1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 2008 ARM Limited
6 * Written by Catalin Marinas <catalin.marinas@arm.com>
8 * For more information on the algorithm and kmemleak usage, please see
9 * Documentation/dev-tools/kmemleak.rst.
14 * The following locks and mutexes are used by kmemleak:
16 * - kmemleak_lock (raw_spinlock_t): protects the object_list modifications and
17 * accesses to the object_tree_root. The object_list is the main list
18 * holding the metadata (struct kmemleak_object) for the allocated memory
19 * blocks. The object_tree_root is a red black tree used to look-up
20 * metadata based on a pointer to the corresponding memory block. The
21 * kmemleak_object structures are added to the object_list and
22 * object_tree_root in the create_object() function called from the
23 * kmemleak_alloc() callback and removed in delete_object() called from the
24 * kmemleak_free() callback
25 * - kmemleak_object.lock (raw_spinlock_t): protects a kmemleak_object.
26 * Accesses to the metadata (e.g. count) are protected by this lock. Note
27 * that some members of this structure may be protected by other means
28 * (atomic or kmemleak_lock). This lock is also held when scanning the
29 * corresponding memory block to avoid the kernel freeing it via the
30 * kmemleak_free() callback. This is less heavyweight than holding a global
31 * lock like kmemleak_lock during scanning.
32 * - scan_mutex (mutex): ensures that only one thread may scan the memory for
33 * unreferenced objects at a time. The gray_list contains the objects which
34 * are already referenced or marked as false positives and need to be
35 * scanned. This list is only modified during a scanning episode when the
36 * scan_mutex is held. At the end of a scan, the gray_list is always empty.
37 * Note that the kmemleak_object.use_count is incremented when an object is
38 * added to the gray_list and therefore cannot be freed. This mutex also
39 * prevents multiple users of the "kmemleak" debugfs file together with
40 * modifications to the memory scanning parameters including the scan_thread
43 * Locks and mutexes are acquired/nested in the following order:
45 * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING)
47 * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex
50 * The kmemleak_object structures have a use_count incremented or decremented
51 * using the get_object()/put_object() functions. When the use_count becomes
52 * 0, this count can no longer be incremented and put_object() schedules the
53 * kmemleak_object freeing via an RCU callback. All calls to the get_object()
54 * function must be protected by rcu_read_lock() to avoid accessing a freed
58 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
60 #include <linux/init.h>
61 #include <linux/kernel.h>
62 #include <linux/list.h>
63 #include <linux/sched/signal.h>
64 #include <linux/sched/task.h>
65 #include <linux/sched/task_stack.h>
66 #include <linux/jiffies.h>
67 #include <linux/delay.h>
68 #include <linux/export.h>
69 #include <linux/kthread.h>
70 #include <linux/rbtree.h>
72 #include <linux/debugfs.h>
73 #include <linux/seq_file.h>
74 #include <linux/cpumask.h>
75 #include <linux/spinlock.h>
76 #include <linux/module.h>
77 #include <linux/mutex.h>
78 #include <linux/rcupdate.h>
79 #include <linux/stacktrace.h>
80 #include <linux/cache.h>
81 #include <linux/percpu.h>
82 #include <linux/memblock.h>
83 #include <linux/pfn.h>
84 #include <linux/mmzone.h>
85 #include <linux/slab.h>
86 #include <linux/thread_info.h>
87 #include <linux/err.h>
88 #include <linux/uaccess.h>
89 #include <linux/string.h>
90 #include <linux/nodemask.h>
92 #include <linux/workqueue.h>
93 #include <linux/crc32.h>
95 #include <asm/sections.h>
96 #include <asm/processor.h>
97 #include <linux/atomic.h>
99 #include <linux/kasan.h>
100 #include <linux/kmemleak.h>
101 #include <linux/memory_hotplug.h>
104 * Kmemleak configuration and common defines.
106 #define MAX_TRACE 16 /* stack trace length */
107 #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
108 #define SECS_FIRST_SCAN 60 /* delay before the first scan */
109 #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
110 #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
112 #define BYTES_PER_POINTER sizeof(void *)
114 /* GFP bitmask for kmemleak internal allocations */
115 #define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC)) | \
116 __GFP_NORETRY | __GFP_NOMEMALLOC | \
119 /* scanning area inside a memory block */
120 struct kmemleak_scan_area {
121 struct hlist_node node;
126 #define KMEMLEAK_GREY 0
127 #define KMEMLEAK_BLACK -1
130 * Structure holding the metadata for each allocated memory block.
131 * Modifications to such objects should be made while holding the
132 * object->lock. Insertions or deletions from object_list, gray_list or
133 * rb_node are already protected by the corresponding locks or mutex (see
134 * the notes on locking above). These objects are reference-counted
135 * (use_count) and freed using the RCU mechanism.
