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 as well as
17 * del_state modifications and accesses to the object_tree_root (or
18 * object_phys_tree_root). The object_list is the main list holding the
19 * metadata (struct kmemleak_object) for the allocated memory blocks.
20 * The object_tree_root and object_phys_tree_root are red
21 * black trees used to look-up metadata based on a pointer to the
22 * corresponding memory block. The object_phys_tree_root is for objects
23 * allocated with physical address. The kmemleak_object structures are
24 * added to the object_list and object_tree_root (or object_phys_tree_root)
25 * in the create_object() function called from the kmemleak_alloc() (or
26 * kmemleak_alloc_phys()) callback and removed in delete_object() called from
27 * the kmemleak_free() callback
28 * - kmemleak_object.lock (raw_spinlock_t): protects a kmemleak_object.
29 * Accesses to the metadata (e.g. count) are protected by this lock. Note
30 * that some members of this structure may be protected by other means
31 * (atomic or kmemleak_lock). This lock is also held when scanning the
32 * corresponding memory block to avoid the kernel freeing it via the
33 * kmemleak_free() callback. This is less heavyweight than holding a global
34 * lock like kmemleak_lock during scanning.
35 * - scan_mutex (mutex): ensures that only one thread may scan the memory for
36 * unreferenced objects at a time. The gray_list contains the objects which
37 * are already referenced or marked as false positives and need to be
38 * scanned. This list is only modified during a scanning episode when the
39 * scan_mutex is held. At the end of a scan, the gray_list is always empty.
40 * Note that the kmemleak_object.use_count is incremented when an object is
41 * added to the gray_list and therefore cannot be freed. This mutex also
42 * prevents multiple users of the "kmemleak" debugfs file together with
43 * modifications to the memory scanning parameters including the scan_thread
46 * Locks and mutexes are acquired/nested in the following order:
48 * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING)
50 * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex
53 * The kmemleak_object structures have a use_count incremented or decremented
54 * using the get_object()/put_object() functions. When the use_count becomes
55 * 0, this count can no longer be incremented and put_object() schedules the
56 * kmemleak_object freeing via an RCU callback. All calls to the get_object()
57 * function must be protected by rcu_read_lock() to avoid accessing a freed
61 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
63 #include <linux/init.h>
64 #include <linux/kernel.h>
65 #include <linux/list.h>
66 #include <linux/sched/signal.h>
67 #include <linux/sched/task.h>
68 #include <linux/sched/task_stack.h>
69 #include <linux/jiffies.h>
70 #include <linux/delay.h>
71 #include <linux/export.h>
72 #include <linux/kthread.h>
73 #include <linux/rbtree.h>
75 #include <linux/debugfs.h>
76 #include <linux/seq_file.h>
77 #include <linux/cpumask.h>
78 #include <linux/spinlock.h>
79 #include <linux/module.h>
80 #include <linux/mutex.h>
81 #include <linux/rcupdate.h>
82 #include <linux/stacktrace.h>
83 #include <linux/stackdepot.h>
84 #include <linux/cache.h>
85 #include <linux/percpu.h>
86 #include <linux/memblock.h>
87 #include <linux/pfn.h>
88 #include <linux/mmzone.h>
89 #include <linux/slab.h>
90 #include <linux/thread_info.h>
91 #include <linux/err.h>
92 #include <linux/uaccess.h>
93 #include <linux/string.h>
94 #include <linux/nodemask.h>
96 #include <linux/workqueue.h>
97 #include <linux/crc32.h>
99 #include <asm/sections.h>
100 #include <asm/processor.h>
101 #include <linux/atomic.h>
103 #include <linux/kasan.h>
104 #include <linux/kfence.h>
105 #include <linux/kmemleak.h>
106 #include <linux/memory_hotplug.h>
109 * Kmemleak configuration and common defines.
111 #define MAX_TRACE 16 /* stack trace length */
112 #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
113 #define SECS_FIRST_SCAN 60 /* delay before the first scan */
114 #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
115 #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
117 #define BYTES_PER_POINTER sizeof(void *)
119 /* GFP bitmask for kmemleak internal allocations */
120 #define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC | \
121 __GFP_NOLOCKDEP)) | \
122 __GFP_NORETRY | __GFP_NOMEMALLOC | \
125 /* scanning area inside a memory block */
126 struct kmemleak_scan_area
{
127 struct hlist_node node
;
132 #define KMEMLEAK_GREY 0
133 #define KMEMLEAK_BLACK -1
136 * Structure holding the metadata for each allocated memory block.
137 * Modifications to such objects should be made while holding the
138 * object->lock. Insertions or deletions from object_list, gray_list or
139 * rb_node are already protected by the corresponding locks or mutex (see
140 * the notes on locking above). These objects are reference-counted
141 * (use_count) and freed using the RCU mechanism.
143 struct kmemleak_object
{
145 unsigned int flags
; /* object status flags */
146 struct list_head object_list
;
147 struct list_head gray_list
;
148 struct rb_node rb_node
;
149 struct rcu_head rcu
; /* object_list lockless traversal */
150 /* object usage count; object freed when use_count == 0 */
152 unsigned int del_state
; /* deletion state */
153 unsigned long pointer
;
155 /* pass surplus references to this pointer */
156 unsigned long excess_ref
;
157 /* minimum number of a pointers found before it is considered leak */
159 /* the total number of pointers found pointing to this object */
161 /* checksum for detecting modified objects */
163 /* memory ranges to be scanned inside an object (empty for all) */
164 struct hlist_head area_list
;
165 depot_stack_handle_t trace_handle
;
166 unsigned long jiffies
; /* creation timestamp */
167 pid_t pid
; /* pid of the current task */
168 char comm
[TASK_COMM_LEN
]; /* executable name */
171 /* flag representing the memory block allocation status */
172 #define OBJECT_ALLOCATED (1 << 0)
173 /* flag set after the first reporting of an unreference object */
174 #define OBJECT_REPORTED (1 << 1)
175 /* flag set to not scan the object */
176 #define OBJECT_NO_SCAN (1 << 2)
177 /* flag set to fully scan the object when scan_area allocation failed */
178 #define OBJECT_FULL_SCAN (1 << 3)
179 /* flag set for object allocated with physical address */
180 #define OBJECT_PHYS (1 << 4)
182 /* set when __remove_object() called */
183 #define DELSTATE_REMOVED (1 << 0)
184 /* set to temporarily prevent deletion from object_list */
185 #define DELSTATE_NO_DELETE (1 << 1)
187 #define HEX_PREFIX " "
188 /* number of bytes to print per line; must be 16 or 32 */
189 #define HEX_ROW_SIZE 16
190 /* number of bytes to print at a time (1, 2, 4, 8) */
191 #define HEX_GROUP_SIZE 1
192 /* include ASCII after the hex output */
194 /* max number of lines to be printed */
195 #define HEX_MAX_LINES 2
197 /* the list of all allocated objects */
198 static LIST_HEAD(object_list
);
199 /* the list of gray-colored objects (see color_gray comment below) */
200 static LIST_HEAD(gray_list
);
201 /* memory pool allocation */
202 static struct kmemleak_object mem_pool
[CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE
];
203 static int mem_pool_free_count
= ARRAY_SIZE(mem_pool
);
204 static LIST_HEAD(mem_pool_free_list
);
205 /* search tree for object boundaries */
206 static struct rb_root object_tree_root
= RB_ROOT
;
207 /* search tree for object (with OBJECT_PHYS flag) boundaries */
208 static struct rb_root object_phys_tree_root
= RB_ROOT
;
209 /* protecting the access to object_list, object_tree_root (or object_phys_tree_root) */
210 static DEFINE_RAW_SPINLOCK(kmemleak_lock
);
212 /* allocation caches for kmemleak internal data */
213 static struct kmem_cache
*object_cache
;
214 static struct kmem_cache
*scan_area_cache
;
216 /* set if tracing memory operations is enabled */
217 static int kmemleak_enabled
= 1;
218 /* same as above but only for the kmemleak_free() callback */
219 static int kmemleak_free_enabled
= 1;
220 /* set in the late_initcall if there were no errors */
221 static int kmemleak_late_initialized
;
222 /* set if a kmemleak warning was issued */
223 static int kmemleak_warning
;
224 /* set if a fatal kmemleak error has occurred */
225 static int kmemleak_error
;
227 /* minimum and maximum address that may be valid pointers */
228 static unsigned long min_addr
= ULONG_MAX
;
229 static unsigned long max_addr
;
231 static struct task_struct
*scan_thread
;
232 /* used to avoid reporting of recently allocated objects */
233 static unsigned long jiffies_min_age
;
234 static unsigned long jiffies_last_scan
;
235 /* delay between automatic memory scannings */
236 static unsigned long jiffies_scan_wait
;
237 /* enables or disables the task stacks scanning */
238 static int kmemleak_stack_scan
= 1;
239 /* protects the memory scanning, parameters and debug/kmemleak file access */
240 static DEFINE_MUTEX(scan_mutex
);
241 /* setting kmemleak=on, will set this var, skipping the disable */
242 static int kmemleak_skip_disable
;
243 /* If there are leaks that can be reported */
244 static bool kmemleak_found_leaks
;
246 static bool kmemleak_verbose
;
247 module_param_named(verbose
, kmemleak_verbose
, bool, 0600);
249 static void kmemleak_disable(void);
252 * Print a warning and dump the stack trace.
