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/dev-tools/kmemleak.rst.
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 red black 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 * Locks and mutexes are acquired/nested in the following order:
58 * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING)
60 * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex
63 * The kmemleak_object structures have a use_count incremented or decremented
64 * using the get_object()/put_object() functions. When the use_count becomes
65 * 0, this count can no longer be incremented and put_object() schedules the
66 * kmemleak_object freeing via an RCU callback. All calls to the get_object()
67 * function must be protected by rcu_read_lock() to avoid accessing a freed
71 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
73 #include <linux/init.h>
74 #include <linux/kernel.h>
75 #include <linux/list.h>
76 #include <linux/sched/signal.h>
77 #include <linux/sched/task.h>
78 #include <linux/sched/task_stack.h>
79 #include <linux/jiffies.h>
80 #include <linux/delay.h>
81 #include <linux/export.h>
82 #include <linux/kthread.h>
83 #include <linux/rbtree.h>
85 #include <linux/debugfs.h>
86 #include <linux/seq_file.h>
87 #include <linux/cpumask.h>
88 #include <linux/spinlock.h>
89 #include <linux/module.h>
90 #include <linux/mutex.h>
91 #include <linux/rcupdate.h>
92 #include <linux/stacktrace.h>
93 #include <linux/cache.h>
94 #include <linux/percpu.h>
95 #include <linux/memblock.h>
96 #include <linux/pfn.h>
97 #include <linux/mmzone.h>
98 #include <linux/slab.h>
99 #include <linux/thread_info.h>
100 #include <linux/err.h>
101 #include <linux/uaccess.h>
102 #include <linux/string.h>
103 #include <linux/nodemask.h>
104 #include <linux/mm.h>
105 #include <linux/workqueue.h>
106 #include <linux/crc32.h>
108 #include <asm/sections.h>
109 #include <asm/processor.h>
110 #include <linux/atomic.h>
112 #include <linux/kasan.h>
113 #include <linux/kmemleak.h>
114 #include <linux/memory_hotplug.h>
117 * Kmemleak configuration and common defines.
119 #define MAX_TRACE 16 /* stack trace length */
120 #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
121 #define SECS_FIRST_SCAN 60 /* delay before the first scan */
122 #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
123 #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
125 #define BYTES_PER_POINTER sizeof(void *)
127 /* GFP bitmask for kmemleak internal allocations */
128 #define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC)) | \
129 __GFP_NORETRY | __GFP_NOMEMALLOC | \
130 __GFP_NOWARN | __GFP_NOFAIL)
132 /* scanning area inside a memory block */
133 struct kmemleak_scan_area {
134 struct hlist_node node;
139 #define KMEMLEAK_GREY 0
140 #define KMEMLEAK_BLACK -1
143 * Structure holding the metadata for each allocated memory block.
144 * Modifications to such objects should be made while holding the
145 * object->lock. Insertions or deletions from object_list, gray_list or
146 * rb_node are already protected by the corresponding locks or mutex (see
147 * the notes on locking above). These objects are reference-counted
148 * (use_count) and freed using the RCU mechanism.
150 struct kmemleak_object {
152 unsigned int flags; /* object status flags */
153 struct list_head object_list;
154 struct list_head gray_list;
155 struct rb_node rb_node;
156 struct rcu_head rcu; /* object_list lockless traversal */
157 /* object usage count; object freed when use_count == 0 */
159 unsigned long pointer;
161 /* pass surplus references to this pointer */
162 unsigned long excess_ref;
163 /* minimum number of a pointers found before it is considered leak */
165 /* the total number of pointers found pointing to this object */
167 /* checksum for detecting modified objects */
169 /* memory ranges to be scanned inside an object (empty for all) */
170 struct hlist_head area_list;
171 unsigned long trace[MAX_TRACE];
172 unsigned int trace_len;
173 unsigned long jiffies; /* creation timestamp */
174 pid_t pid; /* pid of the current task */
175 char comm[TASK_COMM_LEN]; /* executable name */
178 /* flag representing the memory block allocation status */
179 #define OBJECT_ALLOCATED (1 << 0)
180 /* flag set after the first reporting of an unreference object */
181 #define OBJECT_REPORTED (1 << 1)
182 /* flag set to not scan the object */
183 #define OBJECT_NO_SCAN (1 << 2)
185 #define HEX_PREFIX " "
186 /* number of bytes to print per line; must be 16 or 32 */
187 #define HEX_ROW_SIZE 16
188 /* number of bytes to print at a time (1, 2, 4, 8) */
189 #define HEX_GROUP_SIZE 1
190 /* include ASCII after the hex output */
192 /* max number of lines to be printed */
193 #define HEX_MAX_LINES 2
195 /* the list of all allocated objects */
196 static LIST_HEAD(object_list);
197 /* the list of gray-colored objects (see color_gray comment below) */
198 static LIST_HEAD(gray_list);
199 /* search tree for object boundaries */
200 static struct rb_root object_tree_root = RB_ROOT;
201 /* rw_lock protecting the access to object_list and object_tree_root */
202 static DEFINE_RWLOCK(kmemleak_lock);
204 /* allocation caches for kmemleak internal data */
205 static struct kmem_cache *object_cache;
206 static struct kmem_cache *scan_area_cache;
208 /* set if tracing memory operations is enabled */
209 static int kmemleak_enabled;
210 /* same as above but only for the kmemleak_free() callback */
211 static int kmemleak_free_enabled;
212 /* set in the late_initcall if there were no errors */
213 static int kmemleak_initialized;
214 /* enables or disables early logging of the memory operations */
215 static int kmemleak_early_log = 1;
216 /* set if a kmemleak warning was issued */
217 static int kmemleak_warning;
218 /* set if a fatal kmemleak error has occurred */
219 static int kmemleak_error;
221 /* minimum and maximum address that may be valid pointers */
222 static unsigned long min_addr = ULONG_MAX;
223 static unsigned long max_addr;
225 static struct task_struct *scan_thread;
226 /* used to avoid reporting of recently allocated objects */
227 static unsigned long jiffies_min_age;
228 static unsigned long jiffies_last_scan;
229 /* delay between automatic memory scannings */
230 static signed long jiffies_scan_wait;
231 /* enables or disables the task stacks scanning */
232 static int kmemleak_stack_scan = 1;
233 /* protects the memory scanning, parameters and debug/kmemleak file access */
234 static DEFINE_MUTEX(scan_mutex);
235 /* setting kmemleak=on, will set this var, skipping the disable */
236 static int kmemleak_skip_disable;
237 /* If there are leaks that can be reported */
238 static bool kmemleak_found_leaks;
240 static bool kmemleak_verbose;
241 module_param_named(verbose, kmemleak_verbose, bool, 0600);
244 * Early object allocation/freeing logging. Kmemleak is initialized after the
245 * kernel allocator. However, both the kernel allocator and kmemleak may
246 * allocate memory blocks which need to be tracked. Kmemleak defines an
247 * arbitrary buffer to hold the allocation/freeing information before it is
251 /* kmemleak operation type for early logging */
254 KMEMLEAK_ALLOC_PERCPU,
257 KMEMLEAK_FREE_PERCPU,
262 KMEMLEAK_SET_EXCESS_REF
266 * Structure holding the information passed to kmemleak callbacks during the
270 int op_type; /* kmemleak operation type */
271 int min_count; /* minimum reference count */
272 const void *ptr; /* allocated/freed memory block */
274 size_t size; /* memory block size */
275 unsigned long excess_ref; /* surplus reference passing */
277 unsigned long trace[MAX_TRACE]; /* stack trace */
278 unsigned int trace_len; /* stack trace length */
281 /* early logging buffer and current position */
282 static struct early_log
283 early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata;
284 static int crt_early_log __initdata;
286 static void kmemleak_disable(void);
289 * Print a warning and dump the stack trace.
