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 static LIST_HEAD(slab_caches_to_rcu_destroy); 41 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); 42 static DECLARE_WORK(slab_caches_to_rcu_destroy_work, 43 slab_caches_to_rcu_destroy_workfn); 44 45 /* 46 * Set of flags that will prevent slab merging 47 */ 48 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ 49 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ 50 SLAB_FAILSLAB | kasan_never_merge()) 51 52 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ 53 SLAB_CACHE_DMA32 | SLAB_ACCOUNT) 54 55 /* 56 * Merge control. If this is set then no merging of slab caches will occur. 57 */ 58 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); 59 60 static int __init setup_slab_nomerge(char *str) 61 { 62 slab_nomerge = true; 63 return 1; 64 } 65 66 static int __init setup_slab_merge(char *str) 67 { 68 slab_nomerge = false; 69 return 1; 70 } 71 72 #ifdef CONFIG_SLUB 73 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); 74 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0); 75 #endif 76 77 __setup("slab_nomerge", setup_slab_nomerge); 78 __setup("slab_merge", setup_slab_merge); 79 80 /* 81 * Determine the size of a slab object 82 */ 83 unsigned int kmem_cache_size(struct kmem_cache *s) 84 { 85 return s->object_size; 86 } 87 EXPORT_SYMBOL(kmem_cache_size); 88 89 #ifdef CONFIG_DEBUG_VM 90 static int kmem_cache_sanity_check(const char *name, unsigned int size) 91 { 92 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) { 93 pr_err("kmem_cache_create(%s) integrity check failed\n", name); 94 return -EINVAL; 95 } 96 97 WARN_ON(strchr(name, ' ')); /* It confuses parsers */ 98 return 0; 99 } 100 #else 101 static inline int kmem_cache_sanity_check(const char *name, unsigned int size) 102 { 103 return 0; 104 } 105 #endif 106 107 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) 108 { 109 size_t i; 110 111 for (i = 0; i < nr; i++) { 112 if (s) 113 kmem_cache_free(s, p[i]); 114 else 115 kfree(p[i]); 116 } 117 } 118 119 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, 120 void **p) 121 { 122 size_t i; 123 124 for (i = 0; i < nr; i++) { 125 void *x = p[i] = kmem_cache_alloc(s, flags); 126 if (!x) { 127 __kmem_cache_free_bulk(s, i, p); 128 return 0; 129 } 130 } 131 return i; 132 } 133 134 /* 135 * Figure out what the alignment of the objects will be given a set of 136 * flags, a user specified alignment and the size of the objects. 137 */ 138 static unsigned int calculate_alignment(slab_flags_t flags, 139 unsigned int align, unsigned int size) 140 { 141 /* 142 * If the user wants hardware cache aligned objects then follow that 143 * suggestion if the object is sufficiently large. 144 * 145 * The hardware cache alignment cannot override the specified 146 * alignment though. If that is greater then use it. 147 */ 148 if (flags & SLAB_HWCACHE_ALIGN) { 149 unsigned int ralign; 150 151 ralign = cache_line_size(); 152 while (size <= ralign / 2) 153 ralign /= 2; 154 align = max(align, ralign); 155 } 156 157 if (align < ARCH_SLAB_MINALIGN) 158 align = ARCH_SLAB_MINALIGN; 159 160 return ALIGN(align, sizeof(void *)); 161 } 162 163 /* 164 * Find a mergeable slab cache 165 */ 166 int slab_unmergeable(struct kmem_cache *s) 167 { 168 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) 169 return 1; 170 171 if (s->ctor) 172 return 1; 173 174 if (s->usersize) 175 return 1; 176 177 /* 178 * We may have set a slab to be unmergeable during bootstrap. 179 */ 180 if (s->refcount < 0) 181 return 1; 182 183 return 0; 184 } 185 186 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align, 187 slab_flags_t flags, const char *name, void (*ctor)(void *)) 188 { 189 struct kmem_cache *s; 190 191 if (slab_nomerge) 192 return NULL; 193 194 if (ctor) 195 return NULL; 196 197 size = ALIGN(size, sizeof(void *)); 198 align = calculate_alignment(flags, align, size); 199 size = ALIGN(size, align); 200 flags = kmem_cache_flags(size, flags, name); 201 202 if (flags & SLAB_NEVER_MERGE) 203 return NULL; 204 205 list_for_each_entry_reverse(s, &slab_caches, list) { 206 if (slab_unmergeable(s)) 207 continue; 208 209 if (size > s->size) 210 continue; 211 212 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) 213 continue; 214 /* 215 * Check if alignment is compatible. 