254 #define kmemleak_warn(x...) do { \
257 kmemleak_warning = 1; \
261 * Macro invoked when a serious kmemleak condition occurred and cannot be
262 * recovered from. Kmemleak will be disabled and further allocation/freeing
263 * tracing no longer available.
265 #define kmemleak_stop(x...) do { \
267 kmemleak_disable(); \
270 #define warn_or_seq_printf(seq, fmt, ...) do { \
272 seq_printf(seq, fmt, ##__VA_ARGS__); \
274 pr_warn(fmt, ##__VA_ARGS__); \
277 static void warn_or_seq_hex_dump(struct seq_file
*seq
, int prefix_type
,
278 int rowsize
, int groupsize
, const void *buf
,
279 size_t len
, bool ascii
)
282 seq_hex_dump(seq
, HEX_PREFIX
, prefix_type
, rowsize
, groupsize
,
285 print_hex_dump(KERN_WARNING
, pr_fmt(HEX_PREFIX
), prefix_type
,
286 rowsize
, groupsize
, buf
, len
, ascii
);
290 * Printing of the objects hex dump to the seq file. The number of lines to be
291 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
292 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
293 * with the object->lock held.
295 static void hex_dump_object(struct seq_file
*seq
,
296 struct kmemleak_object
*object
)
298 const u8
*ptr
= (const u8
*)object
->pointer
;
301 if (WARN_ON_ONCE(object
->flags
& OBJECT_PHYS
))
304 /* limit the number of lines to HEX_MAX_LINES */
305 len
= min_t(size_t, object
->size
, HEX_MAX_LINES
* HEX_ROW_SIZE
);
307 warn_or_seq_printf(seq
, " hex dump (first %zu bytes):\n", len
);
308 kasan_disable_current();
309 warn_or_seq_hex_dump(seq
, DUMP_PREFIX_NONE
, HEX_ROW_SIZE
,
310 HEX_GROUP_SIZE
, kasan_reset_tag((void *)ptr
), len
, HEX_ASCII
);
311 kasan_enable_current();
315 * Object colors, encoded with count and min_count:
316 * - white - orphan object, not enough references to it (count < min_count)
317 * - gray - not orphan, not marked as false positive (min_count == 0) or
318 * sufficient references to it (count >= min_count)
319 * - black - ignore, it doesn't contain references (e.g. text section)
320 * (min_count == -1). No function defined for this color.
321 * Newly created objects don't have any color assigned (object->count == -1)
322 * before the next memory scan when they become white.
324 static bool color_white(const struct kmemleak_object
*object
)
326 return object
->count
!= KMEMLEAK_BLACK
&&
327 object
->count
< object
->min_count
;
330 static bool color_gray(const struct kmemleak_object
*object
)
332 return object
->min_count
!= KMEMLEAK_BLACK
&&
333 object
->count
>= object
->min_count
;
337 * Objects are considered unreferenced only if their color is white, they have
338 * not be deleted and have a minimum age to avoid false positives caused by
339 * pointers temporarily stored in CPU registers.
341 static bool unreferenced_object(struct kmemleak_object
*object
)
343 return (color_white(object
) && object
->flags
& OBJECT_ALLOCATED
) &&
344 time_before_eq(object
->jiffies
+ jiffies_min_age
,
349 * Printing of the unreferenced objects information to the seq file. The
350 * print_unreferenced function must be called with the object->lock held.
352 static void print_unreferenced(struct seq_file
*seq
,
353 struct kmemleak_object
*object
)
356 unsigned long *entries
;
357 unsigned int nr_entries
;
358 unsigned int msecs_age
= jiffies_to_msecs(jiffies
- object
->jiffies
);
360 nr_entries
= stack_depot_fetch(object
->trace_handle
, &entries
);
361 warn_or_seq_printf(seq
, "unreferenced object 0x%08lx (size %zu):\n",
362 object
->pointer
, object
->size
);
363 warn_or_seq_printf(seq
, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
364 object
->comm
, object
->pid
, object
->jiffies
,
365 msecs_age
/ 1000, msecs_age
% 1000);
366 hex_dump_object(seq
, object
);
367 warn_or_seq_printf(seq
, " backtrace:\n");
369 for (i
= 0; i
< nr_entries
; i
++) {
370 void *ptr
= (void *)entries
[i
];
371 warn_or_seq_printf(seq
, " [<%pK>] %pS\n", ptr
, ptr
);
376 * Print the kmemleak_object information. This function is used mainly for
377 * debugging special cases when kmemleak operations. It must be called with
378 * the object->lock held.
380 static void dump_object_info(struct kmemleak_object
*object
)
382 pr_notice("Object 0x%08lx (size %zu):\n",
383 object
->pointer
, object
->size
);
384 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
385 object
->comm
, object
->pid
, object
->jiffies
);
386 pr_notice(" min_count = %d\n", object
->min_count
);
387 pr_notice(" count = %d\n", object
->count
);
388 pr_notice(" flags = 0x%x\n", object
->flags
);
389 pr_notice(" checksum = %u\n", object
->checksum
);
390 pr_notice(" backtrace:\n");
391 if (object
->trace_handle
)
392 stack_depot_print(object
->trace_handle
);
396 * Look-up a memory block metadata (kmemleak_object) in the object search
397 * tree based on a pointer value. If alias is 0, only values pointing to the
398 * beginning of the memory block are allowed. The kmemleak_lock must be held
399 * when calling this function.
401 static struct kmemleak_object
*__lookup_object(unsigned long ptr
, int alias
,
404 struct rb_node
*rb
= is_phys
? object_phys_tree_root
.rb_node
:
405 object_tree_root
.rb_node
;
406 unsigned long untagged_ptr
= (unsigned long)kasan_reset_tag((void *)ptr
);
409 struct kmemleak_object
*object
;
410 unsigned long untagged_objp
;
412 object
= rb_entry(rb
, struct kmemleak_object
, rb_node
);
413 untagged_objp
= (unsigned long)kasan_reset_tag((void *)object
->pointer
);
415 if (untagged_ptr
< untagged_objp
)
416 rb
= object
->rb_node
.rb_left
;
417 else if (untagged_objp
+ object
->size
<= untagged_ptr
)
418 rb
= object
->rb_node
.rb_right
;
419 else if (untagged_objp
== untagged_ptr
|| alias
)
422 kmemleak_warn("Found object by alias at 0x%08lx\n",
424 dump_object_info(object
);
431 /* Look-up a kmemleak object which allocated with virtual address. */
432 static struct kmemleak_object
*lookup_object(unsigned long ptr
, int alias
)
434 return __lookup_object(ptr
, alias
, false);
438 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
439 * that once an object's use_count reached 0, the RCU freeing was already
440 * registered and the object should no longer be used. This function must be
441 * called under the protection of rcu_read_lock().
443 static int get_object(struct kmemleak_object
*object
)
445 return atomic_inc_not_zero(&object
->use_count
);
449 * Memory pool allocation and freeing. kmemleak_lock must not be held.
