1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Slab allocator functions that are independent of the allocator strategy 4 * 5 * (C) 2012 Christoph Lameter <cl@linux.com> 6 */ 7 #include <linux/slab.h> 8 9 #include <linux/mm.h> 10 #include <linux/poison.h> 11 #include <linux/interrupt.h> 12 #include <linux/memory.h> 13 #include <linux/cache.h> 14 #include <linux/compiler.h> 15 #include <linux/kfence.h> 16 #include <linux/module.h> 17 #include <linux/cpu.h> 18 #include <linux/uaccess.h> 19 #include <linux/seq_file.h> 20 #include <linux/proc_fs.h> 21 #include <linux/debugfs.h> 22 #include <linux/kasan.h> 23 #include <asm/cacheflush.h> 24 #include <asm/tlbflush.h> 25 #include <asm/page.h> 26 #include <linux/memcontrol.h> 27 #include <linux/stackdepot.h> 28 29 #include "internal.h" 30 #include "slab.h" 31 32 #define CREATE_TRACE_POINTS 33 #include <trace/events/kmem.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 /* 108 * Figure out what the alignment of the objects will be given a set of 109 * flags, a user specified alignment and the size of the objects. 110 */ 111 static unsigned int calculate_alignment(slab_flags_t flags, 112 unsigned int align, unsigned int size) 113 { 114 /* 115 * If the user wants hardware cache aligned objects then follow that 116 * suggestion if the object is sufficiently large. 117 * 118 * The hardware cache alignment cannot override the specified 119 * alignment though. If that is greater then use it. 120 */ 121 if (flags & SLAB_HWCACHE_ALIGN) { 122 unsigned int ralign; 123 124 ralign = cache_line_size(); 125 while (size <= ralign / 2) 126 ralign /= 2; 127 align = max(align, ralign); 128 } 129 130 align = max(align, arch_slab_minalign()); 131 132 return ALIGN(align, sizeof(void *)); 133 } 134 135 /* 136 * Find a mergeable slab cache 137 */ 138 int slab_unmergeable(struct kmem_cache *s) 139 { 140 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) 141 return 1; 142 143 if (s->ctor) 144 return 1; 145 146 if (s->usersize) 147 return 1; 148 149 /* 150 * We may have set a slab to be unmergeable during bootstrap. 151 */ 152 if (s->refcount < 0) 153 return 1; 154 155 return 0; 156 } 157 158 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align, 159 slab_flags_t flags, const char *name, void (*ctor)(void *)) 160 { 161 struct kmem_cache *s; 162 163 if (slab_nomerge) 164 return NULL; 165 166 if (ctor) 167 return NULL; 168 169 size = ALIGN(size, sizeof(void *)); 170 align = calculate_alignment(flags, align, size); 171 size = ALIGN(size, align); 172 flags = kmem_cache_flags(size, flags, name); 173 174 if (flags & SLAB_NEVER_MERGE) 175 return NULL; 176 177 list_for_each_entry_reverse(s, &slab_caches, list) { 178 if (slab_unmergeable(s)) 179 continue; 180 181 if (size > s->size) 182 continue; 183 184 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) 185 continue; 186 /* 187 * Check if alignment is compatible. 188 * Courtesy of Adrian Drzewiecki 189 */ 190 if ((s->size & ~(align - 1)) != s->size) 191 continue; 192 193 if (s->size - size >= sizeof(void *)) 194 continue; 195 196 if (IS_ENABLED(CONFIG_SLAB) && align && 197 (align > s->align || s->align % align)) 198 continue; 199 200 return s; 201 } 202 return NULL; 203 } 204 205 static struct kmem_cache *create_cache(const char *name, 206 unsigned int object_size, unsigned int align, 207 slab_flags_t flags, unsigned int useroffset, 208 unsigned int usersize, void (*ctor)(void *), 209 struct kmem_cache *root_cache) 210 { 211 struct kmem_cache *s; 212 int err; 213 214 if (WARN_ON(useroffset + usersize > object_size)) 215 useroffset = usersize = 0; 216 217 err = -ENOMEM; 218 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); 219 if (!s) 220 goto out; 221 222 s->name = name; 223 s->size = s->object_size = object_size; 224 s->align = align; 225 s->ctor = ctor; 226 s->useroffset = useroffset; 227 s->usersize = usersize; 228 229 err = __kmem_cache_create(s, flags); 230 if (err) 231 goto out_free_cache; 232 233 s->refcount = 1; 234 list_add(&s->list, &slab_caches); 235 out: 236 if (err) 237 return ERR_PTR(err); 238 return s; 239 240 out_free_cache: 241 kmem_cache_free(kmem_cache, s); 242 goto out; 243 } 244 245 /** 246 * kmem_cache_create_usercopy - Create a cache with a region suitable 247 * for copying to userspace 248 * @name: A string which is used in /proc/slabinfo to identify this cache. 