137 struct kmemleak_object {
139 unsigned int flags; /* object status flags */
140 struct list_head object_list;
141 struct list_head gray_list;
142 struct rb_node rb_node;
143 struct rcu_head rcu; /* object_list lockless traversal */
144 /* object usage count; object freed when use_count == 0 */
146 unsigned long pointer;
148 /* pass surplus references to this pointer */
149 unsigned long excess_ref;
150 /* minimum number of a pointers found before it is considered leak */
152 /* the total number of pointers found pointing to this object */
154 /* checksum for detecting modified objects */
156 /* memory ranges to be scanned inside an object (empty for all) */
157 struct hlist_head area_list;
158 unsigned long trace[MAX_TRACE];
159 unsigned int trace_len;
160 unsigned long jiffies; /* creation timestamp */
161 pid_t pid; /* pid of the current task */
162 char comm[TASK_COMM_LEN]; /* executable name */
165 /* flag representing the memory block allocation status */
166 #define OBJECT_ALLOCATED (1 << 0)
167 /* flag set after the first reporting of an unreference object */
168 #define OBJECT_REPORTED (1 << 1)
169 /* flag set to not scan the object */
170 #define OBJECT_NO_SCAN (1 << 2)
171 /* flag set to fully scan the object when scan_area allocation failed */
172 #define OBJECT_FULL_SCAN (1 << 3)
174 #define HEX_PREFIX " "
175 /* number of bytes to print per line; must be 16 or 32 */
176 #define HEX_ROW_SIZE 16
177 /* number of bytes to print at a time (1, 2, 4, 8) */
178 #define HEX_GROUP_SIZE 1
179 /* include ASCII after the hex output */
181 /* max number of lines to be printed */
182 #define HEX_MAX_LINES 2
184 /* the list of all allocated objects */
185 static LIST_HEAD(object_list);
186 /* the list of gray-colored objects (see color_gray comment below) */
187 static LIST_HEAD(gray_list);
188 /* memory pool allocation */
189 static struct kmemleak_object mem_pool[CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE];
190 static int mem_pool_free_count = ARRAY_SIZE(mem_pool);
191 static LIST_HEAD(mem_pool_free_list);
192 /* search tree for object boundaries */
193 static struct rb_root object_tree_root = RB_ROOT;
194 /* protecting the access to object_list and object_tree_root */
195 static DEFINE_RAW_SPINLOCK(kmemleak_lock);
197 /* allocation caches for kmemleak internal data */
198 static struct kmem_cache *object_cache;
199 static struct kmem_cache *scan_area_cache;
201 /* set if tracing memory operations is enabled */
202 static int kmemleak_enabled = 1;
203 /* same as above but only for the kmemleak_free() callback */
204 static int kmemleak_free_enabled = 1;
205 /* set in the late_initcall if there were no errors */
206 static int kmemleak_initialized;
207 /* set if a kmemleak warning was issued */
208 static int kmemleak_warning;
209 /* set if a fatal kmemleak error has occurred */
210 static int kmemleak_error;
212 /* minimum and maximum address that may be valid pointers */
213 static unsigned long min_addr = ULONG_MAX;
214 static unsigned long max_addr;
216 static struct task_struct *scan_thread;
217 /* used to avoid reporting of recently allocated objects */
218 static unsigned long jiffies_min_age;
219 static unsigned long jiffies_last_scan;
220 /* delay between automatic memory scannings */
221 static signed long jiffies_scan_wait;
222 /* enables or disables the task stacks scanning */
223 static int kmemleak_stack_scan = 1;
224 /* protects the memory scanning, parameters and debug/kmemleak file access */
225 static DEFINE_MUTEX(scan_mutex);
226 /* setting kmemleak=on, will set this var, skipping the disable */
227 static int kmemleak_skip_disable;
228 /* If there are leaks that can be reported */
229 static bool kmemleak_found_leaks;
231 static bool kmemleak_verbose;
232 module_param_named(verbose, kmemleak_verbose, bool, 0600);
234 static void kmemleak_disable(void);
237 * Print a warning and dump the stack trace.
239 #define kmemleak_warn(x...) do { \
242 kmemleak_warning = 1; \
246 * Macro invoked when a serious kmemleak condition occurred and cannot be
247 * recovered from. Kmemleak will be disabled and further allocation/freeing
248 * tracing no longer available.
250 #define kmemleak_stop(x...) do { \
252 kmemleak_disable(); \
255 #define warn_or_seq_printf(seq, fmt, ...) do { \
257 seq_printf(seq, fmt, ##__VA_ARGS__); \
259 pr_warn(fmt, ##__VA_ARGS__); \
262 static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type,
263 int rowsize, int groupsize, const void *buf,
264 size_t len, bool ascii)
267 seq_hex_dump(seq, HEX_PREFIX, prefix_type, rowsize, groupsize,
270 print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type,
271 rowsize, groupsize, buf, len, ascii);
275 * Printing of the objects hex dump to the seq file. The number of lines to be
276 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
277 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
278 * with the object->lock held.
280 static void hex_dump_object(struct seq_file *seq,
281 struct kmemleak_object *object)
283 const u8 *ptr = (const u8 *)object->pointer;
286 /* limit the number of lines to HEX_MAX_LINES */
287 len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
289 warn_or_seq_printf(seq, " hex dump (first %zu bytes):\n", len);
290 kasan_disable_current();
291 warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE,
292 HEX_GROUP_SIZE, ptr, len, HEX_ASCII);
293 kasan_enable_current();
297 * Object colors, encoded with count and min_count:
298 * - white - orphan object, not enough references to it (count < min_count)
299 * - gray - not orphan, not marked as false positive (min_count == 0) or
300 * sufficient references to it (count >= min_count)
301 * - black - ignore, it doesn't contain references (e.g. text section)
302 * (min_count == -1). No function defined for this color.
303 * Newly created objects don't have any color assigned (object->count == -1)
304 * before the next memory scan when they become white.
306 static bool color_white(const struct kmemleak_object *object)
308 return object->count != KMEMLEAK_BLACK &&
309 object->count < object->min_count;
312 static bool color_gray(const struct kmemleak_object *object)
314 return object->min_count != KMEMLEAK_BLACK &&
315 object->count >= object->min_count;
319 * Objects are considered unreferenced only if their color is white, they have
320 * not be deleted and have a minimum age to avoid false positives caused by
321 * pointers temporarily stored in CPU registers.
323 static bool unreferenced_object(struct kmemleak_object *object)
325 return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
326 time_before_eq(object->jiffies + jiffies_min_age,
331 * Printing of the unreferenced objects information to the seq file. The
332 * print_unreferenced function must be called with the object->lock held.
334 static void print_unreferenced(struct seq_file *seq,
335 struct kmemleak_object *object)
338 unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
340 warn_or_seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
341 object->pointer, object->size);
342 warn_or_seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
343 object->comm, object->pid, object->jiffies,
344 msecs_age / 1000, msecs_age % 1000);
345 hex_dump_object(seq, object);
346 warn_or_seq_printf(seq, " backtrace:\n");
348 for (i = 0; i < object->trace_len; i++) {
349 void *ptr = (void *)object->trace[i];
350 warn_or_seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
355 * Print the kmemleak_object information. This function is used mainly for
356 * debugging special cases when kmemleak operations. It must be called with
357 * the object->lock held.
359 static void dump_object_info(struct kmemleak_object *object)
361 pr_notice("Object 0x%08lx (size %zu):\n",
362 object->pointer, object->size);
363 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
364 object->comm, object->pid, object->jiffies);
365 pr_notice(" min_count = %d\n", object->min_count);
366 pr_notice(" count = %d\n", object->count);
367 pr_notice(" flags = 0x%x\n", object->flags);
368 pr_notice(" checksum = %u\n", object->checksum);
369 pr_notice(" backtrace:\n");
370 stack_trace_print(object->trace, object->trace_len, 4);
374 * Look-up a memory block metadata (kmemleak_object) in the object search
375 * tree based on a pointer value. If alias is 0, only values pointing to the
376 * beginning of the memory block are allowed. The kmemleak_lock must be held
377 * when calling this function.
