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("kmem_cache_create: Failed to create slab '%s'. Error %d\n", 381 name, err); 382 else { 383 pr_warn("kmem_cache_create(%s) failed with error %d\n", 384 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 kfence_shutdown_cache(s); 452 #ifdef SLAB_SUPPORTS_SYSFS 453 sysfs_slab_release(s); 454 #else 455 slab_kmem_cache_release(s); 456 #endif 457 } 458 } 459 460 static int shutdown_cache(struct kmem_cache *s) 461 { 462 /* free asan quarantined objects */ 463 kasan_cache_shutdown(s); 464 465 if (__kmem_cache_shutdown(s) != 0) 466 return -EBUSY; 467 468 list_del(&s->list); 469 470 if (s->flags & SLAB_TYPESAFE_BY_RCU) { 471 #ifdef SLAB_SUPPORTS_SYSFS 472 sysfs_slab_unlink(s); 473 #endif 474 list_add_tail(&s->list, &slab_caches_to_rcu_destroy); 475 schedule_work(&slab_caches_to_rcu_destroy_work); 476 } else { 477 kfence_shutdown_cache(s); 478 #ifdef SLAB_SUPPORTS_SYSFS 479 sysfs_slab_unlink(s); 480 sysfs_slab_release(s); 481 #else 482 slab_kmem_cache_release(s); 483 #endif 484 } 485 486 return 0; 487 } 488 489 void slab_kmem_cache_release(struct kmem_cache *s) 490 { 491 __kmem_cache_release(s); 492 kfree_const(s->name); 493 kmem_cache_free(kmem_cache, s); 494 } 495 496 void kmem_cache_destroy(struct kmem_cache *s) 497 { 498 int err; 499 500 if (unlikely(!s)) 501 return; 502 503 mutex_lock(&slab_mutex); 504 505 s->refcount--; 506 if (s->refcount) 507 goto out_unlock; 508 509 err = shutdown_cache(s); 510 if (err) { 511 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n", 512 s->name); 513 dump_stack(); 514 } 515 out_unlock: 516 mutex_unlock(&slab_mutex); 517 } 518 EXPORT_SYMBOL(kmem_cache_destroy); 519 520 /** 521 * kmem_cache_shrink - Shrink a cache. 522 * @cachep: The cache to shrink. 523 * 524 * Releases as many slabs as possible for a cache. 525 * To help debugging, a zero exit status indicates all slabs were released. 526 * 527 * Return: %0 if all slabs were released, non-zero otherwise 528 */ 529 int kmem_cache_shrink(struct kmem_cache *cachep) 530 { 531 int ret; 532 533 534 kasan_cache_shrink(cachep); 535 ret = __kmem_cache_shrink(cachep); 536 537 return ret; 538 } 539 EXPORT_SYMBOL(kmem_cache_shrink); 540 541 bool slab_is_available(void) 542 { 543 return slab_state >= UP; 544 } 545 546 #ifdef CONFIG_PRINTK 547 /** 548 * kmem_valid_obj - does the pointer reference a valid slab object? 549 * @object: pointer to query. 550 * 551 * Return: %true if the pointer is to a not-yet-freed object from 552 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer 553 * is to an already-freed object, and %false otherwise. 554 */ 555 bool kmem_valid_obj(void *object) 556 { 557 struct page *page; 558 559 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */ 560 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object)) 561 return false; 562 page = virt_to_head_page(object); 563 return PageSlab(page); 564 } 565 EXPORT_SYMBOL_GPL(kmem_valid_obj); 566 567 /** 568 * kmem_dump_obj - Print available slab provenance information 569 * @object: slab object for which to find provenance information. 570 * 571 * This function uses pr_cont(), so that the caller is expected to have 572 * printed out whatever preamble is appropriate. The provenance information 573 * depends on the type of object and on how much debugging is enabled. 574 * For a slab-cache object, the fact that it is a slab object is printed, 575 * and, if available, the slab name, return address, and stack trace from 576 * the allocation of that object. 