1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Stack depot - a stack trace storage that avoids duplication. 4 * 5 * Internally, stack depot maintains a hash table of unique stacktraces. The 6 * stack traces themselves are stored contiguously one after another in a set 7 * of separate page allocations. 8 * 9 * Author: Alexander Potapenko <glider@google.com> 10 * Copyright (C) 2016 Google, Inc. 11 * 12 * Based on the code by Dmitry Chernenkov. 13 */ 14 15 #define pr_fmt(fmt) "stackdepot: " fmt 16 17 #include <linux/gfp.h> 18 #include <linux/jhash.h> 19 #include <linux/kernel.h> 20 #include <linux/kmsan.h> 21 #include <linux/list.h> 22 #include <linux/mm.h> 23 #include <linux/mutex.h> 24 #include <linux/percpu.h> 25 #include <linux/printk.h> 26 #include <linux/refcount.h> 27 #include <linux/slab.h> 28 #include <linux/spinlock.h> 29 #include <linux/stacktrace.h> 30 #include <linux/stackdepot.h> 31 #include <linux/string.h> 32 #include <linux/types.h> 33 #include <linux/memblock.h> 34 #include <linux/kasan-enabled.h> 35 36 #define DEPOT_HANDLE_BITS (sizeof(depot_stack_handle_t) * 8) 37 38 #define DEPOT_POOL_ORDER 2 /* Pool size order, 4 pages */ 39 #define DEPOT_POOL_SIZE (1LL << (PAGE_SHIFT + DEPOT_POOL_ORDER)) 40 #define DEPOT_STACK_ALIGN 4 41 #define DEPOT_OFFSET_BITS (DEPOT_POOL_ORDER + PAGE_SHIFT - DEPOT_STACK_ALIGN) 42 #define DEPOT_POOL_INDEX_BITS (DEPOT_HANDLE_BITS - DEPOT_OFFSET_BITS - \ 43 STACK_DEPOT_EXTRA_BITS) 44 #if IS_ENABLED(CONFIG_KMSAN) && CONFIG_STACKDEPOT_MAX_FRAMES >= 32 45 /* 46 * KMSAN is frequently used in fuzzing scenarios and thus saves a lot of stack 47 * traces. As KMSAN does not support evicting stack traces from the stack 48 * depot, the stack depot capacity might be reached quickly with large stack 49 * records. Adjust the maximum number of stack depot pools for this case. 50 */ 51 #define DEPOT_POOLS_CAP (8192 * (CONFIG_STACKDEPOT_MAX_FRAMES / 16)) 52 #else 53 #define DEPOT_POOLS_CAP 8192 54 #endif 55 #define DEPOT_MAX_POOLS \ 56 (((1LL << (DEPOT_POOL_INDEX_BITS)) < DEPOT_POOLS_CAP) ? \ 57 (1LL << (DEPOT_POOL_INDEX_BITS)) : DEPOT_POOLS_CAP) 58 59 /* Compact structure that stores a reference to a stack. */ 60 union handle_parts { 61 depot_stack_handle_t handle; 62 struct { 63 u32 pool_index : DEPOT_POOL_INDEX_BITS; 64 u32 offset : DEPOT_OFFSET_BITS; 65 u32 extra : STACK_DEPOT_EXTRA_BITS; 66 }; 67 }; 68 69 struct stack_record { 70 struct list_head list; /* Links in hash table or freelist */ 71 u32 hash; /* Hash in hash table */ 72 u32 size; /* Number of stored frames */ 73 union handle_parts handle; 74 refcount_t count; 75 unsigned long entries[CONFIG_STACKDEPOT_MAX_FRAMES]; /* Frames */ 76 }; 77 78 #define DEPOT_STACK_RECORD_SIZE \ 79 ALIGN(sizeof(struct stack_record), 1 << DEPOT_STACK_ALIGN) 80 81 static bool stack_depot_disabled; 82 static bool __stack_depot_early_init_requested __initdata = IS_ENABLED(CONFIG_STACKDEPOT_ALWAYS_INIT); 83 static bool __stack_depot_early_init_passed __initdata; 84 85 /* Use one hash table bucket per 16 KB of memory. */ 86 #define STACK_HASH_TABLE_SCALE 14 87 /* Limit the number of buckets between 4K and 1M. */ 88 #define STACK_BUCKET_NUMBER_ORDER_MIN 12 89 #define STACK_BUCKET_NUMBER_ORDER_MAX 20 90 /* Initial seed for jhash2. */ 91 #define STACK_HASH_SEED 0x9747b28c 92 93 /* Hash table of stored stack records. */ 94 static struct list_head *stack_table; 95 /* Fixed order of the number of table buckets. Used when KASAN is enabled. */ 96 static unsigned