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/proc_fs.h> 21 #include <linux/debugfs.h> 22 #include <linux/kasan.h> 23 #include <asm/cacheflush.h> 24 #include <asm/tlbflush.h> 25 #include <asm/page.h> 26 #include <linux/memcontrol.h> 27 #include <linux/stackdepot.h> 28 29 #define CREATE_TRACE_POINTS 30 #include <trace/events/kmem.h> 31 32 #include "internal.h" 33 34 #include "slab.h" 35 36 enum slab_state slab_state; 37 LIST_HEAD(slab_caches); 38 DEFINE_MUTEX(slab_mutex); 39 struct kmem_cache *kmem_cache; 40 41 static LIST_HEAD(slab_caches_to_rcu_destroy); 42 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); 43 static DECLARE_WORK(slab_caches_to_rcu_destroy_work, 44 slab_caches_to_rcu_destroy_workfn); 45 46 /* 47 * Set of flags that will prevent slab merging 48 */ 49 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ 50 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ 51 SLAB_FAILSLAB | kasan_never_merge()) 52 53 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ 54 SLAB_CACHE_DMA32 | SLAB_ACCOUNT) 55 56 /* 57 * Merge control. If this is set then no merging of slab caches will occur. 58 */ 59 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); 60 61 static int __init setup_slab_nomerge(char *str) 62 { 63 slab_nomerge = true; 64 return 1; 65 } 66 67 static int __init setup_slab_merge(char *str) 68 { 69 slab_nomerge = false; 70 return 1; 71 } 72 73 #ifdef CONFIG_SLUB 74 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); 75 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0); 76 #endif 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 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) 109 { 110 size_t i; 111 112 for (i = 0; i < nr; i++) { 113 if (s) 114 kmem_cache_free(s, p[i]); 115 else 116 kfree(p[i]); 117 } 118 } 119 120 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, 121 void **p) 122 { 123 size_t i; 124 125 for (i = 0; i < nr; i++) { 126 void *x = p[i] = kmem_cache_alloc(s, flags); 127 if (!x) { 128 __kmem_cache_free_bulk(s, i, p); 129 return 0; 130 } 131 } 132 return i; 133 } 134 135 /* 136 * Figure out what the alignment of the objects will be given a set of 137 * flags, a user specified alignment and the size of the objects. 138 */ 139 static unsigned int calculate_alignment(slab_flags_t flags, 140 unsigned int align, unsigned int size) 141 { 142 /* 143 * If the user wants hardware cache aligned objects then follow that 144 * suggestion if the object is sufficiently large. 145 * 146 * The hardware cache alignment cannot override the specified 147 * alignment though. If that is greater then use it. 148 */ 149 if (flags & SLAB_HWCACHE_ALIGN) { 150 unsigned int ralign; 151 152 ralign = cache_line_size(); 153 while (size <= ralign / 2) 154 ralign /= 2; 155 align = max(align, ralign); 156 } 157 158 align = max(align, arch_slab_minalign()); 159 160 return ALIGN(align, sizeof(void *)); 161 } 162 163 /* 164 * Find a mergeable slab cache 165 */ 166 int slab_unmergeable(struct kmem_cache *s) 167 { 168 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) 169 return 1; 170 171 if (s->ctor) 172 return 1; 173 174 if (s->usersize) 175 return 1; 176 177 /* 178 * We may have set a slab to be unmergeable during bootstrap. 179 */ 180 if (s->refcount < 0) 181 return 1; 182 183 return 0; 184 } 185 186 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align, 187 slab_flags_t flags, const char *name, void (*ctor)(void *)) 188 { 189 struct kmem_cache *s; 190 191 if (slab_nomerge) 192 return NULL; 193 194 if (ctor) 195 return NULL; 196 197 size = ALIGN(size, sizeof(void *)); 198 align = calculate_alignment(flags, align, size); 199 size = ALIGN(size, align); 200 flags = kmem_cache_flags(size, flags, name); 201 202 if (flags & SLAB_NEVER_MERGE) 203 return NULL; 204 205 list_for_each_entry_reverse(s, &slab_caches, list) { 206 if (slab_unmergeable(s)) 207 continue; 208 209 if (size > s->size) 210 continue; 211 212 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) 213 continue; 214 /* 215 * Check if alignment is compatible. 