_RAND_7(sz) .name[KMALLOC_RANDOM_START + 8] = "kmalloc-rnd-08-" #sz, 751 #define KMA_RAND_9(sz) KMA_RAND_8(sz) .name[KMALLOC_RANDOM_START + 9] = "kmalloc-rnd-09-" #sz, 752 #define KMA_RAND_10(sz) KMA_RAND_9(sz) .name[KMALLOC_RANDOM_START + 10] = "kmalloc-rnd-10-" #sz, 753 #define KMA_RAND_11(sz) KMA_RAND_10(sz) .name[KMALLOC_RANDOM_START + 11] = "kmalloc-rnd-11-" #sz, 754 #define KMA_RAND_12(sz) KMA_RAND_11(sz) .name[KMALLOC_RANDOM_START + 12] = "kmalloc-rnd-12-" #sz, 755 #define KMA_RAND_13(sz) KMA_RAND_12(sz) .name[KMALLOC_RANDOM_START + 13] = "kmalloc-rnd-13-" #sz, 756 #define KMA_RAND_14(sz) KMA_RAND_13(sz) .name[KMALLOC_RANDOM_START + 14] = "kmalloc-rnd-14-" #sz, 757 #define KMA_RAND_15(sz) KMA_RAND_14(sz) .name[KMALLOC_RANDOM_START + 15] = "kmalloc-rnd-15-" #sz, 758 #else // CONFIG_RANDOM_KMALLOC_CACHES 759 #define KMALLOC_RANDOM_NAME(N, sz) 760 #endif 761 762 #define INIT_KMALLOC_INFO(__size, __short_size) \ 763 { \ 764 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ 765 KMALLOC_RCL_NAME(__short_size) \ 766 KMALLOC_CGROUP_NAME(__short_size) \ 767 KMALLOC_DMA_NAME(__short_size) \ 768 KMALLOC_RANDOM_NAME(RANDOM_KMALLOC_CACHES_NR, __short_size) \ 769 .size = __size, \ 770 } 771 772 /* 773 * kmalloc_info[] is to make slab_debug=,kmalloc-xx option work at boot time. 774 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is 775 * kmalloc-2M. 776 */ 777 const struct kmalloc_info_struct kmalloc_info[] __initconst = { 778 INIT_KMALLOC_INFO(0, 0), 779 INIT_KMALLOC_INFO(96, 96), 780 INIT_KMALLOC_INFO(192, 192), 781 INIT_KMALLOC_INFO(8, 8), 782 INIT_KMALLOC_INFO(16, 16), 783 INIT_KMALLOC_INFO(32, 32), 784 INIT_KMALLOC_INFO(64, 64), 785 INIT_KMALLOC_INFO(128, 128), 786 INIT_KMALLOC_INFO(256, 256), 787 INIT_KMALLOC_INFO(512, 512), 788 INIT_KMALLOC_INFO(1024, 1k), 789 INIT_KMALLOC_INFO(2048, 2k), 790 INIT_KMALLOC_INFO(4096, 4k), 791 INIT_KMALLOC_INFO(8192, 8k), 792 INIT_KMALLOC_INFO(16384, 16k), 793 INIT_KMALLOC_INFO(32768, 32k), 794 INIT_KMALLOC_INFO(65536, 64k), 795 INIT_KMALLOC_INFO(131072, 128k), 796 INIT_KMALLOC_INFO(262144, 256k), 797 INIT_KMALLOC_INFO(524288, 512k), 798 INIT_KMALLOC_INFO(1048576, 1M), 799 INIT_KMALLOC_INFO(2097152, 2M) 800 }; 801 802 /* 803 * Patch up the size_index table if we have strange large alignment 804 * requirements for the kmalloc array. This is only the case for 805 * MIPS it seems. The standard arches will not generate any code here. 806 * 807 * Largest permitted alignment is 256 bytes due to the way we 808 * handle the index determination for the smaller caches. 809 * 810 * Make sure that nothing crazy happens if someone starts tinkering 811 * around with ARCH_KMALLOC_MINALIGN 812 */ 813 void __init setup_kmalloc_cache_index_table(void) 814 { 815 unsigned int i; 816 817 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || 818 !is_power_of_2(KMALLOC_MIN_SIZE)); 819 820 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { 821 unsigned int elem = size_index_elem(i); 822 823 if (elem >= ARRAY_SIZE(kmalloc_size_index)) 824 break; 825 kmalloc_size_index[elem] = KMALLOC_SHIFT_LOW; 826 } 827 828 if (KMALLOC_MIN_SIZE >= 64) { 829 /* 830 * The 96 byte sized cache is not used if the alignment 831 * is 64 byte. 832 */ 833 for (i = 64 + 8; i <= 96; i += 8) 834 kmalloc_size_index[size_index_elem(i)] = 7; 835 836 } 837 838 if (KMALLOC_MIN_SIZE >= 128) { 839 /* 840 * The 192 byte sized cache is not used if the alignment 841 * is 128 byte. Redirect kmalloc to use the 256 byte cache 842 * instead. 