451 static struct kmemleak_object
*mem_pool_alloc(gfp_t gfp
)
454 struct kmemleak_object
*object
;
456 /* try the slab allocator first */
458 object
= kmem_cache_alloc(object_cache
, gfp_kmemleak_mask(gfp
));
463 /* slab allocation failed, try the memory pool */
464 raw_spin_lock_irqsave(&kmemleak_lock
, flags
);
465 object
= list_first_entry_or_null(&mem_pool_free_list
,
466 typeof(*object
), object_list
);
468 list_del(&object
->object_list
);
469 else if (mem_pool_free_count
)
470 object
= &mem_pool
[--mem_pool_free_count
];
472 pr_warn_once("Memory pool empty, consider increasing CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE\n");
473 raw_spin_unlock_irqrestore(&kmemleak_lock
, flags
);
479 * Return the object to either the slab allocator or the memory pool.
481 static void mem_pool_free(struct kmemleak_object
*object
)
485 if (object
< mem_pool
|| object
>= mem_pool
+ ARRAY_SIZE(mem_pool
)) {
486 kmem_cache_free(object_cache
, object
);
490 /* add the object to the memory pool free list */
491 raw_spin_lock_irqsave(&kmemleak_lock
, flags
);
492 list_add(&object
->object_list
, &mem_pool_free_list
);
493 raw_spin_unlock_irqrestore(&kmemleak_lock
, flags
);
497 * RCU callback to free a kmemleak_object.
499 static void free_object_rcu(struct rcu_head
*rcu
)
501 struct hlist_node
*tmp
;
502 struct kmemleak_scan_area
*area
;
503 struct kmemleak_object
*object
=
504 container_of(rcu
, struct kmemleak_object
, rcu
);
507 * Once use_count is 0 (guaranteed by put_object), there is no other
508 * code accessing this object, hence no need for locking.
510 hlist_for_each_entry_safe(area
, tmp
, &object
->area_list
, node
) {
511 hlist_del(&area
->node
);
512 kmem_cache_free(scan_area_cache
, area
);
514 mem_pool_free(object
);
518 * Decrement the object use_count. Once the count is 0, free the object using
519 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
520 * delete_object() path, the delayed RCU freeing ensures that there is no
521 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
524 static void put_object(struct kmemleak_object
*object
)
526 if (!atomic_dec_and_test(&object
->use_count
))
529 /* should only get here after delete_object was called */
530 WARN_ON(object
->flags
& OBJECT_ALLOCATED
);
533 * It may be too early for the RCU callbacks, however, there is no
534 * concurrent object_list traversal when !object_cache and all objects
535 * came from the memory pool. Free the object directly.
538 call_rcu(&object
->rcu
, free_object_rcu
);
540 free_object_rcu(&object
->rcu
);
544 * Look up an object in the object search tree and increase its use_count.
546 static struct kmemleak_object
*__find_and_get_object(unsigned long ptr
, int alias
,
550 struct kmemleak_object
*object
;
553 raw_spin_lock_irqsave(&kmemleak_lock
, flags
);
554 object
= __lookup_object(ptr
, alias
, is_phys
);
555 raw_spin_unlock_irqrestore(&kmemleak_lock
, flags
);
557 /* check whether the object is still available */
558 if (object
&& !get_object(object
))
565 /* Look up and get an object which allocated with virtual address. */
566 static struct kmemleak_object
*find_and_get_object(unsigned long ptr
, int alias
)
568 return __find_and_get_object(ptr
, alias
, false);
572 * Remove an object from the object_tree_root (or object_phys_tree_root)
573 * and object_list. Must be called with the kmemleak_lock held _if_ kmemleak
576 static void __remove_object(struct kmemleak_object
*object
)
578 rb_erase(&object
->rb_node
, object
->flags
& OBJECT_PHYS
?
579 &object_phys_tree_root
:
581 if (!(object
->del_state
& DELSTATE_NO_DELETE
))
582 list_del_rcu(&object
->object_list
);
583 object
->del_state
|= DELSTATE_REMOVED
;
587 * Look up an object in the object search tree and remove it from both
588 * object_tree_root (or object_phys_tree_root) and object_list. The
589 * returned object's use_count should be at least 1, as initially set
590 * by create_object().
592 static struct kmemleak_object
*find_and_remove_object(unsigned long ptr
, int alias
,
596 struct kmemleak_object
*object
;
598 raw_spin_lock_irqsave(&kmemleak_lock
, flags
);
599 object
= __lookup_object(ptr
, alias
, is_phys
);
601 __remove_object(object
);
602 raw_spin_unlock_irqrestore(&kmemleak_lock
, flags
);
607 static noinline depot_stack_handle_t
set_track_prepare(void)
609 depot_stack_handle_t trace_handle
;
610 unsigned long entries
[MAX_TRACE
];
611 unsigned int nr_entries
;
614 * Use object_cache to determine whether kmemleak_init() has
615 * been invoked. stack_depot_early_init() is called before
616 * kmemleak_init() in mm_core_init().
620 nr_entries
= stack_trace_save(entries
, ARRAY_SIZE(entries
), 3);
621 trace_handle
= stack_depot_save(entries
, nr_entries
, GFP_NOWAIT
);
627 * Create the metadata (struct kmemleak_object) corresponding to an allocated
628 * memory block and add it to the object_list and object_tree_root (or
629 * object_phys_tree_root).
631 static void __create_object(unsigned long ptr
, size_t size
,
632 int min_count
, gfp_t gfp
, bool is_phys
)
635 struct kmemleak_object
*object
, *parent
;
636 struct rb_node
**link
, *rb_parent
;
637 unsigned long untagged_ptr
;
638 unsigned long untagged_objp
;
640 object
= mem_pool_alloc(gfp
);
642 pr_warn("Cannot allocate a kmemleak_object structure\n");
647 INIT_LIST_HEAD(&object
->object_list
);
648 INIT_LIST_HEAD(&object
->gray_list
);
649 INIT_HLIST_HEAD(&object
->area_list
);
650 raw_spin_lock_init(&object
->lock
);
651 atomic_set(&object
->use_count
, 1);
652 object
->flags
= OBJECT_ALLOCATED
| (is_phys
? OBJECT_PHYS
: 0);
653 object
->pointer
= ptr
;
654 object
->size
= kfence_ksize((void *)ptr
) ?: size
;
655 object
->excess_ref
= 0;
656 object
->min_count
= min_count
;
657 object
->count
= 0; /* white color initially */
658 object
->jiffies
= jiffies
;
659 object
->checksum
= 0;
660 object
->del_state
= 0;
662 /* task information */
665 strncpy(object
->comm
, "hardirq", sizeof(object
->comm
));
666 } else if (in_serving_softirq()) {
668 strncpy(object
->comm
, "softirq", sizeof(object
->comm
));
670 object
->pid
= current
->pid
;
672 * There is a small chance of a race with set_task_comm(),
673 * however using get_task_comm() here may cause locking
674 * dependency issues with current->alloc_lock. In the worst
675 * case, the command line is not correct.
677 strncpy(object
->comm
, current
->comm
, sizeof(object
->comm
));
680 /* kernel backtrace */
681 object
->trace_handle
= set_track_prepare();
683 raw_spin_lock_irqsave(&kmemleak_lock
, flags
);
685 untagged_ptr
= (unsigned long)kasan_reset_tag((void *)ptr
);
687 * Only update min_addr and max_addr with object
688 * storing virtual address.
691 min_addr
= min(min_addr
, untagged_ptr
);
692 max_addr
= max(max_addr
, untagged_ptr
+ size
);
694 link
= is_phys
? &object_phys_tree_root
.rb_node
:
695 &object_tree_root
.rb_node
;
699 parent
= rb_entry(rb_parent
, struct kmemleak_object
, rb_node
);
700 untagged_objp
= (unsigned long)kasan_reset_tag((void *)parent
->pointer
);
701 if (untagged_ptr
+ size
<= untagged_objp
)
702 link
= &parent
->rb_node
.rb_left
;
703 else if (untagged_objp
+ parent
->size
<= untagged_ptr
)
704 link
= &parent
->rb_node
.rb_right
;
706 kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
709 * No need for parent->lock here since "parent" cannot
710 * be freed while the kmemleak_lock is held.