249 * @size: The size of objects to be created in this cache. 250 * @align: The required alignment for the objects. 251 * @flags: SLAB flags 252 * @useroffset: Usercopy region offset 253 * @usersize: Usercopy region size 254 * @ctor: A constructor for the objects. 255 * 256 * Cannot be called within a interrupt, but can be interrupted. 257 * The @ctor is run when new pages are allocated by the cache. 258 * 259 * The flags are 260 * 261 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) 262 * to catch references to uninitialised memory. 263 * 264 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check 265 * for buffer overruns. 266 * 267 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware 268 * cacheline. This can be beneficial if you're counting cycles as closely 269 * as davem. 270 * 271 * Return: a pointer to the cache on success, NULL on failure. 272 */ 273 struct kmem_cache * 274 kmem_cache_create_usercopy(const char *name, 275 unsigned int size, unsigned int align, 276 slab_flags_t flags, 277 unsigned int useroffset, unsigned int usersize, 278 void (*ctor)(void *)) 279 { 280 struct kmem_cache *s = NULL; 281 const char *cache_name; 282 int err; 283 284 #ifdef CONFIG_SLUB_DEBUG 285 /* 286 * If no slub_debug was enabled globally, the static key is not yet 287 * enabled by setup_slub_debug(). Enable it if the cache is being 288 * created with any of the debugging flags passed explicitly. 289 * It's also possible that this is the first cache created with 290 * SLAB_STORE_USER and we should init stack_depot for it. 291 */ 292 if (flags & SLAB_DEBUG_FLAGS) 293 static_branch_enable(&slub_debug_enabled); 294 if (flags & SLAB_STORE_USER) 295 stack_depot_init(); 296 #endif 297 298 mutex_lock(&slab_mutex); 299 300 err = kmem_cache_sanity_check(name, size); 301 if (err) { 302 goto out_unlock; 303 } 304 305 /* Refuse requests with allocator specific flags */ 306 if (flags & ~SLAB_FLAGS_PERMITTED) { 307 err = -EINVAL; 308 goto out_unlock; 309 } 310 311 /* 312 * Some allocators will constraint the set of valid flags to a subset 313 * of all flags. We expect them to define CACHE_CREATE_MASK in this 314 * case, and we'll just provide them with a sanitized version of the 315 * passed flags. 316 */ 317 flags &= CACHE_CREATE_MASK; 318 319 /* Fail closed on bad usersize of useroffset values. */ 320 if (WARN_ON(!usersize && useroffset) || 321 WARN_ON(size < usersize || size - usersize < useroffset)) 322 usersize = useroffset = 0; 323 324 if (!usersize) 325 s = __kmem_cache_alias(name, size, align, flags, ctor); 326 if (s) 327 goto out_unlock; 328 329 cache_name = kstrdup_const(name, GFP_KERNEL); 330 if (!cache_name) { 331 err = -ENOMEM; 332 goto out_unlock; 333 } 334 335 s = create_cache(cache_name, size, 336 calculate_alignment(flags, align, size), 337 flags, useroffset, usersize, ctor, NULL); 338 if (IS_ERR(s)) { 339 err = PTR_ERR(s); 340 kfree_const(cache_name); 341 } 342 343 out_unlock: 344 mutex_unlock(&slab_mutex); 345 346 if (err) { 347 if (flags & SLAB_PANIC) 348 panic("%s: Failed to create slab '%s'. Error %d\n", 349 __func__, name, err); 350 else { 351 pr_warn("%s(%s) failed with error %d\n", 352 __func__, name, err); 353 dump_stack(); 354 } 355 return NULL; 356 } 357 return s; 358 } 359 EXPORT_SYMBOL(kmem_cache_create_usercopy); 360 361 /** 362 * kmem_cache_create - Create a cache. 363 * @name: A string which is used in /proc/slabinfo to identify this cache. 364 * @size: The size of objects to be created in this cache. 365 * @align: The required alignment for the objects. 366 * @flags: SLAB flags 367 * @ctor: A constructor for the objects. 368 * 369 * Cannot be called within a interrupt, but can be interrupted. 370 * The @ctor is run when new pages are allocated by the cache. 371 * 372 * The flags are 373 * 374 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) 375 * to catch references to uninitialised memory. 376 * 377 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check 378 * for buffer overruns. 379 * 380 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware 381 * cacheline. This can be beneficial if you're counting cycles as closely 382 * as davem. 