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 28 #define CREATE_TRACE_POINTS 29 #include <trace/events/kmem.h> 30 31 #include "internal.h" 32 33 #include "slab.h" 34 35 enum slab_state slab_state; 36 LIST_HEAD(slab_caches); 37 DEFINE_MUTEX(slab_mutex); 38 struct kmem_cache *kmem_cache; 39 40 #ifdef CONFIG_HARDENED_USERCOPY 41 bool usercopy_fallback __ro_after_init = 42 IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK); 43 module_param(usercopy_fallback, bool, 0400); 44 MODULE_PARM_DESC(usercopy_fallback, 45 "WARN instead of reject usercopy whitelist violations"); 46 #endif 47 48 static LIST_HEAD(slab_caches_to_rcu_destroy); 49 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); 50 static DECLARE_WORK(slab_caches_to_rcu_destroy_work, 51 slab_caches_to_rcu_destroy_workfn); 52 53 /* 54 * Set of flags that will prevent slab merging 55 */ 56 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ 57 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ 58 SLAB_FAILSLAB | kasan_never_merge()) 59 60 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ 61 SLAB_CACHE_DMA32 | SLAB_ACCOUNT) 62 63 /* 64 * Merge control. If this is set then no merging of slab caches will occur. 65 */ 66 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); 67 68 static int __init setup_slab_nomerge(char *str) 69 { 70 slab_nomerge = true; 71 return 1; 72 } 73 74 static int __init setup_slab_merge(char *str) 75 { 76 slab_nomerge = false; 77 return 1; 78 } 79 80 #ifdef CONFIG_SLUB 81 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); 82 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0); 83 #endif 84 85 __setup("slab_nomerge", setup_slab_nomerge); 86 __setup("slab_merge", setup_slab_merge); 87 88 /* 89 * Determine the size of a slab object 90 */ 91 unsigned int kmem_cache_size(struct kmem_cache *s) 92 { 93 return s->object_size; 94 } 95 EXPORT_SYMBOL(kmem_cache_size); 96 97 #ifdef CONFIG_DEBUG_VM 98 static int kmem_cache_sanity_check(const char *name, unsigned int size) 99 { 100 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) { 101 pr_err("kmem_cache_create(%s) integrity check failed\n", name); 102 return -EINVAL; 103 } 104 105 WARN_ON(strchr(name, ' ')); /* It confuses parsers */ 106 return 0; 107 } 108 #else 109 static inline int kmem_cache_sanity_check(const char *name, unsigned int size) 110 { 111 return 0; 112 } 113 #endif 114 115 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) 116 { 117 size_t i; 118 119 for (i = 0; i < nr; i++) { 120 if (s) 121 kmem_cache_free(s, p[i]); 122 else 123 kfree(p[i]); 124 } 125 } 126 127 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, 128 void **p) 129 { 130 size_t i; 131 132 for (i = 0; i < nr; i++) { 133 void *x = p[i] = kmem_cache_alloc(s, flags); 134 if (!x) { 135 __kmem_cache_free_bulk(s, i, p); 136 return 0; 137 } 138 } 139 return i; 140 } 141 142 /* 143 * Figure out what the alignment of the objects will be given a set of 144 * flags, a user specified alignment and the size of the objects. 145 */ 146 static unsigned int calculate_alignment(slab_flags_t flags, 147 unsigned int align, unsigned int size) 148 { 149 /* 150 * If the user wants hardware cache aligned objects then follow that 151 * suggestion if the object is sufficiently large. 152 * 153 * The hardware cache alignment cannot override the specified 154 * alignment though. If that is greater then use it. 155 */ 156 if (flags & SLAB_HWCACHE_ALIGN) { 157 unsigned int ralign; 158 159 ralign = cache_line_size(); 160 while (size <= ralign / 2) 161 ralign /= 2; 162 align = max(align, ralign); 163 } 164 165 if (align < ARCH_SLAB_MINALIGN) 166 align = ARCH_SLAB_MINALIGN; 167 168 return ALIGN(align, sizeof(void *)); 169 } 170 171 /* 172 * Find a mergeable slab cache 173 */ 174 int slab_unmergeable(struct kmem_cache *s) 175 { 176 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) 177 return 1; 178 179 if (s->ctor) 180 return 1; 181 182 if (s->usersize) 183 return 1; 184 185 /* 186 * We may have set a slab to be unmergeable during bootstrap. 