1 /* 2 * mm/kmemleak.c 3 * 4 * Copyright (C) 2008 ARM Limited 5 * Written by Catalin Marinas <catalin.marinas@arm.com> 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 * 11 * This program is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 14 * GNU General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public License 17 * along with this program; if not, write to the Free Software 18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 19 * 20 * 21 * For more information on the algorithm and kmemleak usage, please see 22 * Documentation/kmemleak.txt. 23 * 24 * Notes on locking 25 * ---------------- 26 * 27 * The following locks and mutexes are used by kmemleak: 28 * 29 * - kmemleak_lock (rwlock): protects the object_list modifications and 30 * accesses to the object_tree_root. The object_list is the main list 31 * holding the metadata (struct kmemleak_object) for the allocated memory 32 * blocks. The object_tree_root is a priority search tree used to look-up 33 * metadata based on a pointer to the corresponding memory block. The 34 * kmemleak_object structures are added to the object_list and 35 * object_tree_root in the create_object() function called from the 36 * kmemleak_alloc() callback and removed in delete_object() called from the 37 * kmemleak_free() callback 38 * - kmemleak_object.lock (spinlock): protects a kmemleak_object. Accesses to 39 * the metadata (e.g. count) are protected by this lock. Note that some 40 * members of this structure may be protected by other means (atomic or 41 * kmemleak_lock). This lock is also held when scanning the corresponding 42 * memory block to avoid the kernel freeing it via the kmemleak_free() 43 * callback. This is less heavyweight than holding a global lock like 44 * kmemleak_lock during scanning 45 * - scan_mutex (mutex): ensures that only one thread may scan the memory for 46 * unreferenced objects at a time. The gray_list contains the objects which 47 * are already referenced or marked as false positives and need to be 48 * scanned. This list is only modified during a scanning episode when the 49 * scan_mutex is held. At the end of a scan, the gray_list is always empty. 50 * Note that the kmemleak_object.use_count is incremented when an object is 51 * added to the gray_list and therefore cannot be freed. This mutex also 52 * prevents multiple users of the "kmemleak" debugfs file together with 53 * modifications to the memory scanning parameters including the scan_thread 54 * pointer 55 * 56 * The kmemleak_object structures have a use_count incremented or decremented 57 * using the get_object()/put_object() functions. When the use_count becomes 58 * 0, this count can no longer be incremented and put_object() schedules the 59 * kmemleak_object freeing via an RCU callback. All calls to the get_object() 60 * function must be protected by rcu_read_lock() to avoid accessing a freed 61 * structure. 62 */ 63 64 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 65 66 #include <linux/init.h> 67 #include <linux/kernel.h> 68 #include <linux/list.h> 69 #include <linux/sched.h> 70 #include <linux/jiffies.h> 71 #include <linux/delay.h> 72 #include <linux/module.h> 73 #include <linux/kthread.h> 74 #include <linux/prio_tree.h> 75 #include <linux/fs.h> 76 #include <linux/debugfs.h> 77 #include <linux/seq_file.h> 78 #include <linux/cpumask.h> 79 #include <linux/spinlock.h> 80 #include <linux/mutex.h> 81 #include <linux/rcupdate.h> 82 #include <linux/stacktrace.h> 83 #include <linux/cache.h> 84 #include <linux/percpu.h> 85 #include <linux/hardirq.h> 86 #include <linux/mmzone.h> 87 #include <linux/slab.h> 88 #include <linux/thread_info.h> 89 #include <linux/err.h> 90 #include <linux/uaccess.h> 91 #include <linux/string.h> 92 #include <linux/nodemask.h> 93 #include <linux/mm.h> 94 #include <linux/workqueue.h> 95 #include <linux/crc32.h> 96 97 #include <asm/sections.h> 98 #include <asm/processor.h> 99 #include <asm/atomic.h> 100 101 #include <linux/kmemcheck.h> 102 #include <linux/kmemleak.h> 103 104 /* 105 * Kmemleak configuration and common defines. 106 */ 107 #define MAX_TRACE 16 /* stack trace length */ 108 #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */ 109 #define SECS_FIRST_SCAN 60 /* delay before the first scan */ 110 #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */ 111 #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */ 112 113 #define BYTES_PER_POINTER sizeof(void *) 114 115 /* GFP bitmask for kmemleak internal allocations */ 116 #define GFP_KMEMLEAK_MASK (GFP_KERNEL | GFP_ATOMIC) 117 118 /* scanning area inside a memory block */ 119 struct kmemleak_scan_area { 120 struct hlist_node node; 121 unsigned long start; 122 size_t size; 123 }; 124 125 #define KMEMLEAK_GREY 0 126 #define KMEMLEAK_BLACK -1 127 128 /* 129 * Structure holding the metadata for each allocated memory block. 130 * Modifications to such objects should be made while holding the 131 * object->lock. Insertions or deletions from object_list, gray_list or 132 * tree_node are already protected by the corresponding locks or mutex (see 133 * the notes on locking above). These objects are reference-counted 134 * (use_count) and freed using the RCU mechanism. 135 */ 136 struct kmemleak_object { 137 spinlock_t lock; 138 unsigned long flags; /* object status flags */ 139 struct list_head object_list; 140 struct list_head gray_list; 141 struct prio_tree_node tree_node; 142 struct rcu_head rcu; /* object_list lockless traversal */ 143 /* object usage count; object freed when use_count == 0 */ 144 atomic_t use_count; 145 unsigned long pointer; 146 size_t size; 147 /* minimum number of a pointers found before it is considered leak */ 148 int min_count; 149 /* the total number of pointers found pointing to this object */ 150 int count; 151 /* checksum for detecting modified objects */ 152 u32 checksum; 153 /* memory ranges to be scanned inside an object (empty for all) */ 154 struct hlist_head area_list; 155 unsigned long trace[MAX_TRACE]; 156 unsigned int trace_len; 157 unsigned long jiffies; /* creation timestamp */ 158 pid_t pid; /* pid of the current task */ 159 char comm[TASK_COMM_LEN]; /* executable name */ 160 }; 161 162 /* flag representing the memory block allocation status */ 163 #define OBJECT_ALLOCATED (1 << 0) 164 /* flag set after the first reporting of an unreference object */ 165 #define OBJECT_REPORTED (1 << 1) 166 /* flag set to not scan the object */ 167 #define OBJECT_NO_SCAN (1 << 2) 168 169 /* number of bytes to print per line; must be 16 or 32 */ 170 #define HEX_ROW_SIZE 16 171 /* number of bytes to print at a time (1, 2, 4, 8) */ 172 #define HEX_GROUP_SIZE 1 173 /* include ASCII after the hex