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 mutex_lock(&slab_mutex); 506 507 s->refcount--; 508 if (s->refcount) 509 goto out_unlock; 510 511 err = shutdown_cache(s); 512 if (err) { 513 pr_err("%s %s: Slab cache still has objects\n", 514 __func__, s->name); 515 dump_stack(); 516 } 517 out_unlock: 518 mutex_unlock(&slab_mutex); 519 } 520 EXPORT_SYMBOL(kmem_cache_destroy); 521 522 /** 523 * kmem_cache_shrink - Shrink a cache. 524 * @cachep: The cache to shrink. 525 * 526 * Releases as many slabs as possible for a cache. 527 * To help debugging, a zero exit status indicates all slabs were released. 528 * 529 * Return: %0 if all slabs were released, non-zero otherwise 530 */ 531 int kmem_cache_shrink(struct kmem_cache *cachep) 532 { 533 int ret; 534 535 536 kasan_cache_shrink(cachep); 537 ret = __kmem_cache_shrink(cachep); 538 539 return ret; 540 } 541 EXPORT_SYMBOL(kmem_cache_shrink); 542 543 bool slab_is_available(void) 544 { 545 return slab_state >= UP; 546 } 547 548 #ifdef CONFIG_PRINTK 549 /** 550 * kmem_valid_obj - does the pointer reference a valid slab object? 551 * @object: pointer to query. 552 * 553 * Return: %true if the pointer is to a not-yet-freed object from 554 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer 555 * is to an already-freed object, and %false otherwise. 556 */ 557 bool kmem_valid_obj(void *object) 558 { 559 struct page *page; 560 561 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */ 562 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object)) 563 return false; 564 page = virt_to_head_page(object); 565 return PageSlab(page); 566 } 567 EXPORT_SYMBOL_GPL(kmem_valid_obj); 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 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 page *page; 589 unsigned long ptroffset; 590 struct kmem_obj_info kp = { }; 591 592 if (WARN_ON_ONCE(!virt_addr_valid(object))) 593 return; 594 page = virt_to_head_page(object); 595 if (WARN_ON_ONCE(!PageSlab(page))) { 596 pr_cont(" non-slab memory.\n"); 597 return; 598 } 599 kmem_obj_info(&kp, object, page); 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 (kp.kp_objp) 605 pr_cont(" start %px", kp.kp_objp); 606 if (kp.kp_data_offset) 607 pr_cont(" data offset %lu", kp.kp_data_offset); 608 if (kp.kp_objp) { 609 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset; 610 pr_cont(" pointer offset %lu", ptroffset); 611 } 612 if (kp.kp_slab_cache && kp.kp_slab_cache->usersize) 613 pr_cont(" size %u", kp.kp_slab_cache->usersize); 614 if (kp.kp_ret) 615 pr_cont(" allocated at %pS\n", kp.kp_ret); 616 else 617 pr_cont("\n"); 618 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) { 619 if (!kp.kp_stack[i]) 620 break; 621 pr_info(" %pS\n", kp.kp_stack[i]); 622 } 623 } 624 EXPORT_SYMBOL_GPL(kmem_dump_obj); 625 #endif 626 627 #ifndef CONFIG_SLOB 628 /* Create a cache during boot when no slab services are available yet */ 629 void __init create_boot_cache(struct kmem_cache *s, const char *name, 630 unsigned int size, slab_flags_t flags, 631 unsigned int useroffset, unsigned int usersize) 632 { 633 int err; 634 unsigned int align = ARCH_KMALLOC_MINALIGN; 635 636 s->name = name; 637 s->size = s->object_size = size; 638 639 /* 640 * For power of two sizes, guarantee natural alignment for kmalloc 641 * caches, regardless of SL*B debugging options. 