4 * Copyright (C) 2008 ARM Limited
5 * Written by Catalin Marinas <catalin.marinas@arm.com>
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.
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.
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
21 * For more information on the algorithm and kmemleak usage, please see
22 * Documentation/kmemleak.txt.
27 * The following locks and mutexes are used by kmemleak:
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
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
64 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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/gfp.h>
77 #include <linux/debugfs.h>
78 #include <linux/seq_file.h>
79 #include <linux/cpumask.h>
80 #include <linux/spinlock.h>
81 #include <linux/mutex.h>
82 #include <linux/rcupdate.h>
83 #include <linux/stacktrace.h>
84 #include <linux/cache.h>
85 #include <linux/percpu.h>
86 #include <linux/hardirq.h>
87 #include <linux/mmzone.h>
88 #include <linux/slab.h>
89 #include <linux/thread_info.h>
90 #include <linux/err.h>
91 #include <linux/uaccess.h>
92 #include <linux/string.h>
93 #include <linux/nodemask.h>
96 #include <asm/sections.h>
97 #include <asm/processor.h>
98 #include <asm/atomic.h>
100 #include <linux/kmemleak.h>
103 * Kmemleak configuration and common defines.
105 #define MAX_TRACE 16 /* stack trace length */
106 #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
107 #define SECS_FIRST_SCAN 60 /* delay before the first scan */
108 #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
109 #define GRAY_LIST_PASSES 25 /* maximum number of gray list scans */
111 #define BYTES_PER_POINTER sizeof(void *)
113 /* GFP bitmask for kmemleak internal allocations */
114 #define GFP_KMEMLEAK_MASK (GFP_KERNEL | GFP_ATOMIC)
116 /* scanning area inside a memory block */
117 struct kmemleak_scan_area {
118 struct hlist_node node;
119 unsigned long offset;
124 * Structure holding the metadata for each allocated memory block.
125 * Modifications to such objects should be made while holding the
126 * object->lock. Insertions or deletions from object_list, gray_list or
127 * tree_node are already protected by the corresponding locks or mutex (see
128 * the notes on locking above). These objects are reference-counted
129 * (use_count) and freed using the RCU mechanism.
131 struct kmemleak_object {
133 unsigned long flags; /* object status flags */
134 struct list_head object_list;
135 struct list_head gray_list;
136 struct prio_tree_node tree_node;
137 struct rcu_head rcu; /* object_list lockless traversal */
138 /* object usage count; object freed when use_count == 0 */
140 unsigned long pointer;
142 /* minimum number of a pointers found before it is considered leak */
144 /* the total number of pointers found pointing to this object */
146 /* memory ranges to be scanned inside an object (empty for all) */
147 struct hlist_head area_list;
148 unsigned long trace[MAX_TRACE];
149 unsigned int trace_len;
150 unsigned long jiffies; /* creation timestamp */
151 pid_t pid; /* pid of the current task */
152 char comm[TASK_COMM_LEN]; /* executable name */
155 /* flag representing the memory block allocation status */
156 #define OBJECT_ALLOCATED (1 << 0)
157 /* flag set after the first reporting of an unreference object */
158 #define OBJECT_REPORTED (1 << 1)
159 /* flag set to not scan the object */
160 #define OBJECT_NO_SCAN (1 << 2)
161 /* flag set on newly allocated objects */
162 #define OBJECT_NEW (1 << 3)
164 /* the list of all allocated objects */
165 static LIST_HEAD(object_list);
166 /* the list of gray-colored objects (see color_gray comment below) */
167 static LIST_HEAD(gray_list);
168 /* prio search tree for object boundaries */
169 static struct prio_tree_root object_tree_root;
170 /* rw_lock protecting the access to object_list and prio_tree_root */
171 static DEFINE_RWLOCK(kmemleak_lock);
173 /* allocation caches for kmemleak internal data */
174 static struct kmem_cache *object_cache;
175 static struct kmem_cache *scan_area_cache;
177 /* set if tracing memory operations is enabled */
178 static atomic_t kmemleak_enabled = ATOMIC_INIT(0);
179 /* set in the late_initcall if there were no errors */
180 static atomic_t kmemleak_initialized = ATOMIC_INIT(0);
181 /* enables or disables early logging of the memory operations */
182 static atomic_t kmemleak_early_log = ATOMIC_INIT(1);
183 /* set if a fata kmemleak error has occurred */
184 static atomic_t kmemleak_error = ATOMIC_INIT(0);
186 /* minimum and maximum address that may be valid pointers */
187 static unsigned long min_addr = ULONG_MAX;
188 static unsigned long max_addr;
190 static struct task_struct *scan_thread;
191 /* used to avoid reporting of recently allocated objects */
192 static unsigned long jiffies_min_age;
193 static unsigned long jiffies_last_scan;
194 /* delay between automatic memory scannings */
195 static signed long jiffies_scan_wait;
196 /* enables or disables the task stacks scanning */
197 static int kmemleak_stack_scan = 1;
198 /* protects the memory scanning, parameters and debug/kmemleak file access */
199 static DEFINE_MUTEX(scan_mutex);
202 * Early object allocation/freeing logging. Kmemleak is initialized after the
203 * kernel allocator. However, both the kernel allocator and kmemleak may
204 * allocate memory blocks which need to be tracked. Kmemleak defines an
205 * arbitrary buffer to hold the allocation/freeing information before it is
209 /* kmemleak operation type for early logging */
220 * Structure holding the information passed to kmemleak callbacks during the
224 int op_type; /* kmemleak operation type */
225 const void *ptr; /* allocated/freed memory block */
226 size_t size; /* memory block size */
227 int min_count; /* minimum reference count */
228 unsigned long offset; /* scan area offset */
229 size_t length; /* scan area length */
232 /* early logging buffer and current position */
233 static struct early_log early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE];
234 static int crt_early_log;
236 static void kmemleak_disable(void);
239 * Print a warning and dump the stack trace.