output */ 174 #define HEX_ASCII 1 175 /* max number of lines to be printed */ 176 #define HEX_MAX_LINES 2 177 178 /* the list of all allocated objects */ 179 static LIST_HEAD(object_list); 180 /* the list of gray-colored objects (see color_gray comment below) */ 181 static LIST_HEAD(gray_list); 182 /* prio search tree for object boundaries */ 183 static struct prio_tree_root object_tree_root; 184 /* rw_lock protecting the access to object_list and prio_tree_root */ 185 static DEFINE_RWLOCK(kmemleak_lock); 186 187 /* allocation caches for kmemleak internal data */ 188 static struct kmem_cache *object_cache; 189 static struct kmem_cache *scan_area_cache; 190 191 /* set if tracing memory operations is enabled */ 192 static atomic_t kmemleak_enabled = ATOMIC_INIT(0); 193 /* set in the late_initcall if there were no errors */ 194 static atomic_t kmemleak_initialized = ATOMIC_INIT(0); 195 /* enables or disables early logging of the memory operations */ 196 static atomic_t kmemleak_early_log = ATOMIC_INIT(1); 197 /* set if a fata kmemleak error has occurred */ 198 static atomic_t kmemleak_error = ATOMIC_INIT(0); 199 200 /* minimum and maximum address that may be valid pointers */ 201 static unsigned long min_addr = ULONG_MAX; 202 static unsigned long max_addr; 203 204 static struct task_struct *scan_thread; 205 /* used to avoid reporting of recently allocated objects */ 206 static unsigned long jiffies_min_age; 207 static unsigned long jiffies_last_scan; 208 /* delay between automatic memory scannings */ 209 static signed long jiffies_scan_wait; 210 /* enables or disables the task stacks scanning */ 211 static int kmemleak_stack_scan = 1; 212 /* protects the memory scanning, parameters and debug/kmemleak file access */ 213 static DEFINE_MUTEX(scan_mutex); 214 215 /* 216 * Early object allocation/freeing logging. Kmemleak is initialized after the 217 * kernel allocator. However, both the kernel allocator and kmemleak may 218 * allocate memory blocks which need to be tracked. Kmemleak defines an 219 * arbitrary buffer to hold the allocation/freeing information before it is 220 * fully initialized. 221 */ 222 223 /* kmemleak operation type for early logging */ 224 enum { 225 KMEMLEAK_ALLOC, 226 KMEMLEAK_FREE, 227 KMEMLEAK_FREE_PART, 228 KMEMLEAK_NOT_LEAK, 229 KMEMLEAK_IGNORE, 230 KMEMLEAK_SCAN_AREA, 231 KMEMLEAK_NO_SCAN 232 }; 233 234 /* 235 * Structure holding the information passed to kmemleak callbacks during the 236 * early logging. 237 */ 238 struct early_log { 239 int op_type; /* kmemleak operation type */ 240 const void *ptr; /* allocated/freed memory block */ 241 size_t size; /* memory block size */ 242 int min_count; /* minimum reference count */ 243 unsigned long trace[MAX_TRACE]; /* stack trace */ 244 unsigned int trace_len; /* stack trace length */ 245 }; 246 247 /* early logging buffer and current position */ 248 static struct early_log 249 early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata; 250 static int crt_early_log __initdata; 251 252 static void kmemleak_disable(void); 253 254 /* 255 * Print a warning and dump the stack trace. 256 */ 257 #define kmemleak_warn(x...) do { \ 258 pr_warning(x); \ 259 dump_stack(); \ 260 } while (0) 261 262 /* 263 * Macro invoked when a serious kmemleak condition occured and cannot be 264 * recovered from. Kmemleak will be disabled and further allocation/freeing 265 * tracing no longer available. 266 */ 267 #define kmemleak_stop(x...) do { \ 268 kmemleak_warn(x); \ 269 kmemleak_disable(); \ 270 } while (0) 271 272 /* 273 * Printing of the objects hex dump to the seq file. The number of lines to be 274 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The 275 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called 276 * with the object->lock held. 277 */ 278 static void hex_dump_object(struct seq_file *seq, 279 struct kmemleak_object *object) 280 { 281 const u8 *ptr = (const u8 *)object->pointer; 282 int i, len, remaining; 283 unsigned char linebuf[HEX_ROW_SIZE * 5]; 284 285 /* limit the number of lines to HEX_MAX_LINES */ 286 remaining = len = 287 min(object->size, (size_t)(HEX_MAX_LINES * HEX_ROW_SIZE)); 288 289 seq_printf(seq, " hex dump (first %d bytes):\n", len); 290 for (i = 0; i < len; i += HEX_ROW_SIZE) { 291 int linelen = min(remaining, HEX_ROW_SIZE); 292 293 remaining -= HEX_ROW_SIZE; 294 hex_dump_to_buffer(ptr + i, linelen, HEX_ROW_SIZE, 295 HEX_GROUP_SIZE, linebuf, sizeof(linebuf), 296 HEX_ASCII); 297 seq_printf(seq, " %s\n", linebuf); 298 } 299 } 300 301 /* 302 * Object colors, encoded with count and min_count: 303 * - white - orphan object, not enough references to it (count < min_count) 304 * - gray - not orphan, not marked as false positive (min_count == 0) or 305 * sufficient references to it (count >= min_count) 306 * - black - ignore, it doesn't contain references (e.g. text section) 307 * (min_count == -1). No function defined for this color. 308 * Newly created objects don't have any color assigned (object->count == -1) 309 * before the next memory scan when they become white. 310 */ 311 static bool color_white(const struct kmemleak_object *object) 312 { 313 return object->count != KMEMLEAK_BLACK && 314 object->count < object->min_count; 315 } 316 317 static bool color_gray(const struct kmemleak_object *object) 318 { 319 return object->min_count != KMEMLEAK_BLACK && 320 object->count >= object->min_count; 321 } 322 323 /* 324 * Objects are considered unreferenced only if their color is white, they have 325 * not be deleted and have a minimum age to avoid false positives caused by 326 * pointers temporarily stored in CPU registers. 327 */ 328 static bool unreferenced_object(struct kmemleak_object *object) 329 { 330 return (color_white(object) && object->flags & OBJECT_ALLOCATED) && 331 time_before_eq(object->jiffies + jiffies_min_age, 332 jiffies_last_scan); 333 } 334 335 /* 336 * Printing of the unreferenced objects information to the seq file. The 337 * print_unreferenced function must be called with the object->lock held. 338 */ 339 static void print_unreferenced(struct seq_file *seq, 340 struct kmemleak_object *object) 341 { 342 int i; 343 unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies); 344 345 seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n", 346 object->pointer, object->size); 347 seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n", 348 object->comm, object->pid, object->jiffies, 349 msecs_age / 1000, msecs_age % 1000); 350 hex_dump_object(seq, object); 351 seq_printf(seq, " backtrace:\n"); 352 353 for (i = 0; i < object->trace_len; i++) { 354 void *ptr = (void *)object->trace[i]; 355 seq_printf(seq, " [<%p>] %pS\n", ptr, ptr); 356 } 