383 * 384 * Return: a pointer to the cache on success, NULL on failure. 385 */ 386 struct kmem_cache * 387 kmem_cache_create(const char *name, unsigned int size, unsigned int align, 388 slab_flags_t flags, void (*ctor)(void *)) 389 { 390 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0, 391 ctor); 392 } 393 EXPORT_SYMBOL(kmem_cache_create); 394 395 #ifdef SLAB_SUPPORTS_SYSFS 396 /* 397 * For a given kmem_cache, kmem_cache_destroy() should only be called 398 * once or there will be a use-after-free problem. The actual deletion 399 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock 400 * protection. So they are now done without holding those locks. 401 * 402 * Note that there will be a slight delay in the deletion of sysfs files 403 * if kmem_cache_release() is called indrectly from a work function. 404 */ 405 static void kmem_cache_release(struct kmem_cache *s) 406 { 407 sysfs_slab_unlink(s); 408 sysfs_slab_release(s); 409 } 410 #else 411 static void kmem_cache_release(struct kmem_cache *s) 412 { 413 slab_kmem_cache_release(s); 414 } 415 #endif 416 417 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) 418 { 419 LIST_HEAD(to_destroy); 420 struct kmem_cache *s, *s2; 421 422 /* 423 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the 424 * @slab_caches_to_rcu_destroy list. The slab pages are freed 425 * through RCU and the associated kmem_cache are dereferenced 426 * while freeing the pages, so the kmem_caches should be freed only 427 * after the pending RCU operations are finished. As rcu_barrier() 428 * is a pretty slow operation, we batch all pending destructions 429 * asynchronously. 430 */ 431 mutex_lock(&slab_mutex); 432 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); 433 mutex_unlock(&slab_mutex); 434 435 if (list_empty(&to_destroy)) 436 return; 437 438 rcu_barrier(); 439 440 list_for_each_entry_safe(s, s2, &to_destroy, list) { 441 debugfs_slab_release(s); 442 kfence_shutdown_cache(s); 443 kmem_cache_release(s); 444 } 445 } 446 447 static int shutdown_cache(struct kmem_cache *s) 448 { 449 /* free asan quarantined objects */ 450 kasan_cache_shutdown(s); 451 452 if (__kmem_cache_shutdown(s) != 0) 453 return -EBUSY; 454 455 list_del(&s->list); 456 457 if (s->flags & SLAB_TYPESAFE_BY_RCU) { 458 list_add_tail(&s->list, &slab_caches_to_rcu_destroy); 459 schedule_work(&slab_caches_to_rcu_destroy_work); 460 } else { 461 kfence_shutdown_cache(s); 462 debugfs_slab_release(s); 463 } 464 465 return 0; 466 } 467 468 void slab_kmem_cache_release(struct kmem_cache *s) 469 { 470 __kmem_cache_release(s); 471 kfree_const(s->name); 472 kmem_cache_free(kmem_cache, s); 473 } 474 475 void kmem_cache_destroy(struct kmem_cache *s) 476 { 477 int refcnt; 478 479 if (unlikely(!s) || !kasan_check_byte(s)) 480 return; 481 482 cpus_read_lock(); 483 mutex_lock(&slab_mutex); 484 485 refcnt = --s->refcount; 486 if (refcnt) 487 goto out_unlock; 488 489 WARN(shutdown_cache(s), 490 "%s %s: Slab cache still has objects when called from %pS", 491 __func__, s->name, (void *)_RET_IP_); 492 out_unlock: 493 mutex_unlock(&slab_mutex); 494 cpus_read_unlock(); 495 if (!refcnt && !(s->flags & SLAB_TYPESAFE_BY_RCU)) 496 kmem_cache_release(s); 497 } 498 EXPORT_SYMBOL(kmem_cache_destroy); 499 500 /** 501 * kmem_cache_shrink - Shrink a cache. 502 * @cachep: The cache to shrink. 503 * 504 * Releases as many slabs as possible for a cache. 505 * To help debugging, a zero exit status indicates all slabs were released. 506 * 507 * Return: %0 if all slabs were released, non-zero otherwise 508 */ 509 int kmem_cache_shrink(struct kmem_cache *cachep) 510 { 511 int ret; 512 513 514 kasan_cache_shrink(cachep); 515 ret = __kmem_cache_shrink(cachep); 516 517 return ret; 518 } 519 EXPORT_SYMBOL(kmem_cache_shrink); 520 521 bool slab_is_available(void) 522 { 523 return slab_state >= UP; 524 } 525 526 #ifdef CONFIG_PRINTK 527 /** 528 * kmem_valid_obj - does the pointer reference a valid slab object? 529 * @object: pointer to query. 530 * 531 * Return: %true if the pointer is to a not-yet-freed object from 532 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer 533 * is to an already-freed object, and %false otherwise. 