577 * 578 * This function will splat if passed a pointer to a non-slab object. 579 * If you are not sure what type of object you have, you should instead 580 * use mem_dump_obj(). 581 */ 582 void kmem_dump_obj(void *object) 583 { 584 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc"; 585 int i; 586 struct page *page; 587 unsigned long ptroffset; 588 struct kmem_obj_info kp = { }; 589 590 if (WARN_ON_ONCE(!virt_addr_valid(object))) 591 return; 592 page = virt_to_head_page(object); 593 if (WARN_ON_ONCE(!PageSlab(page))) { 594 pr_cont(" non-slab memory.\n"); 595 return; 596 } 597 kmem_obj_info(&kp, object, page); 598 if (kp.kp_slab_cache) 599 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name); 600 else 601 pr_cont(" slab%s", cp); 602 if (kp.kp_objp) 603 pr_cont(" start %px", kp.kp_objp); 604 if (kp.kp_data_offset) 605 pr_cont(" data offset %lu", kp.kp_data_offset); 606 if (kp.kp_objp) { 607 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset; 608 pr_cont(" pointer offset %lu", ptroffset); 609 } 610 if (kp.kp_slab_cache && kp.kp_slab_cache->usersize) 611 pr_cont(" size %u", kp.kp_slab_cache->usersize); 612 if (kp.kp_ret) 613 pr_cont(" allocated at %pS\n", kp.kp_ret); 614 else 615 pr_cont("\n"); 616 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) { 617 if (!kp.kp_stack[i]) 618 break; 619 pr_info(" %pS\n", kp.kp_stack[i]); 620 } 621 } 622 EXPORT_SYMBOL_GPL(kmem_dump_obj); 623 #endif 624 625 #ifndef CONFIG_SLOB 626 /* Create a cache during boot when no slab services are available yet */ 627 void __init create_boot_cache(struct kmem_cache *s, const char *name, 628 unsigned int size, slab_flags_t flags, 629 unsigned int useroffset, unsigned int usersize) 630 { 631 int err; 632 unsigned int align = ARCH_KMALLOC_MINALIGN; 633 634 s->name = name; 635 s->size = s->object_size = size; 636 637 /* 638 * For power of two sizes, guarantee natural alignment for kmalloc 639 * caches, regardless of SL*B debugging options. 640 */ 641 if (is_power_of_2(size)) 642 align = max(align, size); 643 s->align = calculate_alignment(flags, align, size); 644 645 s->useroffset = useroffset; 646 s->usersize = usersize; 647 648 err = __kmem_cache_create(s, flags); 649 650 if (err) 651 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", 652 name, size, err); 653 654 s->refcount = -1; /* Exempt from merging for now */ 655 } 656 657 struct kmem_cache *__init create_kmalloc_cache(const char *name, 658 unsigned int size, slab_flags_t flags, 659 unsigned int useroffset, unsigned int usersize) 660 { 661 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); 662 663 if (!s) 664 panic("Out of memory when creating slab %s\n", name); 665 666 create_boot_cache(s, name, size, flags, useroffset, usersize); 667 kasan_cache_create_kmalloc(s); 668 list_add(&s->list, &slab_caches); 669 s->refcount = 1; 670 return s; 671 } 672 673 struct kmem_cache * 674 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init = 675 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ }; 676 EXPORT_SYMBOL(kmalloc_caches); 677 678 /* 679 * Conversion table for small slabs sizes / 8 to the index in the 680 * kmalloc array. This is necessary for slabs < 192 since we have non power 681 * of two cache sizes there. The size of larger slabs can be determined using 682 * fls. 683 */ 684 static u8 size_index[24] __ro_after_init = { 685 3, /* 8 */ 686 4, /* 16 */ 687 5, /* 24 */ 688 5, /* 32 */ 689 6, /* 40 */ 690 6, /* 48 */ 691 6, /* 56 */ 692 6, /* 64 */ 693 1, /* 72 */ 694 1, /* 80 */ 695 1, /* 88 */ 696 1, /* 96 */ 697 7, /* 104 */ 698 7, /* 112 */ 