241 #define kmemleak_warn(x...) do { \
247 * Macro invoked when a serious kmemleak condition occured and cannot be
248 * recovered from. Kmemleak will be disabled and further allocation/freeing
249 * tracing no longer available.
251 #define kmemleak_stop(x...) do { \
253 kmemleak_disable(); \
257 * Object colors, encoded with count and min_count:
258 * - white - orphan object, not enough references to it (count < min_count)
259 * - gray - not orphan, not marked as false positive (min_count == 0) or
260 * sufficient references to it (count >= min_count)
261 * - black - ignore, it doesn't contain references (e.g. text section)
262 * (min_count == -1). No function defined for this color.
263 * Newly created objects don't have any color assigned (object->count == -1)
264 * before the next memory scan when they become white.
266 static int color_white(const struct kmemleak_object *object)
268 return object->count != -1 && object->count < object->min_count;
271 static int color_gray(const struct kmemleak_object *object)
273 return object->min_count != -1 && object->count >= object->min_count;
276 static int color_black(const struct kmemleak_object *object)
278 return object->min_count == -1;
282 * Objects are considered unreferenced only if their color is white, they have
283 * not be deleted and have a minimum age to avoid false positives caused by
284 * pointers temporarily stored in CPU registers.
286 static int unreferenced_object(struct kmemleak_object *object)
288 return (object->flags & OBJECT_ALLOCATED) && color_white(object) &&
289 time_before_eq(object->jiffies + jiffies_min_age,
294 * Printing of the unreferenced objects information to the seq file. The
295 * print_unreferenced function must be called with the object->lock held.
297 static void print_unreferenced(struct seq_file *seq,
298 struct kmemleak_object *object)
302 seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
303 object->pointer, object->size);
304 seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu\n",
305 object->comm, object->pid, object->jiffies);
306 seq_printf(seq, " backtrace:\n");
308 for (i = 0; i < object->trace_len; i++) {
309 void *ptr = (void *)object->trace[i];
310 seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
315 * Print the kmemleak_object information. This function is used mainly for
316 * debugging special cases when kmemleak operations. It must be called with
317 * the object->lock held.
319 static void dump_object_info(struct kmemleak_object *object)
321 struct stack_trace trace;
323 trace.nr_entries = object->trace_len;
324 trace.entries = object->trace;
326 pr_notice("Object 0x%08lx (size %zu):\n",
327 object->tree_node.start, object->size);
328 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
329 object->comm, object->pid, object->jiffies);
330 pr_notice(" min_count = %d\n", object->min_count);
331 pr_notice(" count = %d\n", object->count);
332 pr_notice(" backtrace:\n");
333 print_stack_trace(&trace, 4);
337 * Look-up a memory block metadata (kmemleak_object) in the priority search
338 * tree based on a pointer value. If alias is 0, only values pointing to the
339 * beginning of the memory block are allowed. The kmemleak_lock must be held
340 * when calling this function.
342 static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
344 struct prio_tree_node *node;
345 struct prio_tree_iter iter;
346 struct kmemleak_object *object;
348 prio_tree_iter_init(&iter, &object_tree_root, ptr, ptr);
349 node = prio_tree_next(&iter);
351 object = prio_tree_entry(node, struct kmemleak_object,
353 if (!alias && object->pointer != ptr) {
354 kmemleak_warn("Found object by alias");
364 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
365 * that once an object's use_count reached 0, the RCU freeing was already
366 * registered and the object should no longer be used. This function must be
367 * called under the protection of rcu_read_lock().