187 */ 188 if (s->refcount < 0) 189 return 1; 190 191 return 0; 192 } 193 194 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align, 195 slab_flags_t flags, const char *name, void (*ctor)(void *)) 196 { 197 struct kmem_cache *s; 198 199 if (slab_nomerge) 200 return NULL; 201 202 if (ctor) 203 return NULL; 204 205 size = ALIGN(size, sizeof(void *)); 206 align = calculate_alignment(flags, align, size); 207 size = ALIGN(size, align); 208 flags = kmem_cache_flags(size, flags, name); 209 210 if (flags & SLAB_NEVER_MERGE) 211 return NULL; 212 213 list_for_each_entry_reverse(s, &slab_caches, list) { 214 if (slab_unmergeable(s)) 215 continue; 216 217 if (size > s->size) 218 continue; 219 220 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) 221 continue; 222 /* 223 * Check if alignment is compatible. 224 * Courtesy of Adrian Drzewiecki 225 */ 226 if ((s->size & ~(align - 1)) != s->size) 227 continue; 228 229 if (s->size - size >= sizeof(void *)) 230 continue; 231 232 if (IS_ENABLED(CONFIG_SLAB) && align && 233 (align > s->align || s->align % align)) 234 continue; 235 236 return s; 237 } 238 return NULL; 239 } 240 241 static struct kmem_cache *create_cache(const char *name, 242 unsigned int object_size, unsigned int align, 243 slab_flags_t flags, unsigned int useroffset, 244 unsigned int usersize, void (*ctor)(void *), 245 struct kmem_cache *root_cache) 246 { 247 struct kmem_cache *s; 248 int err; 249 250 if (WARN_ON(useroffset + usersize > object_size)) 251 useroffset = usersize = 0; 252 253 err = -ENOMEM; 254 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); 255 if (!s) 256 goto out; 257 258 s->name = name; 259 s->size = s->object_size = object_size; 260 s->align = align; 261 s->ctor = ctor; 262 s->useroffset = useroffset; 263 s->usersize = usersize; 264 265 err = __kmem_cache_create(s, flags); 266 if (err) 267 goto out_free_cache; 268 269 s->refcount = 1; 270 list_add(&s->list, &slab_caches); 271 out: 272 if (err) 273 return ERR_PTR(err); 274 return s; 275 276 out_free_cache: 277 kmem_cache_free(kmem_cache, s); 278 goto out; 279 } 280 281 /** 282 * kmem_cache_create_usercopy - Create a cache with a region suitable 283 * for copying to userspace 284 * @name: A string which is used in /proc/slabinfo to identify this cache. 285 * @size: The size of objects to be created in this cache. 286 * @align: The required alignment for the objects. 287 * @flags: SLAB flags 288 * @useroffset: Usercopy region offset 289 * @usersize: Usercopy region size 290 * @ctor: A constructor for the objects. 291 * 292 * Cannot be called within a interrupt, but can be interrupted. 293 * The @ctor is run when new pages are allocated by the cache. 294 * 295 * The flags are 296 * 297 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) 298 * to catch references to uninitialised memory. 299 * 300 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check 301 * for buffer overruns. 302 * 303 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware 304 * cacheline. This can be beneficial if you're counting cycles as closely 305 * as davem. 306 * 307 * Return: a pointer to the cache on success, NULL on failure. 308 */ 309 struct kmem_cache * 310 kmem_cache_create_usercopy(const char *name, 311 unsigned int size, unsigned int align, 312 slab_flags_t flags, 313 unsigned int useroffset, unsigned int usersize, 314 void (*ctor)(void *)) 315 { 316 struct kmem_cache *s = NULL; 317 const char *cache_name; 318 int err; 319 320 #ifdef CONFIG_SLUB_DEBUG 321 /* 322 * If no slub_debug was enabled globally, the static key is not yet 323 * enabled by setup_slub_debug(). Enable it if the cache is being 324 * created with any of the debugging flags passed explicitly. 