699 7, /* 120 */ 700 7, /* 128 */ 701 2, /* 136 */ 702 2, /* 144 */ 703 2, /* 152 */ 704 2, /* 160 */ 705 2, /* 168 */ 706 2, /* 176 */ 707 2, /* 184 */ 708 2 /* 192 */ 709 }; 710 711 static inline unsigned int size_index_elem(unsigned int bytes) 712 { 713 return (bytes - 1) / 8; 714 } 715 716 /* 717 * Find the kmem_cache structure that serves a given size of 718 * allocation 719 */ 720 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) 721 { 722 unsigned int index; 723 724 if (size <= 192) { 725 if (!size) 726 return ZERO_SIZE_PTR; 727 728 index = size_index[size_index_elem(size)]; 729 } else { 730 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE)) 731 return NULL; 732 index = fls(size - 1); 733 } 734 735 return kmalloc_caches[kmalloc_type(flags)][index]; 736 } 737 738 #ifdef CONFIG_ZONE_DMA 739 #define INIT_KMALLOC_INFO(__size, __short_size) \ 740 { \ 741 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ 742 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ 743 .name[KMALLOC_DMA] = "dma-kmalloc-" #__short_size, \ 744 .size = __size, \ 745 } 746 #else 747 #define INIT_KMALLOC_INFO(__size, __short_size) \ 748 { \ 749 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ 750 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ 751 .size = __size, \ 752 } 753 #endif 754 755 /* 756 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. 757 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is 758 * kmalloc-67108864. 759 */ 760 const struct kmalloc_info_struct kmalloc_info[] __initconst = { 761 INIT_KMALLOC_INFO(0, 0), 762 INIT_KMALLOC_INFO(96, 96), 763 INIT_KMALLOC_INFO(192, 192), 764 INIT_KMALLOC_INFO(8, 8), 765 INIT_KMALLOC_INFO(16, 16), 766 INIT_KMALLOC_INFO(32, 32), 767 INIT_KMALLOC_INFO(64, 64), 768 INIT_KMALLOC_INFO(128, 128), 769 INIT_KMALLOC_INFO(256, 256), 770 INIT_KMALLOC_INFO(512, 512), 771 INIT_KMALLOC_INFO(1024, 1k), 772 INIT_KMALLOC_INFO(2048, 2k), 773 INIT_KMALLOC_INFO(4096, 4k), 774 INIT_KMALLOC_INFO(8192, 8k), 775 INIT_KMALLOC_INFO(16384, 16k), 776 INIT_KMALLOC_INFO(32768, 32k), 777 INIT_KMALLOC_INFO(65536, 64k), 778 INIT_KMALLOC_INFO(131072, 128k), 779 INIT_KMALLOC_INFO(262144, 256k), 780 INIT_KMALLOC_INFO(524288, 512k), 781 INIT_KMALLOC_INFO(1048576, 1M), 782 INIT_KMALLOC_INFO(2097152, 2M), 783 INIT_KMALLOC_INFO(4194304, 4M), 784 INIT_KMALLOC_INFO(8388608, 8M), 785 INIT_KMALLOC_INFO(16777216, 16M), 786 INIT_KMALLOC_INFO(33554432, 32M), 787 INIT_KMALLOC_INFO(67108864, 64M) 788 }; 789 790 /* 791 * Patch up the size_index table if we have strange large alignment 792 * requirements for the kmalloc array. This is only the case for 793 * MIPS it seems. The standard arches will not generate any code here. 794 * 795 * Largest permitted alignment is 256 bytes due to the way we 796 * handle the index determination for the smaller caches. 797 * 798 * Make sure that nothing crazy happens if someone starts tinkering 799 * around with ARCH_KMALLOC_MINALIGN 800 */ 801 void __init setup_kmalloc_cache_index_table(void) 802 { 803 unsigned int i; 804 805 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || 806 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); 807 808 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { 809 unsigned int elem = size_index_elem(i); 810 811 if (elem >= ARRAY_SIZE(size_index)) 812 break; 813 size_index[elem] = KMALLOC_SHIFT_LOW; 814 } 815 816 if (KMALLOC_MIN_SIZE >= 64) { 817 /* 818 * The 96 byte size cache is not used if the alignment 819 * is 64 byte. 820 */ 821 for (i = 64 + 8; i <= 96; i += 8) 822 size_index[size_index_elem(i)] = 7; 823 824 } 825 826 if (KMALLOC_MIN_SIZE >= 128) { 827 /* 828 * The 192 byte sized cache is not used if the alignment 829 * is 128 byte. Redirect kmalloc to use the 256 byte cache 830 * instead. 