642 */ 643 if (is_power_of_2(size)) 644 align = max(align, size); 645 s->align = calculate_alignment(flags, align, size); 646 647 s->useroffset = useroffset; 648 s->usersize = usersize; 649 650 err = __kmem_cache_create(s, flags); 651 652 if (err) 653 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", 654 name, size, err); 655 656 s->refcount = -1; /* Exempt from merging for now */ 657 } 658 659 struct kmem_cache *__init create_kmalloc_cache(const char *name, 660 unsigned int size, slab_flags_t flags, 661 unsigned int useroffset, unsigned int usersize) 662 { 663 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); 664 665 if (!s) 666 panic("Out of memory when creating slab %s\n", name); 667 668 create_boot_cache(s, name, size, flags, useroffset, usersize); 669 kasan_cache_create_kmalloc(s); 670 list_add(&s->list, &slab_caches); 671 s->refcount = 1; 672 return s; 673 } 674 675 struct kmem_cache * 676 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init = 677 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ }; 678 EXPORT_SYMBOL(kmalloc_caches); 679 680 /* 681 * Conversion table for small slabs sizes / 8 to the index in the 682 * kmalloc array. This is necessary for slabs < 192 since we have non power 683 * of two cache sizes there. The size of larger slabs can be determined using 684 * fls. 685 */ 686 static u8 size_index[24] __ro_after_init = { 687 3, /* 8 */ 688 4, /* 16 */ 689 5, /* 24 */ 690 5, /* 32 */ 691 6, /* 40 */ 692 6, /* 48 */ 693 6, /* 56 */ 694 6, /* 64 */ 695 1, /* 72 */ 696 1, /* 80 */ 697 1, /* 88 */ 698 1, /* 96 */ 699 7, /* 104 */ 700 7, /* 112 */ 701 7, /* 120 */ 702 7, /* 128 */ 703 2, /* 136 */ 704 2, /* 144 */ 705 2, /* 152 */ 706 2, /* 160 */ 707 2, /* 168 */ 708 2, /* 176 */ 709 2, /* 184 */ 710 2 /* 192 */ 711 }; 712 713 static inline unsigned int size_index_elem(unsigned int bytes) 714 { 715 return (bytes - 1) / 8; 716 } 717 718 /* 719 * Find the kmem_cache structure that serves a given size of 720 * allocation 721 */ 722 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) 723 { 724 unsigned int index; 725 726 if (size <= 192) { 727 if (!size) 728 return ZERO_SIZE_PTR; 729 730 index = size_index[size_index_elem(size)]; 731 } else { 732 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE)) 733 return NULL; 734 index = fls(size - 1); 735 } 736 737 return kmalloc_caches[kmalloc_type(flags)][index]; 738 } 739 740 #ifdef CONFIG_ZONE_DMA 741 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz, 742 #else 743 #define KMALLOC_DMA_NAME(sz) 744 #endif 745 746 #ifdef CONFIG_MEMCG_KMEM 747 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz, 748 #else 749 #define KMALLOC_CGROUP_NAME(sz) 750 #endif 751 752 #define INIT_KMALLOC_INFO(__size, __short_size) \ 753 { \ 754 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ 755 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ 756 KMALLOC_CGROUP_NAME(__short_size) \ 757 KMALLOC_DMA_NAME(__short_size) \ 758 .size = __size, \ 759 } 760 761 /* 762 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. 763 * kmalloc_index() supports up to 2^25=32MB, so the final entry of the table is 764 * kmalloc-32M. 765 */ 766 const struct kmalloc_info_struct kmalloc_info[] __initconst = { 767 INIT_KMALLOC_INFO(0, 0), 768 INIT_KMALLOC_INFO(96, 96), 769 INIT_KMALLOC_INFO(192, 192), 770 INIT_KMALLOC_INFO(8, 8), 771 INIT_KMALLOC_INFO(16, 16), 772 INIT_KMALLOC_INFO(32, 32), 773 INIT_KMALLOC_INFO(64, 64), 774 INIT_KMALLOC_INFO(128, 128), 775 INIT_KMALLOC_INFO(256, 256), 776 INIT_KMALLOC_INFO(512, 512), 777 INIT_KMALLOC_INFO(1024, 1k), 778 INIT_KMALLOC_INFO(2048, 2k), 779 INIT_KMALLOC_INFO(4096, 4k), 780 INIT_KMALLOC_INFO(8192, 8k), 781 INIT_KMALLOC_INFO(16384, 