325 */ 326 if (flags & SLAB_DEBUG_FLAGS) 327 static_branch_enable(&slub_debug_enabled); 328 #endif 329 330 mutex_lock(&slab_mutex); 331 332 err = kmem_cache_sanity_check(name, size); 333 if (err) { 334 goto out_unlock; 335 } 336 337 /* Refuse requests with allocator specific flags */ 338 if (flags & ~SLAB_FLAGS_PERMITTED) { 339 err = -EINVAL; 340 goto out_unlock; 341 } 342 343 /* 344 * Some allocators will constraint the set of valid flags to a subset 345 * of all flags. We expect them to define CACHE_CREATE_MASK in this 346 * case, and we'll just provide them with a sanitized version of the 347 * passed flags. 348 */ 349 flags &= CACHE_CREATE_MASK; 350 351 /* Fail closed on bad usersize of useroffset values. */ 352 if (WARN_ON(!usersize && useroffset) || 353 WARN_ON(size < usersize || size - usersize < useroffset)) 354 usersize = useroffset = 0; 355 356 if (!usersize) 357 s = __kmem_cache_alias(name, size, align, flags, ctor); 358 if (s) 359 goto out_unlock; 360 361 cache_name = kstrdup_const(name, GFP_KERNEL); 362 if (!cache_name) { 363 err = -ENOMEM; 364 goto out_unlock; 365 } 366 367 s = create_cache(cache_name, size, 368 calculate_alignment(flags, align, size), 369 flags, useroffset, usersize, ctor, NULL); 370 if (IS_ERR(s)) { 371 err = PTR_ERR(s); 372 kfree_const(cache_name); 373 } 374 375 out_unlock: 376 mutex_unlock(&slab_mutex); 377 378 if (err) { 379 if (flags & SLAB_PANIC) 380 panic("%s: Failed to create slab '%s'. Error %d\n", 381 __func__, name, err); 382 else { 383 pr_warn("%s(%s) failed with error %d\n", 384 __func__, name, err); 385 dump_stack(); 386 } 387 return NULL; 388 } 389 return s; 390 } 391 EXPORT_SYMBOL(kmem_cache_create_usercopy); 392 393 /** 394 * kmem_cache_create - Create a cache. 395 * @name: A string which is used in /proc/slabinfo to identify this cache. 396 * @size: The size of objects to be created in this cache. 397 * @align: The required alignment for the objects. 398 * @flags: SLAB flags 399 * @ctor: A constructor for the objects. 400 * 401 * Cannot be called within a interrupt, but can be interrupted. 402 * The @ctor is run when new pages are allocated by the cache. 403 * 404 * The flags are 405 * 406 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) 407 * to catch references to uninitialised memory. 408 * 409 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check 410 * for buffer overruns. 411 * 412 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware 413 * cacheline. This can be beneficial if you're counting cycles as closely 414 * as davem. 415 * 416 * Return: a pointer to the cache on success, NULL on failure. 417 */ 418 struct kmem_cache * 419 kmem_cache_create(const char *name, unsigned int size, unsigned int align, 420 slab_flags_t flags, void (*ctor)(void *)) 421 { 422 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0, 423 ctor); 424 } 425 EXPORT_SYMBOL(kmem_cache_create); 426 427 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) 428 { 429 LIST_HEAD(to_destroy); 430 struct kmem_cache *s, *s2; 431 432 /* 433 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the 434 * @slab_caches_to_rcu_destroy list. The slab pages are freed 435 * through RCU and the associated kmem_cache are dereferenced 436 * while freeing the pages, so the kmem_caches should be freed only 437 * after the pending RCU operations are finished. As rcu_barrier() 438 * is a pretty slow operation, we batch all pending destructions 439 * asynchronously. 