369 static int get_object(struct kmemleak_object *object)
371 return atomic_inc_not_zero(&object->use_count);
375 * RCU callback to free a kmemleak_object.
377 static void free_object_rcu(struct rcu_head *rcu)
379 struct hlist_node *elem, *tmp;
380 struct kmemleak_scan_area *area;
381 struct kmemleak_object *object =
382 container_of(rcu, struct kmemleak_object, rcu);
385 * Once use_count is 0 (guaranteed by put_object), there is no other
386 * code accessing this object, hence no need for locking.
388 hlist_for_each_entry_safe(area, elem, tmp, &object->area_list, node) {
390 kmem_cache_free(scan_area_cache, area);
392 kmem_cache_free(object_cache, object);
396 * Decrement the object use_count. Once the count is 0, free the object using
397 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
398 * delete_object() path, the delayed RCU freeing ensures that there is no
399 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
402 static void put_object(struct kmemleak_object *object)
404 if (!atomic_dec_and_test(&object->use_count))
407 /* should only get here after delete_object was called */
408 WARN_ON(object->flags & OBJECT_ALLOCATED);
410 call_rcu(&object->rcu, free_object_rcu);
414 * Look up an object in the prio search tree and increase its use_count.
416 static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
419 struct kmemleak_object *object = NULL;
422 read_lock_irqsave(&kmemleak_lock, flags);
423 if (ptr >= min_addr && ptr < max_addr)
424 object = lookup_object(ptr, alias);
425 read_unlock_irqrestore(&kmemleak_lock, flags);
427 /* check whether the object is still available */
428 if (object && !get_object(object))
436 * Create the metadata (struct kmemleak_object) corresponding to an allocated
437 * memory block and add it to the object_list and object_tree_root.
439 static void create_object(unsigned long ptr, size_t size, int min_count,
443 struct kmemleak_object *object;
444 struct prio_tree_node *node;
445 struct stack_trace trace;
447 object = kmem_cache_alloc(object_cache, gfp & GFP_KMEMLEAK_MASK);
449 kmemleak_stop("Cannot allocate a kmemleak_object structure\n");
453 INIT_LIST_HEAD(&object->object_list);
454 INIT_LIST_HEAD(&object->gray_list);
455 INIT_HLIST_HEAD(&object->area_list);
456 spin_lock_init(&object->lock);
457 atomic_set(&object->use_count, 1);
458 object->flags = OBJECT_ALLOCATED | OBJECT_NEW;
459 object->pointer = ptr;
461 object->min_count = min_count;
462 object->count = -1; /* no color initially */
463 object->jiffies = jiffies;
465 /* task information */
468 strncpy(object->comm, "hardirq", sizeof(object->comm));
469 } else if (in_softirq()) {
471 strncpy(object->comm, "softirq", sizeof(object->comm));
473 object->pid = current->pid;
475 * There is a small chance of a race with set_task_comm(),
476 * however using get_task_comm() here may cause locking
477 * dependency issues with current->alloc_lock. In the worst
478 * case, the command line is not correct.
480 strncpy(object->comm, current->comm, sizeof(object->comm));
483 /* kernel backtrace */
484 trace.max_entries = MAX_TRACE;
485 trace.nr_entries = 0;
486 trace.entries = object->trace;
488 save_stack_trace(&trace);
489 object->trace_len = trace.nr_entries;
491 INIT_PRIO_TREE_NODE(&object->tree_node);
492 object->tree_node.start = ptr;
493 object->tree_node.last = ptr + size - 1;
495 write_lock_irqsave(&kmemleak_lock, flags);
496 min_addr = min(min_addr, ptr);
497 max_addr = max(max_addr, ptr + size);
498 node = prio_tree_insert(&object_tree_root, &object->tree_node);
500 * The code calling the kernel does not yet have the pointer to the
501 * memory block to be able to free it. However, we still hold the
502 * kmemleak_lock here in case parts of the kernel started freeing
503 * random memory blocks.
505 if (node != &object->tree_node) {
508 kmemleak_stop("Cannot insert 0x%lx into the object search tree "
509 "(already existing)\n", ptr);
510 object = lookup_object(ptr, 1);
511 spin_lock_irqsave(&object->lock, flags);
512 dump_object_info(object);
513 spin_unlock_irqrestore(&object->lock, flags);
517 list_add_tail_rcu(&object->object_list, &object_list);
519 write_unlock_irqrestore(&kmemleak_lock, flags);
523 * Remove the metadata (struct kmemleak_object) for a memory block from the
524 * object_list and object_tree_root and decrement its use_count.