843 */ 844 for (i = 128 + 8; i <= 192; i += 8) 845 kmalloc_size_index[size_index_elem(i)] = 8; 846 } 847 } 848 849 static unsigned int __kmalloc_minalign(void) 850 { 851 unsigned int minalign = dma_get_cache_alignment(); 852 853 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && 854 is_swiotlb_allocated()) 855 minalign = ARCH_KMALLOC_MINALIGN; 856 857 return max(minalign, arch_slab_minalign()); 858 } 859 860 static void __init 861 new_kmalloc_cache(int idx, enum kmalloc_cache_type type) 862 { 863 slab_flags_t flags = 0; 864 unsigned int minalign = __kmalloc_minalign(); 865 unsigned int aligned_size = kmalloc_info[idx].size; 866 int aligned_idx = idx; 867 868 if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) { 869 flags |= SLAB_RECLAIM_ACCOUNT; 870 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) { 871 if (mem_cgroup_kmem_disabled()) { 872 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx]; 873 return; 874 } 875 flags |= SLAB_ACCOUNT; 876 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) { 877 flags |= SLAB_CACHE_DMA; 878 } 879 880 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 881 if (type >= KMALLOC_RANDOM_START && type <= KMALLOC_RANDOM_END) 882 flags |= SLAB_NO_MERGE; 883 #endif 884 885 /* 886 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for 887 * KMALLOC_NORMAL caches. 888 */ 889 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL)) 890 flags |= SLAB_NO_MERGE; 891 892 if (minalign > ARCH_KMALLOC_MINALIGN) { 893 aligned_size = ALIGN(aligned_size, minalign); 894 aligned_idx = __kmalloc_index(aligned_size, false); 895 } 896 897 if (!kmalloc_caches[type][aligned_idx]) 898 kmalloc_caches[type][aligned_idx] = create_kmalloc_cache( 899 kmalloc_info[aligned_idx].name[type], 900 aligned_size, flags); 901 if (idx != aligned_idx) 902 kmalloc_caches[type][idx] = kmalloc_caches[type][aligned_idx]; 903 } 904 905 /* 906 * Create the kmalloc array. Some of the regular kmalloc arrays 907 * may already have been created because they were needed to 908 * enable allocations for slab creation. 909 */ 910 void __init create_kmalloc_caches(void) 911 { 912 int i; 913 enum kmalloc_cache_type type; 914 915 /* 916 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined 917 */ 918 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) { 919 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { 920 if (!kmalloc_caches[type][i]) 921 new_kmalloc_cache(i, type); 922 923 /* 924 * Caches that are not of the two-to-the-power-of size. 925 * These have to be created immediately after the 926 * earlier power of two caches 927 */ 928 if (KMALLOC_MIN_SIZE <= 32 && i == 6 && 929 !kmalloc_caches[type][1]) 930 new_kmalloc_cache(1, type); 931 if (KMALLOC_MIN_SIZE <= 64 && i == 7 && 932 !kmalloc_caches[type][2]) 933 new_kmalloc_cache(2, type); 934 } 935 } 936 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 937 random_kmalloc_seed = get_random_u64(); 938 #endif 939 940 /* Kmalloc array is now usable */ 941 slab_state = UP; 942 } 943 944 /** 945 * __ksize -- Report full size of underlying allocation 946 * @object: pointer to the object 947 * 948 * This should only be used internally to query the true size of allocations. 