357 } 358 359 /* 360 * Print the kmemleak_object information. This function is used mainly for 361 * debugging special cases when kmemleak operations. It must be called with 362 * the object->lock held. 363 */ 364 static void dump_object_info(struct kmemleak_object *object) 365 { 366 struct stack_trace trace; 367 368 trace.nr_entries = object->trace_len; 369 trace.entries = object->trace; 370 371 pr_notice("Object 0x%08lx (size %zu):\n", 372 object->tree_node.start, object->size); 373 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n", 374 object->comm, object->pid, object->jiffies); 375 pr_notice(" min_count = %d\n", object->min_count); 376 pr_notice(" count = %d\n", object->count); 377 pr_notice(" flags = 0x%lx\n", object->flags); 378 pr_notice(" checksum = %d\n", object->checksum); 379 pr_notice(" backtrace:\n"); 380 print_stack_trace(&trace, 4); 381 } 382 383 /* 384 * Look-up a memory block metadata (kmemleak_object) in the priority search 385 * tree based on a pointer value. If alias is 0, only values pointing to the 386 * beginning of the memory block are allowed. The kmemleak_lock must be held 387 * when calling this function. 388 */ 389 static struct kmemleak_object *lookup_object(unsigned long ptr, int alias) 390 { 391 struct prio_tree_node *node; 392 struct prio_tree_iter iter; 393 struct kmemleak_object *object; 394 395 prio_tree_iter_init(&iter, &object_tree_root, ptr, ptr); 396 node = prio_tree_next(&iter); 397 if (node) { 398 object = prio_tree_entry(node, struct kmemleak_object, 399 tree_node); 400 if (!alias && object->pointer != ptr) { 401 kmemleak_warn("Found object by alias"); 402 object = NULL; 403 } 404 } else 405 object = NULL; 406 407 return object; 408 } 409 410 /* 411 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note 412 * that once an object's use_count reached 0, the RCU freeing was already 413 * registered and the object should no longer be used. This function must be 414 * called under the protection of rcu_read_lock(). 415 */ 416 static int get_object(struct kmemleak_object *object) 417 { 418 return atomic_inc_not_zero(&object->use_count); 419 } 420 421 /* 422 * RCU callback to free a kmemleak_object. 423 */ 424 static void free_object_rcu(struct rcu_head *rcu) 425 { 426 struct hlist_node *elem, *tmp; 427 struct kmemleak_scan_area *area; 428 struct kmemleak_object *object = 429 container_of(rcu, struct kmemleak_object, rcu); 430 431 /* 432 * Once use_count is 0 (guaranteed by put_object), there is no other 433 * code accessing this object, hence no need for locking. 434 */ 435 hlist_for_each_entry_safe(area, elem, tmp, &object->area_list, node) { 436 hlist_del(elem); 437 kmem_cache_free(scan_area_cache, area); 438 } 439 kmem_cache_free(object_cache, object); 440 } 441 442 /* 443 * Decrement the object use_count. Once the count is 0, free the object using 444 * an RCU callback. Since put_object() may be called via the kmemleak_free() -> 445 * delete_object() path, the delayed RCU freeing ensures that there is no 446 * recursive call to the kernel allocator. Lock-less RCU object_list traversal 447 * is also possible. 448 */ 449 static void put_object(struct kmemleak_object *object) 450 { 451 if (!atomic_dec_and_test(&object->use_count)) 452 return; 453 454 /* should only get here after delete_object was called */ 455 WARN_ON(object->flags & OBJECT_ALLOCATED); 456 457 call_rcu(&object->rcu, free_object_rcu); 458 } 459 460 /* 461 * Look up an object in the prio search tree and increase its use_count. 462 */ 463 static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias) 464 { 465 unsigned long flags; 466 struct kmemleak_object *object = NULL; 467 468 rcu_read_lock(); 469 read_lock_irqsave(&kmemleak_lock, flags); 470 if (ptr >= min_addr && ptr < max_addr) 471 object = lookup_object(ptr, alias); 472 read_unlock_irqrestore(&kmemleak_lock, flags); 473 474 /* check whether the object is still available */ 475 if (object && !get_object(object)) 476 object = NULL; 477 rcu_read_unlock(); 478 479 return object; 480 } 481 482 /* 483 * Save stack trace to the given array of MAX_TRACE size. 484 */ 485 static int __save_stack_trace(unsigned long *trace) 486 { 487 struct stack_trace stack_trace; 488 489 stack_trace.max_entries = MAX_TRACE; 490 stack_trace.nr_entries = 0; 491 stack_trace.entries = trace; 492 stack_trace.skip = 2; 493 save_stack_trace(&stack_trace); 494 495 return stack_trace.nr_entries; 496 } 497 498 /* 499 * Create the metadata (struct kmemleak_object) corresponding to an allocated 500 * memory block and add it to the object_list and object_tree_root. 501 */ 502 static struct kmemleak_object *create_object(unsigned long ptr, size_t size, 503 int min_count, gfp_t gfp) 504 { 505 unsigned long flags; 506 struct kmemleak_object *object; 507 struct prio_tree_node *node; 508 509 object = kmem_cache_alloc(object_cache, gfp & GFP_KMEMLEAK_MASK); 510 if (!object) { 511 kmemleak_stop("Cannot allocate a kmemleak_object structure\n"); 512 return NULL; 513 } 514 515 INIT_LIST_HEAD(&object->object_list); 516 INIT_LIST_HEAD(&object->gray_list); 517 INIT_HLIST_HEAD(&object->area_list); 518 spin_lock_init(&object->lock); 519 atomic_set(&object->use_count, 1); 520 object->flags = OBJECT_ALLOCATED; 521 object->pointer = ptr; 522 object->size = size; 523 object->min_count = min_count; 524 object->count = 0; /* white color initially */ 525 object->jiffies = jiffies; 526 object->checksum = 0; 527 528 /* task information */ 529 if (in_irq()) { 530 object->pid = 0; 531 strncpy(object->comm, "hardirq", sizeof(object->comm)); 532 } else if (in_softirq()) { 533 object->pid = 0; 534 strncpy(object->comm, "softirq", sizeof(object->comm)); 535 } else { 536 object->pid = current->pid; 537 /* 538 * There is a small chance of a race with set_task_comm(), 539 * however using get_task_comm() here may cause locking 540 * dependency issues with current->alloc_lock. In the worst 541 * case, the command line is not correct. 542 */ 543 strncpy(object->comm, current->comm, sizeof(object->comm)); 544 } 545 546 /* kernel backtrace */ 547 object->trace_len = __save_stack_trace(object->trace); 548 549 INIT_PRIO_TREE_NODE(&object->tree_node); 550 object->tree_node.start = ptr; 551 object->tree_node.last = ptr + size - 1; 552 553 write_lock_irqsave(&kmemleak_lock, flags); 554 555 min_addr = min(min_addr, ptr); 556 max_addr = max(max_addr, ptr + size); 557 node = prio_tree_insert(&object_tree_root, &object->tree_node); 558 /* 559 * The code calling the kernel does not yet have the pointer to the 560 * memory block to be able to free it. However, we still hold the 561 * kmemleak_lock here in case parts of the kernel started freeing 562 * random memory blocks. 