534 */ 535 bool kmem_valid_obj(void *object) 536 { 537 struct folio *folio; 538 539 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */ 540 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object)) 541 return false; 542 folio = virt_to_folio(object); 543 return folio_test_slab(folio); 544 } 545 EXPORT_SYMBOL_GPL(kmem_valid_obj); 546 547 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab) 548 { 549 if (__kfence_obj_info(kpp, object, slab)) 550 return; 551 __kmem_obj_info(kpp, object, slab); 552 } 553 554 /** 555 * kmem_dump_obj - Print available slab provenance information 556 * @object: slab object for which to find provenance information. 557 * 558 * This function uses pr_cont(), so that the caller is expected to have 559 * printed out whatever preamble is appropriate. The provenance information 560 * depends on the type of object and on how much debugging is enabled. 561 * For a slab-cache object, the fact that it is a slab object is printed, 562 * and, if available, the slab name, return address, and stack trace from 563 * the allocation and last free path of that object. 564 * 565 * This function will splat if passed a pointer to a non-slab object. 566 * If you are not sure what type of object you have, you should instead 567 * use mem_dump_obj(). 568 */ 569 void kmem_dump_obj(void *object) 570 { 571 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc"; 572 int i; 573 struct slab *slab; 574 unsigned long ptroffset; 575 struct kmem_obj_info kp = { }; 576 577 if (WARN_ON_ONCE(!virt_addr_valid(object))) 578 return; 579 slab = virt_to_slab(object); 580 if (WARN_ON_ONCE(!slab)) { 581 pr_cont(" non-slab memory.\n"); 582 return; 583 } 584 kmem_obj_info(&kp, object, slab); 585 if (kp.kp_slab_cache) 586 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name); 587 else 588 pr_cont(" slab%s", cp); 589 if (is_kfence_address(object)) 590 pr_cont(" (kfence)"); 591 if (kp.kp_objp) 592 pr_cont(" start %px", kp.kp_objp); 593 if (kp.kp_data_offset) 594 pr_cont(" data offset %lu", kp.kp_data_offset); 595 if (kp.kp_objp) { 596 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset; 597 pr_cont(" pointer offset %lu", ptroffset); 598 } 599 if (kp.kp_slab_cache && kp.kp_slab_cache->usersize) 600 pr_cont(" size %u", kp.kp_slab_cache->usersize); 601 if (kp.kp_ret) 602 pr_cont(" allocated at %pS\n", kp.kp_ret); 603 else 604 pr_cont("\n"); 605 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) { 606 if (!kp.kp_stack[i]) 607 break; 608 pr_info(" %pS\n", kp.kp_stack[i]); 609 } 610 611 if (kp.kp_free_stack[0]) 612 pr_cont(" Free path:\n"); 613 614 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) { 615 if (!kp.kp_free_stack[i]) 616 break; 617 pr_info(" %pS\n", kp.kp_free_stack[i]); 618 } 619 620 } 621 EXPORT_SYMBOL_GPL(kmem_dump_obj); 622 #endif 623 624 #ifndef CONFIG_SLOB 625 /* Create a cache during boot when no slab services are available yet */ 626 void __init create_boot_cache(struct kmem_cache *s, const char *name, 627 unsigned int size, slab_flags_t flags, 628 unsigned int useroffset, unsigned int usersize) 629 { 630 int err; 631 unsigned int align = ARCH_KMALLOC_MINALIGN; 632 633 s->name = name; 634 s->size = s->object_size = size; 635 636 /* 637 * For power of two sizes, guarantee natural alignment for kmalloc 638 * caches, regardless of SL*B debugging options. 639 */ 640 if (is_power_of_2(size)) 641 align = max(align, size); 642 s->align = calculate_alignment(flags, align, size); 643 644 s->useroffset = useroffset; 645 s->usersize = usersize; 646 647 err = __kmem_cache_create(s, flags); 648 649 if (err) 650 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", 651 name, size, err); 652 653 s->refcount = -1; /* Exempt from merging for now */ 654 } 655 656 struct kmem_cache *__init create_kmalloc_cache(const char *name, 657 unsigned int size, slab_flags_t flags, 658 unsigned int useroffset, unsigned int usersize) 659 { 660 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); 661 662 if (!s) 663 panic("Out of memory when creating slab %s\n", name); 664 665 create_boot_cache(s, name, size, flags, useroffset, usersize); 666 kasan_cache_create_kmalloc(s); 667 list_add(&s->list, &slab_caches); 668 s->refcount = 1; 669 return s; 670 } 671 672 struct kmem_cache * 673 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init = 674 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ }; 675 EXPORT_SYMBOL(kmalloc_caches); 676 677 /* 678 * Conversion table for small slabs sizes / 8 to the index in the 679 * kmalloc array. This is necessary for slabs < 192 since we have non power 680 * of two cache sizes there. The size of larger slabs can be determined using 681 * fls. 682 */ 683 static u8 size_index[24] __ro_after_init = { 684 3, /* 8 */ 685 4, /* 16 */ 686 5, /* 24 */ 687 5, /* 32 */ 688 6, /* 40 */ 689 6, /* 48 */ 690 6, /* 56 */ 691 6, /* 64 */ 692 1, /* 72 */ 693 1, /* 80 */ 694 1, /* 88 */ 695 1, /* 96 */ 696 7, /* 104 */ 697 7, /* 112 */ 698 7, /* 120 */ 699 7, /* 128 */ 700 2, /* 136 */ 701 2, /* 144 */ 702 2, /* 152 */ 703 2, /* 160 */ 704 2, /* 168 */ 705 2, /* 176 */ 706 2, /* 184 */ 707 2 /* 192 */ 708 }; 709 710 static inline unsigned int size_index_elem(unsigned int bytes) 711 { 712 return (bytes - 1) / 8; 713 } 714 715 /* 716 * Find the kmem_cache structure that serves a given size of 717 * allocation 718 */ 719 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) 720 { 721 unsigned int index; 722 723 if (size <= 192) { 724 if (!size) 725 return ZERO_SIZE_PTR; 726 727 index = size_index[size_index_elem(size)]; 728 } else { 729 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE)) 730 return NULL; 731 index = fls(size - 1); 732 } 733 734 return kmalloc_caches[kmalloc_type(flags)][index]; 735 } 736 737 #ifdef CONFIG_ZONE_DMA 738 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz, 739 #else 740 #define KMALLOC_DMA_NAME(sz) 741 #endif 742 743 #ifdef CONFIG_MEMCG_KMEM 744 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz, 745 #else 746 #define KMALLOC_CGROUP_NAME(sz) 747 #endif 748 749 #define INIT_KMALLOC_INFO(__size, __short_size) \ 750 { \ 751 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ 752 .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ 753 KMALLOC_CGROUP_NAME(__short_size) \ 754 KMALLOC_DMA_NAME(__short_size) \ 755 .size = __size, \ 756 } 757 758 /* 759 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. 760 * kmalloc_index() supports up to 2^25=32MB, so the final entry of the table is 761 * kmalloc-32M. 762 */ 763 const struct kmalloc_info_struct kmalloc_info[] __initconst = { 764 INIT_KMALLOC_INFO(0, 0), 765 INIT_KMALLOC_INFO(96, 96), 766 INIT_KMALLOC_INFO(192, 192), 767 INIT_KMALLOC_INFO(8, 8), 768 INIT_KMALLOC_INFO(16, 16), 769 INIT_KMALLOC_INFO(32, 32), 770 INIT_KMALLOC_INFO(64, 64), 771 INIT_KMALLOC_INFO(128, 128), 772 INIT_KMALLOC_INFO(256, 256), 773 INIT_KMALLOC_INFO(512, 512), 774 INIT_KMALLOC_INFO(1024, 1k), 775 INIT_KMALLOC_INFO(2048, 2k), 776 INIT_KMALLOC_INFO(4096, 4k), 777 INIT_KMALLOC_INFO(8192, 8k), 778 INIT_KMALLOC_INFO(16384, 16k), 779 INIT_KMALLOC_INFO(32768, 32k), 780 INIT_KMALLOC_INFO(65536, 64k), 781 INIT_KMALLOC_INFO(131072, 128k), 782 INIT_KMALLOC_INFO(262144, 256k), 783 INIT_KMALLOC_INFO(524288, 512k), 784 INIT_KMALLOC_INFO(1048576, 1M), 785 INIT_KMALLOC_INFO(2097152, 2M), 786 INIT_KMALLOC_INFO(4194304, 4M), 787 INIT_KMALLOC_INFO(8388608, 8M), 788 INIT_KMALLOC_INFO(16777216, 16M), 789 INIT_KMALLOC_INFO(33554432, 32M) 790 }; 791 792 /* 793 * Patch up the size_index table if we have strange large alignment 794 * requirements for the kmalloc array. This is only the case for 795 * MIPS it seems. The standard arches will not generate any code here. 796 * 797 * Largest permitted alignment is 256 bytes due to the way we 798 * handle the index determination for the smaller caches. 799 * 800 * Make sure that nothing crazy happens if someone starts tinkering 801 * around with ARCH_KMALLOC_MINALIGN 802 */ 803 void __init setup_kmalloc_cache_index_table(void) 804 { 805 unsigned int i; 806 807 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || 808 !is_power_of_2(KMALLOC_MIN_SIZE)); 809 810 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { 811 unsigned int elem = size_index_elem(i); 812 813 if (elem >= ARRAY_SIZE(size_index)) 814 break; 815 size_index[elem] = KMALLOC_SHIFT_LOW; 816 } 817 818 if (KMALLOC_MIN_SIZE >= 64) { 819 /* 820 * The 96 byte sized cache is not used if the alignment 821 * is 64 byte. 822 */ 823 for (i = 64 + 8; i <= 96; i += 8) 824 size_index[size_index_elem(i)] = 7; 825 826 } 827 828 if (KMALLOC_MIN_SIZE >= 128) { 829 /* 830 * The 192 byte sized cache is not used if the alignment 831 * is 128 byte. Redirect kmalloc to use the 256 byte cache 832 * instead. 