949 * It is not meant to be a way to discover the usable size of an allocation 950 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond 951 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS, 952 * and/or FORTIFY_SOURCE. 953 * 954 * Return: size of the actual memory used by @object in bytes 955 */ 956 size_t __ksize(const void *object) 957 { 958 struct folio *folio; 959 960 if (unlikely(object == ZERO_SIZE_PTR)) 961 return 0; 962 963 folio = virt_to_folio(object); 964 965 if (unlikely(!folio_test_slab(folio))) { 966 if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE)) 967 return 0; 968 if (WARN_ON(object != folio_address(folio))) 969 return 0; 970 return folio_size(folio); 971 } 972 973 #ifdef CONFIG_SLUB_DEBUG 974 skip_orig_size_check(folio_slab(folio)->slab_cache, object); 975 #endif 976 977 return slab_ksize(folio_slab(folio)->slab_cache); 978 } 979 980 gfp_t kmalloc_fix_flags(gfp_t flags) 981 { 982 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; 983 984 flags &= ~GFP_SLAB_BUG_MASK; 985 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", 986 invalid_mask, &invalid_mask, flags, &flags); 987 dump_stack(); 988 989 return flags; 990 } 991 992 #ifdef CONFIG_SLAB_FREELIST_RANDOM 993 /* Randomize a generic freelist */ 994 static void freelist_randomize(unsigned int *list, 995 unsigned int count) 996 { 997 unsigned int rand; 998 unsigned int i; 999 1000 for (i = 0; i < count; i++) 1001 list[i] = i; 1002 1003 /* Fisher-Yates shuffle */ 1004 for (i = count - 1; i > 0; i--) { 1005 rand = get_random_u32_below(i + 1); 1006 swap(list[i], list[rand]); 1007 } 1008 } 1009 1010 /* Create a random sequence per cache */ 1011 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, 1012 gfp_t gfp) 1013 { 1014 1015 if (count < 2 || cachep->random_seq) 1016 return 0; 1017 1018 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); 1019 if (!cachep->random_seq) 1020 return -ENOMEM; 1021 1022 freelist_randomize(cachep->random_seq, count); 1023 return 0; 1024 } 1025 1026 /* Destroy the per-cache random freelist sequence */ 1027 void cache_random_seq_destroy(struct kmem_cache *cachep) 1028 { 1029 kfree(cachep->random_seq); 1030 cachep->random_seq = NULL; 1031 } 1032 #endif /* CONFIG_SLAB_FREELIST_RANDOM */ 1033 1034 #ifdef CONFIG_SLUB_DEBUG 1035 #define SLABINFO_RIGHTS (0400) 1036 1037 static void print_slabinfo_header(struct seq_file *m) 1038 { 1039 /* 1040 * Output format version, so at least we can change it 1041 * without _too_ many complaints. 1042 */ 1043 seq_puts(m, "slabinfo - version: 2.1\n"); 1044 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); 1045 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); 1046 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); 1047 seq_putc(m, '\n'); 1048 } 1049 1050 static void *slab_start(struct seq_file *m, loff_t *pos) 1051 { 1052 mutex_lock(&slab_mutex); 1053 return seq_list_start(&slab_caches, *pos); 1054 } 1055 1056 static void *slab_next(struct seq_file *m, void *p, loff_t *pos) 1057 { 1058 return seq_list_next(p, &slab_caches, pos); 1059 } 1060 1061 static void slab_stop(struct seq_file *m, void *p) 1062 { 1063 mutex_unlock(&slab_mutex); 1064 } 1065 1066 static void cache_show(struct kmem_cache *s, struct seq_file *m) 1067 { 1068 struct slabinfo sinfo; 1069 1070 memset(&sinfo, 0, sizeof(sinfo)); 1071 get_slabinfo(s, &sinfo); 1072 1073 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", 1074 s->name, sinfo.active_objs, sinfo.num_objs, s->size, 1075 sinfo.objects_per_slab, (1 << sinfo.cache_order)); 1076 1077 seq_printf(m, " : tunables %4u %4u %4u", 1078 sinfo.limit, sinfo.batchcount, sinfo.shared); 1079 seq_printf(m, " : slabdata %6lu %6lu %6lu", 1080 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); 1081 slabinfo_show_stats(m, s); 1082 seq_putc(m, '\n'); 1083 } 1084 1085 static int slab_show(struct seq_file *m, void *p) 1086 { 1087 struct kmem_cache *s = list_entry(p, struct kmem_cache, list); 1088 1089 if (p == slab_caches.next) 1090 print_slabinfo_header(m); 1091 cache_show(s, m); 1092 return 0; 1093 } 1094 1095 void dump_unreclaimable_slab(void) 1096 { 1097 struct kmem_cache *s; 1098 struct slabinfo sinfo; 1099 1100 /* 1101 * Here acquiring slab_mutex is risky since we don't prefer to get 1102 * sleep in oom path. But, without mutex hold, it may introduce a 1103 * risk of crash. 1104 * Use mutex_trylock to protect the list traverse, dump nothing 1105 * without acquiring the mutex. 1106 */ 1107 if (!mutex_trylock(&slab_mutex)) { 1108 pr_warn("excessive unreclaimable slab but cannot dump stats\n"); 1109 return; 1110 } 1111 1112 pr_info("Unreclaimable slab info:\n"); 1113 pr_info("Name Used Total\n"); 1114 1115 list_for_each_entry(s, &slab_caches, list) { 1116 if (s->flags & SLAB_RECLAIM_ACCOUNT) 1117 continue; 1118 1119 get_slabinfo(s, &sinfo); 1120 1121 if (sinfo.num_objs > 0) 1122 pr_info("%-17s %10luKB %10luKB\n", s->name, 1123 (sinfo.active_objs * s->size) / 1024, 1124 (sinfo.num_objs * s->size) / 1024); 1125 } 1126 mutex_unlock(&slab_mutex); 1127 } 1128 1129 /* 1130 * slabinfo_op - iterator that generates /proc/slabinfo 1131 * 1132 * Output layout: 1133 * cache-name 1134 * num-active-objs 1135 * total-objs 1136 * object size 1137 * num-active-slabs 1138 * total-slabs 1139 * num-pages-per-slab 1140 * + further values on SMP and with statistics enabled 1141 */ 1142 static const struct seq_operations slabinfo_op = { 1143 .start = slab_start, 1144 .next = slab_next, 1145 .stop = slab_stop, 1146 .show = slab_show, 1147 }; 1148 1149 static int slabinfo_open(struct inode *inode, struct file *file) 1150 { 1151 return seq_open(file, &slabinfo_op); 1152 } 1153 1154 static const struct proc_ops slabinfo_proc_ops = { 1155 .proc_flags = PROC_ENTRY_PERMANENT, 1156 .proc_open = slabinfo_open, 1157 .proc_read = seq_read, 1158 .proc_write = slabinfo_write, 1159 .proc_lseek = seq_lseek, 1160 .proc_release = seq_release, 1161 }; 1162 1163 static int __init slab_proc_init(void) 1164 { 1165 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); 1166 return 0; 1167 } 1168 module_init(slab_proc_init); 1169 1170 #endif /* CONFIG_SLUB_DEBUG */ 1171 1172 static __always_inline __realloc_size(2) void * 1173 __do_krealloc(const void *p, size_t new_size, gfp_t flags) 1174 { 1175 void *ret; 1176 size_t ks; 1177 1178 /* Check for double-free before calling ksize. */ 1179 if (likely(!ZERO_OR_NULL_PTR(p))) { 1180 if (!kasan_check_byte(p)) 1181 return NULL; 1182 ks = ksize(p); 1183 } else 1184 ks = 0; 1185 1186 /* If the object still fits, repoison it precisely. */ 1187 if (ks >= new_size) { 1188 p = kasan_krealloc((void *)p, new_size, flags); 1189 return (void *)p; 1190 } 1191 1192 ret = kmalloc_track_caller(new_size, flags); 1193 if (ret && p) { 1194 /* Disable KASAN checks as the object's redzone is accessed. */ 1195 kasan_disable_current(); 1196 memcpy(ret, kasan_reset_tag(p), ks); 1197 kasan_enable_current(); 1198 } 1199 1200 return ret; 1201 } 1202 1203 /** 1204 * krealloc - reallocate memory. The contents will remain unchanged. 