int stack_bucket_number_order; 97 /* Hash mask for indexing the table. */ 98 static unsigned int stack_hash_mask; 99 100 /* Array of memory regions that store stack records. */ 101 static void *stack_pools[DEPOT_MAX_POOLS]; 102 /* Newly allocated pool that is not yet added to stack_pools. */ 103 static void *new_pool; 104 /* Number of pools in stack_pools. */ 105 static int pools_num; 106 /* Freelist of stack records within stack_pools. */ 107 static LIST_HEAD(free_stacks); 108 /* 109 * Stack depot tries to keep an extra pool allocated even before it runs out 110 * of space in the currently used pool. This flag marks whether this extra pool 111 * needs to be allocated. It has the value 0 when either an extra pool is not 112 * yet allocated or if the limit on the number of pools is reached. 113 */ 114 static bool new_pool_required = true; 115 /* Lock that protects the variables above. */ 116 static DEFINE_RWLOCK(pool_rwlock); 117 118 static int __init disable_stack_depot(char *str) 119 { 120 return kstrtobool(str, &stack_depot_disabled); 121 } 122 early_param("stack_depot_disable", disable_stack_depot); 123 124 void __init stack_depot_request_early_init(void) 125 { 126 /* Too late to request early init now. */ 127 WARN_ON(__stack_depot_early_init_passed); 128 129 __stack_depot_early_init_requested = true; 130 } 131 132 /* Initialize list_head's within the hash table. */ 133 static void init_stack_table(unsigned long entries) 134 { 135 unsigned long i; 136 137 for (i = 0; i < entries; i++) 138 INIT_LIST_HEAD(&stack_table[i]); 139 } 140 141 /* Allocates a hash table via memblock. Can only be used during early boot. */ 142 int __init stack_depot_early_init(void) 143 { 144 unsigned long entries = 0; 145 146 /* This function must be called only once, from mm_init(). */ 147 if (WARN_ON(__stack_depot_early_init_passed)) 148 return 0; 149 __stack_depot_early_init_passed = true; 150 151 /* 152 * Print disabled message even if early init has not been requested: 153 * stack_depot_init() will not print one. 154 */ 155 if (stack_depot_disabled) { 156 pr_info("disabled\n"); 157 return 0; 158 } 159 160 /* 161 * If KASAN is enabled, use the maximum order: KASAN is frequently used 162 * in fuzzing scenarios, which leads to a large number of different 163 * stack traces being stored in stack depot. 164 */ 165 if (kasan_enabled() && !stack_bucket_number_order) 166 stack_bucket_number_order = STACK_BUCKET_NUMBER_ORDER_MAX; 167 168 /* 169 * Check if early init has been requested after setting 170 * stack_bucket_number_order: stack_depot_init() uses its value. 171 */ 172 if (!__stack_depot_early_init_requested) 173 return 0; 174 175 /* 176 * If stack_bucket_number_order is not set, leave entries as 0 to rely 177 * on the automatic calculations performed by alloc_large_system_hash(). 178 */ 179 if (stack_bucket_number_order) 180 entries = 1UL << stack_bucket_number_order; 181 pr_info("allocating hash table via alloc_large_system_hash\n"); 182 stack_table = alloc_large_system_hash("stackdepot", 183 sizeof(struct list_head), 184 entries, 185 STACK_HASH_TABLE_SCALE, 186 HASH_EARLY, 187 NULL, 188 &stack_hash_mask, 189 1UL << STACK_BUCKET_NUMBER_ORDER_MIN, 190 1UL << STACK_BUCKET_NUMBER_ORDER_MAX); 191 if (!stack_table) { 192 pr_err("hash table allocation failed, disabling\n"); 193 stack_depot_disabled = true; 194 return -ENOMEM; 195 } 196 if (!entries) { 197 /* 198 * Obtain the number of entries that was calculated by 199 * alloc_large_system_hash(). 