440 */ 441 mutex_lock(&slab_mutex); 442 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); 443 mutex_unlock(&slab_mutex); 444 445 if (list_empty(&to_destroy)) 446 return; 447 448 rcu_barrier(); 449 450 list_for_each_entry_safe(s, s2, &to_destroy, list) { 451 debugfs_slab_release(s); 452 kfence_shutdown_cache(s); 453 #ifdef SLAB_SUPPORTS_SYSFS 454 sysfs_slab_release(s); 455 #else 456 slab_kmem_cache_release(s); 457 #endif 458 } 459 } 460 461 static int shutdown_cache(struct kmem_cache *s) 462 { 463 /* free asan quarantined objects */ 464 kasan_cache_shutdown(s); 465 466 if (__kmem_cache_shutdown(s) != 0) 467 return -EBUSY; 468 469 list_del(&s->list); 470 471 if (s->flags & SLAB_TYPESAFE_BY_RCU) { 472 #ifdef SLAB_SUPPORTS_SYSFS 473 sysfs_slab_unlink(s); 474 #endif 475 list_add_tail(&s->list, &slab_caches_to_rcu_destroy); 476 schedule_work(&slab_caches_to_rcu_destroy_work); 477 } else { 478 kfence_shutdown_cache(s); 479 debugfs_slab_release(s); 480 #ifdef SLAB_SUPPORTS_SYSFS 481 sysfs_slab_unlink(s); 482 sysfs_slab_release(s); 483 #else 484 slab_kmem_cache_release(s); 485 #endif 486 } 487 488 return 0; 489 } 490 491 void slab_kmem_cache_release(struct kmem_cache *s) 492 { 493 __kmem_cache_release(s); 494 kfree_const(s->name); 495 kmem_cache_free(kmem_cache, s); 496 } 497 498 void kmem_cache_destroy(struct kmem_cache *s) 499 { 500 int err; 501 502 if (unlikely(!s)) 503 return; 504 505 cpus_read_lock(); 506 mutex_lock(&slab_mutex); 507 508 s->refcount--; 509 if (s->refcount) 510 goto out_unlock; 511 512 err = shutdown_cache(s); 513 if (err) { 514 pr_err("%s %s: Slab cache still has objects\n", 515 __func__, s->name); 516 dump_stack(); 517 } 518 out_unlock: 519 mutex_unlock(&slab_mutex); 520 cpus_read_unlock(); 521 } 522 EXPORT_SYMBOL(kmem_cache_destroy); 523 524 /** 525 * kmem_cache_shrink - Shrink a cache. 526 * @cachep: The cache to shrink. 527 * 528 * Releases as many slabs as possible for a cache. 529 * To help debugging, a zero exit status indicates all slabs were released. 530 * 531 * Return: %0 if all slabs were released, non-zero otherwise 532 */ 533 int kmem_cache_shrink(struct kmem_cache *cachep) 534 { 535 int ret; 536 537 538 kasan_cache_shrink(cachep); 539 ret = __kmem_cache_shrink(cachep); 540 541 return ret; 542 } 543 EXPORT_SYMBOL(kmem_cache_shrink); 544 545 bool slab_is_available(void) 546 { 547 return slab_state >= UP; 548 } 549 550 #ifdef CONFIG_PRINTK 551 /** 552 * kmem_valid_obj - does the pointer reference a valid slab object? 553 * @object: pointer to query. 554 * 555 * Return: %true if the pointer is to a not-yet-freed object from 556 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer 557 * is to an already-freed object, and %false otherwise. 558 */ 559 bool kmem_valid_obj(void *object) 560 { 561 struct page *page; 562 563 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */ 564 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object)) 565 return false; 566 page = virt_to_head_page(object); 567 return PageSlab(page); 568 } 569 EXPORT_SYMBOL_GPL(kmem_valid_obj); 570 571 /** 572 * kmem_dump_obj - Print available slab provenance information 573 * @object: slab object for which to find provenance information. 574 * 575 * This function uses pr_cont(), so that the caller is expected to have 576 * printed out whatever preamble is appropriate. The provenance information 577 * depends on the type of object and on how much debugging is enabled. 578 * For a slab-cache object, the fact that it is a slab object is printed, 579 * and, if available, the slab name, return address, and stack trace from 580 * the allocation and last free path of that object. 581 * 582 * This function will splat if passed a pointer to a non-slab object. 583 * If you are not sure what type of object you have, you should instead 584 * use mem_dump_obj(). 