216 * Courtesy of Adrian Drzewiecki 217 */ 218 if ((s->size & ~(align - 1)) != s->size) 219 continue; 220 221 if (s->size - size >= sizeof(void *)) 222 continue; 223 224 if (IS_ENABLED(CONFIG_SLAB) && align && 225 (align > s->align || s->align % align)) 226 continue; 227 228 return s; 229 } 230 return NULL; 231 } 232 233 static struct kmem_cache *create_cache(const char *name, 234 unsigned int object_size, unsigned int align, 235 slab_flags_t flags, unsigned int useroffset, 236 unsigned int usersize, void (*ctor)(void *), 237 struct kmem_cache *root_cache) 238 { 239 struct kmem_cache *s; 240 int err; 241 242 if (WARN_ON(useroffset + usersize > object_size)) 243 useroffset = usersize = 0; 244 245 err = -ENOMEM; 246 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); 247 if (!s) 248 goto out; 249 250 s->name = name; 251 s->size = s->object_size = object_size; 252 s->align = align; 253 s->ctor = ctor; 254 s->useroffset = useroffset; 255 s->usersize = usersize; 256 257 err = __kmem_cache_create(s, flags); 258 if (err) 259 goto out_free_cache; 260 261 s->refcount = 1; 262 list_add(&s->list, &slab_caches); 263 out: 264 if (err) 265 return ERR_PTR(err); 266 return s; 267 268 out_free_cache: 269 kmem_cache_free(kmem_cache, s); 270 goto out; 271 } 272 273 /** 274 * kmem_cache_create_usercopy - Create a cache with a region suitable 275 * for copying to userspace 276 * @name: A string which is used in /proc/slabinfo to identify this cache. 277 * @size: The size of objects to be created in this cache. 278 * @align: The required alignment for the objects. 279 * @flags: SLAB flags 280 * @useroffset: Usercopy region offset 281 * @usersize: Usercopy region size 282 * @ctor: A constructor for the objects. 283 * 284 * Cannot be called within a interrupt, but can be interrupted. 285 * The @ctor is run when new pages are allocated by the cache. 286 * 287 * The flags are 288 * 289 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) 290 * to catch references to uninitialised memory. 291 * 292 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check 293 * for buffer overruns. 294 * 295 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware 296 * cacheline. This can be beneficial if you're counting cycles as closely 297 * as davem. 298 * 299 * Return: a pointer to the cache on success, NULL on failure. 300 */ 301 struct kmem_cache * 302 kmem_cache_create_usercopy(const char *name, 303 unsigned int size, unsigned int align, 304 slab_flags_t flags, 305 unsigned int useroffset, unsigned int usersize, 306 void (*ctor)(void *)) 307 { 308 struct kmem_cache *s = NULL; 309 const char *cache_name; 310 int err; 311 312 #ifdef CONFIG_SLUB_DEBUG 313 /* 314 * If no slub_debug was enabled globally, the static key is not yet 315 * enabled by setup_slub_debug(). Enable it if the cache is being 316 * created with any of the debugging flags passed explicitly. 317 */ 318 if (flags & SLAB_DEBUG_FLAGS) 319 static_branch_enable(&slub_debug_enabled); 320 #endif 321 322 mutex_lock(&slab_mutex); 323 324 err = kmem_cache_sanity_check(name, size); 325 if (err) { 326 goto out_unlock; 327 } 328 329 /* Refuse requests with allocator specific flags */ 330 if (flags & ~SLAB_FLAGS_PERMITTED) { 331 err = -EINVAL; 332 goto out_unlock; 333 } 334 335 /* 336 * Some allocators will constraint the set of valid flags to a subset 337 * of all flags. We expect them to define CACHE_CREATE_MASK in this 338 * case, and we'll just provide them with a sanitized version of the 339 * passed flags. 340 */ 341 flags &= CACHE_CREATE_MASK; 342 343 /* Fail closed on bad usersize of useroffset values. */ 344 if (WARN_ON(!usersize && useroffset) || 345 WARN_ON(size < usersize || size - usersize < useroffset)) 346 usersize = useroffset = 0; 347 348 if (!usersize) 349 s = __kmem_cache_alias(name, size, align, flags, ctor); 350 if (s) 351 goto out_unlock; 352 353 cache_name = kstrdup_const(name, GFP_KERNEL); 354 if (!cache_name) { 355 err = -ENOMEM; 356 goto out_unlock; 357 } 358 359 s = create_cache(cache_name, size, 360 calculate_alignment(flags, align, size), 361 flags, useroffset, usersize, ctor, NULL); 362 if (IS_ERR(s)) { 363 err = PTR_ERR(s); 364 kfree_const(cache_name); 365 } 366 367 out_unlock: 368 mutex_unlock(&slab_mutex); 369 370 if (err) { 371 if (flags & SLAB_PANIC) 372 panic("%s: Failed to create slab '%s'. Error %d\n", 373 __func__, name, err); 374 else { 375 pr_warn("%s(%s) failed with error %d\n", 376 __func__, name, err); 377 dump_stack(); 378 } 379 return NULL; 380 } 381 return s; 382 } 383 EXPORT_SYMBOL(kmem_cache_create_usercopy); 384 385 /** 386 * kmem_cache_create - Create a cache. 387 * @name: A string which is used in /proc/slabinfo to identify this cache. 388 * @size: The size of objects to be created in this cache. 