526 static void delete_object(unsigned long ptr)
529 struct kmemleak_object *object;
531 write_lock_irqsave(&kmemleak_lock, flags);
532 object = lookup_object(ptr, 0);
535 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
538 write_unlock_irqrestore(&kmemleak_lock, flags);
541 prio_tree_remove(&object_tree_root, &object->tree_node);
542 list_del_rcu(&object->object_list);
543 write_unlock_irqrestore(&kmemleak_lock, flags);
545 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
546 WARN_ON(atomic_read(&object->use_count) < 1);
549 * Locking here also ensures that the corresponding memory block
550 * cannot be freed when it is being scanned.
552 spin_lock_irqsave(&object->lock, flags);
553 object->flags &= ~OBJECT_ALLOCATED;
554 spin_unlock_irqrestore(&object->lock, flags);
559 * Make a object permanently as gray-colored so that it can no longer be
560 * reported as a leak. This is used in general to mark a false positive.
562 static void make_gray_object(unsigned long ptr)
565 struct kmemleak_object *object;
567 object = find_and_get_object(ptr, 0);
569 kmemleak_warn("Graying unknown object at 0x%08lx\n", ptr);
573 spin_lock_irqsave(&object->lock, flags);
574 object->min_count = 0;
575 spin_unlock_irqrestore(&object->lock, flags);
580 * Mark the object as black-colored so that it is ignored from scans and
583 static void make_black_object(unsigned long ptr)
586 struct kmemleak_object *object;
588 object = find_and_get_object(ptr, 0);
590 kmemleak_warn("Blacking unknown object at 0x%08lx\n", ptr);
594 spin_lock_irqsave(&object->lock, flags);
595 object->min_count = -1;
596 spin_unlock_irqrestore(&object->lock, flags);
601 * Add a scanning area to the object. If at least one such area is added,
602 * kmemleak will only scan these ranges rather than the whole memory block.
604 static void add_scan_area(unsigned long ptr, unsigned long offset,
605 size_t length, gfp_t gfp)
608 struct kmemleak_object *object;
609 struct kmemleak_scan_area *area;
611 object = find_and_get_object(ptr, 0);
613 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
618 area = kmem_cache_alloc(scan_area_cache, gfp & GFP_KMEMLEAK_MASK);
620 kmemleak_warn("Cannot allocate a scan area\n");
624 spin_lock_irqsave(&object->lock, flags);
625 if (offset + length > object->size) {
626 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
627 dump_object_info(object);
628 kmem_cache_free(scan_area_cache, area);
632 INIT_HLIST_NODE(&area->node);
633 area->offset = offset;
634 area->length = length;
636 hlist_add_head(&area->node, &object->area_list);
638 spin_unlock_irqrestore(&object->lock, flags);
644 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
645 * pointer. Such object will not be scanned by kmemleak but references to it
648 static void object_no_scan(unsigned long ptr)
651 struct kmemleak_object *object;
653 object = find_and_get_object(ptr, 0);
655 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
659 spin_lock_irqsave(&object->lock, flags);
660 object->flags |= OBJECT_NO_SCAN;
661 spin_unlock_irqrestore(&object->lock, flags);
666 * Log an early kmemleak_* call to the early_log buffer. These calls will be
667 * processed later once kmemleak is fully initialized.
669 static void log_early(int op_type, const void *ptr, size_t size,
670 int min_count, unsigned long offset, size_t length)
673 struct early_log *log;
675 if (crt_early_log >= ARRAY_SIZE(early_log)) {
676 pr_warning("Early log buffer exceeded\n");
682 * There is no need for locking since the kernel is still in UP mode
683 * at this stage. Disabling the IRQs is enough.
685 local_irq_save(flags);
686 log = &early_log[crt_early_log];
687 log->op_type = op_type;
690 log->min_count = min_count;
691 log->offset = offset;
692 log->length = length;
694 local_irq_restore(flags);
698 * Memory allocation function callback. This function is called from the
699 * kernel allocators when a new block is allocated (kmem_cache_alloc, kmalloc,
702 void kmemleak_alloc(const void *ptr, size_t size, int min_count, gfp_t gfp)
704 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
706 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
707 create_object((unsigned long)ptr, size, min_count, gfp);
708 else if (atomic_read(&kmemleak_early_log))
709 log_early(KMEMLEAK_ALLOC, ptr, size, min_count, 0, 0);
711 EXPORT_SYMBOL_GPL(kmemleak_alloc);
714 * Memory freeing function callback. This function is called from the kernel
715 * allocators when a block is freed (kmem_cache_free, kfree, vfree etc.).