200 */ 201 entries = stack_hash_mask + 1; 202 } 203 init_stack_table(entries); 204 205 return 0; 206 } 207 208 /* Allocates a hash table via kvcalloc. Can be used after boot. */ 209 int stack_depot_init(void) 210 { 211 static DEFINE_MUTEX(stack_depot_init_mutex); 212 unsigned long entries; 213 int ret = 0; 214 215 mutex_lock(&stack_depot_init_mutex); 216 217 if (stack_depot_disabled || stack_table) 218 goto out_unlock; 219 220 /* 221 * Similarly to stack_depot_early_init, use stack_bucket_number_order 222 * if assigned, and rely on automatic scaling otherwise. 223 */ 224 if (stack_bucket_number_order) { 225 entries = 1UL << stack_bucket_number_order; 226 } else { 227 int scale = STACK_HASH_TABLE_SCALE; 228 229 entries = nr_free_buffer_pages(); 230 entries = roundup_pow_of_two(entries); 231 232 if (scale > PAGE_SHIFT) 233 entries >>= (scale - PAGE_SHIFT); 234 else 235 entries <<= (PAGE_SHIFT - scale); 236 } 237 238 if (entries < 1UL << STACK_BUCKET_NUMBER_ORDER_MIN) 239 entries = 1UL << STACK_BUCKET_NUMBER_ORDER_MIN; 240 if (entries > 1UL << STACK_BUCKET_NUMBER_ORDER_MAX) 241 entries = 1UL << STACK_BUCKET_NUMBER_ORDER_MAX; 242 243 pr_info("allocating hash table of %lu entries via kvcalloc\n", entries); 244 stack_table = kvcalloc(entries, sizeof(struct list_head), GFP_KERNEL); 245 if (!stack_table) { 246 pr_err("hash table allocation failed, disabling\n"); 247 stack_depot_disabled = true; 248 ret = -ENOMEM; 249 goto out_unlock; 250 } 251 stack_hash_mask = entries - 1; 252 init_stack_table(entries); 253 254 out_unlock: 255 mutex_unlock(&stack_depot_init_mutex); 256 257 return ret; 258 } 259 EXPORT_SYMBOL_GPL(stack_depot_init); 260 261 /* Initializes a stack depol pool. */ 262 static void depot_init_pool(void *pool) 263 { 264 int offset; 265 266 lockdep_assert_held_write(&pool_rwlock); 267 268 WARN_ON(!list_empty(&free_stacks)); 269 270 /* Initialize handles and link stack records into the freelist. */ 271 for (offset = 0; offset <= DEPOT_POOL_SIZE - DEPOT_STACK_RECORD_SIZE; 272 offset += DEPOT_STACK_RECORD_SIZE) { 273 struct stack_record *stack = pool + offset; 274 275 stack->handle.pool_index = pools_num; 276 stack->handle.offset = offset >> DEPOT_STACK_ALIGN; 277 stack->handle.extra = 0; 278 279 list_add(&stack->list, &free_stacks); 280 } 281 282 /* Save reference to the pool to be used by depot_fetch_stack(). */ 283 stack_pools[pools_num] = pool; 284 pools_num++; 285 } 286 287 /* Keeps the preallocated memory to be used for a new stack depot pool. */ 288 static void depot_keep_new_pool(void **prealloc) 289 { 290 lockdep_assert_held_write(&pool_rwlock); 291 292 /* 293 * If a new pool is already saved or the maximum number of 294 * pools is reached, do not use the preallocated memory. 295 */ 296 if (!new_pool_required) 297 return; 298 299 /* 300 * Use the preallocated memory for the new pool 301 * as long as we do not exceed the maximum number of pools. 302 */ 303 if (pools_num < DEPOT_MAX_POOLS) { 304 new_pool = *prealloc; 305 *prealloc = NULL; 306 } 307 308 /* 309 * At this point, either a new pool is kept or the maximum 310 * number of pools is reached. In either case, take note that 311 * keeping another pool is not required. 