712 dump_object_info(parent
);
713 kmem_cache_free(object_cache
, object
);
717 rb_link_node(&object
->rb_node
, rb_parent
, link
);
718 rb_insert_color(&object
->rb_node
, is_phys
? &object_phys_tree_root
:
720 list_add_tail_rcu(&object
->object_list
, &object_list
);
722 raw_spin_unlock_irqrestore(&kmemleak_lock
, flags
);
725 /* Create kmemleak object which allocated with virtual address. */
726 static void create_object(unsigned long ptr
, size_t size
,
727 int min_count
, gfp_t gfp
)
729 __create_object(ptr
, size
, min_count
, gfp
, false);
732 /* Create kmemleak object which allocated with physical address. */
733 static void create_object_phys(unsigned long ptr
, size_t size
,
734 int min_count
, gfp_t gfp
)
736 __create_object(ptr
, size
, min_count
, gfp
, true);
740 * Mark the object as not allocated and schedule RCU freeing via put_object().
742 static void __delete_object(struct kmemleak_object
*object
)
746 WARN_ON(!(object
->flags
& OBJECT_ALLOCATED
));
747 WARN_ON(atomic_read(&object
->use_count
) < 1);
750 * Locking here also ensures that the corresponding memory block
751 * cannot be freed when it is being scanned.
753 raw_spin_lock_irqsave(&object
->lock
, flags
);
754 object
->flags
&= ~OBJECT_ALLOCATED
;
755 raw_spin_unlock_irqrestore(&object
->lock
, flags
);
760 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
763 static void delete_object_full(unsigned long ptr
)
765 struct kmemleak_object
*object
;
767 object
= find_and_remove_object(ptr
, 0, false);
770 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
775 __delete_object(object
);
779 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
780 * delete it. If the memory block is partially freed, the function may create
781 * additional metadata for the remaining parts of the block.
783 static void delete_object_part(unsigned long ptr
, size_t size
, bool is_phys
)
785 struct kmemleak_object
*object
;
786 unsigned long start
, end
;
788 object
= find_and_remove_object(ptr
, 1, is_phys
);
791 kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
798 * Create one or two objects that may result from the memory block
799 * split. Note that partial freeing is only done by free_bootmem() and
800 * this happens before kmemleak_init() is called.
802 start
= object
->pointer
;
803 end
= object
->pointer
+ object
->size
;
805 __create_object(start
, ptr
- start
, object
->min_count
,
806 GFP_KERNEL
, is_phys
);
807 if (ptr
+ size
< end
)
808 __create_object(ptr
+ size
, end
- ptr
- size
, object
->min_count
,
809 GFP_KERNEL
, is_phys
);
811 __delete_object(object
);
814 static void __paint_it(struct kmemleak_object
*object
, int color
)
816 object
->min_count
= color
;
817 if (color
== KMEMLEAK_BLACK
)
818 object
->flags
|= OBJECT_NO_SCAN
;
821 static void paint_it(struct kmemleak_object
*object
, int color
)
825 raw_spin_lock_irqsave(&object
->lock
, flags
);
826 __paint_it(object
, color
);
827 raw_spin_unlock_irqrestore(&object
->lock
, flags
);
830 static void paint_ptr(unsigned long ptr
, int color
, bool is_phys
)
832 struct kmemleak_object
*object
;
834 object
= __find_and_get_object(ptr
, 0, is_phys
);
836 kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n",
838 (color
== KMEMLEAK_GREY
) ? "Grey" :
839 (color
== KMEMLEAK_BLACK
) ? "Black" : "Unknown");
842 paint_it(object
, color
);
847 * Mark an object permanently as gray-colored so that it can no longer be
848 * reported as a leak. This is used in general to mark a false positive.
850 static void make_gray_object(unsigned long ptr
)
852 paint_ptr(ptr
, KMEMLEAK_GREY
, false);
856 * Mark the object as black-colored so that it is ignored from scans and
859 static void make_black_object(unsigned long ptr
, bool is_phys
)
861 paint_ptr(ptr
, KMEMLEAK_BLACK
, is_phys
);
865 * Add a scanning area to the object. If at least one such area is added,
866 * kmemleak will only scan these ranges rather than the whole memory block.
868 static void add_scan_area(unsigned long ptr
, size_t size
, gfp_t gfp
)
871 struct kmemleak_object
*object
;
872 struct kmemleak_scan_area
*area
= NULL
;
873 unsigned long untagged_ptr
;
874 unsigned long untagged_objp
;
876 object
= find_and_get_object(ptr
, 1);
878 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
883 untagged_ptr
= (unsigned long)kasan_reset_tag((void *)ptr
);
884 untagged_objp
= (unsigned long)kasan_reset_tag((void *)object
->pointer
);
887 area
= kmem_cache_alloc(scan_area_cache
, gfp_kmemleak_mask(gfp
));
889 raw_spin_lock_irqsave(&object
->lock
, flags
);
891 pr_warn_once("Cannot allocate a scan area, scanning the full object\n");
892 /* mark the object for full scan to avoid false positives */
893 object
->flags
|= OBJECT_FULL_SCAN
;
896 if (size
== SIZE_MAX
) {
897 size
= untagged_objp
+ object
->size
- untagged_ptr
;
898 } else if (untagged_ptr
+ size
> untagged_objp
+ object
->size
) {
899 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr
);
900 dump_object_info(object
);
901 kmem_cache_free(scan_area_cache
, area
);
905 INIT_HLIST_NODE(&area
->node
);
909 hlist_add_head(&area
->node
, &object
->area_list
);
911 raw_spin_unlock_irqrestore(&object
->lock
, flags
);
916 * Any surplus references (object already gray) to 'ptr' are passed to
917 * 'excess_ref'. This is used in the vmalloc() case where a pointer to
918 * vm_struct may be used as an alternative reference to the vmalloc'ed object
919 * (see free_thread_stack()).
921 static void object_set_excess_ref(unsigned long ptr
, unsigned long excess_ref
)
924 struct kmemleak_object
*object
;
926 object
= find_and_get_object(ptr
, 0);
928 kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n",
933 raw_spin_lock_irqsave(&object
->lock
, flags
);
934 object
->excess_ref
= excess_ref
;
935 raw_spin_unlock_irqrestore(&object
->lock
, flags
);
940 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
941 * pointer. Such object will not be scanned by kmemleak but references to it
944 static void object_no_scan(unsigned long ptr
)
947 struct kmemleak_object
*object
;
949 object
= find_and_get_object(ptr
, 0);
951 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr
);
955 raw_spin_lock_irqsave(&object
->lock
, flags
);
956 object
->flags
|= OBJECT_NO_SCAN
;
957 raw_spin_unlock_irqrestore(&object
->lock
, flags
);
962 * kmemleak_alloc - register a newly allocated object
963 * @ptr: pointer to beginning of the object
964 * @size: size of the object
965 * @min_count: minimum number of references to this object. If during memory
966 * scanning a number of references less than @min_count is found,
967 * the object is reported as a memory leak. If @min_count is 0,
968 * the object is never reported as a leak. If @min_count is -1,
969 * the object is ignored (not scanned and not reported as a leak)
970 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
972 * This function is called from the kernel allocators when a new object
973 * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.).
975 void __ref
kmemleak_alloc(const void *ptr
, size_t size
, int min_count
,
978 pr_debug("%s(0x%p, %zu, %d)\n", __func__
, ptr
, size
, min_count
);
980 if (kmemleak_enabled
&& ptr
&& !IS_ERR(ptr
))
981 create_object((unsigned long)ptr
, size
, min_count
, gfp
);
983 EXPORT_SYMBOL_GPL(kmemleak_alloc
);
986 * kmemleak_alloc_percpu - register a newly allocated __percpu object
987 * @ptr: __percpu pointer to beginning of the object
988 * @size: size of the object
989 * @gfp: flags used for kmemleak internal memory allocations
991 * This function is called from the kernel percpu allocator when a new object
992 * (memory block) is allocated (alloc_percpu).