216 * Courtesy of Adrian Drzewiecki 217 */ 218 if ((s->size & ~(align - 1)) != s->size) 219 continue; 220 221 if (s->size - size >= sizeof(void *)) 222 continue; 223 224 if (IS_ENABLED(CONFIG_SLAB) && align && 225 (align > s->align || s->align % align)) 226 continue; 227 228 return s; 229 } 230 return NULL; 231 } 232 233 static struct kmem_cache *create_cache(const char *name, 234 unsigned int object_size, unsigned int align, 235 slab_flags_t flags, unsigned int useroffset, 236 unsigned int usersize, void (*ctor)(void *), 237 struct kmem_cache *root_cache) 238 { 239 struct kmem_cache *s; 240 int err; 241 242 if (WARN_ON(useroffset + usersize > object_size)) 243 useroffset = usersize = 0; 244 245 err = -ENOMEM; 246 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); 247 if (!s) 248 goto out; 249 250 s->name = name; 251 s->size = s->object_size = object_size; 252 s->align = align; 253 s->ctor = ctor; 254 s->useroffset = useroffset; 255 s->usersize = usersize; 256 257 err = __kmem_cache_create(s, flags); 258 if (err) 259 goto out_free_cache; 260 261 s->refcount = 1; 262 list_add(&s->list, &slab_caches); 263 out: 264 if (err) 265 return ERR_PTR(err); 266 return s; 267 268 out_free_cache: 269 kmem_cache_free(kmem_cache, s); 270 goto out; 271 } 272 273 /** 274 * kmem_cache_create_usercopy - Create a cache with a region suitable 275 * for copying to userspace 276 * @name: A string which is used in /proc/slabinfo to identify this cache. 277 * @size: The size of objects to be created in this cache. 278 * @align: The required alignment for the objects. 279 * @flags: SLAB flags 280 * @useroffset: Usercopy region offset 281 * @usersize: Usercopy region size 282 * @ctor: A constructor for the objects. 283 * 284 * Cannot be called within a interrupt, but can be interrupted. 285 * The @ctor is run when new pages are allocated by the cache. 286 * 287 * The flags are 288 * 289 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) 290 * to catch references to uninitialised memory. 291 * 292 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check 293 * for buffer overruns. 294 * 295 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware 296 * cacheline. This can be beneficial if you're counting cycles as closely 297 * as davem. 298 * 299 * Return: a pointer to the cache on success, NULL on failure. 300 */ 301 struct kmem_cache * 302 kmem_cache_create_usercopy(const char *name, 303 unsigned int size, unsigned int align, 304 slab_flags_t flags, 305 unsigned int useroffset, unsigned int usersize, 306 void (*ctor)(void *)) 307 { 308 struct kmem_cache *s = NULL; 309 const char *cache_name; 310 int err; 311 312 #ifdef CONFIG_SLUB_DEBUG 313 /* 314 * If no slub_debug was enabled globally, the static key is not yet 315 * enabled by setup_slub_debug(). Enable it if the cache is being 316 * created with any of the debugging flags passed explicitly. 317 * It's also possible that this is the first cache created with 318 * SLAB_STORE_USER and we should init stack_depot for it. 319 */ 320 if (flags & SLAB_DEBUG_FLAGS) 321 static_branch_enable(&slub_debug_enabled); 322 if (flags & SLAB_STORE_USER) 323 stack_depot_init(); 324 #endif 325 326 mutex_lock(&slab_mutex); 327 328 err = kmem_cache_sanity_check(name, size); 329 if (err) { 330 goto out_unlock; 331 } 332 333 /* Refuse requests with allocator specific flags */ 334 if (flags & ~SLAB_FLAGS_PERMITTED) { 335 err = -EINVAL; 336 goto out_unlock; 337 } 338 339 /* 340 * Some allocators will constraint the set of valid flags to a subset 341 * of all flags. We expect them to define CACHE_CREATE_MASK in this 342 * case, and we'll just provide them with a sanitized version of the 343 * passed flags. 344 */ 345 flags &= CACHE_CREATE_MASK; 346 347 /* Fail closed on bad usersize of useroffset values. */ 348 if (WARN_ON(!usersize && useroffset) || 349 WARN_ON(size < usersize || size - usersize < useroffset)) 350 usersize = useroffset = 0; 351 352 if (!usersize) 353 s = __kmem_cache_alias(name, size, align, flags, ctor); 354 if (s) 355 goto out_unlock; 356 357 cache_name = kstrdup_const(name, GFP_KERNEL); 358 if (!cache_name) { 359 err = -ENOMEM; 360 goto out_unlock; 361 } 362 363 s = create_cache(cache_name, size, 364 calculate_alignment(flags, align, size), 365 flags, useroffset, usersize, ctor, NULL); 366 if (IS_ERR(s)) { 367 err = PTR_ERR(s); 368 kfree_const(cache_name); 369 } 370 371 out_unlock: 372 mutex_unlock(&slab_mutex); 373 374 if (err) { 375 if (flags & SLAB_PANIC) 376 panic("%s: Failed to create slab '%s'. Error %d\n", 377 __func__, name, err); 378 else { 379 pr_warn("%s(%s) failed with error %d\n", 380 __func__, name, err); 381 dump_stack(); 382 } 383 return NULL; 384 } 385 return s; 386 } 387 EXPORT_SYMBOL(kmem_cache_create_usercopy); 388 389 /** 390 * kmem_cache_create - Create a cache. 391 * @name: A string which is used in /proc/slabinfo to identify this cache. 392 * @size: The size of objects to be created in this cache. 