1205 * @p: object to reallocate memory for. 1206 * @new_size: how many bytes of memory are required. 1207 * @flags: the type of memory to allocate. 1208 * 1209 * The contents of the object pointed to are preserved up to the 1210 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored). 1211 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size 1212 * is 0 and @p is not a %NULL pointer, the object pointed to is freed. 1213 * 1214 * Return: pointer to the allocated memory or %NULL in case of error 1215 */ 1216 void *krealloc(const void *p, size_t new_size, gfp_t flags) 1217 { 1218 void *ret; 1219 1220 if (unlikely(!new_size)) { 1221 kfree(p); 1222 return ZERO_SIZE_PTR; 1223 } 1224 1225 ret = __do_krealloc(p, new_size, flags); 1226 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) 1227 kfree(p); 1228 1229 return ret; 1230 } 1231 EXPORT_SYMBOL(krealloc); 1232 1233 /** 1234 * kfree_sensitive - Clear sensitive information in memory before freeing 1235 * @p: object to free memory of 1236 * 1237 * The memory of the object @p points to is zeroed before freed. 1238 * If @p is %NULL, kfree_sensitive() does nothing. 1239 * 1240 * Note: this function zeroes the whole allocated buffer which can be a good 1241 * deal bigger than the requested buffer size passed to kmalloc(). So be 1242 * careful when using this function in performance sensitive code. 1243 */ 1244 void kfree_sensitive(const void *p) 1245 { 1246 size_t ks; 1247 void *mem = (void *)p; 1248 1249 ks = ksize(mem); 1250 if (ks) { 1251 kasan_unpoison_range(mem, ks); 1252 memzero_explicit(mem, ks); 1253 } 1254 kfree(mem); 1255 } 1256 EXPORT_SYMBOL(kfree_sensitive); 1257 1258 size_t ksize(const void *objp) 1259 { 1260 /* 1261 * We need to first check that the pointer to the object is valid. 1262 * The KASAN report printed from ksize() is more useful, then when 1263 * it's printed later when the behaviour could be undefined due to 1264 * a potential use-after-free or double-free. 1265 * 1266 * We use kasan_check_byte(), which is supported for the hardware 1267 * tag-based KASAN mode, unlike kasan_check_read/write(). 1268 * 1269 * If the pointed to memory is invalid, we return 0 to avoid users of 1270 * ksize() writing to and potentially corrupting the memory region. 1271 * 1272 * We want to perform the check before __ksize(), to avoid potentially 1273 * crashing in __ksize() due to accessing invalid metadata. 1274 */ 1275 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp)) 1276 return 0; 1277 1278 return kfence_ksize(objp) ?: __ksize(objp); 1279 } 1280 EXPORT_SYMBOL(ksize); 1281 1282 /* Tracepoints definitions. */ 1283 EXPORT_TRACEPOINT_SYMBOL(kmalloc); 1284 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); 1285 EXPORT_TRACEPOINT_SYMBOL(kfree); 1286 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); 1287 1288