16k), 782 INIT_KMALLOC_INFO(32768, 32k), 783 INIT_KMALLOC_INFO(65536, 64k), 784 INIT_KMALLOC_INFO(131072, 128k), 785 INIT_KMALLOC_INFO(262144, 256k), 786 INIT_KMALLOC_INFO(524288, 512k), 787 INIT_KMALLOC_INFO(1048576, 1M), 788 INIT_KMALLOC_INFO(2097152, 2M), 789 INIT_KMALLOC_INFO(4194304, 4M), 790 INIT_KMALLOC_INFO(8388608, 8M), 791 INIT_KMALLOC_INFO(16777216, 16M), 792 INIT_KMALLOC_INFO(33554432, 32M) 793 }; 794 795 /* 796 * Patch up the size_index table if we have strange large alignment 797 * requirements for the kmalloc array. This is only the case for 798 * MIPS it seems. The standard arches will not generate any code here. 799 * 800 * Largest permitted alignment is 256 bytes due to the way we 801 * handle the index determination for the smaller caches. 802 * 803 * Make sure that nothing crazy happens if someone starts tinkering 804 * around with ARCH_KMALLOC_MINALIGN 805 */ 806 void __init setup_kmalloc_cache_index_table(void) 807 { 808 unsigned int i; 809 810 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || 811 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); 812 813 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { 814 unsigned int elem = size_index_elem(i); 815 816 if (elem >= ARRAY_SIZE(size_index)) 817 break; 818 size_index[elem] = KMALLOC_SHIFT_LOW; 819 } 820 821 if (KMALLOC_MIN_SIZE >= 64) { 822 /* 823 * The 96 byte size cache is not used if the alignment 824 * is 64 byte. 825 */ 826 for (i = 64 + 8; i <= 96; i += 8) 827 size_index[size_index_elem(i)] = 7; 828 829 } 830 831 if (KMALLOC_MIN_SIZE >= 128) { 832 /* 833 * The 192 byte sized cache is not used if the alignment 834 * is 128 byte. Redirect kmalloc to use the 256 byte cache 835 * instead. 836 */ 837 for (i = 128 + 8; i <= 192; i += 8) 838 size_index[size_index_elem(i)] = 8; 839 } 840 } 841 842 static void __init 843 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags) 844 { 845 if (type == KMALLOC_RECLAIM) { 846 flags |= SLAB_RECLAIM_ACCOUNT; 847 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) { 848 if (cgroup_memory_nokmem) { 849 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx]; 850 return; 851 } 852 flags |= SLAB_ACCOUNT; 853 } 854 855 kmalloc_caches[type][idx] = create_kmalloc_cache( 856 kmalloc_info[idx].name[type], 857 kmalloc_info[idx].size, flags, 0, 858 kmalloc_info[idx].size); 859 860 /* 861 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for 862 * KMALLOC_NORMAL caches. 863 */ 864 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL)) 865 kmalloc_caches[type][idx]->refcount = -1; 866 } 867 868 /* 869 * Create the kmalloc array. Some of the regular kmalloc arrays 870 * may already have been created because they were needed to 871 * enable allocations for slab creation. 872 */ 873 void __init create_kmalloc_caches(slab_flags_t flags) 874 { 875 int i; 876 enum kmalloc_cache_type type; 877 878 /* 879 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined 880 */ 881 for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) { 882 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { 883 if (!kmalloc_caches[type][i]) 884 new_kmalloc_cache(i, type, flags); 885 886 /* 887 * Caches that are not of the two-to-the-power-of size. 888 * These have to be created immediately after the 889 * earlier power of two caches 890 */ 891 if (KMALLOC_MIN_SIZE <= 32 && i == 6 && 892 !kmalloc_caches[type][1]) 893 new_kmalloc_cache(1, type, flags); 894 if (KMALLOC_MIN_SIZE <= 64 && i == 7 && 895 !kmalloc_caches[type][2]) 896 new_kmalloc_cache(2, type, flags); 