379 static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
381 struct rb_node *rb = object_tree_root.rb_node;
384 struct kmemleak_object *object =
385 rb_entry(rb, struct kmemleak_object, rb_node);
386 if (ptr < object->pointer)
387 rb = object->rb_node.rb_left;
388 else if (object->pointer + object->size <= ptr)
389 rb = object->rb_node.rb_right;
390 else if (object->pointer == ptr || alias)
393 kmemleak_warn("Found object by alias at 0x%08lx\n",
395 dump_object_info(object);
403 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
404 * that once an object's use_count reached 0, the RCU freeing was already
405 * registered and the object should no longer be used. This function must be
406 * called under the protection of rcu_read_lock().
408 static int get_object(struct kmemleak_object *object)
410 return atomic_inc_not_zero(&object->use_count);
414 * Memory pool allocation and freeing. kmemleak_lock must not be held.
416 static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
419 struct kmemleak_object *object;
421 /* try the slab allocator first */
423 object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
428 /* slab allocation failed, try the memory pool */
429 raw_spin_lock_irqsave(&kmemleak_lock, flags);
430 object = list_first_entry_or_null(&mem_pool_free_list,
431 typeof(*object), object_list);
433 list_del(&object->object_list);
434 else if (mem_pool_free_count)
435 object = &mem_pool[--mem_pool_free_count];
437 pr_warn_once("Memory pool empty, consider increasing CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE\n");
438 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
444 * Return the object to either the slab allocator or the memory pool.
446 static void mem_pool_free(struct kmemleak_object *object)
450 if (object < mem_pool || object >= mem_pool + ARRAY_SIZE(mem_pool)) {
451 kmem_cache_free(object_cache, object);
455 /* add the object to the memory pool free list */
456 raw_spin_lock_irqsave(&kmemleak_lock, flags);
457 list_add(&object->object_list, &mem_pool_free_list);
458 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
462 * RCU callback to free a kmemleak_object.
464 static void free_object_rcu(struct rcu_head *rcu)
466 struct hlist_node *tmp;
467 struct kmemleak_scan_area *area;
468 struct kmemleak_object *object =
469 container_of(rcu, struct kmemleak_object, rcu);
472 * Once use_count is 0 (guaranteed by put_object), there is no other
473 * code accessing this object, hence no need for locking.
475 hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
476 hlist_del(&area->node);
477 kmem_cache_free(scan_area_cache, area);
479 mem_pool_free(object);
483 * Decrement the object use_count. Once the count is 0, free the object using
484 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
485 * delete_object() path, the delayed RCU freeing ensures that there is no
486 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
489 static void put_object(struct kmemleak_object *object)
491 if (!atomic_dec_and_test(&object->use_count))
494 /* should only get here after delete_object was called */
495 WARN_ON(object->flags & OBJECT_ALLOCATED);
498 * It may be too early for the RCU callbacks, however, there is no
499 * concurrent object_list traversal when !object_cache and all objects
500 * came from the memory pool. Free the object directly.
503 call_rcu(&object->rcu, free_object_rcu);
505 free_object_rcu(&object->rcu);
509 * Look up an object in the object search tree and increase its use_count.
511 static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
514 struct kmemleak_object *object;
517 raw_spin_lock_irqsave(&kmemleak_lock, flags);
518 object = lookup_object(ptr, alias);
519 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
521 /* check whether the object is still available */
522 if (object && !get_object(object))
530 * Remove an object from the object_tree_root and object_list. Must be called
531 * with the kmemleak_lock held _if_ kmemleak is still enabled.
533 static void __remove_object(struct kmemleak_object *object)
535 rb_erase(&object->rb_node, &object_tree_root);
536 list_del_rcu(&object->object_list);
540 * Look up an object in the object search tree and remove it from both
541 * object_tree_root and object_list. The returned object's use_count should be
542 * at least 1, as initially set by create_object().
544 static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias)
547 struct kmemleak_object *object;
549 raw_spin_lock_irqsave(&kmemleak_lock, flags);
550 object = lookup_object(ptr, alias);
552 __remove_object(object);
553 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
559 * Save stack trace to the given array of MAX_TRACE size.
561 static int __save_stack_trace(unsigned long *trace)
563 return stack_trace_save(trace, MAX_TRACE, 2);
567 * Create the metadata (struct kmemleak_object) corresponding to an allocated
568 * memory block and add it to the object_list and object_tree_root.
570 static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
571 int min_count, gfp_t gfp)
574 struct kmemleak_object *object, *parent;
575 struct rb_node **link, *rb_parent;
576 unsigned long untagged_ptr;
578 object = mem_pool_alloc(gfp);
580 pr_warn("Cannot allocate a kmemleak_object structure\n");
585 INIT_LIST_HEAD(&object->object_list);
586 INIT_LIST_HEAD(&object->gray_list);
587 INIT_HLIST_HEAD(&object->area_list);
588 raw_spin_lock_init(&object->lock);
589 atomic_set(&object->use_count, 1);
590 object->flags = OBJECT_ALLOCATED;
591 object->pointer = ptr;
593 object->excess_ref = 0;
594 object->min_count = min_count;
595 object->count = 0; /* white color initially */
596 object->jiffies = jiffies;
597 object->checksum = 0;
599 /* task information */
602 strncpy(object->comm, "hardirq", sizeof(object->comm));
603 } else if (in_serving_softirq()) {
605 strncpy(object->comm, "softirq", sizeof(object->comm));
607 object->pid = current->pid;
609 * There is a small chance of a race with set_task_comm(),
610 * however using get_task_comm() here may cause locking
611 * dependency issues with current->alloc_lock. In the worst
612 * case, the command line is not correct.
614 strncpy(object->comm, current->comm, sizeof(object->comm));
617 /* kernel backtrace */
618 object->trace_len = __save_stack_trace(object->trace);
620 raw_spin_lock_irqsave(&kmemleak_lock, flags);
622 untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
623 min_addr = min(min_addr, untagged_ptr);
624 max_addr = max(max_addr, untagged_ptr + size);
625 link = &object_tree_root.rb_node;
629 parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
630 if (ptr + size <= parent->pointer)
631 link = &parent->rb_node.rb_left;
632 else if (parent->pointer + parent->size <= ptr)
633 link = &parent->rb_node.rb_right;
635 kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
638 * No need for parent->lock here since "parent" cannot
639 * be freed while the kmemleak_lock is held.