393 * @align: The required alignment for the objects. 394 * @flags: SLAB flags 395 * @ctor: A constructor for the objects. 396 * 397 * Cannot be called within a interrupt, but can be interrupted. 398 * The @ctor is run when new pages are allocated by the cache. 399 * 400 * The flags are 401 * 402 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) 403 * to catch references to uninitialised memory. 404 * 405 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check 406 * for buffer overruns. 407 * 408 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware 409 * cacheline. This can be beneficial if you're counting cycles as closely 410 * as davem. 411 * 412 * Return: a pointer to the cache on success, NULL on failure. 413 */ 414 struct kmem_cache * 415 kmem_cache_create(const char *name, unsigned int size, unsigned int align, 416 slab_flags_t flags, void (*ctor)(void *)) 417 { 418 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0, 419 ctor); 420 } 421 EXPORT_SYMBOL(kmem_cache_create); 422 423 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) 424 { 425 LIST_HEAD(to_destroy); 426 struct kmem_cache *s, *s2; 427 428 /* 429 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the 430 * @slab_caches_to_rcu_destroy list. The slab pages are freed 431 * through RCU and the associated kmem_cache are dereferenced 432 * while freeing the pages, so the kmem_caches should be freed only 433 * after the pending RCU operations are finished. As rcu_barrier() 434 * is a pretty slow operation, we batch all pending destructions 435 * asynchronously. 436 */ 437 mutex_lock(&slab_mutex); 438 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); 439 mutex_unlock(&slab_mutex); 440 441 if (list_empty(&to_destroy)) 442 return; 443 444 rcu_barrier(); 445 446 list_for_each_entry_safe(s, s2, &to_destroy, list) { 447 debugfs_slab_release(s); 448 kfence_shutdown_cache(s); 449 #ifdef SLAB_SUPPORTS_SYSFS 450 sysfs_slab_release(s); 451 #else 452 slab_kmem_cache_release(s); 453 #endif 454 } 455 } 456 457 static int shutdown_cache(struct kmem_cache *s) 458 { 459 /* free asan quarantined objects */ 460 kasan_cache_shutdown(s); 461 462 if (__kmem_cache_shutdown(s) != 0) 463 return -EBUSY; 464 465 list_del(&s->list); 466 467 if (s->flags & SLAB_TYPESAFE_BY_RCU) { 468 #ifdef SLAB_SUPPORTS_SYSFS 469 sysfs_slab_unlink(s); 470 #endif 471 list_add_tail(&s->list, &slab_caches_to_rcu_destroy); 472 schedule_work(&slab_caches_to_rcu_destroy_work); 473 } else { 474 kfence_shutdown_cache(s); 475 debugfs_slab_release(s); 476 #ifdef SLAB_SUPPORTS_SYSFS 477 sysfs_slab_unlink(s); 478 sysfs_slab_release(s); 479 #else 480 slab_kmem_cache_release(s); 481 #endif 482 } 483 484 return 0; 485 } 486 487 void slab_kmem_cache_release(struct kmem_cache *s) 488 { 489 __kmem_cache_release(s); 490 kfree_const(s->name); 491 kmem_cache_free(kmem_cache, s); 492 } 493 494 void kmem_cache_destroy(struct kmem_cache *s) 495 { 496 if (unlikely(!s) || !kasan_check_byte(s)) 497 return; 498 499 cpus_read_lock(); 500 mutex_lock(&slab_mutex); 501 502 s->refcount--; 503 if (s->refcount) 504 goto out_unlock; 505 506 WARN(shutdown_cache(s), 507 "%s %s: Slab cache still has objects when called from %pS", 508 __func__, s->name, (void *)_RET_IP_); 509 out_unlock: 510 mutex_unlock(&slab_mutex); 511 cpus_read_unlock(); 512 } 513 EXPORT_SYMBOL(kmem_cache_destroy); 514 515 /** 516 * kmem_cache_shrink - Shrink a cache. 517 * @cachep: The cache to shrink. 518 * 519 * Releases as many slabs as possible for a cache. 520 * To help debugging, a zero exit status indicates all slabs were released. 521 * 522 * Return: %0 if all slabs were released, non-zero otherwise 523 */ 524 int kmem_cache_shrink(struct kmem_cache *cachep) 525 { 526 int ret; 527 528 529 kasan_cache_shrink(cachep); 530 ret = __kmem_cache_shrink(cachep); 531 532 return ret; 533 } 534 EXPORT_SYMBOL(kmem_cache_shrink); 535 536 bool slab_is_available(void) 537 { 538 return slab_state >= UP; 539 } 540 541 #ifdef CONFIG_PRINTK 542 /** 543 * kmem_valid_obj - does the pointer reference a valid slab object? 544 * @object: pointer to query. 545 * 546 * Return: %true if the pointer is to a not-yet-freed object from 547 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer 548 * is to an already-freed object, and %false otherwise. 