563 */ 564 if (node != &object->tree_node) { 565 kmemleak_stop("Cannot insert 0x%lx into the object search tree " 566 "(already existing)\n", ptr); 567 object = lookup_object(ptr, 1); 568 spin_lock(&object->lock); 569 dump_object_info(object); 570 spin_unlock(&object->lock); 571 572 goto out; 573 } 574 list_add_tail_rcu(&object->object_list, &object_list); 575 out: 576 write_unlock_irqrestore(&kmemleak_lock, flags); 577 return object; 578 } 579 580 /* 581 * Remove the metadata (struct kmemleak_object) for a memory block from the 582 * object_list and object_tree_root and decrement its use_count. 583 */ 584 static void __delete_object(struct kmemleak_object *object) 585 { 586 unsigned long flags; 587 588 write_lock_irqsave(&kmemleak_lock, flags); 589 prio_tree_remove(&object_tree_root, &object->tree_node); 590 list_del_rcu(&object->object_list); 591 write_unlock_irqrestore(&kmemleak_lock, flags); 592 593 WARN_ON(!(object->flags & OBJECT_ALLOCATED)); 594 WARN_ON(atomic_read(&object->use_count) < 2); 595 596 /* 597 * Locking here also ensures that the corresponding memory block 598 * cannot be freed when it is being scanned. 599 */ 600 spin_lock_irqsave(&object->lock, flags); 601 object->flags &= ~OBJECT_ALLOCATED; 602 spin_unlock_irqrestore(&object->lock, flags); 603 put_object(object); 604 } 605 606 /* 607 * Look up the metadata (struct kmemleak_object) corresponding to ptr and 608 * delete it. 609 */ 610 static void delete_object_full(unsigned long ptr) 611 { 612 struct kmemleak_object *object; 613 614 object = find_and_get_object(ptr, 0); 615 if (!object) { 616 #ifdef DEBUG 617 kmemleak_warn("Freeing unknown object at 0x%08lx\n", 618 ptr); 619 #endif 620 return; 621 } 622 __delete_object(object); 623 put_object(object); 624 } 625 626 /* 627 * Look up the metadata (struct kmemleak_object) corresponding to ptr and 628 * delete it. If the memory block is partially freed, the function may create 629 * additional metadata for the remaining parts of the block. 630 */ 631 static void delete_object_part(unsigned long ptr, size_t size) 632 { 633 struct kmemleak_object *object; 634 unsigned long start, end; 635 636 object = find_and_get_object(ptr, 1); 637 if (!object) { 638 #ifdef DEBUG 639 kmemleak_warn("Partially freeing unknown object at 0x%08lx " 640 "(size %zu)\n", ptr, size); 641 #endif 642 return; 643 } 644 __delete_object(object); 645 646 /* 647 * Create one or two objects that may result from the memory block 648 * split. Note that partial freeing is only done by free_bootmem() and 649 * this happens before kmemleak_init() is called. The path below is 650 * only executed during early log recording in kmemleak_init(), so 651 * GFP_KERNEL is enough. 652 */ 653 start = object->pointer; 654 end = object->pointer + object->size; 655 if (ptr > start) 656 create_object(start, ptr - start, object->min_count, 657 GFP_KERNEL); 658 if (ptr + size < end) 659 create_object(ptr + size, end - ptr - size, object->min_count, 660 GFP_KERNEL); 661 662 put_object(object); 663 } 664 665 static void __paint_it(struct kmemleak_object *object, int color) 666 { 667 object->min_count = color; 668 if (color == KMEMLEAK_BLACK) 669 object->flags |= OBJECT_NO_SCAN; 670 } 671 672 static void paint_it(struct kmemleak_object *object, int color) 673 { 674 unsigned long flags; 675 676 spin_lock_irqsave(&object->lock, flags); 677 __paint_it(object, color); 678 spin_unlock_irqrestore(&object->lock, flags); 679 } 680 681 static void paint_ptr(unsigned long ptr, int color) 682 { 683 struct kmemleak_object *object; 684 685 object = find_and_get_object(ptr, 0); 686 if (!object) { 687 kmemleak_warn("Trying to color unknown object " 688 "at 0x%08lx as %s\n", ptr, 689 (color == KMEMLEAK_GREY) ? "Grey" : 690 (color == KMEMLEAK_BLACK) ? "Black" : "Unknown"); 691 return; 692 } 693 paint_it(object, color); 694 put_object(object); 695 } 696 697 /* 698 * Make a object permanently as gray-colored so that it can no longer be 699 * reported as a leak. This is used in general to mark a false positive. 700 */ 701 static void make_gray_object(unsigned long ptr) 702 { 703 paint_ptr(ptr, KMEMLEAK_GREY); 704 } 705 706 /* 707 * Mark the object as black-colored so that it is ignored from scans and 708 * reporting. 709 */ 710 static void make_black_object(unsigned long ptr) 711 { 712 paint_ptr(ptr, KMEMLEAK_BLACK); 713 } 714 715 /* 716 * Add a scanning area to the object. If at least one such area is added, 717 * kmemleak will only scan these ranges rather than the whole memory block. 718 */ 719 static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp) 720 { 721 unsigned long flags; 722 struct kmemleak_object *object; 723 struct kmemleak_scan_area *area; 724 725 object = find_and_get_object(ptr, 1); 726 if (!object) { 727 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n", 728 ptr); 729 return; 730 } 731 732 area = kmem_cache_alloc(scan_area_cache, gfp & GFP_KMEMLEAK_MASK); 733 if (!area) { 734 kmemleak_warn("Cannot allocate a scan area\n"); 735 goto out; 736 } 737 738 spin_lock_irqsave(&object->lock, flags); 739 if (ptr + size > object->pointer + object->size) { 740 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr); 741 dump_object_info(object); 742 kmem_cache_free(scan_area_cache, area); 743 goto out_unlock; 744 } 745 746 INIT_HLIST_NODE(&area->node); 747 area->start = ptr; 748 area->size = size; 749 750 hlist_add_head(&area->node, &object->area_list); 751 out_unlock: 752 spin_unlock_irqrestore(&object->lock, flags); 753 out: 754 put_object(object); 755 } 756 757 /* 758 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give 759 * pointer. Such object will not be scanned by kmemleak but references to it 760 * are searched. 761 */ 762 static void object_no_scan(unsigned long ptr) 763 { 764 unsigned long flags; 765 struct kmemleak_object *object; 766 767 object = find_and_get_object(ptr, 0); 768 if (!object) { 769 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr); 770 return; 771 } 772 773 spin_lock_irqsave(&object->lock, flags); 774 object->flags |= OBJECT_NO_SCAN; 775 spin_unlock_irqrestore(&object->lock, flags); 776 put_object(object); 777 } 778 779 /* 780 * Log an early kmemleak_* call to the early_log buffer. These calls will be 781 * processed later once kmemleak is fully initialized. 782 */ 783 static void __init log_early(int op_type, const void *ptr, size_t size, 784 int min_count) 785 { 786 unsigned long flags; 787 struct early_log *log; 788 789 if (crt_early_log >= ARRAY_SIZE(early_log)) { 790 pr_warning("Early log buffer exceeded, " 791 "please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n"); 792 kmemleak_disable(); 793 return; 794 } 795 796 /* 797 * There is no need for locking since the kernel is still in UP mode 798 * at this stage. Disabling the IRQs is enough. 