389 * @align: The required alignment for the objects. 390 * @flags: SLAB flags 391 * @ctor: A constructor for the objects. 392 * 393 * Cannot be called within a interrupt, but can be interrupted. 394 * The @ctor is run when new pages are allocated by the cache. 395 * 396 * The flags are 397 * 398 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) 399 * to catch references to uninitialised memory. 400 * 401 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check 402 * for buffer overruns. 403 * 404 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware 405 * cacheline. This can be beneficial if you're counting cycles as closely 406 * as davem. 407 * 408 * Return: a pointer to the cache on success, NULL on failure. 409 */ 410 struct kmem_cache * 411 kmem_cache_create(const char *name, unsigned int size, unsigned int align, 412 slab_flags_t flags, void (*ctor)(void *)) 413 { 414 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0, 415 ctor); 416 } 417 EXPORT_SYMBOL(kmem_cache_create); 418 419 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) 420 { 421 LIST_HEAD(to_destroy); 422 struct kmem_cache *s, *s2; 423 424 /* 425 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the 426 * @slab_caches_to_rcu_destroy list. The slab pages are freed 427 * through RCU and the associated kmem_cache are dereferenced 428 * while freeing the pages, so the kmem_caches should be freed only 429 * after the pending RCU operations are finished. As rcu_barrier() 430 * is a pretty slow operation, we batch all pending destructions 431 * asynchronously. 432 */ 433 mutex_lock(&slab_mutex); 434 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); 435 mutex_unlock(&slab_mutex); 436 437 if (list_empty(&to_destroy)) 438 return; 439 440 rcu_barrier(); 441 442 list_for_each_entry_safe(s, s2, &to_destroy, list) { 443 debugfs_slab_release(s); 444 kfence_shutdown_cache(s); 445 #ifdef SLAB_SUPPORTS_SYSFS 446 sysfs_slab_release(s); 447 #else 448 slab_kmem_cache_release(s); 449 #endif 450 } 451 } 452 453 static int shutdown_cache(struct kmem_cache *s) 454 { 455 /* free asan quarantined objects */ 456 kasan_cache_shutdown(s); 457 458 if (__kmem_cache_shutdown(s) != 0) 459 return -EBUSY; 460 461 list_del(&s->list); 462 463 if (s->flags & SLAB_TYPESAFE_BY_RCU) { 464 #ifdef SLAB_SUPPORTS_SYSFS 465 sysfs_slab_unlink(s); 466 #endif 467 list_add_tail(&s->list, &slab_caches_to_rcu_destroy); 468 schedule_work(&slab_caches_to_rcu_destroy_work); 469 } else { 470 kfence_shutdown_cache(s); 471 debugfs_slab_release(s); 472 #ifdef SLAB_SUPPORTS_SYSFS 473 sysfs_slab_unlink(s); 474 sysfs_slab_release(s); 475 #else 476 slab_kmem_cache_release(s); 477 #endif 478 } 479 480 return 0; 481 } 482 483 void slab_kmem_cache_release(struct kmem_cache *s) 484 { 485 __kmem_cache_release(s); 486 kfree_const(s->name); 487 kmem_cache_free(kmem_cache, s); 488 } 489 490 void kmem_cache_destroy(struct kmem_cache *s) 491 { 492 if (unlikely(!s) || !kasan_check_byte(s)) 493 return; 494 495 cpus_read_lock(); 496 mutex_lock(&slab_mutex); 497 498 s->refcount--; 499 if (s->refcount) 500 goto out_unlock; 501 502 WARN(shutdown_cache(s), 503 "%s %s: Slab cache still has objects when called from %pS", 504 __func__, s->name, (void *)_RET_IP_); 505 out_unlock: 506 mutex_unlock(&slab_mutex); 507 cpus_read_unlock(); 508 } 509 EXPORT_SYMBOL(kmem_cache_destroy); 510 511 /** 512 * kmem_cache_shrink - Shrink a cache. 513 * @cachep: The cache to shrink. 514 * 515 * Releases as many slabs as possible for a cache. 516 * To help debugging, a zero exit status indicates all slabs were released. 517 * 518 * Return: %0 if all slabs were released, non-zero otherwise 519 */ 520 int kmem_cache_shrink(struct kmem_cache *cachep) 521 { 522 int ret; 523 524 525 kasan_cache_shrink(cachep); 526 ret = __kmem_cache_shrink(cachep); 527 528 return ret; 529 } 530 EXPORT_SYMBOL(kmem_cache_shrink); 531 532 bool slab_is_available(void) 533 { 534 return slab_state >= UP; 535 } 536 537 #ifdef CONFIG_PRINTK 538 /** 539 * kmem_valid_obj - does the pointer reference a valid slab object? 540 * @object: pointer to query. 541 * 542 * Return: %true if the pointer is to a not-yet-freed object from 543 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer 544 * is to an already-freed object, and %false otherwise. 