585 */ 586 void kmem_dump_obj(void *object) 587 { 588 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc"; 589 int i; 590 struct page *page; 591 unsigned long ptroffset; 592 struct kmem_obj_info kp = { }; 593 594 if (WARN_ON_ONCE(!virt_addr_valid(object))) 595 return; 596 page = virt_to_head_page(object); 597 if (WARN_ON_ONCE(!PageSlab(page))) { 598 pr_cont(" non-slab memory.\n"); 599 return; 600 } 601 kmem_obj_info(&kp, object, page); 602 if (kp.kp_slab_cache) 603 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name); 604 else 605 pr_cont(" slab%s", cp); 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 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); 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 size 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 (cgroup_memory_nokmem) { 861 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx]; 862 return; 863 } 864 flags |= SLAB_ACCOUNT; 865 } 866 867 kmalloc_caches[type][idx] = create_kmalloc_cache( 868 kmalloc_info[idx].name[type], 869 kmalloc_info[idx].size, flags, 0, 870 kmalloc_info[idx].size); 871 872 /* 873 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for 874 * KMALLOC_NORMAL caches. 875 */ 876 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL)) 877 kmalloc_caches[type][idx]->refcount = -1; 878 } 879 880 /* 881 * Create the kmalloc array. Some of the regular kmalloc arrays 882 * may already have been created because they were needed to 883 * enable allocations for slab creation. 884 */ 885 void __init create_kmalloc_caches(slab_flags_t flags) 886 { 887 int i; 888 enum kmalloc_cache_type type; 889 890 /* 891 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined 892 */ 893 for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) { 894 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { 895 if (!kmalloc_caches[type][i]) 896 new_kmalloc_cache(i, type, flags); 897 898 /* 899 * Caches that are not of the two-to-the-power-of size. 900 * These have to be created immediately after the 901 * earlier power of two caches 902 */ 903 if (KMALLOC_MIN_SIZE <= 32 && i == 6 && 904 !kmalloc_caches[type][1]) 905 new_kmalloc_cache(1, type, flags); 906 if (KMALLOC_MIN_SIZE <= 64 && i == 7 && 907 !kmalloc_caches[type][2]) 908 new_kmalloc_cache(2, type, flags); 909 } 910 } 911 912 /* Kmalloc array is now usable */ 913 slab_state = UP; 914 915 #ifdef CONFIG_ZONE_DMA 916 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { 917 struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i]; 918 919 if (s) { 920 kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache( 921 kmalloc_info[i].name[KMALLOC_DMA], 922 kmalloc_info[i].size, 923 SLAB_CACHE_DMA | flags, 0, 924 kmalloc_info[i].size); 925 } 926 } 927 #endif 928 } 929 #endif /* !CONFIG_SLOB */ 930 931 gfp_t kmalloc_fix_flags(gfp_t flags) 932 { 933 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; 934 935 flags &= ~GFP_SLAB_BUG_MASK; 936 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", 937 invalid_mask, &invalid_mask, flags, &flags); 938 dump_stack(); 939 940 return flags; 941 } 942 943 /* 944 * To avoid unnecessary overhead, we pass through large allocation requests 945 * directly to the page allocator. We use __GFP_COMP, because we will need to 946 * know the allocation order to free the pages properly in kfree. 