717 void kmemleak_free(const void *ptr)
719 pr_debug("%s(0x%p)\n", __func__, ptr);
721 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
722 delete_object((unsigned long)ptr);
723 else if (atomic_read(&kmemleak_early_log))
724 log_early(KMEMLEAK_FREE, ptr, 0, 0, 0, 0);
726 EXPORT_SYMBOL_GPL(kmemleak_free);
729 * Mark an already allocated memory block as a false positive. This will cause
730 * the block to no longer be reported as leak and always be scanned.
732 void kmemleak_not_leak(const void *ptr)
734 pr_debug("%s(0x%p)\n", __func__, ptr);
736 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
737 make_gray_object((unsigned long)ptr);
738 else if (atomic_read(&kmemleak_early_log))
739 log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0, 0, 0);
741 EXPORT_SYMBOL(kmemleak_not_leak);
744 * Ignore a memory block. This is usually done when it is known that the
745 * corresponding block is not a leak and does not contain any references to
746 * other allocated memory blocks.
748 void kmemleak_ignore(const void *ptr)
750 pr_debug("%s(0x%p)\n", __func__, ptr);
752 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
753 make_black_object((unsigned long)ptr);
754 else if (atomic_read(&kmemleak_early_log))
755 log_early(KMEMLEAK_IGNORE, ptr, 0, 0, 0, 0);
757 EXPORT_SYMBOL(kmemleak_ignore);
760 * Limit the range to be scanned in an allocated memory block.
762 void kmemleak_scan_area(const void *ptr, unsigned long offset, size_t length,
765 pr_debug("%s(0x%p)\n", __func__, ptr);
767 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
768 add_scan_area((unsigned long)ptr, offset, length, gfp);
769 else if (atomic_read(&kmemleak_early_log))
770 log_early(KMEMLEAK_SCAN_AREA, ptr, 0, 0, offset, length);
772 EXPORT_SYMBOL(kmemleak_scan_area);
775 * Inform kmemleak not to scan the given memory block.
777 void kmemleak_no_scan(const void *ptr)
779 pr_debug("%s(0x%p)\n", __func__, ptr);
781 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
782 object_no_scan((unsigned long)ptr);
783 else if (atomic_read(&kmemleak_early_log))
784 log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0, 0, 0);
786 EXPORT_SYMBOL(kmemleak_no_scan);
789 * Memory scanning is a long process and it needs to be interruptable. This
790 * function checks whether such interrupt condition occured.
792 static int scan_should_stop(void)
794 if (!atomic_read(&kmemleak_enabled))
798 * This function may be called from either process or kthread context,
799 * hence the need to check for both stop conditions.
802 return signal_pending(current);
804 return kthread_should_stop();
810 * Scan a memory block (exclusive range) for valid pointers and add those
811 * found to the gray list.
813 static void scan_block(void *_start, void *_end,
814 struct kmemleak_object *scanned, int allow_resched)
817 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
818 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
820 for (ptr = start; ptr < end; ptr++) {
822 unsigned long pointer = *ptr;
823 struct kmemleak_object *object;
827 if (scan_should_stop())
830 object = find_and_get_object(pointer, 1);
833 if (object == scanned) {
834 /* self referenced, ignore */
840 * Avoid the lockdep recursive warning on object->lock being
841 * previously acquired in scan_object(). These locks are
842 * enclosed by scan_mutex.
844 spin_lock_irqsave_nested(&object->lock, flags,
845 SINGLE_DEPTH_NESTING);
846 if (!color_white(object)) {
847 /* non-orphan, ignored or new */
848 spin_unlock_irqrestore(&object->lock, flags);
854 * Increase the object's reference count (number of pointers
855 * to the memory block). If this count reaches the required
856 * minimum, the object's color will become gray and it will be
857 * added to the gray_list.
860 if (color_gray(object))
861 list_add_tail(&object->gray_list, &gray_list);
864 spin_unlock_irqrestore(&object->lock, flags);
869 * Scan a memory block corresponding to a kmemleak_object. A condition is
870 * that object->use_count >= 1.
872 static void scan_object(struct kmemleak_object *object)
874 struct kmemleak_scan_area *area;
875 struct hlist_node *elem;
879 * Once the object->lock is aquired, the corresponding memory block
880 * cannot be freed (the same lock is aquired in delete_object).