799 */ 800 local_irq_save(flags); 801 log = &early_log[crt_early_log]; 802 log->op_type = op_type; 803 log->ptr = ptr; 804 log->size = size; 805 log->min_count = min_count; 806 if (op_type == KMEMLEAK_ALLOC) 807 log->trace_len = __save_stack_trace(log->trace); 808 crt_early_log++; 809 local_irq_restore(flags); 810 } 811 812 /* 813 * Log an early allocated block and populate the stack trace. 814 */ 815 static void early_alloc(struct early_log *log) 816 { 817 struct kmemleak_object *object; 818 unsigned long flags; 819 int i; 820 821 if (!atomic_read(&kmemleak_enabled) || !log->ptr || IS_ERR(log->ptr)) 822 return; 823 824 /* 825 * RCU locking needed to ensure object is not freed via put_object(). 826 */ 827 rcu_read_lock(); 828 object = create_object((unsigned long)log->ptr, log->size, 829 log->min_count, GFP_ATOMIC); 830 if (!object) 831 goto out; 832 spin_lock_irqsave(&object->lock, flags); 833 for (i = 0; i < log->trace_len; i++) 834 object->trace[i] = log->trace[i]; 835 object->trace_len = log->trace_len; 836 spin_unlock_irqrestore(&object->lock, flags); 837 out: 838 rcu_read_unlock(); 839 } 840 841 /* 842 * Memory allocation function callback. This function is called from the 843 * kernel allocators when a new block is allocated (kmem_cache_alloc, kmalloc, 844 * vmalloc etc.). 845 */ 846 void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count, 847 gfp_t gfp) 848 { 849 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count); 850 851 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) 852 create_object((unsigned long)ptr, size, min_count, gfp); 853 else if (atomic_read(&kmemleak_early_log)) 854 log_early(KMEMLEAK_ALLOC, ptr, size, min_count); 855 } 856 EXPORT_SYMBOL_GPL(kmemleak_alloc); 857 858 /* 859 * Memory freeing function callback. This function is called from the kernel 860 * allocators when a block is freed (kmem_cache_free, kfree, vfree etc.). 861 */ 862 void __ref kmemleak_free(const void *ptr) 863 { 864 pr_debug("%s(0x%p)\n", __func__, ptr); 865 866 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) 867 delete_object_full((unsigned long)ptr); 868 else if (atomic_read(&kmemleak_early_log)) 869 log_early(KMEMLEAK_FREE, ptr, 0, 0); 870 } 871 EXPORT_SYMBOL_GPL(kmemleak_free); 872 873 /* 874 * Partial memory freeing function callback. This function is usually called 875 * from bootmem allocator when (part of) a memory block is freed. 876 */ 877 void __ref kmemleak_free_part(const void *ptr, size_t size) 878 { 879 pr_debug("%s(0x%p)\n", __func__, ptr); 880 881 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) 882 delete_object_part((unsigned long)ptr, size); 883 else if (atomic_read(&kmemleak_early_log)) 884 log_early(KMEMLEAK_FREE_PART, ptr, size, 0); 885 } 886 EXPORT_SYMBOL_GPL(kmemleak_free_part); 887 888 /* 889 * Mark an already allocated memory block as a false positive. This will cause 890 * the block to no longer be reported as leak and always be scanned. 891 */ 892 void __ref kmemleak_not_leak(const void *ptr) 893 { 894 pr_debug("%s(0x%p)\n", __func__, ptr); 895 896 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) 897 make_gray_object((unsigned long)ptr); 898 else if (atomic_read(&kmemleak_early_log)) 899 log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0); 900 } 901 EXPORT_SYMBOL(kmemleak_not_leak); 902 903 /* 904 * Ignore a memory block. This is usually done when it is known that the 905 * corresponding block is not a leak and does not contain any references to 906 * other allocated memory blocks. 907 */ 908 void __ref kmemleak_ignore(const void *ptr) 909 { 910 pr_debug("%s(0x%p)\n", __func__, ptr); 911 912 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) 913 make_black_object((unsigned long)ptr); 914 else if (atomic_read(&kmemleak_early_log)) 915 log_early(KMEMLEAK_IGNORE, ptr, 0, 0); 916 } 917 EXPORT_SYMBOL(kmemleak_ignore); 918 919 /* 920 * Limit the range to be scanned in an allocated memory block. 921 */ 922 void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp) 923 { 924 pr_debug("%s(0x%p)\n", __func__, ptr); 925 926 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) 927 add_scan_area((unsigned long)ptr, size, gfp); 928 else if (atomic_read(&kmemleak_early_log)) 929 log_early(KMEMLEAK_SCAN_AREA, ptr, size, 0); 930 } 931 EXPORT_SYMBOL(kmemleak_scan_area); 932 933 /* 934 * Inform kmemleak not to scan the given memory block. 935 */ 936 void __ref kmemleak_no_scan(const void *ptr) 937 { 938 pr_debug("%s(0x%p)\n", __func__, ptr); 939 940 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) 941 object_no_scan((unsigned long)ptr); 942 else if (atomic_read(&kmemleak_early_log)) 943 log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0); 944 } 945 EXPORT_SYMBOL(kmemleak_no_scan); 946 947 /* 948 * Update an object's checksum and return true if it was modified. 949 */ 950 static bool update_checksum(struct kmemleak_object *object) 951 { 952 u32 old_csum = object->checksum; 953 954 if (!kmemcheck_is_obj_initialized(object->pointer, object->size)) 955 return false; 956 957 object->checksum = crc32(0, (void *)object->pointer, object->size); 958 return object->checksum != old_csum; 959 } 960 961 /* 962 * Memory scanning is a long process and it needs to be interruptable. This 963 * function checks whether such interrupt condition occured. 964 */ 965 static int scan_should_stop(void) 966 { 967 if (!atomic_read(&kmemleak_enabled)) 968 return 1; 969 970 /* 971 * This function may be called from either process or kthread context, 972 * hence the need to check for both stop conditions. 973 */ 974 if (current->mm) 975 return signal_pending(current); 976 else 977 return kthread_should_stop(); 978 979 return 0; 980 } 981 982 /* 983 * Scan a memory block (exclusive range) for valid pointers and add those 984 * found to the gray list. 985 */ 986 static void scan_block(void *_start, void *_end, 987 struct kmemleak_object *scanned, int allow_resched) 988 { 989 unsigned long *ptr; 990 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER); 991 unsigned long *end = _end - (BYTES_PER_POINTER - 1); 992 993 for (ptr = start; ptr < end; ptr++) { 994 struct kmemleak_object *object; 995 unsigned long flags; 996 unsigned long pointer; 997 998 if (allow_resched) 999 cond_resched(); 1000 if (scan_should_stop()) 1001 break; 1002 1003 /* don't scan uninitialized memory */ 1004 if (!kmemcheck_is_obj_initialized((unsigned long)ptr, 1005 BYTES_PER_POINTER)) 1006 continue; 1007 1008 pointer = *ptr; 1009 1010 object = find_and_get_object(pointer, 1); 1011 if (!object) 1012 continue; 1013 if (object == scanned) { 1014 /* self referenced, ignore */ 1015 put_object(object); 1016 continue; 1017 } 1018 1019 /* 1020 * Avoid the lockdep recursive warning on object->lock being 1021 * previously acquired in scan_object(). These locks are 1022 * enclosed by scan_mutex. 