897 } 898 } 899 900 /* Kmalloc array is now usable */ 901 slab_state = UP; 902 903 #ifdef CONFIG_ZONE_DMA 904 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { 905 struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i]; 906 907 if (s) { 908 kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache( 909 kmalloc_info[i].name[KMALLOC_DMA], 910 kmalloc_info[i].size, 911 SLAB_CACHE_DMA | flags, 0, 912 kmalloc_info[i].size); 913 } 914 } 915 #endif 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 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 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 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 #if defined(CONFIG_MEMCG_KMEM) 1123 int memcg_slab_show(struct seq_file *m, void *p) 1124 { 1125 /* 1126 * Deprecated. 1127 * Please, take a look at tools/cgroup/slabinfo.py . 1128 */ 1129 return 0; 1130 } 1131 #endif 1132 1133 /* 1134 * slabinfo_op - iterator that generates /proc/slabinfo 1135 * 1136 * Output layout: 1137 * cache-name 1138 * num-active-objs 1139 * total-objs 1140 * object size 1141 * num-active-slabs 1142 * total-slabs 1143 * num-pages-per-slab 1144 * + further values on SMP and with statistics enabled 1145 */ 1146 static const struct seq_operations slabinfo_op = { 1147 .start = slab_start, 1148 .next = slab_next, 1149 .stop = slab_stop, 1150 .show = slab_show, 1151 }; 1152 1153 static int slabinfo_open(struct inode *inode, struct file *file) 1154 { 1155 return seq_open(file, &slabinfo_op); 1156 } 1157 1158 static const struct proc_ops slabinfo_proc_ops = { 1159 .proc_flags = PROC_ENTRY_PERMANENT, 1160 .proc_open = slabinfo_open, 1161 .proc_read = seq_read, 1162 .proc_write = slabinfo_write, 1163 .proc_lseek = seq_lseek, 1164 .proc_release = seq_release, 1165 }; 1166 1167 static int __init slab_proc_init(void) 1168 { 1169 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); 1170 return 0; 1171 } 1172 module_init(slab_proc_init); 1173 1174 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ 1175 1176 static __always_inline void *__do_krealloc(const void *p, size_t new_size, 1177 gfp_t flags) 1178 { 1179 void *ret; 1180 size_t ks; 1181 1182 /* Don't use instrumented ksize to allow precise KASAN poisoning. */ 1183 if (likely(!ZERO_OR_NULL_PTR(p))) { 1184 if (!kasan_check_byte(p)) 1185 return NULL; 1186 ks = kfence_ksize(p) ?: __ksize(p); 1187 } else 1188 ks = 0; 1189 1190 /* If the object still fits, repoison it precisely. */ 1191 if (ks >= new_size) { 1192 p = kasan_krealloc((void *)p, new_size, flags); 1193 return (void *)p; 1194 } 1195 1196 ret = kmalloc_track_caller(new_size, flags); 1197 if (ret && p) { 1198 /* Disable KASAN checks as the object's redzone is accessed. */ 1199 kasan_disable_current(); 1200 memcpy(ret, kasan_reset_tag(p), ks); 1201 kasan_enable_current(); 1202 } 1203 1204 return ret; 1205 } 1206 1207 /** 1208 * krealloc - reallocate memory. The contents will remain unchanged. 1209 * @p: object to reallocate memory for. 1210 * @new_size: how many bytes of memory are required. 1211 * @flags: the type of memory to allocate. 1212 * 1213 * The contents of the object pointed to are preserved up to the 1214 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored). 1215 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size 1216 * is 0 and @p is not a %NULL pointer, the object pointed to is freed. 