994 void __ref
kmemleak_alloc_percpu(const void __percpu
*ptr
, size_t size
,
999 pr_debug("%s(0x%p, %zu)\n", __func__
, ptr
, size
);
1002 * Percpu allocations are only scanned and not reported as leaks
1003 * (min_count is set to 0).
1005 if (kmemleak_enabled
&& ptr
&& !IS_ERR(ptr
))
1006 for_each_possible_cpu(cpu
)
1007 create_object((unsigned long)per_cpu_ptr(ptr
, cpu
),
1010 EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu
);
1013 * kmemleak_vmalloc - register a newly vmalloc'ed object
1014 * @area: pointer to vm_struct
1015 * @size: size of the object
1016 * @gfp: __vmalloc() flags used for kmemleak internal memory allocations
1018 * This function is called from the vmalloc() kernel allocator when a new
1019 * object (memory block) is allocated.
1021 void __ref
kmemleak_vmalloc(const struct vm_struct
*area
, size_t size
, gfp_t gfp
)
1023 pr_debug("%s(0x%p, %zu)\n", __func__
, area
, size
);
1026 * A min_count = 2 is needed because vm_struct contains a reference to
1027 * the virtual address of the vmalloc'ed block.
1029 if (kmemleak_enabled
) {
1030 create_object((unsigned long)area
->addr
, size
, 2, gfp
);
1031 object_set_excess_ref((unsigned long)area
,
1032 (unsigned long)area
->addr
);
1035 EXPORT_SYMBOL_GPL(kmemleak_vmalloc
);
1038 * kmemleak_free - unregister a previously registered object
1039 * @ptr: pointer to beginning of the object
1041 * This function is called from the kernel allocators when an object (memory
1042 * block) is freed (kmem_cache_free, kfree, vfree etc.).
1044 void __ref
kmemleak_free(const void *ptr
)
1046 pr_debug("%s(0x%p)\n", __func__
, ptr
);
1048 if (kmemleak_free_enabled
&& ptr
&& !IS_ERR(ptr
))
1049 delete_object_full((unsigned long)ptr
);
1051 EXPORT_SYMBOL_GPL(kmemleak_free
);
1054 * kmemleak_free_part - partially unregister a previously registered object
1055 * @ptr: pointer to the beginning or inside the object. This also
1056 * represents the start of the range to be freed
1057 * @size: size to be unregistered
1059 * This function is called when only a part of a memory block is freed
1060 * (usually from the bootmem allocator).
1062 void __ref
kmemleak_free_part(const void *ptr
, size_t size
)
1064 pr_debug("%s(0x%p)\n", __func__
, ptr
);
1066 if (kmemleak_enabled
&& ptr
&& !IS_ERR(ptr
))
1067 delete_object_part((unsigned long)ptr
, size
, false);
1069 EXPORT_SYMBOL_GPL(kmemleak_free_part
);
1072 * kmemleak_free_percpu - unregister a previously registered __percpu object
1073 * @ptr: __percpu pointer to beginning of the object
1075 * This function is called from the kernel percpu allocator when an object
1076 * (memory block) is freed (free_percpu).
1078 void __ref
kmemleak_free_percpu(const void __percpu
*ptr
)
1082 pr_debug("%s(0x%p)\n", __func__
, ptr
);
1084 if (kmemleak_free_enabled
&& ptr
&& !IS_ERR(ptr
))
1085 for_each_possible_cpu(cpu
)
1086 delete_object_full((unsigned long)per_cpu_ptr(ptr
,
1089 EXPORT_SYMBOL_GPL(kmemleak_free_percpu
);
1092 * kmemleak_update_trace - update object allocation stack trace
1093 * @ptr: pointer to beginning of the object
1095 * Override the object allocation stack trace for cases where the actual
1096 * allocation place is not always useful.
1098 void __ref
kmemleak_update_trace(const void *ptr
)
1100 struct kmemleak_object
*object
;
1101 unsigned long flags
;
1103 pr_debug("%s(0x%p)\n", __func__
, ptr
);
1105 if (!kmemleak_enabled
|| IS_ERR_OR_NULL(ptr
))
1108 object
= find_and_get_object((unsigned long)ptr
, 1);
1111 kmemleak_warn("Updating stack trace for unknown object at %p\n",
1117 raw_spin_lock_irqsave(&object
->lock
, flags
);
1118 object
->trace_handle
= set_track_prepare();
1119 raw_spin_unlock_irqrestore(&object
->lock
, flags
);
1123 EXPORT_SYMBOL(kmemleak_update_trace
);
1126 * kmemleak_not_leak - mark an allocated object as false positive
1127 * @ptr: pointer to beginning of the object
1129 * Calling this function on an object will cause the memory block to no longer
1130 * be reported as leak and always be scanned.
1132 void __ref
kmemleak_not_leak(const void *ptr
)
1134 pr_debug("%s(0x%p)\n", __func__
, ptr
);
1136 if (kmemleak_enabled
&& ptr
&& !IS_ERR(ptr
))
1137 make_gray_object((unsigned long)ptr
);
1139 EXPORT_SYMBOL(kmemleak_not_leak
);
1142 * kmemleak_ignore - ignore an allocated object
1143 * @ptr: pointer to beginning of the object
1145 * Calling this function on an object will cause the memory block to be
1146 * ignored (not scanned and not reported as a leak). This is usually done when
1147 * it is known that the corresponding block is not a leak and does not contain
1148 * any references to other allocated memory blocks.
1150 void __ref
kmemleak_ignore(const void *ptr
)
1152 pr_debug("%s(0x%p)\n", __func__
, ptr
);
1154 if (kmemleak_enabled
&& ptr
&& !IS_ERR(ptr
))
1155 make_black_object((unsigned long)ptr
, false);
1157 EXPORT_SYMBOL(kmemleak_ignore
);
1160 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1161 * @ptr: pointer to beginning or inside the object. This also
1162 * represents the start of the scan area
1163 * @size: size of the scan area
1164 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1166 * This function is used when it is known that only certain parts of an object
1167 * contain references to other objects. Kmemleak will only scan these areas
1168 * reducing the number false negatives.
1170 void __ref
kmemleak_scan_area(const void *ptr
, size_t size
, gfp_t gfp
)
1172 pr_debug("%s(0x%p)\n", __func__
, ptr
);
1174 if (kmemleak_enabled
&& ptr
&& size
&& !IS_ERR(ptr
))
1175 add_scan_area((unsigned long)ptr
, size
, gfp
);
1177 EXPORT_SYMBOL(kmemleak_scan_area
);
1180 * kmemleak_no_scan - do not scan an allocated object
1181 * @ptr: pointer to beginning of the object
1183 * This function notifies kmemleak not to scan the given memory block. Useful
1184 * in situations where it is known that the given object does not contain any
1185 * references to other objects. Kmemleak will not scan such objects reducing
1186 * the number of false negatives.
1188 void __ref
kmemleak_no_scan(const void *ptr
)
1190 pr_debug("%s(0x%p)\n", __func__
, ptr
);
1192 if (kmemleak_enabled
&& ptr
&& !IS_ERR(ptr
))
1193 object_no_scan((unsigned long)ptr
);
1195 EXPORT_SYMBOL(kmemleak_no_scan
);
1198 * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical
1200 * @phys: physical address of the object
1201 * @size: size of the object
1202 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1204 void __ref
kmemleak_alloc_phys(phys_addr_t phys
, size_t size
, gfp_t gfp
)
1206 pr_debug("%s(0x%pa, %zu)\n", __func__
, &phys
, size
);
1208 if (kmemleak_enabled
)
1210 * Create object with OBJECT_PHYS flag and
1211 * assume min_count 0.
1213 create_object_phys((unsigned long)phys
, size
, 0, 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 pr_debug("%s(0x%pa)\n", __func__
, &phys
);
1228 if (kmemleak_enabled
)
1229 delete_object_part((unsigned long)phys
, size
, true);
1231 EXPORT_SYMBOL(kmemleak_free_part_phys
);
1234 * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical
1236 * @phys: physical address of the object
1238 void __ref
kmemleak_ignore_phys(phys_addr_t phys
)
1240 pr_debug("%s(0x%pa)\n", __func__
, &phys
);
1242 if (kmemleak_enabled
)
1243 make_black_object((unsigned long)phys
, true);
1245 EXPORT_SYMBOL(kmemleak_ignore_phys
);
1248 * Update an object's checksum and return true if it was modified.