882 spin_lock_irqsave(&object->lock, flags);
883 if (object->flags & OBJECT_NO_SCAN)
885 if (!(object->flags & OBJECT_ALLOCATED))
886 /* already freed object */
888 if (hlist_empty(&object->area_list))
889 scan_block((void *)object->pointer,
890 (void *)(object->pointer + object->size), object, 0);
892 hlist_for_each_entry(area, elem, &object->area_list, node)
893 scan_block((void *)(object->pointer + area->offset),
894 (void *)(object->pointer + area->offset
895 + area->length), object, 0);
897 spin_unlock_irqrestore(&object->lock, flags);
901 * Scan data sections and all the referenced memory blocks allocated via the
902 * kernel's standard allocators. This function must be called with the
905 static void kmemleak_scan(void)
908 struct kmemleak_object *object, *tmp;
909 struct task_struct *task;
912 int gray_list_pass = 0;
914 jiffies_last_scan = jiffies;
916 /* prepare the kmemleak_object's */
918 list_for_each_entry_rcu(object, &object_list, object_list) {
919 spin_lock_irqsave(&object->lock, flags);
922 * With a few exceptions there should be a maximum of
923 * 1 reference to any object at this point.
925 if (atomic_read(&object->use_count) > 1) {
926 pr_debug("object->use_count = %d\n",
927 atomic_read(&object->use_count));
928 dump_object_info(object);
931 /* reset the reference count (whiten the object) */
933 object->flags &= ~OBJECT_NEW;
934 if (color_gray(object) && get_object(object))
935 list_add_tail(&object->gray_list, &gray_list);
937 spin_unlock_irqrestore(&object->lock, flags);
941 /* data/bss scanning */
942 scan_block(_sdata, _edata, NULL, 1);
943 scan_block(__bss_start, __bss_stop, NULL, 1);
946 /* per-cpu sections scanning */
947 for_each_possible_cpu(i)
948 scan_block(__per_cpu_start + per_cpu_offset(i),
949 __per_cpu_end + per_cpu_offset(i), NULL, 1);
953 * Struct page scanning for each node. The code below is not yet safe
954 * with MEMORY_HOTPLUG.
956 for_each_online_node(i) {
957 pg_data_t *pgdat = NODE_DATA(i);
958 unsigned long start_pfn = pgdat->node_start_pfn;
959 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
962 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
967 page = pfn_to_page(pfn);
968 /* only scan if page is in use */
969 if (page_count(page) == 0)
971 scan_block(page, page + 1, NULL, 1);
976 * Scanning the task stacks may introduce false negatives and it is
977 * not enabled by default.
979 if (kmemleak_stack_scan) {
980 read_lock(&tasklist_lock);
981 for_each_process(task)
982 scan_block(task_stack_page(task),
983 task_stack_page(task) + THREAD_SIZE,
985 read_unlock(&tasklist_lock);
989 * Scan the objects already referenced from the sections scanned
990 * above. More objects will be referenced and, if there are no memory
991 * leaks, all the objects will be scanned. The list traversal is safe
992 * for both tail additions and removals from inside the loop. The
993 * kmemleak objects cannot be freed from outside the loop because their
994 * use_count was increased.
997 object = list_entry(gray_list.next, typeof(*object), gray_list);
998 while (&object->gray_list != &gray_list) {
1001 /* may add new objects to the list */
1002 if (!scan_should_stop())
1003 scan_object(object);
1005 tmp = list_entry(object->gray_list.next, typeof(*object),
1008 /* remove the object from the list and release it */
1009 list_del(&object->gray_list);
1015 if (scan_should_stop() || ++gray_list_pass >= GRAY_LIST_PASSES)
1019 * Check for new objects allocated during this scanning and add them
1023 list_for_each_entry_rcu(object, &object_list, object_list) {
1024 spin_lock_irqsave(&object->lock, flags);
1025 if ((object->flags & OBJECT_NEW) && !color_black(object) &&
1026 get_object(object)) {
1027 object->flags &= ~OBJECT_NEW;
1028 list_add_tail(&object->gray_list, &gray_list);
1030 spin_unlock_irqrestore(&object->lock, flags);
1034 if (!list_empty(&gray_list))
1038 WARN_ON(!list_empty(&gray_list));
1041 * If scanning was stopped or new objects were being allocated at a
1042 * higher rate than gray list scanning, do not report any new
1043 * unreferenced objects.
1045 if (scan_should_stop() || gray_list_pass >= GRAY_LIST_PASSES)
1049 * Scanning result reporting.
1052 list_for_each_entry_rcu(object, &object_list, object_list) {
1053 spin_lock_irqsave(&object->lock, flags);
1054 if (unreferenced_object(object) &&
1055 !(object->flags & OBJECT_REPORTED)) {
1056 object->flags |= OBJECT_REPORTED;
1059 spin_unlock_irqrestore(&object->lock, flags);
1064 pr_info("%d new suspected memory leaks (see "
1065 "/sys/kernel/debug/kmemleak)\n", new_leaks);
1070 * Thread function performing automatic memory scanning. Unreferenced objects
1071 * at the end of a memory scan are reported but only the first time.