831 */ 832 for (i = 128 + 8; i <= 192; i += 8) 833 size_index[size_index_elem(i)] = 8; 834 } 835 } 836 837 static void __init 838 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags) 839 { 840 if (type == KMALLOC_RECLAIM) 841 flags |= SLAB_RECLAIM_ACCOUNT; 842 843 kmalloc_caches[type][idx] = create_kmalloc_cache( 844 kmalloc_info[idx].name[type], 845 kmalloc_info[idx].size, flags, 0, 846 kmalloc_info[idx].size); 847 } 848 849 /* 850 * Create the kmalloc array. Some of the regular kmalloc arrays 851 * may already have been created because they were needed to 852 * enable allocations for slab creation. 853 */ 854 void __init create_kmalloc_caches(slab_flags_t flags) 855 { 856 int i; 857 enum kmalloc_cache_type type; 858 859 for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) { 860 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { 861 if (!kmalloc_caches[type][i]) 862 new_kmalloc_cache(i, type, flags); 863 864 /* 865 * Caches that are not of the two-to-the-power-of size. 866 * These have to be created immediately after the 867 * earlier power of two caches 868 */ 869 if (KMALLOC_MIN_SIZE <= 32 && i == 6 && 870 !kmalloc_caches[type][1]) 871 new_kmalloc_cache(1, type, flags); 872 if (KMALLOC_MIN_SIZE <= 64 && i == 7 && 873 !kmalloc_caches[type][2]) 874 new_kmalloc_cache(2, type, flags); 875 } 876 } 877 878 /* Kmalloc array is now usable */ 879 slab_state = UP; 880 881 #ifdef CONFIG_ZONE_DMA 882 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { 883 struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i]; 884 885 if (s) { 886 kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache( 887 kmalloc_info[i].name[KMALLOC_DMA], 888 kmalloc_info[i].size, 889 SLAB_CACHE_DMA | flags, 0, 890 kmalloc_info[i].size); 891 } 892 } 893 #endif 894 } 895 #endif /* !CONFIG_SLOB */ 896 897 gfp_t kmalloc_fix_flags(gfp_t flags) 898 { 899 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; 900 901 flags &= ~GFP_SLAB_BUG_MASK; 902 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", 903 invalid_mask, &invalid_mask, flags, &flags); 904 dump_stack(); 905 906 return flags; 907 } 908 909 /* 910 * To avoid unnecessary overhead, we pass through large allocation requests 911 * directly to the page allocator. We use __GFP_COMP, because we will need to 912 * know the allocation order to free the pages properly in kfree. 913 */ 914 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) 915 { 916 void *ret = NULL; 917 struct page *page; 918 919 if (unlikely(flags & GFP_SLAB_BUG_MASK)) 920 flags = kmalloc_fix_flags(flags); 921 922 flags |= __GFP_COMP; 923 page = alloc_pages(flags, order); 924 if (likely(page)) { 925 ret = page_address(page); 926 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, 927 PAGE_SIZE << order); 928 } 929 ret = kasan_kmalloc_large(ret, size, flags); 930 /* As ret might get tagged, call kmemleak hook after KASAN. */ 931 kmemleak_alloc(ret, size, 1, flags); 932 return ret; 933 } 934 EXPORT_SYMBOL(kmalloc_order); 935 936 #ifdef CONFIG_TRACING 937 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) 938 { 939 void *ret = kmalloc_order(size, flags, order); 940 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); 941 return ret; 942 } 943 EXPORT_SYMBOL(kmalloc_order_trace); 944 #endif 945 946 #ifdef CONFIG_SLAB_FREELIST_RANDOM 947 /* Randomize a generic