291 #define kmemleak_warn(x...) do { \
294 kmemleak_warning = 1; \
298 * Macro invoked when a serious kmemleak condition occurred and cannot be
299 * recovered from. Kmemleak will be disabled and further allocation/freeing
300 * tracing no longer available.
302 #define kmemleak_stop(x...) do { \
304 kmemleak_disable(); \
307 #define warn_or_seq_printf(seq, fmt, ...) do { \
309 seq_printf(seq, fmt, ##__VA_ARGS__); \
311 pr_warn(fmt, ##__VA_ARGS__); \
314 static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type,
315 int rowsize, int groupsize, const void *buf,
316 size_t len, bool ascii)
319 seq_hex_dump(seq, HEX_PREFIX, prefix_type, rowsize, groupsize,
322 print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type,
323 rowsize, groupsize, buf, len, ascii);
327 * Printing of the objects hex dump to the seq file. The number of lines to be
328 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
329 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
330 * with the object->lock held.
332 static void hex_dump_object(struct seq_file *seq,
333 struct kmemleak_object *object)
335 const u8 *ptr = (const u8 *)object->pointer;
338 /* limit the number of lines to HEX_MAX_LINES */
339 len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
341 warn_or_seq_printf(seq, " hex dump (first %zu bytes):\n", len);
342 kasan_disable_current();
343 warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE,
344 HEX_GROUP_SIZE, ptr, len, HEX_ASCII);
345 kasan_enable_current();
349 * Object colors, encoded with count and min_count:
350 * - white - orphan object, not enough references to it (count < min_count)
351 * - gray - not orphan, not marked as false positive (min_count == 0) or
352 * sufficient references to it (count >= min_count)
353 * - black - ignore, it doesn't contain references (e.g. text section)
354 * (min_count == -1). No function defined for this color.
355 * Newly created objects don't have any color assigned (object->count == -1)
356 * before the next memory scan when they become white.
358 static bool color_white(const struct kmemleak_object *object)
360 return object->count != KMEMLEAK_BLACK &&
361 object->count < object->min_count;
364 static bool color_gray(const struct kmemleak_object *object)
366 return object->min_count != KMEMLEAK_BLACK &&
367 object->count >= object->min_count;
371 * Objects are considered unreferenced only if their color is white, they have
372 * not be deleted and have a minimum age to avoid false positives caused by
373 * pointers temporarily stored in CPU registers.
375 static bool unreferenced_object(struct kmemleak_object *object)
377 return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
378 time_before_eq(object->jiffies + jiffies_min_age,
383 * Printing of the unreferenced objects information to the seq file. The
384 * print_unreferenced function must be called with the object->lock held.
386 static void print_unreferenced(struct seq_file *seq,
387 struct kmemleak_object *object)
390 unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
392 warn_or_seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
393 object->pointer, object->size);
394 warn_or_seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
395 object->comm, object->pid, object->jiffies,
396 msecs_age / 1000, msecs_age % 1000);
397 hex_dump_object(seq, object);
398 warn_or_seq_printf(seq, " backtrace:\n");
400 for (i = 0; i < object->trace_len; i++) {
401 void *ptr = (void *)object->trace[i];
402 warn_or_seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
407 * Print the kmemleak_object information. This function is used mainly for
408 * debugging special cases when kmemleak operations. It must be called with
409 * the object->lock held.
411 static void dump_object_info(struct kmemleak_object *object)
413 pr_notice("Object 0x%08lx (size %zu):\n",
414 object->pointer, object->size);
415 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
416 object->comm, object->pid, object->jiffies);
417 pr_notice(" min_count = %d\n", object->min_count);
418 pr_notice(" count = %d\n", object->count);
419 pr_notice(" flags = 0x%x\n", object->flags);
420 pr_notice(" checksum = %u\n", object->checksum);
421 pr_notice(" backtrace:\n");
422 stack_trace_print(object->trace, object->trace_len, 4);
426 * Look-up a memory block metadata (kmemleak_object) in the object search
427 * tree based on a pointer value. If alias is 0, only values pointing to the
428 * beginning of the memory block are allowed. The kmemleak_lock must be held
429 * when calling this function.
431 static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
433 struct rb_node *rb = object_tree_root.rb_node;
436 struct kmemleak_object *object =
437 rb_entry(rb, struct kmemleak_object, rb_node);
438 if (ptr < object->pointer)
439 rb = object->rb_node.rb_left;
440 else if (object->pointer + object->size <= ptr)
441 rb = object->rb_node.rb_right;
442 else if (object->pointer == ptr || alias)
445 kmemleak_warn("Found object by alias at 0x%08lx\n",
447 dump_object_info(object);
455 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
456 * that once an object's use_count reached 0, the RCU freeing was already
457 * registered and the object should no longer be used. This function must be
458 * called under the protection of rcu_read_lock().
460 static int get_object(struct kmemleak_object *object)
462 return atomic_inc_not_zero(&object->use_count);
466 * RCU callback to free a kmemleak_object.
468 static void free_object_rcu(struct rcu_head *rcu)
470 struct hlist_node *tmp;
471 struct kmemleak_scan_area *area;
472 struct kmemleak_object *object =
473 container_of(rcu, struct kmemleak_object, rcu);
476 * Once use_count is 0 (guaranteed by put_object), there is no other
477 * code accessing this object, hence no need for locking.
479 hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
480 hlist_del(&area->node);
481 kmem_cache_free(scan_area_cache, area);
483 kmem_cache_free(object_cache, object);
487 * Decrement the object use_count. Once the count is 0, free the object using
488 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
489 * delete_object() path, the delayed RCU freeing ensures that there is no
490 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
493 static void put_object(struct kmemleak_object *object)
495 if (!atomic_dec_and_test(&object->use_count))
498 /* should only get here after delete_object was called */
499 WARN_ON(object->flags & OBJECT_ALLOCATED);
501 call_rcu(&object->rcu, free_object_rcu);
505 * Look up an object in the object search tree and increase its use_count.