1073 static int kmemleak_scan_thread(void *arg)
1075 static int first_run = 1;
1077 pr_info("Automatic memory scanning thread started\n");
1078 set_user_nice(current, 10);
1081 * Wait before the first scan to allow the system to fully initialize.
1085 ssleep(SECS_FIRST_SCAN);
1088 while (!kthread_should_stop()) {
1089 signed long timeout = jiffies_scan_wait;
1091 mutex_lock(&scan_mutex);
1093 mutex_unlock(&scan_mutex);
1095 /* wait before the next scan */
1096 while (timeout && !kthread_should_stop())
1097 timeout = schedule_timeout_interruptible(timeout);
1100 pr_info("Automatic memory scanning thread ended\n");
1106 * Start the automatic memory scanning thread. This function must be called
1107 * with the scan_mutex held.
1109 void start_scan_thread(void)
1113 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1114 if (IS_ERR(scan_thread)) {
1115 pr_warning("Failed to create the scan thread\n");
1121 * Stop the automatic memory scanning thread. This function must be called
1122 * with the scan_mutex held.
1124 void stop_scan_thread(void)
1127 kthread_stop(scan_thread);
1133 * Iterate over the object_list and return the first valid object at or after
1134 * the required position with its use_count incremented. The function triggers
1135 * a memory scanning when the pos argument points to the first position.
1137 static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1139 struct kmemleak_object *object;
1143 err = mutex_lock_interruptible(&scan_mutex);
1145 return ERR_PTR(err);
1148 list_for_each_entry_rcu(object, &object_list, object_list) {
1151 if (get_object(object))
1161 * Return the next object in the object_list. The function decrements the
1162 * use_count of the previous object and increases that of the next one.
1164 static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1166 struct kmemleak_object *prev_obj = v;
1167 struct kmemleak_object *next_obj = NULL;
1168 struct list_head *n = &prev_obj->object_list;
1173 list_for_each_continue_rcu(n, &object_list) {
1174 next_obj = list_entry(n, struct kmemleak_object, object_list);
1175 if (get_object(next_obj))
1180 put_object(prev_obj);
1185 * Decrement the use_count of the last object required, if any.
1187 static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1191 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1192 * waiting was interrupted, so only release it if !IS_ERR.
1194 mutex_unlock(&scan_mutex);
1201 * Print the information for an unreferenced object to the seq file.
1203 static int kmemleak_seq_show(struct seq_file *seq, void *v)
1205 struct kmemleak_object *object = v;
1206 unsigned long flags;
1208 spin_lock_irqsave(&object->lock, flags);
1209 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1210 print_unreferenced(seq, object);
1211 spin_unlock_irqrestore(&object->lock, flags);
1215 static const struct seq_operations kmemleak_seq_ops = {
1216 .start = kmemleak_seq_start,
1217 .next = kmemleak_seq_next,
1218 .stop = kmemleak_seq_stop,
1219 .show = kmemleak_seq_show,
1222 static int kmemleak_open(struct inode *inode, struct file *file)
1224 if (!atomic_read(&kmemleak_enabled))
1227 return seq_open(file, &kmemleak_seq_ops);
1230 static int kmemleak_release(struct inode *inode, struct file *file)
1232 return seq_release(inode, file);
1236 * File write operation to configure kmemleak at run-time. The following
1237 * commands can be written to the /sys/kernel/debug/kmemleak file:
1238 * off - disable kmemleak (irreversible)
1239 * stack=on - enable the task stacks scanning
1240 * stack=off - disable the tasks stacks scanning
1241 * scan=on - start the automatic memory scanning thread
1242 * scan=off - stop the automatic memory scanning thread
1243 * scan=... - set the automatic memory scanning period in seconds (0 to
1245 * scan - trigger a memory scan
1247 static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1248 size_t size, loff_t *ppos)
1254 buf_size = min(size, (sizeof(buf) - 1));
1255 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1259 ret = mutex_lock_interruptible(&scan_mutex);
1263 if (strncmp(buf, "off", 3) == 0)
1265 else if (strncmp(buf, "stack=on", 8) == 0)
1266 kmemleak_stack_scan = 1;
1267 else if (strncmp(buf, "stack=off", 9) == 0)
1268 kmemleak_stack_scan = 0;
1269 else if (strncmp(buf, "scan=on", 7) == 0)
1270 start_scan_thread();
1271 else if (strncmp(buf, "scan=off", 8) == 0)
1273 else if (strncmp(buf, "scan=", 5) == 0) {
1276 ret = strict_strtoul(buf + 5, 0, &secs);
1281 jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1282 start_scan_thread();
1284 } else if (strncmp(buf, "scan", 4) == 0)
1290 mutex_unlock(&scan_mutex);
1294 /* ignore the rest of the buffer, only one command at a time */
1299 static const struct file_operations kmemleak_fops = {
1300 .owner = THIS_MODULE,
1301 .open = kmemleak_open,
1303 .write = kmemleak_write,
1304 .llseek = seq_lseek,
1305 .release = kmemleak_release,
1309 * Perform the freeing of the kmemleak internal objects after waiting for any
1310 * current memory scan to complete.