641 dump_object_info(parent);
642 kmem_cache_free(object_cache, object);
647 rb_link_node(&object->rb_node, rb_parent, link);
648 rb_insert_color(&object->rb_node, &object_tree_root);
650 list_add_tail_rcu(&object->object_list, &object_list);
652 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
657 * Mark the object as not allocated and schedule RCU freeing via put_object().
659 static void __delete_object(struct kmemleak_object *object)
663 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
664 WARN_ON(atomic_read(&object->use_count) < 1);
667 * Locking here also ensures that the corresponding memory block
668 * cannot be freed when it is being scanned.
670 raw_spin_lock_irqsave(&object->lock, flags);
671 object->flags &= ~OBJECT_ALLOCATED;
672 raw_spin_unlock_irqrestore(&object->lock, flags);
677 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
680 static void delete_object_full(unsigned long ptr)
682 struct kmemleak_object *object;
684 object = find_and_remove_object(ptr, 0);
687 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
692 __delete_object(object);
696 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
697 * delete it. If the memory block is partially freed, the function may create
698 * additional metadata for the remaining parts of the block.
700 static void delete_object_part(unsigned long ptr, size_t size)
702 struct kmemleak_object *object;
703 unsigned long start, end;
705 object = find_and_remove_object(ptr, 1);
708 kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
715 * Create one or two objects that may result from the memory block
716 * split. Note that partial freeing is only done by free_bootmem() and
717 * this happens before kmemleak_init() is called.
719 start = object->pointer;
720 end = object->pointer + object->size;
722 create_object(start, ptr - start, object->min_count,
724 if (ptr + size < end)
725 create_object(ptr + size, end - ptr - size, object->min_count,
728 __delete_object(object);
731 static void __paint_it(struct kmemleak_object *object, int color)
733 object->min_count = color;
734 if (color == KMEMLEAK_BLACK)
735 object->flags |= OBJECT_NO_SCAN;
738 static void paint_it(struct kmemleak_object *object, int color)
742 raw_spin_lock_irqsave(&object->lock, flags);
743 __paint_it(object, color);
744 raw_spin_unlock_irqrestore(&object->lock, flags);
747 static void paint_ptr(unsigned long ptr, int color)
749 struct kmemleak_object *object;
751 object = find_and_get_object(ptr, 0);
753 kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n",
755 (color == KMEMLEAK_GREY) ? "Grey" :
756 (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
759 paint_it(object, color);
764 * Mark an object permanently as gray-colored so that it can no longer be
765 * reported as a leak. This is used in general to mark a false positive.
767 static void make_gray_object(unsigned long ptr)
769 paint_ptr(ptr, KMEMLEAK_GREY);
773 * Mark the object as black-colored so that it is ignored from scans and
776 static void make_black_object(unsigned long ptr)
778 paint_ptr(ptr, KMEMLEAK_BLACK);
782 * Add a scanning area to the object. If at least one such area is added,
783 * kmemleak will only scan these ranges rather than the whole memory block.
785 static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
788 struct kmemleak_object *object;
789 struct kmemleak_scan_area *area = NULL;
791 object = find_and_get_object(ptr, 1);
793 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
799 area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
801 raw_spin_lock_irqsave(&object->lock, flags);
803 pr_warn_once("Cannot allocate a scan area, scanning the full object\n");
804 /* mark the object for full scan to avoid false positives */
805 object->flags |= OBJECT_FULL_SCAN;
808 if (size == SIZE_MAX) {
809 size = object->pointer + object->size - ptr;
810 } else if (ptr + size > object->pointer + object->size) {
811 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
812 dump_object_info(object);
813 kmem_cache_free(scan_area_cache, area);
817 INIT_HLIST_NODE(&area->node);
821 hlist_add_head(&area->node, &object->area_list);
823 raw_spin_unlock_irqrestore(&object->lock, flags);
828 * Any surplus references (object already gray) to 'ptr' are passed to
829 * 'excess_ref'. This is used in the vmalloc() case where a pointer to
830 * vm_struct may be used as an alternative reference to the vmalloc'ed object
831 * (see free_thread_stack()).
833 static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref)
836 struct kmemleak_object *object;
838 object = find_and_get_object(ptr, 0);
840 kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n",
845 raw_spin_lock_irqsave(&object->lock, flags);
846 object->excess_ref = excess_ref;
847 raw_spin_unlock_irqrestore(&object->lock, flags);
852 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
853 * pointer. Such object will not be scanned by kmemleak but references to it
856 static void object_no_scan(unsigned long ptr)
859 struct kmemleak_object *object;
861 object = find_and_get_object(ptr, 0);
863 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
867 raw_spin_lock_irqsave(&object->lock, flags);
868 object->flags |= OBJECT_NO_SCAN;
869 raw_spin_unlock_irqrestore(&object->lock, flags);
874 * kmemleak_alloc - register a newly allocated object
875 * @ptr: pointer to beginning of the object
876 * @size: size of the object
877 * @min_count: minimum number of references to this object. If during memory
878 * scanning a number of references less than @min_count is found,
879 * the object is reported as a memory leak. If @min_count is 0,
880 * the object is never reported as a leak. If @min_count is -1,
881 * the object is ignored (not scanned and not reported as a leak)
882 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
884 * This function is called from the kernel allocators when a new object
885 * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.).
887 void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
890 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
892 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
893 create_object((unsigned long)ptr, size, min_count, gfp);
895 EXPORT_SYMBOL_GPL(kmemleak_alloc);
898 * kmemleak_alloc_percpu - register a newly allocated __percpu object
899 * @ptr: __percpu pointer to beginning of the object
900 * @size: size of the object
901 * @gfp: flags used for kmemleak internal memory allocations
903 * This function is called from the kernel percpu allocator when a new object
904 * (memory block) is allocated (alloc_percpu).
906 void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
911 pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size);
914 * Percpu allocations are only scanned and not reported as leaks
915 * (min_count is set to 0).
917 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
918 for_each_possible_cpu(cpu)
919 create_object((unsigned long)per_cpu_ptr(ptr, cpu),
922 EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
925 * kmemleak_vmalloc - register a newly vmalloc'ed object
926 * @area: pointer to vm_struct
927 * @size: size of the object
928 * @gfp: __vmalloc() flags used for kmemleak internal memory allocations
930 * This function is called from the vmalloc() kernel allocator when a new
931 * object (memory block) is allocated.