507 static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
510 struct kmemleak_object *object;
513 read_lock_irqsave(&kmemleak_lock, flags);
514 object = lookup_object(ptr, alias);
515 read_unlock_irqrestore(&kmemleak_lock, flags);
517 /* check whether the object is still available */
518 if (object && !get_object(object))
526 * Look up an object in the object search tree and remove it from both
527 * object_tree_root and object_list. The returned object's use_count should be
528 * at least 1, as initially set by create_object().
530 static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias)
533 struct kmemleak_object *object;
535 write_lock_irqsave(&kmemleak_lock, flags);
536 object = lookup_object(ptr, alias);
538 rb_erase(&object->rb_node, &object_tree_root);
539 list_del_rcu(&object->object_list);
541 write_unlock_irqrestore(&kmemleak_lock, flags);
547 * Save stack trace to the given array of MAX_TRACE size.
549 static int __save_stack_trace(unsigned long *trace)
551 return stack_trace_save(trace, MAX_TRACE, 2);
555 * Create the metadata (struct kmemleak_object) corresponding to an allocated
556 * memory block and add it to the object_list and object_tree_root.
558 static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
559 int min_count, gfp_t gfp)
562 struct kmemleak_object *object, *parent;
563 struct rb_node **link, *rb_parent;
564 unsigned long untagged_ptr;
566 object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
568 pr_warn("Cannot allocate a kmemleak_object structure\n");
573 INIT_LIST_HEAD(&object->object_list);
574 INIT_LIST_HEAD(&object->gray_list);
575 INIT_HLIST_HEAD(&object->area_list);
576 spin_lock_init(&object->lock);
577 atomic_set(&object->use_count, 1);
578 object->flags = OBJECT_ALLOCATED;
579 object->pointer = ptr;
581 object->excess_ref = 0;
582 object->min_count = min_count;
583 object->count = 0; /* white color initially */
584 object->jiffies = jiffies;
585 object->checksum = 0;
587 /* task information */
590 strncpy(object->comm, "hardirq", sizeof(object->comm));
591 } else if (in_softirq()) {
593 strncpy(object->comm, "softirq", sizeof(object->comm));
595 object->pid = current->pid;
597 * There is a small chance of a race with set_task_comm(),
598 * however using get_task_comm() here may cause locking
599 * dependency issues with current->alloc_lock. In the worst
600 * case, the command line is not correct.
602 strncpy(object->comm, current->comm, sizeof(object->comm));
605 /* kernel backtrace */
606 object->trace_len = __save_stack_trace(object->trace);
608 write_lock_irqsave(&kmemleak_lock, flags);
610 untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
611 min_addr = min(min_addr, untagged_ptr);
612 max_addr = max(max_addr, untagged_ptr + size);
613 link = &object_tree_root.rb_node;
617 parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
618 if (ptr + size <= parent->pointer)
619 link = &parent->rb_node.rb_left;
620 else if (parent->pointer + parent->size <= ptr)
621 link = &parent->rb_node.rb_right;
623 kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
626 * No need for parent->lock here since "parent" cannot
627 * be freed while the kmemleak_lock is held.
629 dump_object_info(parent);
630 kmem_cache_free(object_cache, object);
635 rb_link_node(&object->rb_node, rb_parent, link);
636 rb_insert_color(&object->rb_node, &object_tree_root);
638 list_add_tail_rcu(&object->object_list, &object_list);
640 write_unlock_irqrestore(&kmemleak_lock, flags);
645 * Mark the object as not allocated and schedule RCU freeing via put_object().
647 static void __delete_object(struct kmemleak_object *object)
651 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
652 WARN_ON(atomic_read(&object->use_count) < 1);
655 * Locking here also ensures that the corresponding memory block
656 * cannot be freed when it is being scanned.
658 spin_lock_irqsave(&object->lock, flags);
659 object->flags &= ~OBJECT_ALLOCATED;
660 spin_unlock_irqrestore(&object->lock, flags);
665 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
668 static void delete_object_full(unsigned long ptr)
670 struct kmemleak_object *object;
672 object = find_and_remove_object(ptr, 0);
675 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
680 __delete_object(object);
684 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
685 * delete it. If the memory block is partially freed, the function may create
686 * additional metadata for the remaining parts of the block.
688 static void delete_object_part(unsigned long ptr, size_t size)
690 struct kmemleak_object *object;
691 unsigned long start, end;
693 object = find_and_remove_object(ptr, 1);
696 kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
703 * Create one or two objects that may result from the memory block
704 * split. Note that partial freeing is only done by free_bootmem() and
705 * this happens before kmemleak_init() is called. The path below is
706 * only executed during early log recording in kmemleak_init(), so
707 * GFP_KERNEL is enough.
709 start = object->pointer;
710 end = object->pointer + object->size;
712 create_object(start, ptr - start, object->min_count,
714 if (ptr + size < end)
715 create_object(ptr + size, end - ptr - size, object->min_count,
718 __delete_object(object);
721 static void __paint_it(struct kmemleak_object *object, int color)
723 object->min_count = color;
724 if (color == KMEMLEAK_BLACK)
725 object->flags |= OBJECT_NO_SCAN;
728 static void paint_it(struct kmemleak_object *object, int color)
732 spin_lock_irqsave(&object->lock, flags);
733 __paint_it(object, color);
734 spin_unlock_irqrestore(&object->lock, flags);
737 static void paint_ptr(unsigned long ptr, int color)
739 struct kmemleak_object *object;
741 object = find_and_get_object(ptr, 0);
743 kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n",
745 (color == KMEMLEAK_GREY) ? "Grey" :
746 (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
749 paint_it(object, color);
754 * Mark an object permanently as gray-colored so that it can no longer be
755 * reported as a leak. This is used in general to mark a false positive.
757 static void make_gray_object(unsigned long ptr)
759 paint_ptr(ptr, KMEMLEAK_GREY);
763 * Mark the object as black-colored so that it is ignored from scans and
766 static void make_black_object(unsigned long ptr)
768 paint_ptr(ptr, KMEMLEAK_BLACK);
772 * Add a scanning area to the object. If at least one such area is added,
773 * kmemleak will only scan these ranges rather than the whole memory block.