freelist */ 948 static void freelist_randomize(struct rnd_state *state, unsigned int *list, 949 unsigned int count) 950 { 951 unsigned int rand; 952 unsigned int i; 953 954 for (i = 0; i < count; i++) 955 list[i] = i; 956 957 /* Fisher-Yates shuffle */ 958 for (i = count - 1; i > 0; i--) { 959 rand = prandom_u32_state(state); 960 rand %= (i + 1); 961 swap(list[i], list[rand]); 962 } 963 } 964 965 /* Create a random sequence per cache */ 966 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, 967 gfp_t gfp) 968 { 969 struct rnd_state state; 970 971 if (count < 2 || cachep->random_seq) 972 return 0; 973 974 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); 975 if (!cachep->random_seq) 976 return -ENOMEM; 977 978 /* Get best entropy at this stage of boot */ 979 prandom_seed_state(&state, get_random_long()); 980 981 freelist_randomize(&state, cachep->random_seq, count); 982 return 0; 983 } 984 985 /* Destroy the per-cache random freelist sequence */ 986 void cache_random_seq_destroy(struct kmem_cache *cachep) 987 { 988 kfree(cachep->random_seq); 989 cachep->random_seq = NULL; 990 } 991 #endif /* CONFIG_SLAB_FREELIST_RANDOM */ 992 993 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) 994 #ifdef CONFIG_SLAB 995 #define SLABINFO_RIGHTS (0600) 996 #else 997 #define SLABINFO_RIGHTS (0400) 998 #endif 999 1000 static void print_slabinfo_header(struct seq_file *m) 1001 { 1002 /* 1003 * Output format version, so at least we can change it 1004 * without _too_ many complaints. 1005 */ 1006 #ifdef CONFIG_DEBUG_SLAB 1007 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); 1008 #else 1009 seq_puts(m, "slabinfo - version: 2.1\n"); 1010 #endif 1011 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); 1012 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); 1013 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); 1014 #ifdef CONFIG_DEBUG_SLAB 1015 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); 1016 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); 1017 #endif 1018 seq_putc(m, '\n'); 1019 } 1020 1021 void *slab_start(struct seq_file *m, loff_t *pos) 1022 { 1023 mutex_lock(&slab_mutex); 1024 return seq_list_start(&slab_caches, *pos); 1025 } 1026 1027 void *slab_next(struct seq_file *m, void *p, loff_t *pos) 1028 { 1029 return seq_list_next(p, &slab_caches, pos); 1030 } 1031 1032 void slab_stop(struct seq_file *m, void *p) 1033 { 1034 mutex_unlock(&slab_mutex); 1035 } 1036 1037 static void cache_show(struct kmem_cache *s, struct seq_file *m) 1038 { 1039 struct slabinfo sinfo; 1040 1041 memset(&sinfo, 0, sizeof(sinfo)); 1042 get_slabinfo(s, &sinfo); 1043 1044 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", 1045 s->name, sinfo.active_objs, sinfo.num_objs, s->size, 1046 sinfo.objects_per_slab, (1 << sinfo.cache_order)); 1047 1048 seq_printf(m, " : tunables %4u %4u %4u", 1049 sinfo.limit, sinfo.batchcount, sinfo.shared); 1050 seq_printf(m, " : slabdata %6lu %6lu %6lu", 1051 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); 1052 slabinfo_show_stats(m, s); 1053 seq_putc(m, '\n'); 1054 } 1055 1056 static int slab_show(struct seq_file *m, void *p) 1057 { 1058 struct kmem_cache *s = list_entry(p, struct kmem_cache, list); 1059 1060 if (p == slab_caches.next) 1061 print_slabinfo_header(m); 1062 cache_show(s, m); 1063 return 0; 1064 } 1065 1066 void dump_unreclaimable_slab(void) 1067 { 1068 struct kmem_cache *s; 1069 struct slabinfo sinfo; 1070 1071 /* 1072 * Here acquiring slab_mutex is risky since we don't prefer to get 1073 * sleep in oom path. But, without mutex hold, it may introduce a 1074 * risk of crash. 