549 */ 550 bool kmem_valid_obj(void *object) 551 { 552 struct folio *folio; 553 554 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */ 555 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object)) 556 return false; 557 folio = virt_to_folio(object); 558 return folio_test_slab(folio); 559 } 560 EXPORT_SYMBOL_GPL(kmem_valid_obj); 561 562 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab) 563 { 564 if (__kfence_obj_info(kpp, object, slab)) 565 return; 566 __kmem_obj_info(kpp, object, slab); 567 } 568 569 /** 570 * kmem_dump_obj - Print available slab provenance information 571 * @object: slab object for which to find provenance information. 572 * 573 * This function uses pr_cont(), so that the caller is expected to have 574 * printed out whatever preamble is appropriate. The provenance information 575 * depends on the type of object and on how much debugging is enabled. 576 * For a slab-cache object, the fact that it is a slab object is printed, 577 * and, if available, the slab name, return address, and stack trace from 578 * the allocation and last free path of that object. 579 * 580 * This function will splat if passed a pointer to a non-slab object. 581 * If you are not sure what type of object you have, you should instead 582 * use mem_dump_obj(). 583 */ 584 void kmem_dump_obj(void *object) 585 { 586 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc"; 587 int i; 588 struct slab *slab; 589 unsigned long ptroffset; 590 struct kmem_obj_info kp = { }; 591 592 if (WARN_ON_ONCE(!virt_addr_valid(object))) 593 return; 594 slab = virt_to_slab(object); 595 if (WARN_ON_ONCE(!slab)) { 596 pr_cont(" non-slab memory.\n"); 597 return; 598 } 599 kmem_obj_info(&kp, object, slab); 600 if (kp.kp_slab_cache) 601 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name); 602 else 603 pr_cont(" slab%s", cp); 604 if (is_kfence_address(object)) 605 pr_cont(" (kfence)"); 606 if (kp.kp_objp) 607 pr_cont(" start %px", kp.kp_objp); 608 if (kp.kp_data_offset) 609 pr_cont(" data offset %lu", kp.kp_data_offset); 610 if (kp.kp_objp) { 611 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset; 612 pr_cont(" pointer offset %lu", ptroffset); 613 } 614 if (kp.kp_slab_cache && kp.kp_slab_cache->usersize) 615 pr_cont(" size %u", kp.kp_slab_cache->usersize); 616 if (kp.kp_ret) 617 pr_cont(" allocated at %pS\n", kp.kp_ret); 618 else 619 pr_cont("\n"); 620 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) { 621 if (!kp.kp_stack[i]) 622 break; 623 pr_info(" %pS\n", kp.kp_stack[i]); 624 } 625 626 if (kp.kp_free_stack[0]) 627 pr_cont(" Free path:\n"); 628 629 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) { 630 if (!kp.kp_free_stack[i]) 631 break; 632 pr_info(" %pS\n", kp.kp_free_stack[i]); 633 } 634 635 } 636 EXPORT_SYMBOL_GPL(kmem_dump_obj); 637 #endif 638 639 #ifndef CONFIG_SLOB 640 /* Create a cache during boot when no slab services are available yet */ 641 void __init create_boot_cache(struct kmem_cache *s, const char *name, 642 unsigned int size, slab_flags_t flags, 643 unsigned int useroffset, unsigned int usersize) 644 { 645 int err; 646 unsigned int align = ARCH_KMALLOC_MINALIGN; 647 648 s->name = name; 649 s->size = s->object_size = size; 650 651 /* 652 * For power of two sizes, guarantee natural alignment for kmalloc 653 * caches, regardless of SL*B debugging options. 654 */ 655 if (is_power_of_2(size)) 656 align = max(align, size); 657 s->align = calculate_alignment(flags, align, size); 658 659 s->useroffset = useroffset; 660 s->usersize = usersize; 661 662 err = __kmem_cache_create(s, flags); 663 664 if (err) 665 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", 666 name, size, err); 667 668 s->refcount = -1; /* Exempt from merging for now */ 669 } 670 671 struct kmem_cache *__init create_kmalloc_cache(const char *name, 672 unsigned int size, slab_flags_t flags, 673 unsigned int useroffset, unsigned int usersize) 674 { 675 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); 676 677 if (!s) 678 panic("Out of memory when creating slab %s\n", name); 679 680 create_boot_cache(s, name, size, flags, useroffset, usersize); 681 kasan_cache_create_kmalloc(s); 682 list_add(&s->list, &slab_caches); 683 s->refcount = 1; 684 return s; 685 } 686 687 struct kmem_cache * 688 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init = 689 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ }; 690 EXPORT_SYMBOL(kmalloc_caches); 691 692 /* 693 * Conversion table for small slabs sizes / 8 to the index in the 694 * kmalloc array. This is necessary for slabs < 192 since we have non power 695 * of two cache sizes there. The size of larger slabs can be determined using 696 * fls. 697 */ 698 static u8 size_index[24] __ro_after_init = { 699 3, /* 8 */ 700 4, /* 16 */ 701 5, /* 24 */ 702 5, /* 32 */ 703 6, /* 40 */ 704 6, /* 48 */ 705 6, /* 56 */ 706 6, /* 64 */ 707 1, /* 72 */ 708 1, /* 80 */ 709 1, /* 88 */ 710 1, /* 96 */ 711 7, /* 104 */ 712 7, /* 112 */ 713 7, /* 120 */ 714 7, /* 128 */ 715 2, /* 136 */ 716 2, /* 144 */ 717 2, /* 152 */ 718 2, /* 160 */ 719 2, /* 168 */ 720 2, /* 176 */ 721 2, /* 184 */ 722 2 /* 192 */ 723 }; 724 725 static inline unsigned int size_index_elem(unsigned int bytes) 726 { 727 return (bytes - 1) / 8; 728 } 729 730 /* 731 * Find the kmem_cache structure that serves a given size of 732 * allocation 733 */ 734 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) 735 { 736 unsigned int index; 737 738 if (size <= 192) { 739 if (!size) 740 return ZERO_SIZE_PTR; 741 742 index = size_index[size_index_elem(size)]; 743 } else { 744 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE)) 745 return NULL; 746 index = fls(size - 1); 747 } 748 749 return kmalloc_caches[kmalloc_type(flags)][index]; 750 } 751 752 #ifdef CONFIG_ZONE_DMA 753 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz, 754 #else 755 #define KMALLOC_DMA_NAME(sz) 756 #endif 757 758 #ifdef CONFIG_MEMCG_KMEM 759 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz, 760 #else 761 #define KMALLOC_CGROUP_NAME(sz) 762 #endif 763 764 #define INIT_KMALLOC_INFO(__size, __short_size) \ 765 { \ 766 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ 767 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ 768 KMALLOC_CGROUP_NAME(__short_size) \ 769 KMALLOC_DMA_NAME(__short_size) \ 770 .size = __size, \ 771 } 772 773 /* 774 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. 775 * kmalloc_index() supports up to 2^25=32MB, so the final entry of the table is 776 * kmalloc-32M. 777 */ 778 const struct kmalloc_info_struct kmalloc_info[] __initconst = { 779 INIT_KMALLOC_INFO(0, 0), 780 INIT_KMALLOC_INFO(96, 96), 781 INIT_KMALLOC_INFO(192, 192), 782 INIT_KMALLOC_INFO(8, 8), 783 INIT_KMALLOC_INFO(16, 16), 784 INIT_KMALLOC_INFO(32, 32), 785 INIT_KMALLOC_INFO(64, 64), 786 INIT_KMALLOC_INFO(128, 128), 787 INIT_KMALLOC_INFO(256, 256), 788 INIT_KMALLOC_INFO(512, 512), 789 INIT_KMALLOC_INFO(1024, 1k), 790 INIT_KMALLOC_INFO(2048, 2k), 791 INIT_KMALLOC_INFO(4096, 4k), 792 INIT_KMALLOC_INFO(8192, 8k), 793 INIT_KMALLOC_INFO(16384, 16k), 794 INIT_KMALLOC_INFO(32768, 32k), 795 INIT_KMALLOC_INFO(65536, 64k), 796 INIT_KMALLOC_INFO(131072, 128k), 797 INIT_KMALLOC_INFO(262144, 256k), 798 INIT_KMALLOC_INFO(524288, 512k), 799 INIT_KMALLOC_INFO(1048576, 1M), 800 INIT_KMALLOC_INFO(2097152, 2M), 801 INIT_KMALLOC_INFO(4194304, 4M), 802 INIT_KMALLOC_INFO(8388608, 8M), 803 INIT_KMALLOC_INFO(16777216, 16M), 804 INIT_KMALLOC_INFO(33554432, 32M) 805 }; 806 807 /* 808 * Patch up the size_index table if we have strange large alignment 809 * requirements for the kmalloc array. This is only the case for 810 * MIPS it seems. The standard arches will not generate any code here. 811 * 812 * Largest permitted alignment is 256 bytes due to the way we 813 * handle the index determination for the smaller caches. 814 * 815 * Make sure that nothing crazy happens if someone starts tinkering 816 * around with ARCH_KMALLOC_MINALIGN 817 */ 818 void __init setup_kmalloc_cache_index_table(void) 819 { 820 unsigned int i; 821 822 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || 823 !is_power_of_2(KMALLOC_MIN_SIZE)); 824 825 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { 826 unsigned int elem = size_index_elem(i); 827 828 if (elem >= ARRAY_SIZE(size_index)) 829 break; 830 size_index[elem] = KMALLOC_SHIFT_LOW; 831 } 832 833 if (KMALLOC_MIN_SIZE >= 64) { 834 /* 835 * The 96 byte sized cache is not used if the alignment 836 * is 64 byte. 837 */ 838 for (i = 64 + 8; i <= 96; i += 8) 839 size_index[size_index_elem(i)] = 7; 840 841 } 842 843 if (KMALLOC_MIN_SIZE >= 128) { 844 /* 845 * The 192 byte sized cache is not used if the alignment 846 * is 128 byte. Redirect kmalloc to use the 256 byte cache 847 * instead. 848 */ 849 for (i = 128 + 8; i <= 192; i += 8) 850 size_index[size_index_elem(i)] = 8; 851 } 852 } 853 854 static void __init 855 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags) 856 { 857 if (type == KMALLOC_RECLAIM) { 858 flags |= SLAB_RECLAIM_ACCOUNT; 859 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) { 860 if (mem_cgroup_kmem_disabled()) { 861 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx]; 862 return; 863 } 864 flags |= SLAB_ACCOUNT; 865 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) { 866 flags |= SLAB_CACHE_DMA; 867 } 868 869 kmalloc_caches[type][idx] = create_kmalloc_cache( 870 kmalloc_info[idx].name[type], 871 kmalloc_info[idx].size, flags, 0, 872 kmalloc_info[idx].size); 873 874 /* 875 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for 876 * KMALLOC_NORMAL caches. 