1250 static bool update_checksum(struct kmemleak_object
*object
)
1252 u32 old_csum
= object
->checksum
;
1254 if (WARN_ON_ONCE(object
->flags
& OBJECT_PHYS
))
1257 kasan_disable_current();
1258 kcsan_disable_current();
1259 object
->checksum
= crc32(0, kasan_reset_tag((void *)object
->pointer
), object
->size
);
1260 kasan_enable_current();
1261 kcsan_enable_current();
1263 return object
->checksum
!= old_csum
;
1267 * Update an object's references. object->lock must be held by the caller.
1269 static void update_refs(struct kmemleak_object
*object
)
1271 if (!color_white(object
)) {
1272 /* non-orphan, ignored or new */
1277 * Increase the object's reference count (number of pointers to the
1278 * memory block). If this count reaches the required minimum, the
1279 * object's color will become gray and it will be added to the
1283 if (color_gray(object
)) {
1284 /* put_object() called when removing from gray_list */
1285 WARN_ON(!get_object(object
));
1286 list_add_tail(&object
->gray_list
, &gray_list
);
1291 * Memory scanning is a long process and it needs to be interruptible. This
1292 * function checks whether such interrupt condition occurred.
1294 static int scan_should_stop(void)
1296 if (!kmemleak_enabled
)
1300 * This function may be called from either process or kthread context,
1301 * hence the need to check for both stop conditions.
1304 return signal_pending(current
);
1306 return kthread_should_stop();
1312 * Scan a memory block (exclusive range) for valid pointers and add those
1313 * found to the gray list.
1315 static void scan_block(void *_start
, void *_end
,
1316 struct kmemleak_object
*scanned
)
1319 unsigned long *start
= PTR_ALIGN(_start
, BYTES_PER_POINTER
);
1320 unsigned long *end
= _end
- (BYTES_PER_POINTER
- 1);
1321 unsigned long flags
;
1322 unsigned long untagged_ptr
;
1324 raw_spin_lock_irqsave(&kmemleak_lock
, flags
);
1325 for (ptr
= start
; ptr
< end
; ptr
++) {
1326 struct kmemleak_object
*object
;
1327 unsigned long pointer
;
1328 unsigned long excess_ref
;
1330 if (scan_should_stop())
1333 kasan_disable_current();
1334 pointer
= *(unsigned long *)kasan_reset_tag((void *)ptr
);
1335 kasan_enable_current();
1337 untagged_ptr
= (unsigned long)kasan_reset_tag((void *)pointer
);
1338 if (untagged_ptr
< min_addr
|| untagged_ptr
>= max_addr
)
1342 * No need for get_object() here since we hold kmemleak_lock.
1343 * object->use_count cannot be dropped to 0 while the object
1344 * is still present in object_tree_root and object_list
1345 * (with updates protected by kmemleak_lock).
1347 object
= lookup_object(pointer
, 1);
1350 if (object
== scanned
)
1351 /* self referenced, ignore */
1355 * Avoid the lockdep recursive warning on object->lock being
1356 * previously acquired in scan_object(). These locks are
1357 * enclosed by scan_mutex.
1359 raw_spin_lock_nested(&object
->lock
, SINGLE_DEPTH_NESTING
);
1360 /* only pass surplus references (object already gray) */
1361 if (color_gray(object
)) {
1362 excess_ref
= object
->excess_ref
;
1363 /* no need for update_refs() if object already gray */
1366 update_refs(object
);
1368 raw_spin_unlock(&object
->lock
);
1371 object
= lookup_object(excess_ref
, 0);
1374 if (object
== scanned
)
1375 /* circular reference, ignore */
1377 raw_spin_lock_nested(&object
->lock
, SINGLE_DEPTH_NESTING
);
1378 update_refs(object
);
1379 raw_spin_unlock(&object
->lock
);
1382 raw_spin_unlock_irqrestore(&kmemleak_lock
, flags
);
1386 * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
1389 static void scan_large_block(void *start
, void *end
)
1393 while (start
< end
) {
1394 next
= min(start
+ MAX_SCAN_SIZE
, end
);
1395 scan_block(start
, next
, NULL
);
1403 * Scan a memory block corresponding to a kmemleak_object. A condition is
1404 * that object->use_count >= 1.
1406 static void scan_object(struct kmemleak_object
*object
)
1408 struct kmemleak_scan_area
*area
;
1409 unsigned long flags
;
1413 * Once the object->lock is acquired, the corresponding memory block
1414 * cannot be freed (the same lock is acquired in delete_object).
1416 raw_spin_lock_irqsave(&object
->lock
, flags
);
1417 if (object
->flags
& OBJECT_NO_SCAN
)
1419 if (!(object
->flags
& OBJECT_ALLOCATED
))
1420 /* already freed object */
1423 obj_ptr
= object
->flags
& OBJECT_PHYS
?
1424 __va((phys_addr_t
)object
->pointer
) :
1425 (void *)object
->pointer
;
1427 if (hlist_empty(&object
->area_list
) ||
1428 object
->flags
& OBJECT_FULL_SCAN
) {
1429 void *start
= obj_ptr
;
1430 void *end
= obj_ptr
+ object
->size
;
1434 next
= min(start
+ MAX_SCAN_SIZE
, end
);
1435 scan_block(start
, next
, object
);
1441 raw_spin_unlock_irqrestore(&object
->lock
, flags
);
1443 raw_spin_lock_irqsave(&object
->lock
, flags
);
1444 } while (object
->flags
& OBJECT_ALLOCATED
);
1446 hlist_for_each_entry(area
, &object
->area_list
, node
)
1447 scan_block((void *)area
->start
,
1448 (void *)(area
->start
+ area
->size
),
1451 raw_spin_unlock_irqrestore(&object
->lock
, flags
);
1455 * Scan the objects already referenced (gray objects). More objects will be
1456 * referenced and, if there are no memory leaks, all the objects are scanned.
1458 static void scan_gray_list(void)
1460 struct kmemleak_object
*object
, *tmp
;
1463 * The list traversal is safe for both tail additions and removals
1464 * from inside the loop. The kmemleak objects cannot be freed from
1465 * outside the loop because their use_count was incremented.
1467 object
= list_entry(gray_list
.next
, typeof(*object
), gray_list
);
1468 while (&object
->gray_list
!= &gray_list
) {
1471 /* may add new objects to the list */
1472 if (!scan_should_stop())
1473 scan_object(object
);
1475 tmp
= list_entry(object
->gray_list
.next
, typeof(*object
),
1478 /* remove the object from the list and release it */
1479 list_del(&object
->gray_list
);
1484 WARN_ON(!list_empty(&gray_list
));
1488 * Conditionally call resched() in an object iteration loop while making sure
1489 * that the given object won't go away without RCU read lock by performing a
1490 * get_object() if necessaary.
1492 static void kmemleak_cond_resched(struct kmemleak_object
*object
)
1494 if (!get_object(object
))
1495 return; /* Try next object */
1497 raw_spin_lock_irq(&kmemleak_lock
);
1498 if (object
->del_state
& DELSTATE_REMOVED
)
1499 goto unlock_put
; /* Object removed */
1500 object
->del_state
|= DELSTATE_NO_DELETE
;
1501 raw_spin_unlock_irq(&kmemleak_lock
);
1507 raw_spin_lock_irq(&kmemleak_lock
);
1508 if (object
->del_state
& DELSTATE_REMOVED
)
1509 list_del_rcu(&object
->object_list
);
1510 object
->del_state
&= ~DELSTATE_NO_DELETE
;
1512 raw_spin_unlock_irq(&kmemleak_lock
);
1517 * Scan data sections and all the referenced memory blocks allocated via the
1518 * kernel's standard allocators. This function must be called with the
1521 static void kmemleak_scan(void)
1523 struct kmemleak_object
*object
;
1525 int __maybe_unused i
;
1528 jiffies_last_scan
= jiffies
;
1530 /* prepare the kmemleak_object's */
1532 list_for_each_entry_rcu(object
, &object_list
, object_list
) {
1533 raw_spin_lock_irq(&object
->lock
);
1536 * With a few exceptions there should be a maximum of
1537 * 1 reference to any object at this point.