775 static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
778 struct kmemleak_object *object;
779 struct kmemleak_scan_area *area;
781 object = find_and_get_object(ptr, 1);
783 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
788 area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
790 pr_warn("Cannot allocate a scan area\n");
794 spin_lock_irqsave(&object->lock, flags);
795 if (size == SIZE_MAX) {
796 size = object->pointer + object->size - ptr;
797 } else if (ptr + size > object->pointer + object->size) {
798 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
799 dump_object_info(object);
800 kmem_cache_free(scan_area_cache, area);
804 INIT_HLIST_NODE(&area->node);
808 hlist_add_head(&area->node, &object->area_list);
810 spin_unlock_irqrestore(&object->lock, flags);
816 * Any surplus references (object already gray) to 'ptr' are passed to
817 * 'excess_ref'. This is used in the vmalloc() case where a pointer to
818 * vm_struct may be used as an alternative reference to the vmalloc'ed object
819 * (see free_thread_stack()).
821 static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref)
824 struct kmemleak_object *object;
826 object = find_and_get_object(ptr, 0);
828 kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n",
833 spin_lock_irqsave(&object->lock, flags);
834 object->excess_ref = excess_ref;
835 spin_unlock_irqrestore(&object->lock, flags);
840 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
841 * pointer. Such object will not be scanned by kmemleak but references to it
844 static void object_no_scan(unsigned long ptr)
847 struct kmemleak_object *object;
849 object = find_and_get_object(ptr, 0);
851 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
855 spin_lock_irqsave(&object->lock, flags);
856 object->flags |= OBJECT_NO_SCAN;
857 spin_unlock_irqrestore(&object->lock, flags);
862 * Log an early kmemleak_* call to the early_log buffer. These calls will be
863 * processed later once kmemleak is fully initialized.
865 static void __init log_early(int op_type, const void *ptr, size_t size,
869 struct early_log *log;
871 if (kmemleak_error) {
872 /* kmemleak stopped recording, just count the requests */
877 if (crt_early_log >= ARRAY_SIZE(early_log)) {
884 * There is no need for locking since the kernel is still in UP mode
885 * at this stage. Disabling the IRQs is enough.
887 local_irq_save(flags);
888 log = &early_log[crt_early_log];
889 log->op_type = op_type;
892 log->min_count = min_count;
893 log->trace_len = __save_stack_trace(log->trace);
895 local_irq_restore(flags);
899 * Log an early allocated block and populate the stack trace.
901 static void early_alloc(struct early_log *log)
903 struct kmemleak_object *object;
907 if (!kmemleak_enabled || !log->ptr || IS_ERR(log->ptr))
911 * RCU locking needed to ensure object is not freed via put_object().
914 object = create_object((unsigned long)log->ptr, log->size,
915 log->min_count, GFP_ATOMIC);
918 spin_lock_irqsave(&object->lock, flags);
919 for (i = 0; i < log->trace_len; i++)
920 object->trace[i] = log->trace[i];
921 object->trace_len = log->trace_len;
922 spin_unlock_irqrestore(&object->lock, flags);
928 * Log an early allocated block and populate the stack trace.
930 static void early_alloc_percpu(struct early_log *log)
933 const void __percpu *ptr = log->ptr;
935 for_each_possible_cpu(cpu) {
936 log->ptr = per_cpu_ptr(ptr, cpu);
942 * kmemleak_alloc - register a newly allocated object
943 * @ptr: pointer to beginning of the object
944 * @size: size of the object
945 * @min_count: minimum number of references to this object. If during memory
946 * scanning a number of references less than @min_count is found,
947 * the object is reported as a memory leak. If @min_count is 0,
948 * the object is never reported as a leak. If @min_count is -1,
949 * the object is ignored (not scanned and not reported as a leak)
950 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
952 * This function is called from the kernel allocators when a new object
953 * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.).
955 void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
958 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
960 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
961 create_object((unsigned long)ptr, size, min_count, gfp);
962 else if (kmemleak_early_log)
963 log_early(KMEMLEAK_ALLOC, ptr, size, min_count);
965 EXPORT_SYMBOL_GPL(kmemleak_alloc);
968 * kmemleak_alloc_percpu - register a newly allocated __percpu object
969 * @ptr: __percpu pointer to beginning of the object
970 * @size: size of the object
971 * @gfp: flags used for kmemleak internal memory allocations
973 * This function is called from the kernel percpu allocator when a new object
974 * (memory block) is allocated (alloc_percpu).
976 void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
981 pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size);
984 * Percpu allocations are only scanned and not reported as leaks
985 * (min_count is set to 0).
987 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
988 for_each_possible_cpu(cpu)
989 create_object((unsigned long)per_cpu_ptr(ptr, cpu),
991 else if (kmemleak_early_log)
992 log_early(KMEMLEAK_ALLOC_PERCPU, ptr, size, 0);
994 EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
997 * kmemleak_vmalloc - register a newly vmalloc'ed object
998 * @area: pointer to vm_struct
999 * @size: size of the object
1000 * @gfp: __vmalloc() flags used for kmemleak internal memory allocations
1002 * This function is called from the vmalloc() kernel allocator when a new
1003 * object (memory block) is allocated.
1005 void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp)
1007 pr_debug("%s(0x%p, %zu)\n", __func__, area, size);
1010 * A min_count = 2 is needed because vm_struct contains a reference to
1011 * the virtual address of the vmalloc'ed block.
1013 if (kmemleak_enabled) {
1014 create_object((unsigned long)area->addr, size, 2, gfp);
1015 object_set_excess_ref((unsigned long)area,
1016 (unsigned long)area->addr);
1017 } else if (kmemleak_early_log) {
1018 log_early(KMEMLEAK_ALLOC, area->addr, size, 2);
1019 /* reusing early_log.size for storing area->addr */
1020 log_early(KMEMLEAK_SET_EXCESS_REF,
1021 area, (unsigned long)area->addr, 0);
1024 EXPORT_SYMBOL_GPL(kmemleak_vmalloc);
1027 * kmemleak_free - unregister a previously registered object
1028 * @ptr: pointer to beginning of the object
1030 * This function is called from the kernel allocators when an object (memory
1031 * block) is freed (kmem_cache_free, kfree, vfree etc.).
1033 void __ref kmemleak_free(const void *ptr)
1035 pr_debug("%s(0x%p)\n", __func__, ptr);
1037 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
1038 delete_object_full((unsigned long)ptr);
1039 else if (kmemleak_early_log)
1040 log_early(KMEMLEAK_FREE, ptr, 0, 0);
1042 EXPORT_SYMBOL_GPL(kmemleak_free);
1045 * kmemleak_free_part - partially unregister a previously registered object
1046 * @ptr: pointer to the beginning or inside the object. This also
1047 * represents the start of the range to be freed
1048 * @size: size to be unregistered
1050 * This function is called when only a part of a memory block is freed
1051 * (usually from the bootmem allocator).