1023 */ 1024 spin_lock_irqsave_nested(&object->lock, flags, 1025 SINGLE_DEPTH_NESTING); 1026 if (!color_white(object)) { 1027 /* non-orphan, ignored or new */ 1028 spin_unlock_irqrestore(&object->lock, flags); 1029 put_object(object); 1030 continue; 1031 } 1032 1033 /* 1034 * Increase the object's reference count (number of pointers 1035 * to the memory block). If this count reaches the required 1036 * minimum, the object's color will become gray and it will be 1037 * added to the gray_list. 1038 */ 1039 object->count++; 1040 if (color_gray(object)) { 1041 list_add_tail(&object->gray_list, &gray_list); 1042 spin_unlock_irqrestore(&object->lock, flags); 1043 continue; 1044 } 1045 1046 spin_unlock_irqrestore(&object->lock, flags); 1047 put_object(object); 1048 } 1049 } 1050 1051 /* 1052 * Scan a memory block corresponding to a kmemleak_object. A condition is 1053 * that object->use_count >= 1. 1054 */ 1055 static void scan_object(struct kmemleak_object *object) 1056 { 1057 struct kmemleak_scan_area *area; 1058 struct hlist_node *elem; 1059 unsigned long flags; 1060 1061 /* 1062 * Once the object->lock is acquired, the corresponding memory block 1063 * cannot be freed (the same lock is acquired in delete_object). 1064 */ 1065 spin_lock_irqsave(&object->lock, flags); 1066 if (object->flags & OBJECT_NO_SCAN) 1067 goto out; 1068 if (!(object->flags & OBJECT_ALLOCATED)) 1069 /* already freed object */ 1070 goto out; 1071 if (hlist_empty(&object->area_list)) { 1072 void *start = (void *)object->pointer; 1073 void *end = (void *)(object->pointer + object->size); 1074 1075 while (start < end && (object->flags & OBJECT_ALLOCATED) && 1076 !(object->flags & OBJECT_NO_SCAN)) { 1077 scan_block(start, min(start + MAX_SCAN_SIZE, end), 1078 object, 0); 1079 start += MAX_SCAN_SIZE; 1080 1081 spin_unlock_irqrestore(&object->lock, flags); 1082 cond_resched(); 1083 spin_lock_irqsave(&object->lock, flags); 1084 } 1085 } else 1086 hlist_for_each_entry(area, elem, &object->area_list, node) 1087 scan_block((void *)area->start, 1088 (void *)(area->start + area->size), 1089 object, 0); 1090 out: 1091 spin_unlock_irqrestore(&object->lock, flags); 1092 } 1093 1094 /* 1095 * Scan the objects already referenced (gray objects). More objects will be 1096 * referenced and, if there are no memory leaks, all the objects are scanned. 1097 */ 1098 static void scan_gray_list(void) 1099 { 1100 struct kmemleak_object *object, *tmp; 1101 1102 /* 1103 * The list traversal is safe for both tail additions and removals 1104 * from inside the loop. The kmemleak objects cannot be freed from 1105 * outside the loop because their use_count was incremented. 1106 */ 1107 object = list_entry(gray_list.next, typeof(*object), gray_list); 1108 while (&object->gray_list != &gray_list) { 1109 cond_resched(); 1110 1111 /* may add new objects to the list */ 1112 if (!scan_should_stop()) 1113 scan_object(object); 1114 1115 tmp = list_entry(object->gray_list.next, typeof(*object), 1116 gray_list); 1117 1118 /* remove the object from the list and release it */ 1119 list_del(&object->gray_list); 1120 put_object(object); 1121 1122 object = tmp; 1123 } 1124 WARN_ON(!list_empty(&gray_list)); 1125 } 1126 1127 /* 1128 * Scan data sections and all the referenced memory blocks allocated via the 1129 * kernel's standard allocators. This function must be called with the 1130 * scan_mutex held. 1131 */ 1132 static void kmemleak_scan(void) 1133 { 1134 unsigned long flags; 1135 struct kmemleak_object *object; 1136 int i; 1137 int new_leaks = 0; 1138 1139 jiffies_last_scan = jiffies; 1140 1141 /* prepare the kmemleak_object's */ 1142 rcu_read_lock(); 1143 list_for_each_entry_rcu(object, &object_list, object_list) { 1144 spin_lock_irqsave(&object->lock, flags); 1145 #ifdef DEBUG 1146 /* 1147 * With a few exceptions there should be a maximum of 1148 * 1 reference to any object at this point. 1149 */ 1150 if (atomic_read(&object->use_count) > 1) { 1151 pr_debug("object->use_count = %d\n", 1152 atomic_read(&object->use_count)); 1153 dump_object_info(object); 1154 } 1155 #endif 1156 /* reset the reference count (whiten the object) */ 1157 object->count = 0; 1158 if (color_gray(object) && get_object(object)) 1159 list_add_tail(&object->gray_list, &gray_list); 1160 1161 spin_unlock_irqrestore(&object->lock, flags); 1162 } 1163 rcu_read_unlock(); 1164 1165 /* data/bss scanning */ 1166 scan_block(_sdata, _edata, NULL, 1); 1167 scan_block(__bss_start, __bss_stop, NULL, 1); 1168 1169 #ifdef CONFIG_SMP 1170 /* per-cpu sections scanning */ 1171 for_each_possible_cpu(i) 1172 scan_block(__per_cpu_start + per_cpu_offset(i), 1173 __per_cpu_end + per_cpu_offset(i), NULL, 1); 1174 #endif 1175 1176 /* 1177 * Struct page scanning for each node. The code below is not yet safe 1178 * with MEMORY_HOTPLUG. 1179 */ 1180 for_each_online_node(i) { 1181 pg_data_t *pgdat = NODE_DATA(i); 1182 unsigned long start_pfn = pgdat->node_start_pfn; 1183 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; 1184 unsigned long pfn; 1185 1186 for (pfn = start_pfn; pfn < end_pfn; pfn++) { 1187 struct page *page; 1188 1189 if (!pfn_valid(pfn)) 1190 continue; 1191 page = pfn_to_page(pfn); 1192 /* only scan if page is in use */ 1193 if (page_count(page) == 0) 1194 continue; 1195 scan_block(page, page + 1, NULL, 1); 1196 } 1197 } 1198 1199 /* 1200 * Scanning the task stacks (may introduce false negatives). 1201 */ 1202 if (kmemleak_stack_scan) { 1203 struct task_struct *p, *g; 1204 1205 read_lock(&tasklist_lock); 1206 do_each_thread(g, p) { 1207 scan_block(task_stack_page(p), task_stack_page(p) + 1208 THREAD_SIZE, NULL, 0); 1209 } while_each_thread(g, p); 1210 read_unlock(&tasklist_lock); 1211 } 1212 1213 /* 1214 * Scan the objects already referenced from the sections scanned 1215 * above. 1216 */ 1217 scan_gray_list(); 1218 1219 /* 1220 * Check for new or unreferenced objects modified since the previous 1221 * scan and color them gray until the next scan. 1222 */ 1223 rcu_read_lock(); 1224 list_for_each_entry_rcu(object, &object_list, object_list) { 1225 spin_lock_irqsave(&object->lock, flags); 1226 if (color_white(object) && (object->flags & OBJECT_ALLOCATED) 1227 && update_checksum(object) && get_object(object)) { 1228 /* color it gray temporarily */ 1229 object->count = object->min_count; 1230 list_add_tail(&object->gray_list, &gray_list); 1231 } 1232 spin_unlock_irqrestore(&object->lock, flags); 1233 } 1234 rcu_read_unlock(); 1235 1236 /* 1237 * Re-scan the gray list for modified unreferenced objects. 