947 */ 948 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) 949 { 950 void *ret = NULL; 951 struct page *page; 952 953 if (unlikely(flags & GFP_SLAB_BUG_MASK)) 954 flags = kmalloc_fix_flags(flags); 955 956 flags |= __GFP_COMP; 957 page = alloc_pages(flags, order); 958 if (likely(page)) { 959 ret = page_address(page); 960 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, 961 PAGE_SIZE << order); 962 } 963 ret = kasan_kmalloc_large(ret, size, flags); 964 /* As ret might get tagged, call kmemleak hook after KASAN. */ 965 kmemleak_alloc(ret, size, 1, flags); 966 return ret; 967 } 968 EXPORT_SYMBOL(kmalloc_order); 969 970 #ifdef CONFIG_TRACING 971 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) 972 { 973 void *ret = kmalloc_order(size, flags, order); 974 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); 975 return ret; 976 } 977 EXPORT_SYMBOL(kmalloc_order_trace); 978 #endif 979 980 #ifdef CONFIG_SLAB_FREELIST_RANDOM 981 /* Randomize a generic freelist */ 982 static void freelist_randomize(struct rnd_state *state, unsigned int *list, 983 unsigned int count) 984 { 985 unsigned int rand; 986 unsigned int i; 987 988 for (i = 0; i < count; i++) 989 list[i] = i; 990 991 /* Fisher-Yates shuffle */ 992 for (i = count - 1; i > 0; i--) { 993 rand = prandom_u32_state(state); 994 rand %= (i + 1); 995 swap(list[i], list[rand]); 996 } 997 } 998 999 /* Create a random sequence per cache */ 1000 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, 1001 gfp_t gfp) 1002 { 1003 struct rnd_state state; 1004 1005 if (count < 2 || cachep->random_seq) 1006 return 0; 1007 1008 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); 1009 if (!cachep->random_seq) 1010 return -ENOMEM; 1011 1012 /* Get best entropy at this stage of boot */ 1013 prandom_seed_state(&state, get_random_long()); 1014 1015 freelist_randomize(&state, cachep->random_seq, count); 1016 return 0; 1017 } 1018 1019 /* Destroy the per-cache random freelist sequence */ 1020 void cache_random_seq_destroy(struct kmem_cache *cachep) 1021 { 1022 kfree(cachep->random_seq); 1023 cachep->random_seq = NULL; 1024 } 1025 #endif /* CONFIG_SLAB_FREELIST_RANDOM */ 1026 1027 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) 1028 #ifdef CONFIG_SLAB 1029 #define SLABINFO_RIGHTS (0600) 1030 #else 1031 #define SLABINFO_RIGHTS (0400) 1032 #endif 1033 1034 static void print_slabinfo_header(struct seq_file *m) 1035 { 1036 /* 1037 * Output format version, so at least we can change it 1038 * without _too_ many complaints. 1039 */ 1040 #ifdef CONFIG_DEBUG_SLAB 1041 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); 1042 #else 1043 seq_puts(m, "slabinfo - version: 2.1\n"); 1044 #endif 1045 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); 1046 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); 1047 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); 1048 #ifdef CONFIG_DEBUG_SLAB 1049 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); 1050 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); 1051 #endif 1052 seq_putc(m, '\n'); 1053 } 1054 1055 void *slab_start(struct seq_file *m, loff_t *pos) 1056 { 1057 mutex_lock(&slab_mutex); 1058 return seq_list_start(&slab_caches, *pos); 1059 } 1060 1061 void *slab_next(struct seq_file *m, void *p, loff_t *pos) 1062 { 1063 return seq_list_next(p, &slab_caches, pos); 1064 } 1065 1066 void slab_stop(struct seq_file *m, void *p) 1067 { 1068 mutex_unlock(&slab_mutex); 1069 } 1070 1071 static void cache_show(struct kmem_cache *s, struct seq_file *m) 1072 { 1073 struct slabinfo sinfo; 1074 1075 memset(&sinfo, 0, sizeof(sinfo)); 1076 get_slabinfo(s, &sinfo); 1077 1078 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", 1079 s->name, sinfo.active_objs, sinfo.num_objs, s->size, 1080 sinfo.objects_per_slab, (1 << sinfo.cache_order)); 1081 1082 seq_printf(m, " : tunables %4u %4u %4u", 1083 sinfo.limit, sinfo.batchcount, sinfo.shared); 1084 seq_printf(m, " : slabdata %6lu %6lu %6lu", 1085 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); 1086 slabinfo_show_stats(m, s); 1087 seq_putc(m, '\n'); 1088 } 1089 1090 static int slab_show(struct seq_file *m, void *p) 1091 { 1092 struct kmem_cache *s = list_entry(p, struct kmem_cache, list); 1093 1094 if (p == slab_caches.next) 1095 print_slabinfo_header(m); 1096 cache_show(s, m); 1097 return 0; 1098 } 1099 1100 void dump_unreclaimable_slab(void) 1101 { 1102 struct kmem_cache *s; 1103 struct slabinfo sinfo; 1104 1105 /* 1106 * Here acquiring slab_mutex is risky since we don't prefer to get 1107 * sleep in oom path. But, without mutex hold, it may introduce a 1108 * risk of crash. 