545 */ 546 bool kmem_valid_obj(void *object) 547 { 548 struct folio *folio; 549 550 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */ 551 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object)) 552 return false; 553 folio = virt_to_folio(object); 554 return folio_test_slab(folio); 555 } 556 EXPORT_SYMBOL_GPL(kmem_valid_obj); 557 558 /** 559 * kmem_dump_obj - Print available slab provenance information 560 * @object: slab object for which to find provenance information. 561 * 562 * This function uses pr_cont(), so that the caller is expected to have 563 * printed out whatever preamble is appropriate. The provenance information 564 * depends on the type of object and on how much debugging is enabled. 565 * For a slab-cache object, the fact that it is a slab object is printed, 566 * and, if available, the slab name, return address, and stack trace from 567 * the allocation and last free path of that object. 568 * 569 * This function will splat if passed a pointer to a non-slab object. 570 * If you are not sure what type of object you have, you should instead 571 * use mem_dump_obj(). 572 */ 573 void kmem_dump_obj(void *object) 574 { 575 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc"; 576 int i; 577 struct slab *slab; 578 unsigned long ptroffset; 579 struct kmem_obj_info kp = { }; 580 581 if (WARN_ON_ONCE(!virt_addr_valid(object))) 582 return; 583 slab = virt_to_slab(object); 584 if (WARN_ON_ONCE(!slab)) { 585 pr_cont(" non-slab memory.\n"); 586 return; 587 } 588 kmem_obj_info(&kp, object, slab); 589 if (kp.kp_slab_cache) 590 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name); 591 else 592 pr_cont(" slab%s", cp); 593 if (kp.kp_objp) 594 pr_cont(" start %px", kp.kp_objp); 595 if (kp.kp_data_offset) 596 pr_cont(" data offset %lu", kp.kp_data_offset); 597 if (kp.kp_objp) { 598 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset; 599 pr_cont(" pointer offset %lu", ptroffset); 600 } 601 if (kp.kp_slab_cache && kp.kp_slab_cache->usersize) 602 pr_cont(" size %u", kp.kp_slab_cache->usersize); 603 if (kp.kp_ret) 604 pr_cont(" allocated at %pS\n", kp.kp_ret); 605 else 606 pr_cont("\n"); 607 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) { 608 if (!kp.kp_stack[i]) 609 break; 610 pr_info(" %pS\n", kp.kp_stack[i]); 611 } 612 613 if (kp.kp_free_stack[0]) 614 pr_cont(" Free path:\n"); 615 616 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) { 617 if (!kp.kp_free_stack[i]) 618 break; 619 pr_info(" %pS\n", kp.kp_free_stack[i]); 620 } 621 622 } 623 EXPORT_SYMBOL_GPL(kmem_dump_obj); 624 #endif 625 626 #ifndef CONFIG_SLOB 627 /* Create a cache during boot when no slab services are available yet */ 628 void __init create_boot_cache(struct kmem_cache *s, const char *name, 629 unsigned int size, slab_flags_t flags, 630 unsigned int useroffset, unsigned int usersize) 631 { 632 int err; 633 unsigned int align = ARCH_KMALLOC_MINALIGN; 634 635 s->name = name; 636 s->size = s->object_size = size; 637 638 /* 639 * For power of two sizes, guarantee natural alignment for kmalloc 640 * caches, regardless of SL*B debugging options. 641 */ 642 if (is_power_of_2(size)) 643 align = max(align, size); 644 s->align = calculate_alignment(flags, align, size); 645 646 s->useroffset = useroffset; 647 s->usersize = usersize; 648 649 err = __kmem_cache_create(s, flags); 650 651 if (err) 652 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", 653 name, size, err); 654 655 s->refcount = -1; /* Exempt from merging for now */ 656 } 657 658 struct kmem_cache *__init create_kmalloc_cache(const char *name, 659 unsigned int size, slab_flags_t flags, 660 unsigned int useroffset, unsigned int usersize) 661 { 662 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); 663 664 if (!s) 665 panic("Out of memory when creating slab %s\n", name); 666 667 create_boot_cache(s, name, size, flags, useroffset, usersize); 668 kasan_cache_create_kmalloc(s); 669 list_add(&s->list, &slab_caches); 670 s->refcount = 1; 671 return s; 672 } 673 674 struct kmem_cache * 675 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init = 676 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ }; 677 EXPORT_SYMBOL(kmalloc_caches); 678 679 /* 680 * Conversion table for small slabs sizes / 8 to the index in the 681 * kmalloc array. This is necessary for slabs < 192 since we have non power 682 * of two cache sizes there. The size of larger slabs can be determined using 683 * fls. 684 */ 685 static u8 size_index[24] __ro_after_init = { 686 3, /* 8 */ 687 4, /* 16 */ 688 5, /* 24 */ 689 5, /* 32 */ 690 6, /* 40 */ 691 6, /* 48 */ 692 6, /* 56 */ 693 6, /* 64 */ 694 1, /* 72 */ 695 1, /* 80 */ 696 1, /* 88 */ 697 1, /* 96 */ 698 7, /* 104 */ 699 7, /* 112 */ 700 7, /* 120 */ 701 7, /* 128 */ 702 2, /* 136 */ 703 2, /* 144 */ 704 2, /* 152 */ 705 2, /* 160 */ 706 2, /* 168 */ 707 2, /* 176 */ 708 2, /* 184 */ 709 2 /* 192 */ 710 }; 711 712 static inline unsigned int size_index_elem(unsigned int bytes) 713 { 714 return (bytes - 1) / 8; 715 } 716 717 /* 718 * Find the kmem_cache structure that serves a given size of 719 * allocation 720 */ 721 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) 722 { 723 unsigned int index; 724 725 if (size <= 192) { 726 if (!size) 727 return ZERO_SIZE_PTR; 728 729 index = size_index[size_index_elem(size)]; 730 } else { 731 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE)) 732 return NULL; 733 index = fls(size - 1); 734 } 735 736 return kmalloc_caches[kmalloc_type(flags)][index]; 737 } 738 739 #ifdef CONFIG_ZONE_DMA 740 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz, 741 #else 742 #define KMALLOC_DMA_NAME(sz) 743 #endif 744 745 #ifdef CONFIG_MEMCG_KMEM 746 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz, 747 #else 748 #define KMALLOC_CGROUP_NAME(sz) 749 #endif 750 751 #define INIT_KMALLOC_INFO(__size, __short_size) \ 752 { \ 753 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ 754 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ 755 KMALLOC_CGROUP_NAME(__short_size) \ 756 KMALLOC_DMA_NAME(__short_size) \ 757 .size = __size, \ 758 } 759 760 /* 761 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. 762 * kmalloc_index() supports up to 2^25=32MB, so the final entry of the table is 763 * kmalloc-32M. 764 */ 765 const struct kmalloc_info_struct kmalloc_info[] __initconst = { 766 INIT_KMALLOC_INFO(0, 0), 767 INIT_KMALLOC_INFO(96, 96), 768 INIT_KMALLOC_INFO(192, 192), 769 INIT_KMALLOC_INFO(8, 8), 770 INIT_KMALLOC_INFO(16, 16), 771 INIT_KMALLOC_INFO(32, 32), 772 INIT_KMALLOC_INFO(64, 64), 773 INIT_KMALLOC_INFO(128, 128), 774 INIT_KMALLOC_INFO(256, 256), 775 INIT_KMALLOC_INFO(512, 512), 776 INIT_KMALLOC_INFO(1024, 1k), 777 INIT_KMALLOC_INFO(2048, 2k), 778 INIT_KMALLOC_INFO(4096, 4k), 779 INIT_KMALLOC_INFO(8192, 8k), 780 INIT_KMALLOC_INFO(16384, 16k), 781 INIT_KMALLOC_INFO(32768, 32k), 782 INIT_KMALLOC_INFO(65536, 64k), 783 INIT_KMALLOC_INFO(131072, 128k), 784 INIT_KMALLOC_INFO(262144, 256k), 785 INIT_KMALLOC_INFO(524288, 512k), 786 INIT_KMALLOC_INFO(1048576, 1M), 787 INIT_KMALLOC_INFO(2097152, 2M), 788 INIT_KMALLOC_INFO(4194304, 4M), 789 INIT_KMALLOC_INFO(8388608, 8M), 790 INIT_KMALLOC_INFO(16777216, 16M), 791 INIT_KMALLOC_INFO(33554432, 32M) 792 }; 793 794 /* 795 * Patch up the size_index table if we have strange large alignment 796 * requirements for the kmalloc array. This is only the case for 797 * MIPS it seems. The standard arches will not generate any code here. 798 * 799 * Largest permitted alignment is 256 bytes due to the way we 800 * handle the index determination for the smaller caches. 801 * 802 * Make sure that nothing crazy happens if someone starts tinkering 803 * around with ARCH_KMALLOC_MINALIGN 804 */ 805 void __init setup_kmalloc_cache_index_table(void) 806 { 807 unsigned int i; 808 809 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || 810 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); 811 812 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { 813 unsigned int elem = size_index_elem(i); 814 815 if (elem >= ARRAY_SIZE(size_index)) 816 break; 817 size_index[elem] = KMALLOC_SHIFT_LOW; 818 } 819 820 if (KMALLOC_MIN_SIZE >= 64) { 821 /* 822 * The 96 byte sized cache is not used if the alignment 823 * is 64 byte. 824 */ 825 for (i = 64 + 8; i <= 96; i += 8) 826 size_index[size_index_elem(i)] = 7; 827 828 } 829 830 if (KMALLOC_MIN_SIZE >= 128) { 831 /* 832 * The 192 byte sized cache is not used if the alignment 833 * is 128 byte. Redirect kmalloc to use the 256 byte cache 834 * instead. 835 */ 836 for (i = 128 + 8; i <= 192; i += 8) 837 size_index[size_index_elem(i)] = 8; 838 } 839 } 840 841 static void __init 842 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags) 843 { 844 if (type == KMALLOC_RECLAIM) { 845 flags |= SLAB_RECLAIM_ACCOUNT; 846 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) { 847 if (mem_cgroup_kmem_disabled()) { 848 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx]; 849 return; 850 } 851 flags |= SLAB_ACCOUNT; 852 } 853 854 kmalloc_caches[type][idx] = create_kmalloc_cache( 855 kmalloc_info[idx].name[type], 856 kmalloc_info[idx].size, flags, 0, 857 kmalloc_info[idx].size); 858 859 /* 860 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for 861 * KMALLOC_NORMAL caches. 