1075 * Use mutex_trylock to protect the list traverse, dump nothing 1076 * without acquiring the mutex. 1077 */ 1078 if (!mutex_trylock(&slab_mutex)) { 1079 pr_warn("excessive unreclaimable slab but cannot dump stats\n"); 1080 return; 1081 } 1082 1083 pr_info("Unreclaimable slab info:\n"); 1084 pr_info("Name Used Total\n"); 1085 1086 list_for_each_entry(s, &slab_caches, list) { 1087 if (s->flags & SLAB_RECLAIM_ACCOUNT) 1088 continue; 1089 1090 get_slabinfo(s, &sinfo); 1091 1092 if (sinfo.num_objs > 0) 1093 pr_info("%-17s %10luKB %10luKB\n", s->name, 1094 (sinfo.active_objs * s->size) / 1024, 1095 (sinfo.num_objs * s->size) / 1024); 1096 } 1097 mutex_unlock(&slab_mutex); 1098 } 1099 1100 #if defined(CONFIG_MEMCG_KMEM) 1101 int memcg_slab_show(struct seq_file *m, void *p) 1102 { 1103 /* 1104 * Deprecated. 1105 * Please, take a look at tools/cgroup/slabinfo.py . 1106 */ 1107 return 0; 1108 } 1109 #endif 1110 1111 /* 1112 * slabinfo_op - iterator that generates /proc/slabinfo 1113 * 1114 * Output layout: 1115 * cache-name 1116 * num-active-objs 1117 * total-objs 1118 * object size 1119 * num-active-slabs 1120 * total-slabs 1121 * num-pages-per-slab 1122 * + further values on SMP and with statistics enabled 1123 */ 1124 static const struct seq_operations slabinfo_op = { 1125 .start = slab_start, 1126 .next = slab_next, 1127 .stop = slab_stop, 1128 .show = slab_show, 1129 }; 1130 1131 static int slabinfo_open(struct inode *inode, struct file *file) 1132 { 1133 return seq_open(file, &slabinfo_op); 1134 } 1135 1136 static const struct proc_ops slabinfo_proc_ops = { 1137 .proc_flags = PROC_ENTRY_PERMANENT, 1138 .proc_open = slabinfo_open, 1139 .proc_read = seq_read, 1140 .proc_write = slabinfo_write, 1141 .proc_lseek = seq_lseek, 1142 .proc_release = seq_release, 1143 }; 1144 1145 static int __init slab_proc_init(void) 1146 { 1147 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); 1148 return 0; 1149 } 1150 module_init(slab_proc_init); 1151 1152 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ 1153 1154 static __always_inline void *__do_krealloc(const void *p, size_t new_size, 1155 gfp_t flags) 1156 { 1157 void *ret; 1158 size_t ks; 1159 1160 /* Don't use instrumented ksize to allow precise KASAN poisoning. */ 1161 if (likely(!ZERO_OR_NULL_PTR(p))) { 1162 if (!kasan_check_byte(p)) 1163 return NULL; 1164 ks = kfence_ksize(p) ?: __ksize(p); 1165 } else 1166 ks = 0; 1167 1168 /* If the object still fits, repoison it precisely. */ 1169 if (ks >= new_size) { 1170 p = kasan_krealloc((void *)p, new_size, flags); 1171 return (void *)p; 1172 } 1173 1174 ret = kmalloc_track_caller(new_size, flags); 1175 if (ret && p) { 1176 /* Disable KASAN checks as the object's redzone is accessed. */ 1177 kasan_disable_current(); 1178 memcpy(ret, kasan_reset_tag(p), ks); 1179 kasan_enable_current(); 1180 } 1181 1182 return ret; 1183 } 1184 1185 /** 1186 * krealloc - reallocate memory. The contents will remain unchanged. 1187 * @p: object to reallocate memory for. 1188 * @new_size: how many bytes of memory are required. 1189 * @flags: the type of memory to allocate. 1190 * 1191 * The contents of the object pointed to are preserved up to the 1192 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored). 1193 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size 1194 * is 0 and @p is not a %NULL pointer, the object pointed to is freed. 