1053 void __ref kmemleak_free_part(const void *ptr, size_t size)
1055 pr_debug("%s(0x%p)\n", __func__, ptr);
1057 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1058 delete_object_part((unsigned long)ptr, size);
1059 else if (kmemleak_early_log)
1060 log_early(KMEMLEAK_FREE_PART, ptr, size, 0);
1062 EXPORT_SYMBOL_GPL(kmemleak_free_part);
1065 * kmemleak_free_percpu - unregister a previously registered __percpu object
1066 * @ptr: __percpu pointer to beginning of the object
1068 * This function is called from the kernel percpu allocator when an object
1069 * (memory block) is freed (free_percpu).
1071 void __ref kmemleak_free_percpu(const void __percpu *ptr)
1075 pr_debug("%s(0x%p)\n", __func__, ptr);
1077 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
1078 for_each_possible_cpu(cpu)
1079 delete_object_full((unsigned long)per_cpu_ptr(ptr,
1081 else if (kmemleak_early_log)
1082 log_early(KMEMLEAK_FREE_PERCPU, ptr, 0, 0);
1084 EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
1087 * kmemleak_update_trace - update object allocation stack trace
1088 * @ptr: pointer to beginning of the object
1090 * Override the object allocation stack trace for cases where the actual
1091 * allocation place is not always useful.
1093 void __ref kmemleak_update_trace(const void *ptr)
1095 struct kmemleak_object *object;
1096 unsigned long flags;
1098 pr_debug("%s(0x%p)\n", __func__, ptr);
1100 if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
1103 object = find_and_get_object((unsigned long)ptr, 1);
1106 kmemleak_warn("Updating stack trace for unknown object at %p\n",
1112 spin_lock_irqsave(&object->lock, flags);
1113 object->trace_len = __save_stack_trace(object->trace);
1114 spin_unlock_irqrestore(&object->lock, flags);
1118 EXPORT_SYMBOL(kmemleak_update_trace);
1121 * kmemleak_not_leak - mark an allocated object as false positive
1122 * @ptr: pointer to beginning of the object
1124 * Calling this function on an object will cause the memory block to no longer
1125 * be reported as leak and always be scanned.
1127 void __ref kmemleak_not_leak(const void *ptr)
1129 pr_debug("%s(0x%p)\n", __func__, ptr);
1131 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1132 make_gray_object((unsigned long)ptr);
1133 else if (kmemleak_early_log)
1134 log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0);
1136 EXPORT_SYMBOL(kmemleak_not_leak);
1139 * kmemleak_ignore - ignore an allocated object
1140 * @ptr: pointer to beginning of the object
1142 * Calling this function on an object will cause the memory block to be
1143 * ignored (not scanned and not reported as a leak). This is usually done when
1144 * it is known that the corresponding block is not a leak and does not contain
1145 * any references to other allocated memory blocks.
1147 void __ref kmemleak_ignore(const void *ptr)
1149 pr_debug("%s(0x%p)\n", __func__, ptr);
1151 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1152 make_black_object((unsigned long)ptr);
1153 else if (kmemleak_early_log)
1154 log_early(KMEMLEAK_IGNORE, ptr, 0, 0);
1156 EXPORT_SYMBOL(kmemleak_ignore);
1159 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1160 * @ptr: pointer to beginning or inside the object. This also
1161 * represents the start of the scan area
1162 * @size: size of the scan area
1163 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1165 * This function is used when it is known that only certain parts of an object
1166 * contain references to other objects. Kmemleak will only scan these areas
1167 * reducing the number false negatives.
1169 void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1171 pr_debug("%s(0x%p)\n", __func__, ptr);
1173 if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
1174 add_scan_area((unsigned long)ptr, size, gfp);
1175 else if (kmemleak_early_log)
1176 log_early(KMEMLEAK_SCAN_AREA, ptr, size, 0);
1178 EXPORT_SYMBOL(kmemleak_scan_area);
1181 * kmemleak_no_scan - do not scan an allocated object
1182 * @ptr: pointer to beginning of the object
1184 * This function notifies kmemleak not to scan the given memory block. Useful
1185 * in situations where it is known that the given object does not contain any
1186 * references to other objects. Kmemleak will not scan such objects reducing
1187 * the number of false negatives.
1189 void __ref kmemleak_no_scan(const void *ptr)
1191 pr_debug("%s(0x%p)\n", __func__, ptr);
1193 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1194 object_no_scan((unsigned long)ptr);
1195 else if (kmemleak_early_log)
1196 log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0);
1198 EXPORT_SYMBOL(kmemleak_no_scan);
1201 * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical
1203 * @phys: physical address of the object
1204 * @size: size of the object
1205 * @min_count: minimum number of references to this object.
1206 * See kmemleak_alloc()
1207 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1209 void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, int min_count,
1212 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1213 kmemleak_alloc(__va(phys), size, min_count, gfp);
1215 EXPORT_SYMBOL(kmemleak_alloc_phys);
1218 * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a
1219 * physical address argument
1220 * @phys: physical address if the beginning or inside an object. This
1221 * also represents the start of the range to be freed
1222 * @size: size to be unregistered
1224 void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size)
1226 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1227 kmemleak_free_part(__va(phys), size);
1229 EXPORT_SYMBOL(kmemleak_free_part_phys);
1232 * kmemleak_not_leak_phys - similar to kmemleak_not_leak but taking a physical
1234 * @phys: physical address of the object
1236 void __ref kmemleak_not_leak_phys(phys_addr_t phys)
1238 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1239 kmemleak_not_leak(__va(phys));
1241 EXPORT_SYMBOL(kmemleak_not_leak_phys);
1244 * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical
1246 * @phys: physical address of the object
1248 void __ref kmemleak_ignore_phys(phys_addr_t phys)
1250 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1251 kmemleak_ignore(__va(phys));
1253 EXPORT_SYMBOL(kmemleak_ignore_phys);
1256 * Update an object's checksum and return true if it was modified.
1258 static bool update_checksum(struct kmemleak_object *object)
1260 u32 old_csum = object->checksum;
1262 kasan_disable_current();
1263 object->checksum = crc32(0, (void *)object->pointer, object->size);
1264 kasan_enable_current();
1266 return object->checksum != old_csum;
1270 * Update an object's references. object->lock must be held by the caller.
1272 static void update_refs(struct kmemleak_object *object)
1274 if (!color_white(object)) {
1275 /* non-orphan, ignored or new */
1280 * Increase the object's reference count (number of pointers to the
1281 * memory block). If this count reaches the required minimum, the
1282 * object's color will become gray and it will be added to the
1286 if (color_gray(object)) {
1287 /* put_object() called when removing from gray_list */
1288 WARN_ON(!get_object(object));
1289 list_add_tail(&object->gray_list, &gray_list);
1294 * Memory scanning is a long process and it needs to be interruptable. This
1295 * function checks whether such interrupt condition occurred.