862 */ 863 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL)) 864 kmalloc_caches[type][idx]->refcount = -1; 865 } 866 867 /* 868 * Create the kmalloc array. Some of the regular kmalloc arrays 869 * may already have been created because they were needed to 870 * enable allocations for slab creation. 871 */ 872 void __init create_kmalloc_caches(slab_flags_t flags) 873 { 874 int i; 875 enum kmalloc_cache_type type; 876 877 /* 878 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined 879 */ 880 for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) { 881 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { 882 if (!kmalloc_caches[type][i]) 883 new_kmalloc_cache(i, type, flags); 884 885 /* 886 * Caches that are not of the two-to-the-power-of size. 887 * These have to be created immediately after the 888 * earlier power of two caches 889 */ 890 if (KMALLOC_MIN_SIZE <= 32 && i == 6 && 891 !kmalloc_caches[type][1]) 892 new_kmalloc_cache(1, type, flags); 893 if (KMALLOC_MIN_SIZE <= 64 && i == 7 && 894 !kmalloc_caches[type][2]) 895 new_kmalloc_cache(2, type, flags); 896 } 897 } 898 899 /* Kmalloc array is now usable */ 900 slab_state = UP; 901 902 #ifdef CONFIG_ZONE_DMA 903 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { 904 struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i]; 905 906 if (s) { 907 kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache( 908 kmalloc_info[i].name[KMALLOC_DMA], 909 kmalloc_info[i].size, 910 SLAB_CACHE_DMA | flags, 0, 911 kmalloc_info[i].size); 912 } 913 } 914 #endif 915 } 916 #endif /* !CONFIG_SLOB */ 917 918 gfp_t kmalloc_fix_flags(gfp_t flags) 919 { 920 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; 921 922 flags &= ~GFP_SLAB_BUG_MASK; 923 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", 924 invalid_mask, &invalid_mask, flags, &flags); 925 dump_stack(); 926 927 return flags; 928 } 929 930 /* 931 * To avoid unnecessary overhead, we pass through large allocation requests 932 * directly to the page allocator. We use __GFP_COMP, because we will need to 933 * know the allocation order to free the pages properly in kfree. 934 */ 935 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) 936 { 937 void *ret = NULL; 938 struct page *page; 939 940 if (unlikely(flags & GFP_SLAB_BUG_MASK)) 941 flags = kmalloc_fix_flags(flags); 942 943 flags |= __GFP_COMP; 944 page = alloc_pages(flags, order); 945 if (likely(page)) { 946 ret = page_address(page); 947 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, 948 PAGE_SIZE << order); 949 } 950 ret = kasan_kmalloc_large(ret, size, flags); 951 /* As ret might get tagged, call kmemleak hook after KASAN. */ 952 kmemleak_alloc(ret, size, 1, flags); 953 return ret; 954 } 955 EXPORT_SYMBOL(kmalloc_order); 956 957 #ifdef CONFIG_TRACING 958 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) 959 { 960 void *ret = kmalloc_order(size, flags, order); 961 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); 962 return ret; 963 } 964 EXPORT_SYMBOL(kmalloc_order_trace); 965 #endif 966 967 #ifdef CONFIG_SLAB_FREELIST_RANDOM 968 /* Randomize a generic freelist */ 969 static void freelist_randomize(struct rnd_state *state, unsigned int *list, 970 unsigned int count) 971 { 972 unsigned int rand; 973 unsigned int i; 974 975 for (i = 0; i < count; i++) 976 list[i] = i; 977 978 /* Fisher-Yates shuffle */ 979 for (i = count - 1; i > 0; i--) { 980 rand = prandom_u32_state(state); 981 rand %= (i + 1); 982 swap(list[i], list[rand]); 983 } 984 } 985 986 /* Create a random sequence per cache */ 987 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, 988 gfp_t gfp) 989 { 990 struct rnd_state state; 991 992 if (count < 2 || cachep->random_seq) 993 return 0; 994 995 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); 996 if (!cachep->random_seq) 997 return -ENOMEM; 998 999 /* Get best entropy at this stage of boot */ 1000 prandom_seed_state(&state, get_random_long()); 1001 1002 freelist_randomize(&state, cachep->random_seq, count); 1003 return 0; 1004 } 1005 1006 /* Destroy the per-cache random freelist sequence */ 1007 void cache_random_seq_destroy(struct kmem_cache *cachep) 1008 { 1009 kfree(cachep->random_seq); 1010 cachep->random_seq = NULL; 1011 } 1012 #endif /* CONFIG_SLAB_FREELIST_RANDOM */ 1013 1014 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) 1015 #ifdef CONFIG_SLAB 1016 #define SLABINFO_RIGHTS (0600) 1017 #else 1018 #define SLABINFO_RIGHTS (0400) 1019 #endif 1020 1021 static void print_slabinfo_header(struct seq_file *m) 1022 { 1023 /* 1024 * Output format version, so at least we can change it 1025 * without _too_ many complaints. 