877 */ 878 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL)) 879 kmalloc_caches[type][idx]->refcount = -1; 880 } 881 882 /* 883 * Create the kmalloc array. Some of the regular kmalloc arrays 884 * may already have been created because they were needed to 885 * enable allocations for slab creation. 886 */ 887 void __init create_kmalloc_caches(slab_flags_t flags) 888 { 889 int i; 890 enum kmalloc_cache_type type; 891 892 /* 893 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined 894 */ 895 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) { 896 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { 897 if (!kmalloc_caches[type][i]) 898 new_kmalloc_cache(i, type, flags); 899 900 /* 901 * Caches that are not of the two-to-the-power-of size. 902 * These have to be created immediately after the 903 * earlier power of two caches 904 */ 905 if (KMALLOC_MIN_SIZE <= 32 && i == 6 && 906 !kmalloc_caches[type][1]) 907 new_kmalloc_cache(1, type, flags); 908 if (KMALLOC_MIN_SIZE <= 64 && i == 7 && 909 !kmalloc_caches[type][2]) 910 new_kmalloc_cache(2, type, flags); 911 } 912 } 913 914 /* Kmalloc array is now usable */ 915 slab_state = UP; 916 } 917 #endif /* !CONFIG_SLOB */ 918 919 gfp_t kmalloc_fix_flags(gfp_t flags) 920 { 921 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; 922 923 flags &= ~GFP_SLAB_BUG_MASK; 924 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", 925 invalid_mask, &invalid_mask, flags, &flags); 926 dump_stack(); 927 928 return flags; 929 } 930 931 /* 932 * To avoid unnecessary overhead, we pass through large allocation requests 933 * directly to the page allocator. We use __GFP_COMP, because we will need to 934 * know the allocation order to free the pages properly in kfree. 935 */ 936 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) 937 { 938 void *ret = NULL; 939 struct page *page; 940 941 if (unlikely(flags & GFP_SLAB_BUG_MASK)) 942 flags = kmalloc_fix_flags(flags); 943 944 flags |= __GFP_COMP; 945 page = alloc_pages(flags, order); 946 if (likely(page)) { 947 ret = page_address(page); 948 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, 949 PAGE_SIZE << order); 950 } 951 ret = kasan_kmalloc_large(ret, size, flags); 952 /* As ret might get tagged, call kmemleak hook after KASAN. */ 953 kmemleak_alloc(ret, size, 1, flags); 954 return ret; 955 } 956 EXPORT_SYMBOL(kmalloc_order); 957 958 #ifdef CONFIG_TRACING 959 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) 960 { 961 void *ret = kmalloc_order(size, flags, order); 962 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); 963 return ret; 964 } 965 EXPORT_SYMBOL(kmalloc_order_trace); 966 #endif 967 968 #ifdef CONFIG_SLAB_FREELIST_RANDOM 969 /* Randomize a generic freelist */ 970 static void freelist_randomize(struct rnd_state *state, unsigned int *list, 971 unsigned int count) 972 { 973 unsigned int rand; 974 unsigned int i; 975 976 for (i = 0; i < count; i++) 977 list[i] = i; 978 979 /* Fisher-Yates shuffle */ 980 for (i = count - 1; i > 0; i--) { 981 rand = prandom_u32_state(state); 982 rand %= (i + 1); 983 swap(list[i], list[rand]); 984 } 985 } 986 987 /* Create a random sequence per cache */ 988 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, 989 gfp_t gfp) 990 { 991 struct rnd_state state; 992 993 if (count < 2 || cachep->random_seq) 994 return 0; 995 996 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); 997 if (!cachep->random_seq) 998 return -ENOMEM; 999 1000 /* Get best entropy at this stage of boot */ 1001 prandom_seed_state(&state, get_random_long()); 1002 1003 freelist_randomize(&state, cachep->random_seq, count); 1004 return 0; 1005 } 1006 1007 /* Destroy the per-cache random freelist sequence */ 1008 void cache_random_seq_destroy(struct kmem_cache *cachep) 1009 { 1010 kfree(cachep->random_seq); 1011 cachep->random_seq = NULL; 1012 } 1013 #endif /* CONFIG_SLAB_FREELIST_RANDOM */ 1014 1015 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) 1016 #ifdef CONFIG_SLAB 1017 #define SLABINFO_RIGHTS (0600) 1018 #else 1019 #define SLABINFO_RIGHTS (0400) 1020 #endif 1021 1022 static void print_slabinfo_header(struct seq_file *m) 1023 { 1024 /* 1025 * Output format version, so at least we can change it 1026 * without _too_ many complaints. 