1297 static int scan_should_stop(void)
1299 if (!kmemleak_enabled)
1303 * This function may be called from either process or kthread context,
1304 * hence the need to check for both stop conditions.
1307 return signal_pending(current);
1309 return kthread_should_stop();
1315 * Scan a memory block (exclusive range) for valid pointers and add those
1316 * found to the gray list.
1318 static void scan_block(void *_start, void *_end,
1319 struct kmemleak_object *scanned)
1322 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1323 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1324 unsigned long flags;
1325 unsigned long untagged_ptr;
1327 read_lock_irqsave(&kmemleak_lock, flags);
1328 for (ptr = start; ptr < end; ptr++) {
1329 struct kmemleak_object *object;
1330 unsigned long pointer;
1331 unsigned long excess_ref;
1333 if (scan_should_stop())
1336 kasan_disable_current();
1338 kasan_enable_current();
1340 untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer);
1341 if (untagged_ptr < min_addr || untagged_ptr >= max_addr)
1345 * No need for get_object() here since we hold kmemleak_lock.
1346 * object->use_count cannot be dropped to 0 while the object
1347 * is still present in object_tree_root and object_list
1348 * (with updates protected by kmemleak_lock).
1350 object = lookup_object(pointer, 1);
1353 if (object == scanned)
1354 /* self referenced, ignore */
1358 * Avoid the lockdep recursive warning on object->lock being
1359 * previously acquired in scan_object(). These locks are
1360 * enclosed by scan_mutex.
1362 spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1363 /* only pass surplus references (object already gray) */
1364 if (color_gray(object)) {
1365 excess_ref = object->excess_ref;
1366 /* no need for update_refs() if object already gray */
1369 update_refs(object);
1371 spin_unlock(&object->lock);
1374 object = lookup_object(excess_ref, 0);
1377 if (object == scanned)
1378 /* circular reference, ignore */
1380 spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1381 update_refs(object);
1382 spin_unlock(&object->lock);
1385 read_unlock_irqrestore(&kmemleak_lock, flags);
1389 * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
1392 static void scan_large_block(void *start, void *end)
1396 while (start < end) {
1397 next = min(start + MAX_SCAN_SIZE, end);
1398 scan_block(start, next, NULL);
1406 * Scan a memory block corresponding to a kmemleak_object. A condition is
1407 * that object->use_count >= 1.
1409 static void scan_object(struct kmemleak_object *object)
1411 struct kmemleak_scan_area *area;
1412 unsigned long flags;
1415 * Once the object->lock is acquired, the corresponding memory block
1416 * cannot be freed (the same lock is acquired in delete_object).
1418 spin_lock_irqsave(&object->lock, flags);
1419 if (object->flags & OBJECT_NO_SCAN)
1421 if (!(object->flags & OBJECT_ALLOCATED))
1422 /* already freed object */
1424 if (hlist_empty(&object->area_list)) {
1425 void *start = (void *)object->pointer;
1426 void *end = (void *)(object->pointer + object->size);
1430 next = min(start + MAX_SCAN_SIZE, end);
1431 scan_block(start, next, object);
1437 spin_unlock_irqrestore(&object->lock, flags);
1439 spin_lock_irqsave(&object->lock, flags);
1440 } while (object->flags & OBJECT_ALLOCATED);
1442 hlist_for_each_entry(area, &object->area_list, node)
1443 scan_block((void *)area->start,
1444 (void *)(area->start + area->size),
1447 spin_unlock_irqrestore(&object->lock, flags);
1451 * Scan the objects already referenced (gray objects). More objects will be
1452 * referenced and, if there are no memory leaks, all the objects are scanned.
1454 static void scan_gray_list(void)
1456 struct kmemleak_object *object, *tmp;
1459 * The list traversal is safe for both tail additions and removals
1460 * from inside the loop. The kmemleak objects cannot be freed from
1461 * outside the loop because their use_count was incremented.
1463 object = list_entry(gray_list.next, typeof(*object), gray_list);
1464 while (&object->gray_list != &gray_list) {
1467 /* may add new objects to the list */
1468 if (!scan_should_stop())
1469 scan_object(object);
1471 tmp = list_entry(object->gray_list.next, typeof(*object),
1474 /* remove the object from the list and release it */
1475 list_del(&object->gray_list);
1480 WARN_ON(!list_empty(&gray_list));
1484 * Scan data sections and all the referenced memory blocks allocated via the
1485 * kernel's standard allocators. This function must be called with the
1488 static void kmemleak_scan(void)
1490 unsigned long flags;
1491 struct kmemleak_object *object;
1495 jiffies_last_scan = jiffies;
1497 /* prepare the kmemleak_object's */
1499 list_for_each_entry_rcu(object, &object_list, object_list) {
1500 spin_lock_irqsave(&object->lock, flags);
1503 * With a few exceptions there should be a maximum of
1504 * 1 reference to any object at this point.
1506 if (atomic_read(&object->use_count) > 1) {
1507 pr_debug("object->use_count = %d\n",
1508 atomic_read(&object->use_count));
1509 dump_object_info(object);
1512 /* reset the reference count (whiten the object) */
1514 if (color_gray(object) && get_object(object))
1515 list_add_tail(&object->gray_list, &gray_list);
1517 spin_unlock_irqrestore(&object->lock, flags);
1522 /* per-cpu sections scanning */
1523 for_each_possible_cpu(i)
1524 scan_large_block(__per_cpu_start + per_cpu_offset(i),
1525 __per_cpu_end + per_cpu_offset(i));
1529 * Struct page scanning for each node.
1532 for_each_online_node(i) {
1533 unsigned long start_pfn = node_start_pfn(i);
1534 unsigned long end_pfn = node_end_pfn(i);
1537 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1538 struct page *page = pfn_to_online_page(pfn);
1543 /* only scan pages belonging to this node */
1544 if (page_to_nid(page) != i)
1546 /* only scan if page is in use */
1547 if (page_count(page) == 0)
1549 scan_block(page, page + 1, NULL);
1557 * Scanning the task stacks (may introduce false negatives).
1559 if (kmemleak_stack_scan) {
1560 struct task_struct *p, *g;
1562 read_lock(&tasklist_lock);
1563 do_each_thread(g, p) {
1564 void *stack = try_get_task_stack(p);
1566 scan_block(stack, stack + THREAD_SIZE, NULL);
1569 } while_each_thread(g, p);
1570 read_unlock(&tasklist_lock);
1574 * Scan the objects already referenced from the sections scanned
1580 * Check for new or unreferenced objects modified since the previous
1581 * scan and color them gray until the next scan.