312 */ 313 new_pool_required = false; 314 } 315 316 /* Updates references to the current and the next stack depot pools. */ 317 static bool depot_update_pools(void **prealloc) 318 { 319 lockdep_assert_held_write(&pool_rwlock); 320 321 /* Check if we still have objects in the freelist. */ 322 if (!list_empty(&free_stacks)) 323 goto out_keep_prealloc; 324 325 /* Check if we have a new pool saved and use it. */ 326 if (new_pool) { 327 depot_init_pool(new_pool); 328 new_pool = NULL; 329 330 /* Take note that we might need a new new_pool. */ 331 if (pools_num < DEPOT_MAX_POOLS) 332 new_pool_required = true; 333 334 /* Try keeping the preallocated memory for new_pool. */ 335 goto out_keep_prealloc; 336 } 337 338 /* Bail out if we reached the pool limit. */ 339 if (unlikely(pools_num >= DEPOT_MAX_POOLS)) { 340 WARN_ONCE(1, "Stack depot reached limit capacity"); 341 return false; 342 } 343 344 /* Check if we have preallocated memory and use it. */ 345 if (*prealloc) { 346 depot_init_pool(*prealloc); 347 *prealloc = NULL; 348 return true; 349 } 350 351 return false; 352 353 out_keep_prealloc: 354 /* Keep the preallocated memory for a new pool if required. */ 355 if (*prealloc) 356 depot_keep_new_pool(prealloc); 357 return true; 358 } 359 360 /* Allocates a new stack in a stack depot pool. */ 361 static struct stack_record * 362 depot_alloc_stack(unsigned long *entries, int size, u32 hash, void **prealloc) 363 { 364 struct stack_record *stack; 365 366 lockdep_assert_held_write(&pool_rwlock); 367 368 /* Update current and new pools if required and possible. */ 369 if (!depot_update_pools(prealloc)) 370 return NULL; 371 372 /* Check if we have a stack record to save the stack trace. */ 373 if (list_empty(&free_stacks)) 374 return NULL; 375 376 /* Get and unlink the first entry from the freelist. */ 377 stack = list_first_entry(&free_stacks, struct stack_record, list); 378 list_del(&stack->list); 379 380 /* Limit number of saved frames to CONFIG_STACKDEPOT_MAX_FRAMES. */ 381 if (size > CONFIG_STACKDEPOT_MAX_FRAMES) 382 size = CONFIG_STACKDEPOT_MAX_FRAMES; 383 384 /* Save the stack trace. */ 385 stack->hash = hash; 386 stack->size = size; 387 /* stack->handle is already filled in by depot_init_pool(). */ 388 refcount_set(&stack->count, 1); 389 memcpy(stack->entries, entries, flex_array_size(stack, entries, size)); 390 391 /* 392 * Let KMSAN know the stored stack record is initialized. This shall 393 * prevent false positive reports if instrumented code accesses it. 394 */ 395 kmsan_unpoison_memory(stack, DEPOT_STACK_RECORD_SIZE); 396 397 return stack; 398 } 399 400 static struct stack_record *depot_fetch_stack(depot_stack_handle_t handle) 401 { 402 union handle_parts parts = { .handle = handle }; 403 void *pool; 404 size_t offset = parts.offset << DEPOT_STACK_ALIGN; 405 struct stack_record *stack; 406 407 lockdep_assert_held(&pool_rwlock); 408 409 if (parts.pool_index > pools_num) { 410 WARN(1, "pool index %d out of bounds (%d) for stack id %08x\n", 411 parts.pool_index, pools_num, handle); 412 return NULL; 413 } 414 415 pool = stack_pools[parts.pool_index]; 416 if (!pool) 417 return NULL; 418 419 stack = pool + offset; 420 return stack; 421 } 422 423 /* Links stack into the freelist. */ 424 static void depot_free_stack(struct stack_record *stack) 425 { 426 lockdep_assert_held_write(&pool_rwlock); 427 428 list_add(&stack->list, &free_stacks); 429 } 430 431 /* Calculates the hash for a stack. */ 432 static inline u32 hash_stack(unsigned long *entries, unsigned int size) 433 { 434 return jhash2((u32 *)entries, 435 array_size(size, sizeof(*entries)) / sizeof(u32), 436 STACK_HASH_SEED); 437 } 438 439 /* 440 * Non-instrumented version of memcmp(). 441 * Does not check the lexicographical order, only the equality. 