1027 */ 1028 #ifdef CONFIG_DEBUG_SLAB 1029 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); 1030 #else 1031 seq_puts(m, "slabinfo - version: 2.1\n"); 1032 #endif 1033 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); 1034 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); 1035 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); 1036 #ifdef CONFIG_DEBUG_SLAB 1037 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); 1038 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); 1039 #endif 1040 seq_putc(m, '\n'); 1041 } 1042 1043 static void *slab_start(struct seq_file *m, loff_t *pos) 1044 { 1045 mutex_lock(&slab_mutex); 1046 return seq_list_start(&slab_caches, *pos); 1047 } 1048 1049 static void *slab_next(struct seq_file *m, void *p, loff_t *pos) 1050 { 1051 return seq_list_next(p, &slab_caches, pos); 1052 } 1053 1054 static void slab_stop(struct seq_file *m, void *p) 1055 { 1056 mutex_unlock(&slab_mutex); 1057 } 1058 1059 static void cache_show(struct kmem_cache *s, struct seq_file *m) 1060 { 1061 struct slabinfo sinfo; 1062 1063 memset(&sinfo, 0, sizeof(sinfo)); 1064 get_slabinfo(s, &sinfo); 1065 1066 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", 1067 s->name, sinfo.active_objs, sinfo.num_objs, s->size, 1068 sinfo.objects_per_slab, (1 << sinfo.cache_order)); 1069 1070 seq_printf(m, " : tunables %4u %4u %4u", 1071 sinfo.limit, sinfo.batchcount, sinfo.shared); 1072 seq_printf(m, " : slabdata %6lu %6lu %6lu", 1073 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); 1074 slabinfo_show_stats(m, s); 1075 seq_putc(m, '\n'); 1076 } 1077 1078 static int slab_show(struct seq_file *m, void *p) 1079 { 1080 struct kmem_cache *s = list_entry(p, struct kmem_cache, list); 1081 1082 if (p == slab_caches.next) 1083 print_slabinfo_header(m); 1084 cache_show(s, m); 1085 return 0; 1086 } 1087 1088 void dump_unreclaimable_slab(void) 1089 { 1090 struct kmem_cache *s; 1091 struct slabinfo sinfo; 1092 1093 /* 1094 * Here acquiring slab_mutex is risky since we don't prefer to get 1095 * sleep in oom path. But, without mutex hold, it may introduce a 1096 * risk of crash. 1097 * Use mutex_trylock to protect the list traverse, dump nothing 1098 * without acquiring the mutex. 1099 */ 1100 if (!mutex_trylock(&slab_mutex)) { 1101 pr_warn("excessive unreclaimable slab but cannot dump stats\n"); 1102 return; 1103 } 1104 1105 pr_info("Unreclaimable slab info:\n"); 1106 pr_info("Name Used Total\n"); 1107 1108 list_for_each_entry(s, &slab_caches, list) { 1109 if (s->flags & SLAB_RECLAIM_ACCOUNT) 1110 continue; 1111 1112 get_slabinfo(s, &sinfo); 1113 1114 if (sinfo.num_objs > 0) 1115 pr_info("%-17s %10luKB %10luKB\n", s->name, 1116 (sinfo.active_objs * s->size) / 1024, 1117 (sinfo.num_objs * s->size) / 1024); 1118 } 1119 mutex_unlock(&slab_mutex); 1120 } 1121 1122 /* 1123 * slabinfo_op - iterator that generates /proc/slabinfo 1124 * 1125 * Output layout: 1126 * cache-name 1127 * num-active-objs 1128 * total-objs 1129 * object size 1130 * num-active-slabs 1131 * total-slabs 1132 * num-pages-per-slab 1133 * + further values on SMP and with statistics enabled 1134 */ 1135 static const struct seq_operations slabinfo_op = { 1136 .start = slab_start, 1137 .next = slab_next, 1138 .stop = slab_stop, 1139 .show = slab_show, 1140 }; 1141 1142 static int slabinfo_open(struct inode *inode, struct file *file) 1143 { 1144 return seq_open(file, &slabinfo_op); 1145 } 1146 1147 static const struct proc_ops slabinfo_proc_ops = { 1148 .proc_flags = PROC_ENTRY_PERMANENT, 1149 .proc_open = slabinfo_open, 1150 .proc_read = seq_read, 1151 .proc_write = slabinfo_write, 1152 .proc_lseek = seq_lseek, 1153 .proc_release = seq_release, 1154 }; 1155 1156 static int __init slab_proc_init(void) 1157 { 1158 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); 1159 return 0; 1160 } 1161 module_init(slab_proc_init); 1162 1163 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ 1164 1165 static __always_inline void *__do_krealloc(const void *p, size_t new_size, 1166 gfp_t flags) 1167 { 1168 void *ret; 1169 size_t ks; 1170 1171 /* Don't use instrumented ksize to allow precise KASAN poisoning. */ 1172 if (likely(!ZERO_OR_NULL_PTR(p))) { 1173 if (!kasan_check_byte(p)) 1174 return NULL; 1175 ks = kfence_ksize(p) ?: __ksize(p); 1176 } else 1177 ks = 0; 1178 1179 /* If the object still fits, repoison it precisely. */ 1180 if (ks >= new_size) { 1181 p = kasan_krealloc((void *)p, new_size, flags); 1182 return (void *)p; 1183 } 1184 1185 ret = kmalloc_track_caller(new_size, flags); 1186 if (ret && p) { 1187 /* Disable KASAN checks as the object's redzone is accessed. */ 1188 kasan_disable_current(); 1189 memcpy(ret, kasan_reset_tag(p), ks); 1190 kasan_enable_current(); 1191 } 1192 1193 return ret; 1194 } 1195 1196 /** 1197 * krealloc - reallocate memory. The contents will remain unchanged. 