1238 */ 1239 scan_gray_list(); 1240 1241 /* 1242 * If scanning was stopped do not report any new unreferenced objects. 1243 */ 1244 if (scan_should_stop()) 1245 return; 1246 1247 /* 1248 * Scanning result reporting. 1249 */ 1250 rcu_read_lock(); 1251 list_for_each_entry_rcu(object, &object_list, object_list) { 1252 spin_lock_irqsave(&object->lock, flags); 1253 if (unreferenced_object(object) && 1254 !(object->flags & OBJECT_REPORTED)) { 1255 object->flags |= OBJECT_REPORTED; 1256 new_leaks++; 1257 } 1258 spin_unlock_irqrestore(&object->lock, flags); 1259 } 1260 rcu_read_unlock(); 1261 1262 if (new_leaks) 1263 pr_info("%d new suspected memory leaks (see " 1264 "/sys/kernel/debug/kmemleak)\n", new_leaks); 1265 1266 } 1267 1268 /* 1269 * Thread function performing automatic memory scanning. Unreferenced objects 1270 * at the end of a memory scan are reported but only the first time. 1271 */ 1272 static int kmemleak_scan_thread(void *arg) 1273 { 1274 static int first_run = 1; 1275 1276 pr_info("Automatic memory scanning thread started\n"); 1277 set_user_nice(current, 10); 1278 1279 /* 1280 * Wait before the first scan to allow the system to fully initialize. 1281 */ 1282 if (first_run) { 1283 first_run = 0; 1284 ssleep(SECS_FIRST_SCAN); 1285 } 1286 1287 while (!kthread_should_stop()) { 1288 signed long timeout = jiffies_scan_wait; 1289 1290 mutex_lock(&scan_mutex); 1291 kmemleak_scan(); 1292 mutex_unlock(&scan_mutex); 1293 1294 /* wait before the next scan */ 1295 while (timeout && !kthread_should_stop()) 1296 timeout = schedule_timeout_interruptible(timeout); 1297 } 1298 1299 pr_info("Automatic memory scanning thread ended\n"); 1300 1301 return 0; 1302 } 1303 1304 /* 1305 * Start the automatic memory scanning thread. This function must be called 1306 * with the scan_mutex held. 1307 */ 1308 static void start_scan_thread(void) 1309 { 1310 if (scan_thread) 1311 return; 1312 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak"); 1313 if (IS_ERR(scan_thread)) { 1314 pr_warning("Failed to create the scan thread\n"); 1315 scan_thread = NULL; 1316 } 1317 } 1318 1319 /* 1320 * Stop the automatic memory scanning thread. This function must be called 1321 * with the scan_mutex held. 1322 */ 1323 static void stop_scan_thread(void) 1324 { 1325 if (scan_thread) { 1326 kthread_stop(scan_thread); 1327 scan_thread = NULL; 1328 } 1329 } 1330 1331 /* 1332 * Iterate over the object_list and return the first valid object at or after 1333 * the required position with its use_count incremented. The function triggers 1334 * a memory scanning when the pos argument points to the first position. 1335 */ 1336 static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos) 1337 { 1338 struct kmemleak_object *object; 1339 loff_t n = *pos; 1340 int err; 1341 1342 err = mutex_lock_interruptible(&scan_mutex); 1343 if (err < 0) 1344 return ERR_PTR(err); 1345 1346 rcu_read_lock(); 1347 list_for_each_entry_rcu(object, &object_list, object_list) { 1348 if (n-- > 0) 1349 continue; 1350 if (get_object(object)) 1351 goto out; 1352 } 1353 object = NULL; 1354 out: 1355 return object; 1356 } 1357 1358 /* 1359 * Return the next object in the object_list. The function decrements the 1360 * use_count of the previous object and increases that of the next one. 1361 */ 1362 static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1363 { 1364 struct kmemleak_object *prev_obj = v; 1365 struct kmemleak_object *next_obj = NULL; 1366 struct list_head *n = &prev_obj->object_list; 1367 1368 ++(*pos); 1369 1370 list_for_each_continue_rcu(n, &object_list) { 1371 next_obj = list_entry(n, struct kmemleak_object, object_list); 1372 if (get_object(next_obj)) 1373 break; 1374 } 1375 1376 put_object(prev_obj); 1377 return next_obj; 1378 } 1379 1380 /* 1381 * Decrement the use_count of the last object required, if any. 1382 */ 1383 static void kmemleak_seq_stop(struct seq_file *seq, void *v) 1384 { 1385 if (!IS_ERR(v)) { 1386 /* 1387 * kmemleak_seq_start may return ERR_PTR if the scan_mutex 1388 * waiting was interrupted, so only release it if !IS_ERR. 1389 */ 1390 rcu_read_unlock(); 1391 mutex_unlock(&scan_mutex); 1392 if (v) 1393 put_object(v); 1394 } 1395 } 1396 1397 /* 1398 * Print the information for an unreferenced object to the seq file. 1399 */ 1400 static int kmemleak_seq_show(struct seq_file *seq, void *v) 1401 { 1402 struct kmemleak_object *object = v; 1403 unsigned long flags; 1404 1405 spin_lock_irqsave(&object->lock, flags); 1406 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object)) 1407 print_unreferenced(seq, object); 1408 spin_unlock_irqrestore(&object->lock, flags); 1409 return 0; 1410 } 1411 1412 static const struct seq_operations kmemleak_seq_ops = { 1413 .start = kmemleak_seq_start, 1414 .next = kmemleak_seq_next, 1415 .stop = kmemleak_seq_stop, 1416 .show = kmemleak_seq_show, 1417 }; 1418 1419 static int kmemleak_open(struct inode *inode, struct file *file) 1420 { 1421 if (!atomic_read(&kmemleak_enabled)) 1422 return -EBUSY; 1423 1424 return seq_open(file, &kmemleak_seq_ops); 1425 } 1426 1427 static int kmemleak_release(struct inode *inode, struct file *file) 1428 { 1429 return seq_release(inode, file); 1430 } 1431 1432 static int dump_str_object_info(const char *str) 1433 { 1434 unsigned long flags; 1435 struct kmemleak_object *object; 1436 unsigned long addr; 1437 1438 addr= simple_strtoul(str, NULL, 0); 1439 object = find_and_get_object(addr, 0); 1440 if (!object) { 1441 pr_info("Unknown object at 0x%08lx\n", addr); 1442 return -EINVAL; 1443 } 1444 1445 spin_lock_irqsave(&object->lock, flags); 1446 dump_object_info(object); 1447 spin_unlock_irqrestore(&object->lock, flags); 1448 1449 put_object(object); 1450 return 0; 1451 } 1452 1453 /* 1454 * We use grey instead of black to ensure we can do future scans on the same 1455 * objects. If we did not do future scans these black objects could 1456 * potentially contain references to newly allocated objects in the future and 1457 * we'd end up with false positives. 