1026 */ 1027 #ifdef CONFIG_DEBUG_SLAB 1028 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); 1029 #else 1030 seq_puts(m, "slabinfo - version: 2.1\n"); 1031 #endif 1032 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); 1033 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); 1034 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); 1035 #ifdef CONFIG_DEBUG_SLAB 1036 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); 1037 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); 1038 #endif 1039 seq_putc(m, '\n'); 1040 } 1041 1042 static void *slab_start(struct seq_file *m, loff_t *pos) 1043 { 1044 mutex_lock(&slab_mutex); 1045 return seq_list_start(&slab_caches, *pos); 1046 } 1047 1048 static void *slab_next(struct seq_file *m, void *p, loff_t *pos) 1049 { 1050 return seq_list_next(p, &slab_caches, pos); 1051 } 1052 1053 static void slab_stop(struct seq_file *m, void *p) 1054 { 1055 mutex_unlock(&slab_mutex); 1056 } 1057 1058 static void cache_show(struct kmem_cache *s, struct seq_file *m) 1059 { 1060 struct slabinfo sinfo; 1061 1062 memset(&sinfo, 0, sizeof(sinfo)); 1063 get_slabinfo(s, &sinfo); 1064 1065 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", 1066 s->name, sinfo.active_objs, sinfo.num_objs, s->size, 1067 sinfo.objects_per_slab, (1 << sinfo.cache_order)); 1068 1069 seq_printf(m, " : tunables %4u %4u %4u", 1070 sinfo.limit, sinfo.batchcount, sinfo.shared); 1071 seq_printf(m, " : slabdata %6lu %6lu %6lu", 1072 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); 1073 slabinfo_show_stats(m, s); 1074 seq_putc(m, '\n'); 1075 } 1076 1077 static int slab_show(struct seq_file *m, void *p) 1078 { 1079 struct kmem_cache *s = list_entry(p, struct kmem_cache, list); 1080 1081 if (p == slab_caches.next) 1082 print_slabinfo_header(m); 1083 cache_show(s, m); 1084 return 0; 1085 } 1086 1087 void dump_unreclaimable_slab(void) 1088 { 1089 struct kmem_cache *s; 1090 struct slabinfo sinfo; 1091 1092 /* 1093 * Here acquiring slab_mutex is risky since we don't prefer to get 1094 * sleep in oom path. But, without mutex hold, it may introduce a 1095 * risk of crash. 1096 * Use mutex_trylock to protect the list traverse, dump nothing 1097 * without acquiring the mutex. 1098 */ 1099 if (!mutex_trylock(&slab_mutex)) { 1100 pr_warn("excessive unreclaimable slab but cannot dump stats\n"); 1101 return; 1102 } 1103 1104 pr_info("Unreclaimable slab info:\n"); 1105 pr_info("Name Used Total\n"); 1106 1107 list_for_each_entry(s, &slab_caches, list) { 1108 if (s->flags & SLAB_RECLAIM_ACCOUNT) 1109 continue; 1110 1111 get_slabinfo(s, &sinfo); 1112 1113 if (sinfo.num_objs > 0) 1114 pr_info("%-17s %10luKB %10luKB\n", s->name, 1115 (sinfo.active_objs * s->size) / 1024, 1116 (sinfo.num_objs * s->size) / 1024); 1117 } 1118 mutex_unlock(&slab_mutex); 1119 } 1120 1121 /* 1122 * slabinfo_op - iterator that generates /proc/slabinfo 1123 * 1124 * Output layout: 1125 * cache-name 1126 * num-active-objs 1127 * total-objs 1128 * object size 1129 * num-active-slabs 1130 * total-slabs 1131 * num-pages-per-slab 1132 * + further values on SMP and with statistics enabled 1133 */ 1134 static const struct seq_operations slabinfo_op = { 1135 .start = slab_start, 1136 .next = slab_next, 1137 .stop = slab_stop, 1138 .show = slab_show, 1139 }; 1140 1141 static int slabinfo_open(struct inode *inode, struct file *file) 1142 { 1143 return seq_open(file, &slabinfo_op); 1144 } 1145 1146 static const struct proc_ops slabinfo_proc_ops = { 1147 .proc_flags = PROC_ENTRY_PERMANENT, 1148 .proc_open = slabinfo_open, 1149 .proc_read = seq_read, 1150 .proc_write = slabinfo_write, 1151 .proc_lseek = seq_lseek, 1152 .proc_release = seq_release, 1153 }; 1154 1155 static int __init slab_proc_init(void) 1156 { 1157 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); 1158 return 0; 1159 } 1160 module_init(slab_proc_init); 1161 1162 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ 1163 1164 static __always_inline void *__do_krealloc(const void *p, size_t new_size, 1165 gfp_t flags) 1166 { 1167 void *ret; 1168 size_t ks; 1169 1170 /* Don't use instrumented ksize to allow precise KASAN poisoning. */ 1171 if (likely(!ZERO_OR_NULL_PTR(p))) { 1172 if (!kasan_check_byte(p)) 1173 return NULL; 1174 ks = kfence_ksize(p) ?: __ksize(p); 1175 } else 1176 ks = 0; 1177 1178 /* If the object still fits, repoison it precisely. */ 1179 if (ks >= new_size) { 1180 p = kasan_krealloc((void *)p, new_size, flags); 1181 return (void *)p; 1182 } 1183 1184 ret = kmalloc_track_caller(new_size, flags); 1185 if (ret && p) { 1186 /* Disable KASAN checks as the object's redzone is accessed. */ 1187 kasan_disable_current(); 1188 memcpy(ret, kasan_reset_tag(p), ks); 1189 kasan_enable_current(); 1190 } 1191 1192 return ret; 1193 } 1194 1195 /** 1196 * krealloc - reallocate memory. The contents will remain unchanged. 