833 */ 834 for (i = 128 + 8; i <= 192; i += 8) 835 size_index[size_index_elem(i)] = 8; 836 } 837 } 838 839 static void __init 840 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags) 841 { 842 if (type == KMALLOC_RECLAIM) { 843 flags |= SLAB_RECLAIM_ACCOUNT; 844 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) { 845 if (mem_cgroup_kmem_disabled()) { 846 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx]; 847 return; 848 } 849 flags |= SLAB_ACCOUNT; 850 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) { 851 flags |= SLAB_CACHE_DMA; 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 < NR_KMALLOC_TYPES; 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 #endif /* !CONFIG_SLOB */ 903 904 gfp_t kmalloc_fix_flags(gfp_t flags) 905 { 906 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; 907 908 flags &= ~GFP_SLAB_BUG_MASK; 909 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", 910 invalid_mask, &invalid_mask, flags, &flags); 911 dump_stack(); 912 913 return flags; 914 } 915 916 /* 917 * To avoid unnecessary overhead, we pass through large allocation requests 918 * directly to the page allocator. We use __GFP_COMP, because we will need to 919 * know the allocation order to free the pages properly in kfree. 920 */ 921 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) 922 { 923 void *ret = NULL; 924 struct page *page; 925 926 if (unlikely(flags & GFP_SLAB_BUG_MASK)) 927 flags = kmalloc_fix_flags(flags); 928 929 flags |= __GFP_COMP; 930 page = alloc_pages(flags, order); 931 if (likely(page)) { 932 ret = page_address(page); 933 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, 934 PAGE_SIZE << order); 935 } 936 ret = kasan_kmalloc_large(ret, size, flags); 937 /* As ret might get tagged, call kmemleak hook after KASAN. */ 938 kmemleak_alloc(ret, size, 1, flags); 939 return ret; 940 } 941 EXPORT_SYMBOL(kmalloc_order); 942 943 #ifdef CONFIG_TRACING 944 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) 945 { 946 void *ret = kmalloc_order(size, flags, order); 947 trace_kmalloc(_RET_IP_, ret, NULL, size, PAGE_SIZE << order, flags); 948 return ret; 949 } 950 EXPORT_SYMBOL(kmalloc_order_trace); 951 #endif 952 953 #ifdef CONFIG_SLAB_FREELIST_RANDOM 954 /* Randomize a generic freelist */ 955 static void freelist_randomize(struct rnd_state *state, unsigned int *list, 956 unsigned int count) 957 { 958 unsigned int rand; 959 unsigned int i; 960 961 for (i = 0; i < count; i++) 962 list[i] = i; 963 964 /* Fisher-Yates shuffle */ 965 for (i = count - 1; i > 0; i--) { 966 rand = prandom_u32_state(state); 967 rand %= (i + 1); 968 swap(list[i], list[rand]); 969 } 970 } 971 972 /* Create a random sequence per cache */ 973 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, 974 gfp_t gfp) 975 { 976 struct rnd_state state; 977 978 if (count < 2 || cachep->random_seq) 979 return 0; 980 981 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); 982 if (!cachep->random_seq) 983 return -ENOMEM; 984 985 /* Get best entropy at this stage of boot */ 986 prandom_seed_state(&state, get_random_long()); 987 988 freelist_randomize(&state, cachep->random_seq, count); 989 return 0; 990 } 991 992 /* Destroy the per-cache random freelist sequence */ 993 void cache_random_seq_destroy(struct kmem_cache *cachep) 994 { 995 kfree(cachep->random_seq); 996 cachep->random_seq = NULL; 997 } 998 #endif /* CONFIG_SLAB_FREELIST_RANDOM */ 999 1000 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) 1001 #ifdef CONFIG_SLAB 1002 #define SLABINFO_RIGHTS (0600) 1003 #else 1004 #define SLABINFO_RIGHTS (0400) 1005 #endif 1006 1007 static void print_slabinfo_header(struct seq_file *m) 1008 { 1009 /* 1010 * Output format version, so at least we can change it 1011 * without _too_ many complaints. 