933 void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp)
935 pr_debug("%s(0x%p, %zu)\n", __func__, area, size);
938 * A min_count = 2 is needed because vm_struct contains a reference to
939 * the virtual address of the vmalloc'ed block.
941 if (kmemleak_enabled) {
942 create_object((unsigned long)area->addr, size, 2, gfp);
943 object_set_excess_ref((unsigned long)area,
944 (unsigned long)area->addr);
947 EXPORT_SYMBOL_GPL(kmemleak_vmalloc);
950 * kmemleak_free - unregister a previously registered object
951 * @ptr: pointer to beginning of the object
953 * This function is called from the kernel allocators when an object (memory
954 * block) is freed (kmem_cache_free, kfree, vfree etc.).
956 void __ref kmemleak_free(const void *ptr)
958 pr_debug("%s(0x%p)\n", __func__, ptr);
960 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
961 delete_object_full((unsigned long)ptr);
963 EXPORT_SYMBOL_GPL(kmemleak_free);
966 * kmemleak_free_part - partially unregister a previously registered object
967 * @ptr: pointer to the beginning or inside the object. This also
968 * represents the start of the range to be freed
969 * @size: size to be unregistered
971 * This function is called when only a part of a memory block is freed
972 * (usually from the bootmem allocator).
974 void __ref kmemleak_free_part(const void *ptr, size_t size)
976 pr_debug("%s(0x%p)\n", __func__, ptr);
978 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
979 delete_object_part((unsigned long)ptr, size);
981 EXPORT_SYMBOL_GPL(kmemleak_free_part);
984 * kmemleak_free_percpu - unregister a previously registered __percpu object
985 * @ptr: __percpu pointer to beginning of the object
987 * This function is called from the kernel percpu allocator when an object
988 * (memory block) is freed (free_percpu).
990 void __ref kmemleak_free_percpu(const void __percpu *ptr)
994 pr_debug("%s(0x%p)\n", __func__, ptr);
996 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
997 for_each_possible_cpu(cpu)
998 delete_object_full((unsigned long)per_cpu_ptr(ptr,
1001 EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
1004 * kmemleak_update_trace - update object allocation stack trace
1005 * @ptr: pointer to beginning of the object
1007 * Override the object allocation stack trace for cases where the actual
1008 * allocation place is not always useful.
1010 void __ref kmemleak_update_trace(const void *ptr)
1012 struct kmemleak_object *object;
1013 unsigned long flags;
1015 pr_debug("%s(0x%p)\n", __func__, ptr);
1017 if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
1020 object = find_and_get_object((unsigned long)ptr, 1);
1023 kmemleak_warn("Updating stack trace for unknown object at %p\n",
1029 raw_spin_lock_irqsave(&object->lock, flags);
1030 object->trace_len = __save_stack_trace(object->trace);
1031 raw_spin_unlock_irqrestore(&object->lock, flags);
1035 EXPORT_SYMBOL(kmemleak_update_trace);
1038 * kmemleak_not_leak - mark an allocated object as false positive
1039 * @ptr: pointer to beginning of the object
1041 * Calling this function on an object will cause the memory block to no longer
1042 * be reported as leak and always be scanned.
1044 void __ref kmemleak_not_leak(const void *ptr)
1046 pr_debug("%s(0x%p)\n", __func__, ptr);
1048 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1049 make_gray_object((unsigned long)ptr);
1051 EXPORT_SYMBOL(kmemleak_not_leak);
1054 * kmemleak_ignore - ignore an allocated object
1055 * @ptr: pointer to beginning of the object
1057 * Calling this function on an object will cause the memory block to be
1058 * ignored (not scanned and not reported as a leak). This is usually done when
1059 * it is known that the corresponding block is not a leak and does not contain
1060 * any references to other allocated memory blocks.
1062 void __ref kmemleak_ignore(const void *ptr)
1064 pr_debug("%s(0x%p)\n", __func__, ptr);
1066 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1067 make_black_object((unsigned long)ptr);
1069 EXPORT_SYMBOL(kmemleak_ignore);
1072 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1073 * @ptr: pointer to beginning or inside the object. This also
1074 * represents the start of the scan area
1075 * @size: size of the scan area
1076 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1078 * This function is used when it is known that only certain parts of an object
1079 * contain references to other objects. Kmemleak will only scan these areas
1080 * reducing the number false negatives.
1082 void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1084 pr_debug("%s(0x%p)\n", __func__, ptr);
1086 if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
1087 add_scan_area((unsigned long)ptr, size, gfp);
1089 EXPORT_SYMBOL(kmemleak_scan_area);
1092 * kmemleak_no_scan - do not scan an allocated object
1093 * @ptr: pointer to beginning of the object
1095 * This function notifies kmemleak not to scan the given memory block. Useful
1096 * in situations where it is known that the given object does not contain any
1097 * references to other objects. Kmemleak will not scan such objects reducing
1098 * the number of false negatives.
1100 void __ref kmemleak_no_scan(const void *ptr)
1102 pr_debug("%s(0x%p)\n", __func__, ptr);
1104 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1105 object_no_scan((unsigned long)ptr);
1107 EXPORT_SYMBOL(kmemleak_no_scan);
1110 * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical
1112 * @phys: physical address of the object
1113 * @size: size of the object
1114 * @min_count: minimum number of references to this object.
1115 * See kmemleak_alloc()
1116 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1118 void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, int min_count,
1121 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1122 kmemleak_alloc(__va(phys), size, min_count, gfp);
1124 EXPORT_SYMBOL(kmemleak_alloc_phys);
1127 * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a
1128 * physical address argument
1129 * @phys: physical address if the beginning or inside an object. This
1130 * also represents the start of the range to be freed
1131 * @size: size to be unregistered
1133 void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size)
1135 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1136 kmemleak_free_part(__va(phys), size);
1138 EXPORT_SYMBOL(kmemleak_free_part_phys);
1141 * kmemleak_not_leak_phys - similar to kmemleak_not_leak but taking a physical
1143 * @phys: physical address of the object
1145 void __ref kmemleak_not_leak_phys(phys_addr_t phys)
1147 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1148 kmemleak_not_leak(__va(phys));
1150 EXPORT_SYMBOL(kmemleak_not_leak_phys);
1153 * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical
1155 * @phys: physical address of the object
1157 void __ref kmemleak_ignore_phys(phys_addr_t phys)
1159 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1160 kmemleak_ignore(__va(phys));
1162 EXPORT_SYMBOL(kmemleak_ignore_phys);
1165 * Update an object's checksum and return true if it was modified.