1198 * @p: object to reallocate memory for. 1199 * @new_size: how many bytes of memory are required. 1200 * @flags: the type of memory to allocate. 1201 * 1202 * The contents of the object pointed to are preserved up to the 1203 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored). 1204 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size 1205 * is 0 and @p is not a %NULL pointer, the object pointed to is freed. 1206 * 1207 * Return: pointer to the allocated memory or %NULL in case of error 1208 */ 1209 void *krealloc(const void *p, size_t new_size, gfp_t flags) 1210 { 1211 void *ret; 1212 1213 if (unlikely(!new_size)) { 1214 kfree(p); 1215 return ZERO_SIZE_PTR; 1216 } 1217 1218 ret = __do_krealloc(p, new_size, flags); 1219 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) 1220 kfree(p); 1221 1222 return ret; 1223 } 1224 EXPORT_SYMBOL(krealloc); 1225 1226 /** 1227 * kfree_sensitive - Clear sensitive information in memory before freeing 1228 * @p: object to free memory of 1229 * 1230 * The memory of the object @p points to is zeroed before freed. 1231 * If @p is %NULL, kfree_sensitive() does nothing. 1232 * 1233 * Note: this function zeroes the whole allocated buffer which can be a good 1234 * deal bigger than the requested buffer size passed to kmalloc(). So be 1235 * careful when using this function in performance sensitive code. 1236 */ 1237 void kfree_sensitive(const void *p) 1238 { 1239 size_t ks; 1240 void *mem = (void *)p; 1241 1242 ks = ksize(mem); 1243 if (ks) 1244 memzero_explicit(mem, ks); 1245 kfree(mem); 1246 } 1247 EXPORT_SYMBOL(kfree_sensitive); 1248 1249 /** 1250 * ksize - get the actual amount of memory allocated for a given object 1251 * @objp: Pointer to the object 1252 * 1253 * kmalloc may internally round up allocations and return more memory 1254 * than requested. ksize() can be used to determine the actual amount of 1255 * memory allocated. The caller may use this additional memory, even though 1256 * a smaller amount of memory was initially specified with the kmalloc call. 1257 * The caller must guarantee that objp points to a valid object previously 1258 * allocated with either kmalloc() or kmem_cache_alloc(). The object 1259 * must not be freed during the duration of the call. 1260 * 1261 * Return: size of the actual memory used by @objp in bytes 1262 */ 1263 size_t ksize(const void *objp) 1264 { 1265 size_t size; 1266 1267 /* 1268 * We need to first check that the pointer to the object is valid, and 1269 * only then unpoison the memory. The report printed from ksize() is 1270 * more useful, then when it's printed later when the behaviour could 1271 * be undefined due to a potential use-after-free or double-free. 1272 * 1273 * We use kasan_check_byte(), which is supported for the hardware 1274 * tag-based KASAN mode, unlike kasan_check_read/write(). 1275 * 1276 * If the pointed to memory is invalid, we return 0 to avoid users of 1277 * ksize() writing to and potentially corrupting the memory region. 1278 * 1279 * We want to perform the check before __ksize(), to avoid potentially 1280 * crashing in __ksize() due to accessing invalid metadata. 1281 */ 1282 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp)) 1283 return 0; 1284 1285 size = kfence_ksize(objp) ?: __ksize(objp); 1286 /* 1287 * We assume that ksize callers could use whole allocated area, 1288 * so we need to unpoison this area. 1289 */ 1290 kasan_unpoison_range(objp, size); 1291 return size; 1292 } 1293 EXPORT_SYMBOL(ksize); 1294 1295 /* Tracepoints definitions. */ 1296 EXPORT_TRACEPOINT_SYMBOL(kmalloc); 1297 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); 1298 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); 1299 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); 1300 EXPORT_TRACEPOINT_SYMBOL(kfree); 1301 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); 1302 1303 int should_failslab(struct kmem_cache *s, gfp_t gfpflags) 1304 { 1305 if (__should_failslab(s, gfpflags)) 1306 return -ENOMEM; 1307 return 0; 1308 } 1309 ALLOW_ERROR_INJECTION(should_failslab, ERRNO); 1310