1012 */ 1013 #ifdef CONFIG_DEBUG_SLAB 1014 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); 1015 #else 1016 seq_puts(m, "slabinfo - version: 2.1\n"); 1017 #endif 1018 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); 1019 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); 1020 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); 1021 #ifdef CONFIG_DEBUG_SLAB 1022 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); 1023 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); 1024 #endif 1025 seq_putc(m, '\n'); 1026 } 1027 1028 static void *slab_start(struct seq_file *m, loff_t *pos) 1029 { 1030 mutex_lock(&slab_mutex); 1031 return seq_list_start(&slab_caches, *pos); 1032 } 1033 1034 static void *slab_next(struct seq_file *m, void *p, loff_t *pos) 1035 { 1036 return seq_list_next(p, &slab_caches, pos); 1037 } 1038 1039 static void slab_stop(struct seq_file *m, void *p) 1040 { 1041 mutex_unlock(&slab_mutex); 1042 } 1043 1044 static void cache_show(struct kmem_cache *s, struct seq_file *m) 1045 { 1046 struct slabinfo sinfo; 1047 1048 memset(&sinfo, 0, sizeof(sinfo)); 1049 get_slabinfo(s, &sinfo); 1050 1051 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", 1052 s->name, sinfo.active_objs, sinfo.num_objs, s->size, 1053 sinfo.objects_per_slab, (1 << sinfo.cache_order)); 1054 1055 seq_printf(m, " : tunables %4u %4u %4u", 1056 sinfo.limit, sinfo.batchcount, sinfo.shared); 1057 seq_printf(m, " : slabdata %6lu %6lu %6lu", 1058 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); 1059 slabinfo_show_stats(m, s); 1060 seq_putc(m, '\n'); 1061 } 1062 1063 static int slab_show(struct seq_file *m, void *p) 1064 { 1065 struct kmem_cache *s = list_entry(p, struct kmem_cache, list); 1066 1067 if (p == slab_caches.next) 1068 print_slabinfo_header(m); 1069 cache_show(s, m); 1070 return 0; 1071 } 1072 1073 void dump_unreclaimable_slab(void) 1074 { 1075 struct kmem_cache *s; 1076 struct slabinfo sinfo; 1077 1078 /* 1079 * Here acquiring slab_mutex is risky since we don't prefer to get 1080 * sleep in oom path. But, without mutex hold, it may introduce a 1081 * risk of crash. 1082 * Use mutex_trylock to protect the list traverse, dump nothing 1083 * without acquiring the mutex. 1084 */ 1085 if (!mutex_trylock(&slab_mutex)) { 1086 pr_warn("excessive unreclaimable slab but cannot dump stats\n"); 1087 return; 1088 } 1089 1090 pr_info("Unreclaimable slab info:\n"); 1091 pr_info("Name Used Total\n"); 1092 1093 list_for_each_entry(s, &slab_caches, list) { 1094 if (s->flags & SLAB_RECLAIM_ACCOUNT) 1095 continue; 1096 1097 get_slabinfo(s, &sinfo); 1098 1099 if (sinfo.num_objs > 0) 1100 pr_info("%-17s %10luKB %10luKB\n", s->name, 1101 (sinfo.active_objs * s->size) / 1024, 1102 (sinfo.num_objs * s->size) / 1024); 1103 } 1104 mutex_unlock(&slab_mutex); 1105 } 1106 1107 /* 1108 * slabinfo_op - iterator that generates /proc/slabinfo 1109 * 1110 * Output layout: 1111 * cache-name 1112 * num-active-objs 1113 * total-objs 1114 * object size 1115 * num-active-slabs 1116 * total-slabs 1117 * num-pages-per-slab 1118 * + further values on SMP and with statistics enabled 1119 */ 1120 static const struct seq_operations slabinfo_op = { 1121 .start = slab_start, 1122 .next = slab_next, 1123 .stop = slab_stop, 1124 .show = slab_show, 1125 }; 1126 1127 static int slabinfo_open(struct inode *inode, struct file *file) 1128 { 1129 return seq_open(file, &slabinfo_op); 1130 } 1131 1132 static const struct proc_ops slabinfo_proc_ops = { 1133 .proc_flags = PROC_ENTRY_PERMANENT, 1134 .proc_open = slabinfo_open, 1135 .proc_read = seq_read, 1136 .proc_write = slabinfo_write, 1137 .proc_lseek = seq_lseek, 1138 .proc_release = seq_release, 1139 }; 1140 1141 static int __init slab_proc_init(void) 1142 { 1143 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); 1144 return 0; 1145 } 1146 module_init(slab_proc_init); 1147 1148 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ 1149 1150 static __always_inline void *__do_krealloc(const void *p, size_t new_size, 1151 gfp_t flags) 1152 { 1153 void *ret; 1154 size_t ks; 1155 1156 /* Don't use instrumented ksize to allow precise KASAN poisoning. */ 1157 if (likely(!ZERO_OR_NULL_PTR(p))) { 1158 if (!kasan_check_byte(p)) 1159 return NULL; 1160 ks = kfence_ksize(p) ?: __ksize(p); 1161 } else 1162 ks = 0; 1163 1164 /* If the object still fits, repoison it precisely. */ 1165 if (ks >= new_size) { 1166 p = kasan_krealloc((void *)p, new_size, flags); 1167 return (void *)p; 1168 } 1169 1170 ret = kmalloc_track_caller(new_size, flags); 1171 if (ret && p) { 1172 /* Disable KASAN checks as the object's redzone is accessed. */ 1173 kasan_disable_current(); 1174 memcpy(ret, kasan_reset_tag(p), ks); 1175 kasan_enable_current(); 1176 } 1177 1178 return ret; 1179 } 1180 1181 /** 1182 * krealloc - reallocate memory. The contents will remain unchanged. 