1167 static bool update_checksum(struct kmemleak_object *object)
1169 u32 old_csum = object->checksum;
1171 kasan_disable_current();
1172 kcsan_disable_current();
1173 object->checksum = crc32(0, (void *)object->pointer, object->size);
1174 kasan_enable_current();
1175 kcsan_enable_current();
1177 return object->checksum != old_csum;
1181 * Update an object's references. object->lock must be held by the caller.
1183 static void update_refs(struct kmemleak_object *object)
1185 if (!color_white(object)) {
1186 /* non-orphan, ignored or new */
1191 * Increase the object's reference count (number of pointers to the
1192 * memory block). If this count reaches the required minimum, the
1193 * object's color will become gray and it will be added to the
1197 if (color_gray(object)) {
1198 /* put_object() called when removing from gray_list */
1199 WARN_ON(!get_object(object));
1200 list_add_tail(&object->gray_list, &gray_list);
1205 * Memory scanning is a long process and it needs to be interruptable. This
1206 * function checks whether such interrupt condition occurred.
1208 static int scan_should_stop(void)
1210 if (!kmemleak_enabled)
1214 * This function may be called from either process or kthread context,
1215 * hence the need to check for both stop conditions.
1218 return signal_pending(current);
1220 return kthread_should_stop();
1226 * Scan a memory block (exclusive range) for valid pointers and add those
1227 * found to the gray list.
1229 static void scan_block(void *_start, void *_end,
1230 struct kmemleak_object *scanned)
1233 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1234 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1235 unsigned long flags;
1236 unsigned long untagged_ptr;
1238 raw_spin_lock_irqsave(&kmemleak_lock, flags);
1239 for (ptr = start; ptr < end; ptr++) {
1240 struct kmemleak_object *object;
1241 unsigned long pointer;
1242 unsigned long excess_ref;
1244 if (scan_should_stop())
1247 kasan_disable_current();
1249 kasan_enable_current();
1251 untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer);
1252 if (untagged_ptr < min_addr || untagged_ptr >= max_addr)
1256 * No need for get_object() here since we hold kmemleak_lock.
1257 * object->use_count cannot be dropped to 0 while the object
1258 * is still present in object_tree_root and object_list
1259 * (with updates protected by kmemleak_lock).
1261 object = lookup_object(pointer, 1);
1264 if (object == scanned)
1265 /* self referenced, ignore */
1269 * Avoid the lockdep recursive warning on object->lock being
1270 * previously acquired in scan_object(). These locks are
1271 * enclosed by scan_mutex.
1273 raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1274 /* only pass surplus references (object already gray) */
1275 if (color_gray(object)) {
1276 excess_ref = object->excess_ref;
1277 /* no need for update_refs() if object already gray */
1280 update_refs(object);
1282 raw_spin_unlock(&object->lock);
1285 object = lookup_object(excess_ref, 0);
1288 if (object == scanned)
1289 /* circular reference, ignore */
1291 raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1292 update_refs(object);
1293 raw_spin_unlock(&object->lock);
1296 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
1300 * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
1303 static void scan_large_block(void *start, void *end)
1307 while (start < end) {
1308 next = min(start + MAX_SCAN_SIZE, end);
1309 scan_block(start, next, NULL);
1317 * Scan a memory block corresponding to a kmemleak_object. A condition is
1318 * that object->use_count >= 1.
1320 static void scan_object(struct kmemleak_object *object)
1322 struct kmemleak_scan_area *area;
1323 unsigned long flags;
1326 * Once the object->lock is acquired, the corresponding memory block
1327 * cannot be freed (the same lock is acquired in delete_object).
1329 raw_spin_lock_irqsave(&object->lock, flags);
1330 if (object->flags & OBJECT_NO_SCAN)
1332 if (!(object->flags & OBJECT_ALLOCATED))
1333 /* already freed object */
1335 if (hlist_empty(&object->area_list) ||
1336 object->flags & OBJECT_FULL_SCAN) {
1337 void *start = (void *)object->pointer;
1338 void *end = (void *)(object->pointer + object->size);
1342 next = min(start + MAX_SCAN_SIZE, end);
1343 scan_block(start, next, object);
1349 raw_spin_unlock_irqrestore(&object->lock, flags);
1351 raw_spin_lock_irqsave(&object->lock, flags);
1352 } while (object->flags & OBJECT_ALLOCATED);
1354 hlist_for_each_entry(area, &object->area_list, node)
1355 scan_block((void *)area->start,
1356 (void *)(area->start + area->size),
1359 raw_spin_unlock_irqrestore(&object->lock, flags);
1363 * Scan the objects already referenced (gray objects). More objects will be
1364 * referenced and, if there are no memory leaks, all the objects are scanned.
1366 static void scan_gray_list(void)
1368 struct kmemleak_object *object, *tmp;
1371 * The list traversal is safe for both tail additions and removals
1372 * from inside the loop. The kmemleak objects cannot be freed from
1373 * outside the loop because their use_count was incremented.
1375 object = list_entry(gray_list.next, typeof(*object), gray_list);
1376 while (&object->gray_list != &gray_list) {
1379 /* may add new objects to the list */
1380 if (!scan_should_stop())
1381 scan_object(object);
1383 tmp = list_entry(object->gray_list.next, typeof(*object),
1386 /* remove the object from the list and release it */
1387 list_del(&object->gray_list);
1392 WARN_ON(!list_empty(&gray_list));
1396 * Scan data sections and all the referenced memory blocks allocated via the
1397 * kernel's standard allocators. This function must be called with the
1400 static void kmemleak_scan(void)
1402 unsigned long flags;
1403 struct kmemleak_object *object;
1407 jiffies_last_scan = jiffies;
1409 /* prepare the kmemleak_object's */
1411 list_for_each_entry_rcu(object, &object_list, object_list) {
1412 raw_spin_lock_irqsave(&object->lock, flags);
1415 * With a few exceptions there should be a maximum of
1416 * 1 reference to any object at this point.