1195 * 1196 * Return: pointer to the allocated memory or %NULL in case of error 1197 */ 1198 void *krealloc(const void *p, size_t new_size, gfp_t flags) 1199 { 1200 void *ret; 1201 1202 if (unlikely(!new_size)) { 1203 kfree(p); 1204 return ZERO_SIZE_PTR; 1205 } 1206 1207 ret = __do_krealloc(p, new_size, flags); 1208 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) 1209 kfree(p); 1210 1211 return ret; 1212 } 1213 EXPORT_SYMBOL(krealloc); 1214 1215 /** 1216 * kfree_sensitive - Clear sensitive information in memory before freeing 1217 * @p: object to free memory of 1218 * 1219 * The memory of the object @p points to is zeroed before freed. 1220 * If @p is %NULL, kfree_sensitive() does nothing. 1221 * 1222 * Note: this function zeroes the whole allocated buffer which can be a good 1223 * deal bigger than the requested buffer size passed to kmalloc(). So be 1224 * careful when using this function in performance sensitive code. 1225 */ 1226 void kfree_sensitive(const void *p) 1227 { 1228 size_t ks; 1229 void *mem = (void *)p; 1230 1231 ks = ksize(mem); 1232 if (ks) 1233 memzero_explicit(mem, ks); 1234 kfree(mem); 1235 } 1236 EXPORT_SYMBOL(kfree_sensitive); 1237 1238 /** 1239 * ksize - get the actual amount of memory allocated for a given object 1240 * @objp: Pointer to the object 1241 * 1242 * kmalloc may internally round up allocations and return more memory 1243 * than requested. ksize() can be used to determine the actual amount of 1244 * memory allocated. The caller may use this additional memory, even though 1245 * a smaller amount of memory was initially specified with the kmalloc call. 1246 * The caller must guarantee that objp points to a valid object previously 1247 * allocated with either kmalloc() or kmem_cache_alloc(). The object 1248 * must not be freed during the duration of the call. 1249 * 1250 * Return: size of the actual memory used by @objp in bytes 1251 */ 1252 size_t ksize(const void *objp) 1253 { 1254 size_t size; 1255 1256 /* 1257 * We need to first check that the pointer to the object is valid, and 1258 * only then unpoison the memory. The report printed from ksize() is 1259 * more useful, then when it's printed later when the behaviour could 1260 * be undefined due to a potential use-after-free or double-free. 1261 * 1262 * We use kasan_check_byte(), which is supported for the hardware 1263 * tag-based KASAN mode, unlike kasan_check_read/write(). 1264 * 1265 * If the pointed to memory is invalid, we return 0 to avoid users of 1266 * ksize() writing to and potentially corrupting the memory region. 1267 * 1268 * We want to perform the check before __ksize(), to avoid potentially 1269 * crashing in __ksize() due to accessing invalid metadata. 1270 */ 1271 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp)) 1272 return 0; 1273 1274 size = kfence_ksize(objp) ?: __ksize(objp); 1275 /* 1276 * We assume that ksize callers could use whole allocated area, 1277 * so we need to unpoison this area. 1278 */ 1279 kasan_unpoison_range(objp, size); 1280 return size; 1281 } 1282 EXPORT_SYMBOL(ksize); 1283 1284 /* Tracepoints definitions. */ 1285 EXPORT_TRACEPOINT_SYMBOL(kmalloc); 1286 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); 1287 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); 1288 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); 1289 EXPORT_TRACEPOINT_SYMBOL(kfree); 1290 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); 1291 1292 int should_failslab(struct kmem_cache *s, gfp_t gfpflags) 1293 { 1294 if (__should_failslab(s, gfpflags)) 1295 return -ENOMEM; 1296 return 0; 1297 } 1298 ALLOW_ERROR_INJECTION(should_failslab, ERRNO); 1299