1312 static int kmemleak_cleanup_thread(void *arg)
1314 struct kmemleak_object *object;
1316 mutex_lock(&scan_mutex);
1320 list_for_each_entry_rcu(object, &object_list, object_list)
1321 delete_object(object->pointer);
1323 mutex_unlock(&scan_mutex);
1329 * Start the clean-up thread.
1331 static void kmemleak_cleanup(void)
1333 struct task_struct *cleanup_thread;
1335 cleanup_thread = kthread_run(kmemleak_cleanup_thread, NULL,
1337 if (IS_ERR(cleanup_thread))
1338 pr_warning("Failed to create the clean-up thread\n");
1342 * Disable kmemleak. No memory allocation/freeing will be traced once this
1343 * function is called. Disabling kmemleak is an irreversible operation.
1345 static void kmemleak_disable(void)
1347 /* atomically check whether it was already invoked */
1348 if (atomic_cmpxchg(&kmemleak_error, 0, 1))
1351 /* stop any memory operation tracing */
1352 atomic_set(&kmemleak_early_log, 0);
1353 atomic_set(&kmemleak_enabled, 0);
1355 /* check whether it is too early for a kernel thread */
1356 if (atomic_read(&kmemleak_initialized))
1359 pr_info("Kernel memory leak detector disabled\n");
1363 * Allow boot-time kmemleak disabling (enabled by default).
1365 static int kmemleak_boot_config(char *str)
1369 if (strcmp(str, "off") == 0)
1371 else if (strcmp(str, "on") != 0)
1375 early_param("kmemleak", kmemleak_boot_config);
1378 * Kmemleak initialization.
1380 void __init kmemleak_init(void)
1383 unsigned long flags;
1385 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1386 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1388 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1389 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1390 INIT_PRIO_TREE_ROOT(&object_tree_root);
1392 /* the kernel is still in UP mode, so disabling the IRQs is enough */
1393 local_irq_save(flags);
1394 if (!atomic_read(&kmemleak_error)) {
1395 atomic_set(&kmemleak_enabled, 1);
1396 atomic_set(&kmemleak_early_log, 0);
1398 local_irq_restore(flags);
1401 * This is the point where tracking allocations is safe. Automatic
1402 * scanning is started during the late initcall. Add the early logged
1403 * callbacks to the kmemleak infrastructure.
1405 for (i = 0; i < crt_early_log; i++) {
1406 struct early_log *log = &early_log[i];
1408 switch (log->op_type) {
1409 case KMEMLEAK_ALLOC:
1410 kmemleak_alloc(log->ptr, log->size, log->min_count,
1414 kmemleak_free(log->ptr);
1416 case KMEMLEAK_NOT_LEAK:
1417 kmemleak_not_leak(log->ptr);
1419 case KMEMLEAK_IGNORE:
1420 kmemleak_ignore(log->ptr);
1422 case KMEMLEAK_SCAN_AREA:
1423 kmemleak_scan_area(log->ptr, log->offset, log->length,
1426 case KMEMLEAK_NO_SCAN:
1427 kmemleak_no_scan(log->ptr);
1436 * Late initialization function.
1438 static int __init kmemleak_late_init(void)
1440 struct dentry *dentry;
1442 atomic_set(&kmemleak_initialized, 1);
1444 if (atomic_read(&kmemleak_error)) {
1446 * Some error occured and kmemleak was disabled. There is a
1447 * small chance that kmemleak_disable() was called immediately
1448 * after setting kmemleak_initialized and we may end up with
1449 * two clean-up threads but serialized by scan_mutex.
1455 dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL,
1458 pr_warning("Failed to create the debugfs kmemleak file\n");
1459 mutex_lock(&scan_mutex);
1460 start_scan_thread();
1461 mutex_unlock(&scan_mutex);
1463 pr_info("Kernel memory leak detector initialized\n");
1467 late_initcall(kmemleak_late_init);