1584 list_for_each_entry_rcu(object, &object_list, object_list) {
1585 spin_lock_irqsave(&object->lock, flags);
1586 if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1587 && update_checksum(object) && get_object(object)) {
1588 /* color it gray temporarily */
1589 object->count = object->min_count;
1590 list_add_tail(&object->gray_list, &gray_list);
1592 spin_unlock_irqrestore(&object->lock, flags);
1597 * Re-scan the gray list for modified unreferenced objects.
1602 * If scanning was stopped do not report any new unreferenced objects.
1604 if (scan_should_stop())
1608 * Scanning result reporting.
1611 list_for_each_entry_rcu(object, &object_list, object_list) {
1612 spin_lock_irqsave(&object->lock, flags);
1613 if (unreferenced_object(object) &&
1614 !(object->flags & OBJECT_REPORTED)) {
1615 object->flags |= OBJECT_REPORTED;
1617 if (kmemleak_verbose)
1618 print_unreferenced(NULL, object);
1622 spin_unlock_irqrestore(&object->lock, flags);
1627 kmemleak_found_leaks = true;
1629 pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
1636 * Thread function performing automatic memory scanning. Unreferenced objects
1637 * at the end of a memory scan are reported but only the first time.
1639 static int kmemleak_scan_thread(void *arg)
1641 static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN);
1643 pr_info("Automatic memory scanning thread started\n");
1644 set_user_nice(current, 10);
1647 * Wait before the first scan to allow the system to fully initialize.
1650 signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000);
1652 while (timeout && !kthread_should_stop())
1653 timeout = schedule_timeout_interruptible(timeout);
1656 while (!kthread_should_stop()) {
1657 signed long timeout = jiffies_scan_wait;
1659 mutex_lock(&scan_mutex);
1661 mutex_unlock(&scan_mutex);
1663 /* wait before the next scan */
1664 while (timeout && !kthread_should_stop())
1665 timeout = schedule_timeout_interruptible(timeout);
1668 pr_info("Automatic memory scanning thread ended\n");
1674 * Start the automatic memory scanning thread. This function must be called
1675 * with the scan_mutex held.
1677 static void start_scan_thread(void)
1681 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1682 if (IS_ERR(scan_thread)) {
1683 pr_warn("Failed to create the scan thread\n");
1689 * Stop the automatic memory scanning thread.
1691 static void stop_scan_thread(void)
1694 kthread_stop(scan_thread);
1700 * Iterate over the object_list and return the first valid object at or after
1701 * the required position with its use_count incremented. The function triggers
1702 * a memory scanning when the pos argument points to the first position.
1704 static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1706 struct kmemleak_object *object;
1710 err = mutex_lock_interruptible(&scan_mutex);
1712 return ERR_PTR(err);
1715 list_for_each_entry_rcu(object, &object_list, object_list) {
1718 if (get_object(object))
1727 * Return the next object in the object_list. The function decrements the
1728 * use_count of the previous object and increases that of the next one.
1730 static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1732 struct kmemleak_object *prev_obj = v;
1733 struct kmemleak_object *next_obj = NULL;
1734 struct kmemleak_object *obj = prev_obj;
1738 list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
1739 if (get_object(obj)) {
1745 put_object(prev_obj);
1750 * Decrement the use_count of the last object required, if any.
1752 static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1756 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1757 * waiting was interrupted, so only release it if !IS_ERR.
1760 mutex_unlock(&scan_mutex);
1767 * Print the information for an unreferenced object to the seq file.
1769 static int kmemleak_seq_show(struct seq_file *seq, void *v)
1771 struct kmemleak_object *object = v;
1772 unsigned long flags;
1774 spin_lock_irqsave(&object->lock, flags);
1775 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1776 print_unreferenced(seq, object);
1777 spin_unlock_irqrestore(&object->lock, flags);
1781 static const struct seq_operations kmemleak_seq_ops = {
1782 .start = kmemleak_seq_start,
1783 .next = kmemleak_seq_next,
1784 .stop = kmemleak_seq_stop,
1785 .show = kmemleak_seq_show,
1788 static int kmemleak_open(struct inode *inode, struct file *file)
1790 return seq_open(file, &kmemleak_seq_ops);
1793 static int dump_str_object_info(const char *str)
1795 unsigned long flags;
1796 struct kmemleak_object *object;
1799 if (kstrtoul(str, 0, &addr))
1801 object = find_and_get_object(addr, 0);
1803 pr_info("Unknown object at 0x%08lx\n", addr);
1807 spin_lock_irqsave(&object->lock, flags);
1808 dump_object_info(object);
1809 spin_unlock_irqrestore(&object->lock, flags);
1816 * We use grey instead of black to ensure we can do future scans on the same
1817 * objects. If we did not do future scans these black objects could
1818 * potentially contain references to newly allocated objects in the future and
1819 * we'd end up with false positives.
1821 static void kmemleak_clear(void)
1823 struct kmemleak_object *object;
1824 unsigned long flags;
1827 list_for_each_entry_rcu(object, &object_list, object_list) {
1828 spin_lock_irqsave(&object->lock, flags);
1829 if ((object->flags & OBJECT_REPORTED) &&
1830 unreferenced_object(object))
1831 __paint_it(object, KMEMLEAK_GREY);
1832 spin_unlock_irqrestore(&object->lock, flags);
1836 kmemleak_found_leaks = false;
1839 static void __kmemleak_do_cleanup(void);
1842 * File write operation to configure kmemleak at run-time. The following
1843 * commands can be written to the /sys/kernel/debug/kmemleak file:
1844 * off - disable kmemleak (irreversible)
1845 * stack=on - enable the task stacks scanning
1846 * stack=off - disable the tasks stacks scanning
1847 * scan=on - start the automatic memory scanning thread
1848 * scan=off - stop the automatic memory scanning thread
1849 * scan=... - set the automatic memory scanning period in seconds (0 to
1851 * scan - trigger a memory scan
1852 * clear - mark all current reported unreferenced kmemleak objects as
1853 * grey to ignore printing them, or free all kmemleak objects
1854 * if kmemleak has been disabled.