1418 if (atomic_read(&object->use_count) > 1) {
1419 pr_debug("object->use_count = %d\n",
1420 atomic_read(&object->use_count));
1421 dump_object_info(object);
1424 /* reset the reference count (whiten the object) */
1426 if (color_gray(object) && get_object(object))
1427 list_add_tail(&object->gray_list, &gray_list);
1429 raw_spin_unlock_irqrestore(&object->lock, flags);
1434 /* per-cpu sections scanning */
1435 for_each_possible_cpu(i)
1436 scan_large_block(__per_cpu_start + per_cpu_offset(i),
1437 __per_cpu_end + per_cpu_offset(i));
1441 * Struct page scanning for each node.
1444 for_each_online_node(i) {
1445 unsigned long start_pfn = node_start_pfn(i);
1446 unsigned long end_pfn = node_end_pfn(i);
1449 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1450 struct page *page = pfn_to_online_page(pfn);
1455 /* only scan pages belonging to this node */
1456 if (page_to_nid(page) != i)
1458 /* only scan if page is in use */
1459 if (page_count(page) == 0)
1461 scan_block(page, page + 1, NULL);
1469 * Scanning the task stacks (may introduce false negatives).
1471 if (kmemleak_stack_scan) {
1472 struct task_struct *p, *g;
1475 for_each_process_thread(g, p) {
1476 void *stack = try_get_task_stack(p);
1478 scan_block(stack, stack + THREAD_SIZE, NULL);
1486 * Scan the objects already referenced from the sections scanned
1492 * Check for new or unreferenced objects modified since the previous
1493 * scan and color them gray until the next scan.
1496 list_for_each_entry_rcu(object, &object_list, object_list) {
1497 raw_spin_lock_irqsave(&object->lock, flags);
1498 if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1499 && update_checksum(object) && get_object(object)) {
1500 /* color it gray temporarily */
1501 object->count = object->min_count;
1502 list_add_tail(&object->gray_list, &gray_list);
1504 raw_spin_unlock_irqrestore(&object->lock, flags);
1509 * Re-scan the gray list for modified unreferenced objects.
1514 * If scanning was stopped do not report any new unreferenced objects.
1516 if (scan_should_stop())
1520 * Scanning result reporting.
1523 list_for_each_entry_rcu(object, &object_list, object_list) {
1524 raw_spin_lock_irqsave(&object->lock, flags);
1525 if (unreferenced_object(object) &&
1526 !(object->flags & OBJECT_REPORTED)) {
1527 object->flags |= OBJECT_REPORTED;
1529 if (kmemleak_verbose)
1530 print_unreferenced(NULL, object);
1534 raw_spin_unlock_irqrestore(&object->lock, flags);
1539 kmemleak_found_leaks = true;
1541 pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
1548 * Thread function performing automatic memory scanning. Unreferenced objects
1549 * at the end of a memory scan are reported but only the first time.
1551 static int kmemleak_scan_thread(void *arg)
1553 static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN);
1555 pr_info("Automatic memory scanning thread started\n");
1556 set_user_nice(current, 10);
1559 * Wait before the first scan to allow the system to fully initialize.
1562 signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000);
1564 while (timeout && !kthread_should_stop())
1565 timeout = schedule_timeout_interruptible(timeout);
1568 while (!kthread_should_stop()) {
1569 signed long timeout = jiffies_scan_wait;
1571 mutex_lock(&scan_mutex);
1573 mutex_unlock(&scan_mutex);
1575 /* wait before the next scan */
1576 while (timeout && !kthread_should_stop())
1577 timeout = schedule_timeout_interruptible(timeout);
1580 pr_info("Automatic memory scanning thread ended\n");
1586 * Start the automatic memory scanning thread. This function must be called
1587 * with the scan_mutex held.
1589 static void start_scan_thread(void)
1593 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1594 if (IS_ERR(scan_thread)) {
1595 pr_warn("Failed to create the scan thread\n");
1601 * Stop the automatic memory scanning thread.
1603 static void stop_scan_thread(void)
1606 kthread_stop(scan_thread);
1612 * Iterate over the object_list and return the first valid object at or after
1613 * the required position with its use_count incremented. The function triggers
1614 * a memory scanning when the pos argument points to the first position.
1616 static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1618 struct kmemleak_object *object;
1622 err = mutex_lock_interruptible(&scan_mutex);
1624 return ERR_PTR(err);
1627 list_for_each_entry_rcu(object, &object_list, object_list) {
1630 if (get_object(object))
1639 * Return the next object in the object_list. The function decrements the
1640 * use_count of the previous object and increases that of the next one.
1642 static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1644 struct kmemleak_object *prev_obj = v;
1645 struct kmemleak_object *next_obj = NULL;
1646 struct kmemleak_object *obj = prev_obj;
1650 list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
1651 if (get_object(obj)) {
1657 put_object(prev_obj);
1662 * Decrement the use_count of the last object required, if any.
1664 static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1668 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1669 * waiting was interrupted, so only release it if !IS_ERR.
1672 mutex_unlock(&scan_mutex);
1679 * Print the information for an unreferenced object to the seq file.
1681 static int kmemleak_seq_show(struct seq_file *seq, void *v)
1683 struct kmemleak_object *object = v;
1684 unsigned long flags;
1686 raw_spin_lock_irqsave(&object->lock, flags);
1687 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1688 print_unreferenced(seq, object);
1689 raw_spin_unlock_irqrestore(&object->lock, flags);
1693 static const struct seq_operations kmemleak_seq_ops = {
1694 .start = kmemleak_seq_start,
1695 .next = kmemleak_seq_next,
1696 .stop = kmemleak_seq_stop,
1697 .show = kmemleak_seq_show,
1700 static int kmemleak_open(struct inode *inode, struct file *file)
1702 return seq_open(file, &kmemleak_seq_ops);
1705 static int dump_str_object_info(const char *str)
1707 unsigned long flags;
1708 struct kmemleak_object *object;
1711 if (kstrtoul(str, 0, &addr))
1713 object = find_and_get_object(addr, 0);
1715 pr_info("Unknown object at 0x%08lx\n", addr);
1719 raw_spin_lock_irqsave(&object->lock, flags);
1720 dump_object_info(object);
1721 raw_spin_unlock_irqrestore(&object->lock, flags);
1728 * We use grey instead of black to ensure we can do future scans on the same
1729 * objects. If we did not do future scans these black objects could
1730 * potentially contain references to newly allocated objects in the future and
1731 * we'd end up with false positives.