1109 * Use mutex_trylock to protect the list traverse, dump nothing 1110 * without acquiring the mutex. 1111 */ 1112 if (!mutex_trylock(&slab_mutex)) { 1113 pr_warn("excessive unreclaimable slab but cannot dump stats\n"); 1114 return; 1115 } 1116 1117 pr_info("Unreclaimable slab info:\n"); 1118 pr_info("Name Used Total\n"); 1119 1120 list_for_each_entry(s, &slab_caches, list) { 1121 if (s->flags & SLAB_RECLAIM_ACCOUNT) 1122 continue; 1123 1124 get_slabinfo(s, &sinfo); 1125 1126 if (sinfo.num_objs > 0) 1127 pr_info("%-17s %10luKB %10luKB\n", s->name, 1128 (sinfo.active_objs * s->size) / 1024, 1129 (sinfo.num_objs * s->size) / 1024); 1130 } 1131 mutex_unlock(&slab_mutex); 1132 } 1133 1134 #if defined(CONFIG_MEMCG_KMEM) 1135 int memcg_slab_show(struct seq_file *m, void *p) 1136 { 1137 /* 1138 * Deprecated. 1139 * Please, take a look at tools/cgroup/slabinfo.py . 1140 */ 1141 return 0; 1142 } 1143 #endif 1144 1145 /* 1146 * slabinfo_op - iterator that generates /proc/slabinfo 1147 * 1148 * Output layout: 1149 * cache-name 1150 * num-active-objs 1151 * total-objs 1152 * object size 1153 * num-active-slabs 1154 * total-slabs 1155 * num-pages-per-slab 1156 * + further values on SMP and with statistics enabled 1157 */ 1158 static const struct seq_operations slabinfo_op = { 1159 .start = slab_start, 1160 .next = slab_next, 1161 .stop = slab_stop, 1162 .show = slab_show, 1163 }; 1164 1165 static int slabinfo_open(struct inode *inode, struct file *file) 1166 { 1167 return seq_open(file, &slabinfo_op); 1168 } 1169 1170 static const struct proc_ops slabinfo_proc_ops = { 1171 .proc_flags = PROC_ENTRY_PERMANENT, 1172 .proc_open = slabinfo_open, 1173 .proc_read = seq_read, 1174 .proc_write = slabinfo_write, 1175 .proc_lseek = seq_lseek, 1176 .proc_release = seq_release, 1177 }; 1178 1179 static int __init slab_proc_init(void) 1180 { 1181 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); 1182 return 0; 1183 } 1184 module_init(slab_proc_init); 1185 1186 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ 1187 1188 static __always_inline void *__do_krealloc(const void *p, size_t new_size, 1189 gfp_t flags) 1190 { 1191 void *ret; 1192 size_t ks; 1193 1194 /* Don't use instrumented ksize to allow precise KASAN poisoning. */ 1195 if (likely(!ZERO_OR_NULL_PTR(p))) { 1196 if (!kasan_check_byte(p)) 1197 return NULL; 1198 ks = kfence_ksize(p) ?: __ksize(p); 1199 } else 1200 ks = 0; 1201 1202 /* If the object still fits, repoison it precisely. */ 1203 if (ks >= new_size) { 1204 p = kasan_krealloc((void *)p, new_size, flags); 1205 return (void *)p; 1206 } 1207 1208 ret = kmalloc_track_caller(new_size, flags); 1209 if (ret && p) { 1210 /* Disable KASAN checks as the object's redzone is accessed. */ 1211 kasan_disable_current(); 1212 memcpy(ret, kasan_reset_tag(p), ks); 1213 kasan_enable_current(); 1214 } 1215 1216 return ret; 1217 } 1218 1219 /** 1220 * krealloc - reallocate memory. The contents will remain unchanged. 1221 * @p: object to reallocate memory for. 1222 * @new_size: how many bytes of memory are required. 1223 * @flags: the type of memory to allocate. 1224 * 1225 * The contents of the object pointed to are preserved up to the 1226 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored). 