1217 * 1218 * Return: pointer to the allocated memory or %NULL in case of error 1219 */ 1220 void *krealloc(const void *p, size_t new_size, gfp_t flags) 1221 { 1222 void *ret; 1223 1224 if (unlikely(!new_size)) { 1225 kfree(p); 1226 return ZERO_SIZE_PTR; 1227 } 1228 1229 ret = __do_krealloc(p, new_size, flags); 1230 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) 1231 kfree(p); 1232 1233 return ret; 1234 } 1235 EXPORT_SYMBOL(krealloc); 1236 1237 /** 1238 * kfree_sensitive - Clear sensitive information in memory before freeing 1239 * @p: object to free memory of 1240 * 1241 * The memory of the object @p points to is zeroed before freed. 1242 * If @p is %NULL, kfree_sensitive() does nothing. 1243 * 1244 * Note: this function zeroes the whole allocated buffer which can be a good 1245 * deal bigger than the requested buffer size passed to kmalloc(). So be 1246 * careful when using this function in performance sensitive code. 1247 */ 1248 void kfree_sensitive(const void *p) 1249 { 1250 size_t ks; 1251 void *mem = (void *)p; 1252 1253 ks = ksize(mem); 1254 if (ks) 1255 memzero_explicit(mem, ks); 1256 kfree(mem); 1257 } 1258 EXPORT_SYMBOL(kfree_sensitive); 1259 1260 /** 1261 * ksize - get the actual amount of memory allocated for a given object 1262 * @objp: Pointer to the object 1263 * 1264 * kmalloc may internally round up allocations and return more memory 1265 * than requested. ksize() can be used to determine the actual amount of 1266 * memory allocated. The caller may use this additional memory, even though 1267 * a smaller amount of memory was initially specified with the kmalloc call. 1268 * The caller must guarantee that objp points to a valid object previously 1269 * allocated with either kmalloc() or kmem_cache_alloc(). The object 1270 * must not be freed during the duration of the call. 1271 * 1272 * Return: size of the actual memory used by @objp in bytes 1273 */ 1274 size_t ksize(const void *objp) 1275 { 1276 size_t size; 1277 1278 /* 1279 * We need to first check that the pointer to the object is valid, and 1280 * only then unpoison the memory. The report printed from ksize() is 1281 * more useful, then when it's printed later when the behaviour could 1282 * be undefined due to a potential use-after-free or double-free. 1283 * 1284 * We use kasan_check_byte(), which is supported for the hardware 1285 * tag-based KASAN mode, unlike kasan_check_read/write(). 1286 * 1287 * If the pointed to memory is invalid, we return 0 to avoid users of 1288 * ksize() writing to and potentially corrupting the memory region. 1289 * 1290 * We want to perform the check before __ksize(), to avoid potentially 1291 * crashing in __ksize() due to accessing invalid metadata. 1292 */ 1293 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp)) 1294 return 0; 1295 1296 size = kfence_ksize(objp) ?: __ksize(objp); 1297 /* 1298 * We assume that ksize callers could use whole allocated area, 1299 * so we need to unpoison this area. 1300 */ 1301 kasan_unpoison_range(objp, size); 1302 return size; 1303 } 1304 EXPORT_SYMBOL(ksize); 1305 1306 /* Tracepoints definitions. */ 1307 EXPORT_TRACEPOINT_SYMBOL(kmalloc); 1308 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); 1309 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); 1310 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); 1311 EXPORT_TRACEPOINT_SYMBOL(kfree); 1312 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); 1313 1314 int should_failslab(struct kmem_cache *s, gfp_t gfpflags) 1315 { 1316 if (__should_failslab(s, gfpflags)) 1317 return -ENOMEM; 1318 return 0; 1319 } 1320 ALLOW_ERROR_INJECTION(should_failslab, ERRNO); 1321