442 */ 443 static inline 444 int stackdepot_memcmp(const unsigned long *u1, const unsigned long *u2, 445 unsigned int n) 446 { 447 for ( ; n-- ; u1++, u2++) { 448 if (*u1 != *u2) 449 return 1; 450 } 451 return 0; 452 } 453 454 /* Finds a stack in a bucket of the hash table. */ 455 static inline struct stack_record *find_stack(struct list_head *bucket, 456 unsigned long *entries, int size, 457 u32 hash) 458 { 459 struct list_head *pos; 460 struct stack_record *found; 461 462 lockdep_assert_held(&pool_rwlock); 463 464 list_for_each(pos, bucket) { 465 found = list_entry(pos, struct stack_record, list); 466 if (found->hash == hash && 467 found->size == size && 468 !stackdepot_memcmp(entries, found->entries, size)) 469 return found; 470 } 471 return NULL; 472 } 473 474 depot_stack_handle_t stack_depot_save_flags(unsigned long *entries, 475 unsigned int nr_entries, 476 gfp_t alloc_flags, 477 depot_flags_t depot_flags) 478 { 479 struct list_head *bucket; 480 struct stack_record *found = NULL; 481 depot_stack_handle_t handle = 0; 482 struct page *page = NULL; 483 void *prealloc = NULL; 484 bool can_alloc = depot_flags & STACK_DEPOT_FLAG_CAN_ALLOC; 485 bool need_alloc = false; 486 unsigned long flags; 487 u32 hash; 488 489 if (WARN_ON(depot_flags & ~STACK_DEPOT_FLAGS_MASK)) 490 return 0; 491 492 /* 493 * If this stack trace is from an interrupt, including anything before 494 * interrupt entry usually leads to unbounded stack depot growth. 495 * 496 * Since use of filter_irq_stacks() is a requirement to ensure stack 497 * depot can efficiently deduplicate interrupt stacks, always 498 * filter_irq_stacks() to simplify all callers' use of stack depot. 499 */ 500 nr_entries = filter_irq_stacks(entries, nr_entries); 501 502 if (unlikely(nr_entries == 0) || stack_depot_disabled) 503 return 0; 504 505 hash = hash_stack(entries, nr_entries); 506 bucket = &stack_table[hash & stack_hash_mask]; 507 508 read_lock_irqsave(&pool_rwlock, flags); 509 printk_deferred_enter(); 510 511 /* Fast path: look the stack trace up without full locking. */ 512 found = find_stack(bucket, entries, nr_entries, hash); 513 if (found) { 514 if (depot_flags & STACK_DEPOT_FLAG_GET) 515 refcount_inc(&found->count); 516 printk_deferred_exit(); 517 read_unlock_irqrestore(&pool_rwlock, flags); 518 goto exit; 519 } 520 521 /* Take note if another stack pool needs to be allocated. */ 522 if (new_pool_required) 523 need_alloc = true; 524 525 printk_deferred_exit(); 526 read_unlock_irqrestore(&pool_rwlock, flags); 527 528 /* 529 * Allocate memory for a new pool if required now: 530 * we won't be able to do that under the lock. 531 */ 532 if (unlikely(can_alloc && need_alloc)) { 533 /* 534 * Zero out zone modifiers, as we don't have specific zone 535 * requirements. Keep the flags related to allocation in atomic 536 * contexts and I/O. 537 */ 538 alloc_flags &= ~GFP_ZONEMASK; 539 alloc_flags &= (GFP_ATOMIC | GFP_KERNEL); 540 alloc_flags |= __GFP_NOWARN; 541 page = alloc_pages(alloc_flags, DEPOT_POOL_ORDER); 542 if (page) 543 prealloc = page_address(page); 544 } 545 546 write_lock_irqsave(&pool_rwlock, flags); 547 printk_deferred_enter(); 548 549 found = find_stack(bucket, entries, nr_entries, hash); 550 if (!found) { 551 struct stack_record *new = 552 depot_alloc_stack(entries, nr_entries, hash, &prealloc); 553 554 if (new) { 555 list_add(&new->list, bucket); 556 found = new; 557 } 558 } else { 559 if (depot_flags & STACK_DEPOT_FLAG_GET) 560 refcount_inc(&found->count); 561 /* 562 * Stack depot already contains this stack trace, but let's 563 * keep the preallocated memory for future. 