1539 if (atomic_read(&object
->use_count
) > 1) {
1540 pr_debug("object->use_count = %d\n",
1541 atomic_read(&object
->use_count
));
1542 dump_object_info(object
);
1546 /* ignore objects outside lowmem (paint them black) */
1547 if ((object
->flags
& OBJECT_PHYS
) &&
1548 !(object
->flags
& OBJECT_NO_SCAN
)) {
1549 unsigned long phys
= object
->pointer
;
1551 if (PHYS_PFN(phys
) < min_low_pfn
||
1552 PHYS_PFN(phys
+ object
->size
) >= max_low_pfn
)
1553 __paint_it(object
, KMEMLEAK_BLACK
);
1556 /* reset the reference count (whiten the object) */
1558 if (color_gray(object
) && get_object(object
))
1559 list_add_tail(&object
->gray_list
, &gray_list
);
1561 raw_spin_unlock_irq(&object
->lock
);
1564 kmemleak_cond_resched(object
);
1569 /* per-cpu sections scanning */
1570 for_each_possible_cpu(i
)
1571 scan_large_block(__per_cpu_start
+ per_cpu_offset(i
),
1572 __per_cpu_end
+ per_cpu_offset(i
));
1576 * Struct page scanning for each node.
1579 for_each_populated_zone(zone
) {
1580 unsigned long start_pfn
= zone
->zone_start_pfn
;
1581 unsigned long end_pfn
= zone_end_pfn(zone
);
1584 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1585 struct page
*page
= pfn_to_online_page(pfn
);
1590 /* only scan pages belonging to this zone */
1591 if (page_zone(page
) != zone
)
1593 /* only scan if page is in use */
1594 if (page_count(page
) == 0)
1596 scan_block(page
, page
+ 1, NULL
);
1604 * Scanning the task stacks (may introduce false negatives).
1606 if (kmemleak_stack_scan
) {
1607 struct task_struct
*p
, *g
;
1610 for_each_process_thread(g
, p
) {
1611 void *stack
= try_get_task_stack(p
);
1613 scan_block(stack
, stack
+ THREAD_SIZE
, NULL
);
1621 * Scan the objects already referenced from the sections scanned
1627 * Check for new or unreferenced objects modified since the previous
1628 * scan and color them gray until the next scan.
1631 list_for_each_entry_rcu(object
, &object_list
, object_list
) {
1633 kmemleak_cond_resched(object
);
1636 * This is racy but we can save the overhead of lock/unlock
1637 * calls. The missed objects, if any, should be caught in
1640 if (!color_white(object
))
1642 raw_spin_lock_irq(&object
->lock
);
1643 if (color_white(object
) && (object
->flags
& OBJECT_ALLOCATED
)
1644 && update_checksum(object
) && get_object(object
)) {
1645 /* color it gray temporarily */
1646 object
->count
= object
->min_count
;
1647 list_add_tail(&object
->gray_list
, &gray_list
);
1649 raw_spin_unlock_irq(&object
->lock
);
1654 * Re-scan the gray list for modified unreferenced objects.
1659 * If scanning was stopped do not report any new unreferenced objects.
1661 if (scan_should_stop())
1665 * Scanning result reporting.
1668 list_for_each_entry_rcu(object
, &object_list
, object_list
) {
1670 kmemleak_cond_resched(object
);
1673 * This is racy but we can save the overhead of lock/unlock
1674 * calls. The missed objects, if any, should be caught in
1677 if (!color_white(object
))
1679 raw_spin_lock_irq(&object
->lock
);
1680 if (unreferenced_object(object
) &&
1681 !(object
->flags
& OBJECT_REPORTED
)) {
1682 object
->flags
|= OBJECT_REPORTED
;
1684 if (kmemleak_verbose
)
1685 print_unreferenced(NULL
, object
);
1689 raw_spin_unlock_irq(&object
->lock
);
1694 kmemleak_found_leaks
= true;
1696 pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
1703 * Thread function performing automatic memory scanning. Unreferenced objects
1704 * at the end of a memory scan are reported but only the first time.
1706 static int kmemleak_scan_thread(void *arg
)
1708 static int first_run
= IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN
);
1710 pr_info("Automatic memory scanning thread started\n");
1711 set_user_nice(current
, 10);
1714 * Wait before the first scan to allow the system to fully initialize.
1717 signed long timeout
= msecs_to_jiffies(SECS_FIRST_SCAN
* 1000);
1719 while (timeout
&& !kthread_should_stop())
1720 timeout
= schedule_timeout_interruptible(timeout
);
1723 while (!kthread_should_stop()) {
1724 signed long timeout
= READ_ONCE(jiffies_scan_wait
);
1726 mutex_lock(&scan_mutex
);
1728 mutex_unlock(&scan_mutex
);
1730 /* wait before the next scan */
1731 while (timeout
&& !kthread_should_stop())
1732 timeout
= schedule_timeout_interruptible(timeout
);
1735 pr_info("Automatic memory scanning thread ended\n");
1741 * Start the automatic memory scanning thread. This function must be called
1742 * with the scan_mutex held.
1744 static void start_scan_thread(void)
1748 scan_thread
= kthread_run(kmemleak_scan_thread
, NULL
, "kmemleak");
1749 if (IS_ERR(scan_thread
)) {
1750 pr_warn("Failed to create the scan thread\n");
1756 * Stop the automatic memory scanning thread.
1758 static void stop_scan_thread(void)
1761 kthread_stop(scan_thread
);
1767 * Iterate over the object_list and return the first valid object at or after
1768 * the required position with its use_count incremented. The function triggers
1769 * a memory scanning when the pos argument points to the first position.
1771 static void *kmemleak_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1773 struct kmemleak_object
*object
;
1777 err
= mutex_lock_interruptible(&scan_mutex
);
1779 return ERR_PTR(err
);
1782 list_for_each_entry_rcu(object
, &object_list
, object_list
) {
1785 if (get_object(object
))
1794 * Return the next object in the object_list. The function decrements the
1795 * use_count of the previous object and increases that of the next one.
1797 static void *kmemleak_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1799 struct kmemleak_object
*prev_obj
= v
;
1800 struct kmemleak_object
*next_obj
= NULL
;
1801 struct kmemleak_object
*obj
= prev_obj
;
1805 list_for_each_entry_continue_rcu(obj
, &object_list
, object_list
) {
1806 if (get_object(obj
)) {
1812 put_object(prev_obj
);
1817 * Decrement the use_count of the last object required, if any.
1819 static void kmemleak_seq_stop(struct seq_file
*seq
, void *v
)
1823 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1824 * waiting was interrupted, so only release it if !IS_ERR.
1827 mutex_unlock(&scan_mutex
);
1834 * Print the information for an unreferenced object to the seq file.
1836 static int kmemleak_seq_show(struct seq_file
*seq
, void *v
)
1838 struct kmemleak_object
*object
= v
;
1839 unsigned long flags
;
1841 raw_spin_lock_irqsave(&object
->lock
, flags
);
1842 if ((object
->flags
& OBJECT_REPORTED
) && unreferenced_object(object
))
1843 print_unreferenced(seq
, object
);
1844 raw_spin_unlock_irqrestore(&object
->lock
, flags
);
1848 static const struct seq_operations kmemleak_seq_ops
= {
1849 .start
= kmemleak_seq_start
,
1850 .next
= kmemleak_seq_next
,
1851 .stop
= kmemleak_seq_stop
,
1852 .show
= kmemleak_seq_show
,
1855 static int kmemleak_open(struct inode
*inode
, struct file
*file
)
1857 return seq_open(file
, &kmemleak_seq_ops
);
1860 static int dump_str_object_info(const char *str
)
1862 unsigned long flags
;
1863 struct kmemleak_object
*object
;
1866 if (kstrtoul(str
, 0, &addr
))
1868 object
= find_and_get_object(addr
, 0);
1870 pr_info("Unknown object at 0x%08lx\n", addr
);
1874 raw_spin_lock_irqsave(&object
->lock
, flags
);
1875 dump_object_info(object
);
1876 raw_spin_unlock_irqrestore(&object
->lock
, flags
);
1883 * We use grey instead of black to ensure we can do future scans on the same
1884 * objects. If we did not do future scans these black objects could
1885 * potentially contain references to newly allocated objects in the future and
1886 * we'd end up with false positives.