1227 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size 1228 * is 0 and @p is not a %NULL pointer, the object pointed to is freed. 1229 * 1230 * Return: pointer to the allocated memory or %NULL in case of error 1231 */ 1232 void *krealloc(const void *p, size_t new_size, gfp_t flags) 1233 { 1234 void *ret; 1235 1236 if (unlikely(!new_size)) { 1237 kfree(p); 1238 return ZERO_SIZE_PTR; 1239 } 1240 1241 ret = __do_krealloc(p, new_size, flags); 1242 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) 1243 kfree(p); 1244 1245 return ret; 1246 } 1247 EXPORT_SYMBOL(krealloc); 1248 1249 /** 1250 * kfree_sensitive - Clear sensitive information in memory before freeing 1251 * @p: object to free memory of 1252 * 1253 * The memory of the object @p points to is zeroed before freed. 1254 * If @p is %NULL, kfree_sensitive() does nothing. 1255 * 1256 * Note: this function zeroes the whole allocated buffer which can be a good 1257 * deal bigger than the requested buffer size passed to kmalloc(). So be 1258 * careful when using this function in performance sensitive code. 1259 */ 1260 void kfree_sensitive(const void *p) 1261 { 1262 size_t ks; 1263 void *mem = (void *)p; 1264 1265 ks = ksize(mem); 1266 if (ks) 1267 memzero_explicit(mem, ks); 1268 kfree(mem); 1269 } 1270 EXPORT_SYMBOL(kfree_sensitive); 1271 1272 /** 1273 * ksize - get the actual amount of memory allocated for a given object 1274 * @objp: Pointer to the object 1275 * 1276 * kmalloc may internally round up allocations and return more memory 1277 * than requested. ksize() can be used to determine the actual amount of 1278 * memory allocated. The caller may use this additional memory, even though 1279 * a smaller amount of memory was initially specified with the kmalloc call. 1280 * The caller must guarantee that objp points to a valid object previously 1281 * allocated with either kmalloc() or kmem_cache_alloc(). The object 1282 * must not be freed during the duration of the call. 1283 * 1284 * Return: size of the actual memory used by @objp in bytes 1285 */ 1286 size_t ksize(const void *objp) 1287 { 1288 size_t size; 1289 1290 /* 1291 * We need to first check that the pointer to the object is valid, and 1292 * only then unpoison the memory. The report printed from ksize() is 1293 * more useful, then when it's printed later when the behaviour could 1294 * be undefined due to a potential use-after-free or double-free. 1295 * 1296 * We use kasan_check_byte(), which is supported for the hardware 1297 * tag-based KASAN mode, unlike kasan_check_read/write(). 1298 * 1299 * If the pointed to memory is invalid, we return 0 to avoid users of 1300 * ksize() writing to and potentially corrupting the memory region. 1301 * 1302 * We want to perform the check before __ksize(), to avoid potentially 1303 * crashing in __ksize() due to accessing invalid metadata. 1304 */ 1305 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp)) 1306 return 0; 1307 1308 size = kfence_ksize(objp) ?: __ksize(objp); 1309 /* 1310 * We assume that ksize callers could use whole allocated area, 1311 * so we need to unpoison this area. 1312 */ 1313 kasan_unpoison_range(objp, size); 1314 return size; 1315 } 1316 EXPORT_SYMBOL(ksize); 1317 1318 /* Tracepoints definitions. */ 1319 EXPORT_TRACEPOINT_SYMBOL(kmalloc); 1320 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); 1321 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); 1322 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); 1323 EXPORT_TRACEPOINT_SYMBOL(kfree); 1324 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); 1325 1326 int should_failslab(struct kmem_cache *s, gfp_t gfpflags) 1327 { 1328 if (__should_failslab(s, gfpflags)) 1329 return -ENOMEM; 1330 return 0; 1331 } 1332 ALLOW_ERROR_INJECTION(should_failslab, ERRNO); 1333