564 */ 565 if (prealloc) 566 depot_keep_new_pool(&prealloc); 567 } 568 569 printk_deferred_exit(); 570 write_unlock_irqrestore(&pool_rwlock, flags); 571 exit: 572 if (prealloc) { 573 /* Stack depot didn't use this memory, free it. */ 574 free_pages((unsigned long)prealloc, DEPOT_POOL_ORDER); 575 } 576 if (found) 577 handle = found->handle.handle; 578 return handle; 579 } 580 EXPORT_SYMBOL_GPL(stack_depot_save_flags); 581 582 depot_stack_handle_t stack_depot_save(unsigned long *entries, 583 unsigned int nr_entries, 584 gfp_t alloc_flags) 585 { 586 return stack_depot_save_flags(entries, nr_entries, alloc_flags, 587 STACK_DEPOT_FLAG_CAN_ALLOC); 588 } 589 EXPORT_SYMBOL_GPL(stack_depot_save); 590 591 unsigned int stack_depot_fetch(depot_stack_handle_t handle, 592 unsigned long **entries) 593 { 594 struct stack_record *stack; 595 unsigned long flags; 596 597 *entries = NULL; 598 /* 599 * Let KMSAN know *entries is initialized. This shall prevent false 600 * positive reports if instrumented code accesses it. 601 */ 602 kmsan_unpoison_memory(entries, sizeof(*entries)); 603 604 if (!handle || stack_depot_disabled) 605 return 0; 606 607 read_lock_irqsave(&pool_rwlock, flags); 608 printk_deferred_enter(); 609 610 stack = depot_fetch_stack(handle); 611 612 printk_deferred_exit(); 613 read_unlock_irqrestore(&pool_rwlock, flags); 614 615 *entries = stack->entries; 616 return stack->size; 617 } 618 EXPORT_SYMBOL_GPL(stack_depot_fetch); 619 620 void stack_depot_put(depot_stack_handle_t handle) 621 { 622 struct stack_record *stack; 623 unsigned long flags; 624 625 if (!handle || stack_depot_disabled) 626 return; 627 628 write_lock_irqsave(&pool_rwlock, flags); 629 printk_deferred_enter(); 630 631 stack = depot_fetch_stack(handle); 632 if (WARN_ON(!stack)) 633 goto out; 634 635 if (refcount_dec_and_test(&stack->count)) { 636 /* Unlink stack from the hash table. */ 637 list_del(&stack->list); 638 639 /* Free stack. */ 640 depot_free_stack(stack); 641 } 642 643 out: 644 printk_deferred_exit(); 645 write_unlock_irqrestore(&pool_rwlock, flags); 646 } 647 EXPORT_SYMBOL_GPL(stack_depot_put); 648 649 void stack_depot_print(depot_stack_handle_t stack) 650 { 651 unsigned long *entries; 652 unsigned int nr_entries; 653 654 nr_entries = stack_depot_fetch(stack, &entries); 655 if (nr_entries > 0) 656 stack_trace_print(entries, nr_entries, 0); 657 } 658 EXPORT_SYMBOL_GPL(stack_depot_print); 659 660 int stack_depot_snprint(depot_stack_handle_t handle, char *buf, size_t size, 661 int spaces) 662 { 663 unsigned long *entries; 664 unsigned int nr_entries; 665 666 nr_entries = stack_depot_fetch(handle, &entries); 667 return nr_entries ? stack_trace_snprint(buf, size, entries, nr_entries, 668 spaces) : 0; 669 } 670 EXPORT_SYMBOL_GPL(stack_depot_snprint); 671 672 depot_stack_handle_t __must_check stack_depot_set_extra_bits( 673 depot_stack_handle_t handle, unsigned int extra_bits) 674 { 675 union handle_parts parts = { .handle = handle }; 676 677 /* Don't set extra bits on empty handles. */ 678 if (!handle) 679 return 0; 680 681 parts.extra = extra_bits; 682 return parts.handle; 683 } 684 EXPORT_SYMBOL(stack_depot_set_extra_bits); 685 686 unsigned int stack_depot_get_extra_bits(depot_stack_handle_t handle) 687 { 688 union handle_parts parts = { .handle = handle }; 689 690 return parts.extra; 691 } 692 EXPORT_SYMBOL(stack_depot_get_extra_bits); 693