1855 * dump=... - dump information about the object found at the given address
1857 static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1858 size_t size, loff_t *ppos)
1864 buf_size = min(size, (sizeof(buf) - 1));
1865 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1869 ret = mutex_lock_interruptible(&scan_mutex);
1873 if (strncmp(buf, "clear", 5) == 0) {
1874 if (kmemleak_enabled)
1877 __kmemleak_do_cleanup();
1881 if (!kmemleak_enabled) {
1886 if (strncmp(buf, "off", 3) == 0)
1888 else if (strncmp(buf, "stack=on", 8) == 0)
1889 kmemleak_stack_scan = 1;
1890 else if (strncmp(buf, "stack=off", 9) == 0)
1891 kmemleak_stack_scan = 0;
1892 else if (strncmp(buf, "scan=on", 7) == 0)
1893 start_scan_thread();
1894 else if (strncmp(buf, "scan=off", 8) == 0)
1896 else if (strncmp(buf, "scan=", 5) == 0) {
1899 ret = kstrtoul(buf + 5, 0, &secs);
1904 jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1905 start_scan_thread();
1907 } else if (strncmp(buf, "scan", 4) == 0)
1909 else if (strncmp(buf, "dump=", 5) == 0)
1910 ret = dump_str_object_info(buf + 5);
1915 mutex_unlock(&scan_mutex);
1919 /* ignore the rest of the buffer, only one command at a time */
1924 static const struct file_operations kmemleak_fops = {
1925 .owner = THIS_MODULE,
1926 .open = kmemleak_open,
1928 .write = kmemleak_write,
1929 .llseek = seq_lseek,
1930 .release = seq_release,
1933 static void __kmemleak_do_cleanup(void)
1935 struct kmemleak_object *object;
1938 list_for_each_entry_rcu(object, &object_list, object_list)
1939 delete_object_full(object->pointer);
1944 * Stop the memory scanning thread and free the kmemleak internal objects if
1945 * no previous scan thread (otherwise, kmemleak may still have some useful
1946 * information on memory leaks).
1948 static void kmemleak_do_cleanup(struct work_struct *work)
1952 mutex_lock(&scan_mutex);
1954 * Once it is made sure that kmemleak_scan has stopped, it is safe to no
1955 * longer track object freeing. Ordering of the scan thread stopping and
1956 * the memory accesses below is guaranteed by the kthread_stop()
1959 kmemleak_free_enabled = 0;
1960 mutex_unlock(&scan_mutex);
1962 if (!kmemleak_found_leaks)
1963 __kmemleak_do_cleanup();
1965 pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
1968 static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
1971 * Disable kmemleak. No memory allocation/freeing will be traced once this
1972 * function is called. Disabling kmemleak is an irreversible operation.
1974 static void kmemleak_disable(void)
1976 /* atomically check whether it was already invoked */
1977 if (cmpxchg(&kmemleak_error, 0, 1))
1980 /* stop any memory operation tracing */
1981 kmemleak_enabled = 0;
1983 /* check whether it is too early for a kernel thread */
1984 if (kmemleak_initialized)
1985 schedule_work(&cleanup_work);
1987 kmemleak_free_enabled = 0;
1989 pr_info("Kernel memory leak detector disabled\n");
1993 * Allow boot-time kmemleak disabling (enabled by default).
1995 static int __init kmemleak_boot_config(char *str)
1999 if (strcmp(str, "off") == 0)
2001 else if (strcmp(str, "on") == 0)
2002 kmemleak_skip_disable = 1;
2007 early_param("kmemleak", kmemleak_boot_config);
2009 static void __init print_log_trace(struct early_log *log)
2011 pr_notice("Early log backtrace:\n");
2012 stack_trace_print(log->trace, log->trace_len, 2);
2016 * Kmemleak initialization.
2018 void __init kmemleak_init(void)
2021 unsigned long flags;
2023 #ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
2024 if (!kmemleak_skip_disable) {
2025 kmemleak_early_log = 0;
2031 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
2032 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
2034 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
2035 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
2037 if (crt_early_log > ARRAY_SIZE(early_log))
2038 pr_warn("Early log buffer exceeded (%d), please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n",
2041 /* the kernel is still in UP mode, so disabling the IRQs is enough */
2042 local_irq_save(flags);
2043 kmemleak_early_log = 0;
2044 if (kmemleak_error) {
2045 local_irq_restore(flags);
2048 kmemleak_enabled = 1;
2049 kmemleak_free_enabled = 1;
2051 local_irq_restore(flags);
2053 /* register the data/bss sections */
2054 create_object((unsigned long)_sdata, _edata - _sdata,
2055 KMEMLEAK_GREY, GFP_ATOMIC);
2056 create_object((unsigned long)__bss_start, __bss_stop - __bss_start,
2057 KMEMLEAK_GREY, GFP_ATOMIC);
2058 /* only register .data..ro_after_init if not within .data */
2059 if (__start_ro_after_init < _sdata || __end_ro_after_init > _edata)
2060 create_object((unsigned long)__start_ro_after_init,
2061 __end_ro_after_init - __start_ro_after_init,
2062 KMEMLEAK_GREY, GFP_ATOMIC);
2065 * This is the point where tracking allocations is safe. Automatic
2066 * scanning is started during the late initcall. Add the early logged
2067 * callbacks to the kmemleak infrastructure.
2069 for (i = 0; i < crt_early_log; i++) {
2070 struct early_log *log = &early_log[i];
2072 switch (log->op_type) {
2073 case KMEMLEAK_ALLOC:
2076 case KMEMLEAK_ALLOC_PERCPU:
2077 early_alloc_percpu(log);
2080 kmemleak_free(log->ptr);
2082 case KMEMLEAK_FREE_PART:
2083 kmemleak_free_part(log->ptr, log->size);
2085 case KMEMLEAK_FREE_PERCPU:
2086 kmemleak_free_percpu(log->ptr);
2088 case KMEMLEAK_NOT_LEAK:
2089 kmemleak_not_leak(log->ptr);
2091 case KMEMLEAK_IGNORE:
2092 kmemleak_ignore(log->ptr);
2094 case KMEMLEAK_SCAN_AREA:
2095 kmemleak_scan_area(log->ptr, log->size, GFP_KERNEL);
2097 case KMEMLEAK_NO_SCAN:
2098 kmemleak_no_scan(log->ptr);
2100 case KMEMLEAK_SET_EXCESS_REF:
2101 object_set_excess_ref((unsigned long)log->ptr,
2105 kmemleak_warn("Unknown early log operation: %d\n",
2109 if (kmemleak_warning) {
2110 print_log_trace(log);
2111 kmemleak_warning = 0;
2117 * Late initialization function.
2119 static int __init kmemleak_late_init(void)
2121 struct dentry *dentry;
2123 kmemleak_initialized = 1;
2125 dentry = debugfs_create_file("kmemleak", 0644, NULL, NULL,
2128 pr_warn("Failed to create the debugfs kmemleak file\n");
2130 if (kmemleak_error) {
2132 * Some error occurred and kmemleak was disabled. There is a
2133 * small chance that kmemleak_disable() was called immediately
2134 * after setting kmemleak_initialized and we may end up with
2135 * two clean-up threads but serialized by scan_mutex.
2137 schedule_work(&cleanup_work);
2141 if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) {
2142 mutex_lock(&scan_mutex);
2143 start_scan_thread();
2144 mutex_unlock(&scan_mutex);
2147 pr_info("Kernel memory leak detector initialized\n");
2151 late_initcall(kmemleak_late_init);