1888 static void kmemleak_clear(void)
1890 struct kmemleak_object
*object
;
1893 list_for_each_entry_rcu(object
, &object_list
, object_list
) {
1894 raw_spin_lock_irq(&object
->lock
);
1895 if ((object
->flags
& OBJECT_REPORTED
) &&
1896 unreferenced_object(object
))
1897 __paint_it(object
, KMEMLEAK_GREY
);
1898 raw_spin_unlock_irq(&object
->lock
);
1902 kmemleak_found_leaks
= false;
1905 static void __kmemleak_do_cleanup(void);
1908 * File write operation to configure kmemleak at run-time. The following
1909 * commands can be written to the /sys/kernel/debug/kmemleak file:
1910 * off - disable kmemleak (irreversible)
1911 * stack=on - enable the task stacks scanning
1912 * stack=off - disable the tasks stacks scanning
1913 * scan=on - start the automatic memory scanning thread
1914 * scan=off - stop the automatic memory scanning thread
1915 * scan=... - set the automatic memory scanning period in seconds (0 to
1917 * scan - trigger a memory scan
1918 * clear - mark all current reported unreferenced kmemleak objects as
1919 * grey to ignore printing them, or free all kmemleak objects
1920 * if kmemleak has been disabled.
1921 * dump=... - dump information about the object found at the given address
1923 static ssize_t
kmemleak_write(struct file
*file
, const char __user
*user_buf
,
1924 size_t size
, loff_t
*ppos
)
1930 buf_size
= min(size
, (sizeof(buf
) - 1));
1931 if (strncpy_from_user(buf
, user_buf
, buf_size
) < 0)
1935 ret
= mutex_lock_interruptible(&scan_mutex
);
1939 if (strncmp(buf
, "clear", 5) == 0) {
1940 if (kmemleak_enabled
)
1943 __kmemleak_do_cleanup();
1947 if (!kmemleak_enabled
) {
1952 if (strncmp(buf
, "off", 3) == 0)
1954 else if (strncmp(buf
, "stack=on", 8) == 0)
1955 kmemleak_stack_scan
= 1;
1956 else if (strncmp(buf
, "stack=off", 9) == 0)
1957 kmemleak_stack_scan
= 0;
1958 else if (strncmp(buf
, "scan=on", 7) == 0)
1959 start_scan_thread();
1960 else if (strncmp(buf
, "scan=off", 8) == 0)
1962 else if (strncmp(buf
, "scan=", 5) == 0) {
1964 unsigned long msecs
;
1966 ret
= kstrtouint(buf
+ 5, 0, &secs
);
1970 msecs
= secs
* MSEC_PER_SEC
;
1971 if (msecs
> UINT_MAX
)
1976 WRITE_ONCE(jiffies_scan_wait
, msecs_to_jiffies(msecs
));
1977 start_scan_thread();
1979 } else if (strncmp(buf
, "scan", 4) == 0)
1981 else if (strncmp(buf
, "dump=", 5) == 0)
1982 ret
= dump_str_object_info(buf
+ 5);
1987 mutex_unlock(&scan_mutex
);
1991 /* ignore the rest of the buffer, only one command at a time */
1996 static const struct file_operations kmemleak_fops
= {
1997 .owner
= THIS_MODULE
,
1998 .open
= kmemleak_open
,
2000 .write
= kmemleak_write
,
2001 .llseek
= seq_lseek
,
2002 .release
= seq_release
,
2005 static void __kmemleak_do_cleanup(void)
2007 struct kmemleak_object
*object
, *tmp
;
2010 * Kmemleak has already been disabled, no need for RCU list traversal
2011 * or kmemleak_lock held.
2013 list_for_each_entry_safe(object
, tmp
, &object_list
, object_list
) {
2014 __remove_object(object
);
2015 __delete_object(object
);
2020 * Stop the memory scanning thread and free the kmemleak internal objects if
2021 * no previous scan thread (otherwise, kmemleak may still have some useful
2022 * information on memory leaks).
2024 static void kmemleak_do_cleanup(struct work_struct
*work
)
2028 mutex_lock(&scan_mutex
);
2030 * Once it is made sure that kmemleak_scan has stopped, it is safe to no
2031 * longer track object freeing. Ordering of the scan thread stopping and
2032 * the memory accesses below is guaranteed by the kthread_stop()
2035 kmemleak_free_enabled
= 0;
2036 mutex_unlock(&scan_mutex
);
2038 if (!kmemleak_found_leaks
)
2039 __kmemleak_do_cleanup();
2041 pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
2044 static DECLARE_WORK(cleanup_work
, kmemleak_do_cleanup
);
2047 * Disable kmemleak. No memory allocation/freeing will be traced once this
2048 * function is called. Disabling kmemleak is an irreversible operation.
2050 static void kmemleak_disable(void)
2052 /* atomically check whether it was already invoked */
2053 if (cmpxchg(&kmemleak_error
, 0, 1))
2056 /* stop any memory operation tracing */
2057 kmemleak_enabled
= 0;
2059 /* check whether it is too early for a kernel thread */
2060 if (kmemleak_late_initialized
)
2061 schedule_work(&cleanup_work
);
2063 kmemleak_free_enabled
= 0;
2065 pr_info("Kernel memory leak detector disabled\n");
2069 * Allow boot-time kmemleak disabling (enabled by default).
2071 static int __init
kmemleak_boot_config(char *str
)
2075 if (strcmp(str
, "off") == 0)
2077 else if (strcmp(str
, "on") == 0) {
2078 kmemleak_skip_disable
= 1;
2079 stack_depot_request_early_init();
2085 early_param("kmemleak", kmemleak_boot_config
);
2088 * Kmemleak initialization.
2090 void __init
kmemleak_init(void)
2092 #ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
2093 if (!kmemleak_skip_disable
) {
2102 jiffies_min_age
= msecs_to_jiffies(MSECS_MIN_AGE
);
2103 jiffies_scan_wait
= msecs_to_jiffies(SECS_SCAN_WAIT
* 1000);
2105 object_cache
= KMEM_CACHE(kmemleak_object
, SLAB_NOLEAKTRACE
);
2106 scan_area_cache
= KMEM_CACHE(kmemleak_scan_area
, SLAB_NOLEAKTRACE
);
2108 /* register the data/bss sections */
2109 create_object((unsigned long)_sdata
, _edata
- _sdata
,
2110 KMEMLEAK_GREY
, GFP_ATOMIC
);
2111 create_object((unsigned long)__bss_start
, __bss_stop
- __bss_start
,
2112 KMEMLEAK_GREY
, GFP_ATOMIC
);
2113 /* only register .data..ro_after_init if not within .data */
2114 if (&__start_ro_after_init
< &_sdata
|| &__end_ro_after_init
> &_edata
)
2115 create_object((unsigned long)__start_ro_after_init
,
2116 __end_ro_after_init
- __start_ro_after_init
,
2117 KMEMLEAK_GREY
, GFP_ATOMIC
);
2121 * Late initialization function.
2123 static int __init
kmemleak_late_init(void)
2125 kmemleak_late_initialized
= 1;
2127 debugfs_create_file("kmemleak", 0644, NULL
, NULL
, &kmemleak_fops
);
2129 if (kmemleak_error
) {
2131 * Some error occurred and kmemleak was disabled. There is a
2132 * small chance that kmemleak_disable() was called immediately
2133 * after setting kmemleak_late_initialized and we may end up with
2134 * two clean-up threads but serialized by scan_mutex.
2136 schedule_work(&cleanup_work
);
2140 if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN
)) {
2141 mutex_lock(&scan_mutex
);
2142 start_scan_thread();
2143 mutex_unlock(&scan_mutex
);
2146 pr_info("Kernel memory leak detector initialized (mem pool available: %d)\n",
2147 mem_pool_free_count
);
2151 late_initcall(kmemleak_late_init
);