1197 * @p: object to reallocate memory for. 1198 * @new_size: how many bytes of memory are required. 1199 * @flags: the type of memory to allocate. 1200 * 1201 * The contents of the object pointed to are preserved up to the 1202 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored). 1203 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size 1204 * is 0 and @p is not a %NULL pointer, the object pointed to is freed. 1205 * 1206 * Return: pointer to the allocated memory or %NULL in case of error 1207 */ 1208 void *krealloc(const void *p, size_t new_size, gfp_t flags) 1209 { 1210 void *ret; 1211 1212 if (unlikely(!new_size)) { 1213 kfree(p); 1214 return ZERO_SIZE_PTR; 1215 } 1216 1217 ret = __do_krealloc(p, new_size, flags); 1218 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) 1219 kfree(p); 1220 1221 return ret; 1222 } 1223 EXPORT_SYMBOL(krealloc); 1224 1225 /** 1226 * kfree_sensitive - Clear sensitive information in memory before freeing 1227 * @p: object to free memory of 1228 * 1229 * The memory of the object @p points to is zeroed before freed. 1230 * If @p is %NULL, kfree_sensitive() does nothing. 1231 * 1232 * Note: this function zeroes the whole allocated buffer which can be a good 1233 * deal bigger than the requested buffer size passed to kmalloc(). So be 1234 * careful when using this function in performance sensitive code. 1235 */ 1236 void kfree_sensitive(const void *p) 1237 { 1238 size_t ks; 1239 void *mem = (void *)p; 1240 1241 ks = ksize(mem); 1242 if (ks) 1243 memzero_explicit(mem, ks); 1244 kfree(mem); 1245 } 1246 EXPORT_SYMBOL(kfree_sensitive); 1247 1248 /** 1249 * ksize - get the actual amount of memory allocated for a given object 1250 * @objp: Pointer to the object 1251 * 1252 * kmalloc may internally round up allocations and return more memory 1253 * than requested. ksize() can be used to determine the actual amount of 1254 * memory allocated. The caller may use this additional memory, even though 1255 * a smaller amount of memory was initially specified with the kmalloc call. 1256 * The caller must guarantee that objp points to a valid object previously 1257 * allocated with either kmalloc() or kmem_cache_alloc(). The object 1258 * must not be freed during the duration of the call. 1259 * 1260 * Return: size of the actual memory used by @objp in bytes 1261 */ 1262 size_t ksize(const void *objp) 1263 { 1264 size_t size; 1265 1266 /* 1267 * We need to first check that the pointer to the object is valid, and 1268 * only then unpoison the memory. The report printed from ksize() is 1269 * more useful, then when it's printed later when the behaviour could 1270 * be undefined due to a potential use-after-free or double-free. 1271 * 1272 * We use kasan_check_byte(), which is supported for the hardware 1273 * tag-based KASAN mode, unlike kasan_check_read/write(). 1274 * 1275 * If the pointed to memory is invalid, we return 0 to avoid users of 1276 * ksize() writing to and potentially corrupting the memory region. 1277 * 1278 * We want to perform the check before __ksize(), to avoid potentially 1279 * crashing in __ksize() due to accessing invalid metadata. 1280 */ 1281 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp)) 1282 return 0; 1283 1284 size = kfence_ksize(objp) ?: __ksize(objp); 1285 /* 1286 * We assume that ksize callers could use whole allocated area, 1287 * so we need to unpoison this area. 1288 */ 1289 kasan_unpoison_range(objp, size); 1290 return size; 1291 } 1292 EXPORT_SYMBOL(ksize); 1293 1294 /* Tracepoints definitions. */ 1295 EXPORT_TRACEPOINT_SYMBOL(kmalloc); 1296 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); 1297 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); 1298 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); 1299 EXPORT_TRACEPOINT_SYMBOL(kfree); 1300 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); 1301 1302 int should_failslab(struct kmem_cache *s, gfp_t gfpflags) 1303 { 1304 if (__should_failslab(s, gfpflags)) 1305 return -ENOMEM; 1306 return 0; 1307 } 1308 ALLOW_ERROR_INJECTION(should_failslab, ERRNO); 1309