1183 * @p: object to reallocate memory for. 1184 * @new_size: how many bytes of memory are required. 1185 * @flags: the type of memory to allocate. 1186 * 1187 * The contents of the object pointed to are preserved up to the 1188 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored). 1189 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size 1190 * is 0 and @p is not a %NULL pointer, the object pointed to is freed. 1191 * 1192 * Return: pointer to the allocated memory or %NULL in case of error 1193 */ 1194 void *krealloc(const void *p, size_t new_size, gfp_t flags) 1195 { 1196 void *ret; 1197 1198 if (unlikely(!new_size)) { 1199 kfree(p); 1200 return ZERO_SIZE_PTR; 1201 } 1202 1203 ret = __do_krealloc(p, new_size, flags); 1204 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) 1205 kfree(p); 1206 1207 return ret; 1208 } 1209 EXPORT_SYMBOL(krealloc); 1210 1211 /** 1212 * kfree_sensitive - Clear sensitive information in memory before freeing 1213 * @p: object to free memory of 1214 * 1215 * The memory of the object @p points to is zeroed before freed. 1216 * If @p is %NULL, kfree_sensitive() does nothing. 1217 * 1218 * Note: this function zeroes the whole allocated buffer which can be a good 1219 * deal bigger than the requested buffer size passed to kmalloc(). So be 1220 * careful when using this function in performance sensitive code. 1221 */ 1222 void kfree_sensitive(const void *p) 1223 { 1224 size_t ks; 1225 void *mem = (void *)p; 1226 1227 ks = ksize(mem); 1228 if (ks) 1229 memzero_explicit(mem, ks); 1230 kfree(mem); 1231 } 1232 EXPORT_SYMBOL(kfree_sensitive); 1233 1234 /** 1235 * ksize - get the actual amount of memory allocated for a given object 1236 * @objp: Pointer to the object 1237 * 1238 * kmalloc may internally round up allocations and return more memory 1239 * than requested. ksize() can be used to determine the actual amount of 1240 * memory allocated. The caller may use this additional memory, even though 1241 * a smaller amount of memory was initially specified with the kmalloc call. 1242 * The caller must guarantee that objp points to a valid object previously 1243 * allocated with either kmalloc() or kmem_cache_alloc(). The object 1244 * must not be freed during the duration of the call. 1245 * 1246 * Return: size of the actual memory used by @objp in bytes 1247 */ 1248 size_t ksize(const void *objp) 1249 { 1250 size_t size; 1251 1252 /* 1253 * We need to first check that the pointer to the object is valid, and 1254 * only then unpoison the memory. The report printed from ksize() is 1255 * more useful, then when it's printed later when the behaviour could 1256 * be undefined due to a potential use-after-free or double-free. 1257 * 1258 * We use kasan_check_byte(), which is supported for the hardware 1259 * tag-based KASAN mode, unlike kasan_check_read/write(). 1260 * 1261 * If the pointed to memory is invalid, we return 0 to avoid users of 1262 * ksize() writing to and potentially corrupting the memory region. 1263 * 1264 * We want to perform the check before __ksize(), to avoid potentially 1265 * crashing in __ksize() due to accessing invalid metadata. 1266 */ 1267 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp)) 1268 return 0; 1269 1270 size = kfence_ksize(objp) ?: __ksize(objp); 1271 /* 1272 * We assume that ksize callers could use whole allocated area, 1273 * so we need to unpoison this area. 1274 */ 1275 kasan_unpoison_range(objp, size); 1276 return size; 1277 } 1278 EXPORT_SYMBOL(ksize); 1279 1280 /* Tracepoints definitions. */ 1281 EXPORT_TRACEPOINT_SYMBOL(kmalloc); 1282 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); 1283 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); 1284 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); 1285 EXPORT_TRACEPOINT_SYMBOL(kfree); 1286 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); 1287 1288 int should_failslab(struct kmem_cache *s, gfp_t gfpflags) 1289 { 1290 if (__should_failslab(s, gfpflags)) 1291 return -ENOMEM; 1292 return 0; 1293 } 1294 ALLOW_ERROR_INJECTION(should_failslab, ERRNO); 1295