1458 */ 1459 static void kmemleak_clear(void) 1460 { 1461 struct kmemleak_object *object; 1462 unsigned long flags; 1463 1464 rcu_read_lock(); 1465 list_for_each_entry_rcu(object, &object_list, object_list) { 1466 spin_lock_irqsave(&object->lock, flags); 1467 if ((object->flags & OBJECT_REPORTED) && 1468 unreferenced_object(object)) 1469 __paint_it(object, KMEMLEAK_GREY); 1470 spin_unlock_irqrestore(&object->lock, flags); 1471 } 1472 rcu_read_unlock(); 1473 } 1474 1475 /* 1476 * File write operation to configure kmemleak at run-time. The following 1477 * commands can be written to the /sys/kernel/debug/kmemleak file: 1478 * off - disable kmemleak (irreversible) 1479 * stack=on - enable the task stacks scanning 1480 * stack=off - disable the tasks stacks scanning 1481 * scan=on - start the automatic memory scanning thread 1482 * scan=off - stop the automatic memory scanning thread 1483 * scan=... - set the automatic memory scanning period in seconds (0 to 1484 * disable it) 1485 * scan - trigger a memory scan 1486 * clear - mark all current reported unreferenced kmemleak objects as 1487 * grey to ignore printing them 1488 * dump=... - dump information about the object found at the given address 1489 */ 1490 static ssize_t kmemleak_write(struct file *file, const char __user *user_buf, 1491 size_t size, loff_t *ppos) 1492 { 1493 char buf[64]; 1494 int buf_size; 1495 int ret; 1496 1497 buf_size = min(size, (sizeof(buf) - 1)); 1498 if (strncpy_from_user(buf, user_buf, buf_size) < 0) 1499 return -EFAULT; 1500 buf[buf_size] = 0; 1501 1502 ret = mutex_lock_interruptible(&scan_mutex); 1503 if (ret < 0) 1504 return ret; 1505 1506 if (strncmp(buf, "off", 3) == 0) 1507 kmemleak_disable(); 1508 else if (strncmp(buf, "stack=on", 8) == 0) 1509 kmemleak_stack_scan = 1; 1510 else if (strncmp(buf, "stack=off", 9) == 0) 1511 kmemleak_stack_scan = 0; 1512 else if (strncmp(buf, "scan=on", 7) == 0) 1513 start_scan_thread(); 1514 else if (strncmp(buf, "scan=off", 8) == 0) 1515 stop_scan_thread(); 1516 else if (strncmp(buf, "scan=", 5) == 0) { 1517 unsigned long secs; 1518 1519 ret = strict_strtoul(buf + 5, 0, &secs); 1520 if (ret < 0) 1521 goto out; 1522 stop_scan_thread(); 1523 if (secs) { 1524 jiffies_scan_wait = msecs_to_jiffies(secs * 1000); 1525 start_scan_thread(); 1526 } 1527 } else if (strncmp(buf, "scan", 4) == 0) 1528 kmemleak_scan(); 1529 else if (strncmp(buf, "clear", 5) == 0) 1530 kmemleak_clear(); 1531 else if (strncmp(buf, "dump=", 5) == 0) 1532 ret = dump_str_object_info(buf + 5); 1533 else 1534 ret = -EINVAL; 1535 1536 out: 1537 mutex_unlock(&scan_mutex); 1538 if (ret < 0) 1539 return ret; 1540 1541 /* ignore the rest of the buffer, only one command at a time */ 1542 *ppos += size; 1543 return size; 1544 } 1545 1546 static const struct file_operations kmemleak_fops = { 1547 .owner = THIS_MODULE, 1548 .open = kmemleak_open, 1549 .read = seq_read, 1550 .write = kmemleak_write, 1551 .llseek = seq_lseek, 1552 .release = kmemleak_release, 1553 }; 1554 1555 /* 1556 * Perform the freeing of the kmemleak internal objects after waiting for any 1557 * current memory scan to complete. 1558 */ 1559 static void kmemleak_do_cleanup(struct work_struct *work) 1560 { 1561 struct kmemleak_object *object; 1562 1563 mutex_lock(&scan_mutex); 1564 stop_scan_thread(); 1565 1566 rcu_read_lock(); 1567 list_for_each_entry_rcu(object, &object_list, object_list) 1568 delete_object_full(object->pointer); 1569 rcu_read_unlock(); 1570 mutex_unlock(&scan_mutex); 1571 } 1572 1573 static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup); 1574 1575 /* 1576 * Disable kmemleak. No memory allocation/freeing will be traced once this 1577 * function is called. Disabling kmemleak is an irreversible operation. 1578 */ 1579 static void kmemleak_disable(void) 1580 { 1581 /* atomically check whether it was already invoked */ 1582 if (atomic_cmpxchg(&kmemleak_error, 0, 1)) 1583 return; 1584 1585 /* stop any memory operation tracing */ 1586 atomic_set(&kmemleak_early_log, 0); 1587 atomic_set(&kmemleak_enabled, 0); 1588 1589 /* check whether it is too early for a kernel thread */ 1590 if (atomic_read(&kmemleak_initialized)) 1591 schedule_work(&cleanup_work); 1592 1593 pr_info("Kernel memory leak detector disabled\n"); 1594 } 1595 1596 /* 1597 * Allow boot-time kmemleak disabling (enabled by default). 1598 */ 1599 static int kmemleak_boot_config(char *str) 1600 { 1601 if (!str) 1602 return -EINVAL; 1603 if (strcmp(str, "off") == 0) 1604 kmemleak_disable(); 1605 else if (strcmp(str, "on") != 0) 1606 return -EINVAL; 1607 return 0; 1608 } 1609 early_param("kmemleak", kmemleak_boot_config); 1610 1611 /* 1612 * Kmemleak initialization. 1613 */ 1614 void __init kmemleak_init(void) 1615 { 1616 int i; 1617 unsigned long flags; 1618 1619 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE); 1620 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000); 1621 1622 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE); 1623 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE); 1624 INIT_PRIO_TREE_ROOT(&object_tree_root); 1625 1626 /* the kernel is still in UP mode, so disabling the IRQs is enough */ 1627 local_irq_save(flags); 1628 if (!atomic_read(&kmemleak_error)) { 1629 atomic_set(&kmemleak_enabled, 1); 1630 atomic_set(&kmemleak_early_log, 0); 1631 } 1632 local_irq_restore(flags); 1633 1634 /* 1635 * This is the point where tracking allocations is safe. Automatic 1636 * scanning is started during the late initcall. Add the early logged 1637 * callbacks to the kmemleak infrastructure. 1638 */ 1639 for (i = 0; i < crt_early_log; i++) { 1640 struct early_log *log = &early_log[i]; 1641 1642 switch (log->op_type) { 1643 case KMEMLEAK_ALLOC: 1644 early_alloc(log); 1645 break; 1646 case KMEMLEAK_FREE: 1647 kmemleak_free(log->ptr); 1648 break; 1649 case KMEMLEAK_FREE_PART: 1650 kmemleak_free_part(log->ptr, log->size); 1651 break; 1652 case KMEMLEAK_NOT_LEAK: 1653 kmemleak_not_leak(log->ptr); 1654 break; 1655 case KMEMLEAK_IGNORE: 1656 kmemleak_ignore(log->ptr); 1657 break; 1658 case KMEMLEAK_SCAN_AREA: 1659 kmemleak_scan_area(log->ptr, log->size, GFP_KERNEL); 1660 break; 1661 case KMEMLEAK_NO_SCAN: 1662 kmemleak_no_scan(log->ptr); 1663 break; 1664 default: 1665 WARN_ON(1); 1666 } 1667 } 1668 } 1669 1670 /* 1671 * Late initialization function. 1672 */ 1673 static int __init kmemleak_late_init(void) 1674 { 1675 struct dentry *dentry; 1676 1677 atomic_set(&kmemleak_initialized, 1); 1678 1679 if (atomic_read(&kmemleak_error)) { 1680 /* 1681 * Some error occured and kmemleak was disabled. There is a 1682 * small chance that kmemleak_disable() was called immediately 1683 * after setting kmemleak_initialized and we may end up with 1684 * two clean-up threads but serialized by scan_mutex. 1685 */ 1686 schedule_work(&cleanup_work); 1687 return -ENOMEM; 1688 } 1689 1690 dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL, 1691 &kmemleak_fops); 1692 if (!dentry) 1693 pr_warning("Failed to create the debugfs kmemleak file\n"); 1694 mutex_lock(&scan_mutex); 1695 start_scan_thread(); 1696 mutex_unlock(&scan_mutex); 1697 1698 pr_info("Kernel memory leak detector initialized\n"); 1699 1700 return 0; 1701 } 1702 late_initcall(kmemleak_late_init); 1703