1733 static void kmemleak_clear(void)
1735 struct kmemleak_object *object;
1736 unsigned long flags;
1739 list_for_each_entry_rcu(object, &object_list, object_list) {
1740 raw_spin_lock_irqsave(&object->lock, flags);
1741 if ((object->flags & OBJECT_REPORTED) &&
1742 unreferenced_object(object))
1743 __paint_it(object, KMEMLEAK_GREY);
1744 raw_spin_unlock_irqrestore(&object->lock, flags);
1748 kmemleak_found_leaks = false;
1751 static void __kmemleak_do_cleanup(void);
1754 * File write operation to configure kmemleak at run-time. The following
1755 * commands can be written to the /sys/kernel/debug/kmemleak file:
1756 * off - disable kmemleak (irreversible)
1757 * stack=on - enable the task stacks scanning
1758 * stack=off - disable the tasks stacks scanning
1759 * scan=on - start the automatic memory scanning thread
1760 * scan=off - stop the automatic memory scanning thread
1761 * scan=... - set the automatic memory scanning period in seconds (0 to
1763 * scan - trigger a memory scan
1764 * clear - mark all current reported unreferenced kmemleak objects as
1765 * grey to ignore printing them, or free all kmemleak objects
1766 * if kmemleak has been disabled.
1767 * dump=... - dump information about the object found at the given address
1769 static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1770 size_t size, loff_t *ppos)
1776 buf_size = min(size, (sizeof(buf) - 1));
1777 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1781 ret = mutex_lock_interruptible(&scan_mutex);
1785 if (strncmp(buf, "clear", 5) == 0) {
1786 if (kmemleak_enabled)
1789 __kmemleak_do_cleanup();
1793 if (!kmemleak_enabled) {
1798 if (strncmp(buf, "off", 3) == 0)
1800 else if (strncmp(buf, "stack=on", 8) == 0)
1801 kmemleak_stack_scan = 1;
1802 else if (strncmp(buf, "stack=off", 9) == 0)
1803 kmemleak_stack_scan = 0;
1804 else if (strncmp(buf, "scan=on", 7) == 0)
1805 start_scan_thread();
1806 else if (strncmp(buf, "scan=off", 8) == 0)
1808 else if (strncmp(buf, "scan=", 5) == 0) {
1811 ret = kstrtoul(buf + 5, 0, &secs);
1816 jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1817 start_scan_thread();
1819 } else if (strncmp(buf, "scan", 4) == 0)
1821 else if (strncmp(buf, "dump=", 5) == 0)
1822 ret = dump_str_object_info(buf + 5);
1827 mutex_unlock(&scan_mutex);
1831 /* ignore the rest of the buffer, only one command at a time */
1836 static const struct file_operations kmemleak_fops = {
1837 .owner = THIS_MODULE,
1838 .open = kmemleak_open,
1840 .write = kmemleak_write,
1841 .llseek = seq_lseek,
1842 .release = seq_release,
1845 static void __kmemleak_do_cleanup(void)
1847 struct kmemleak_object *object, *tmp;
1850 * Kmemleak has already been disabled, no need for RCU list traversal
1851 * or kmemleak_lock held.
1853 list_for_each_entry_safe(object, tmp, &object_list, object_list) {
1854 __remove_object(object);
1855 __delete_object(object);
1860 * Stop the memory scanning thread and free the kmemleak internal objects if
1861 * no previous scan thread (otherwise, kmemleak may still have some useful
1862 * information on memory leaks).
1864 static void kmemleak_do_cleanup(struct work_struct *work)
1868 mutex_lock(&scan_mutex);
1870 * Once it is made sure that kmemleak_scan has stopped, it is safe to no
1871 * longer track object freeing. Ordering of the scan thread stopping and
1872 * the memory accesses below is guaranteed by the kthread_stop()
1875 kmemleak_free_enabled = 0;
1876 mutex_unlock(&scan_mutex);
1878 if (!kmemleak_found_leaks)
1879 __kmemleak_do_cleanup();
1881 pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
1884 static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
1887 * Disable kmemleak. No memory allocation/freeing will be traced once this
1888 * function is called. Disabling kmemleak is an irreversible operation.
1890 static void kmemleak_disable(void)
1892 /* atomically check whether it was already invoked */
1893 if (cmpxchg(&kmemleak_error, 0, 1))
1896 /* stop any memory operation tracing */
1897 kmemleak_enabled = 0;
1899 /* check whether it is too early for a kernel thread */
1900 if (kmemleak_initialized)
1901 schedule_work(&cleanup_work);
1903 kmemleak_free_enabled = 0;
1905 pr_info("Kernel memory leak detector disabled\n");
1909 * Allow boot-time kmemleak disabling (enabled by default).
1911 static int __init kmemleak_boot_config(char *str)
1915 if (strcmp(str, "off") == 0)
1917 else if (strcmp(str, "on") == 0)
1918 kmemleak_skip_disable = 1;
1923 early_param("kmemleak", kmemleak_boot_config);
1926 * Kmemleak initialization.
1928 void __init kmemleak_init(void)
1930 #ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
1931 if (!kmemleak_skip_disable) {
1940 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1941 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1943 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1944 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1946 /* register the data/bss sections */
1947 create_object((unsigned long)_sdata, _edata - _sdata,
1948 KMEMLEAK_GREY, GFP_ATOMIC);
1949 create_object((unsigned long)__bss_start, __bss_stop - __bss_start,
1950 KMEMLEAK_GREY, GFP_ATOMIC);
1951 /* only register .data..ro_after_init if not within .data */
1952 if (&__start_ro_after_init < &_sdata || &__end_ro_after_init > &_edata)
1953 create_object((unsigned long)__start_ro_after_init,
1954 __end_ro_after_init - __start_ro_after_init,
1955 KMEMLEAK_GREY, GFP_ATOMIC);
1959 * Late initialization function.
1961 static int __init kmemleak_late_init(void)
1963 kmemleak_initialized = 1;
1965 debugfs_create_file("kmemleak", 0644, NULL, NULL, &kmemleak_fops);
1967 if (kmemleak_error) {
1969 * Some error occurred and kmemleak was disabled. There is a
1970 * small chance that kmemleak_disable() was called immediately
1971 * after setting kmemleak_initialized and we may end up with
1972 * two clean-up threads but serialized by scan_mutex.
1974 schedule_work(&cleanup_work);
1978 if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) {
1979 mutex_lock(&scan_mutex);
1980 start_scan_thread();
1981 mutex_unlock(&scan_mutex);
1984 pr_info("Kernel memory leak detector initialized (mem pool available: %d)\n",
1985 mem_pool_free_count);
1989 late_initcall(kmemleak_late_init);