1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * mm/kmemleak.c
4 *
5 * Copyright (C) 2008 ARM Limited
6 * Written by Catalin Marinas <catalin.marinas@arm.com>
7 *
8 * For more information on the algorithm and kmemleak usage, please see
9 * Documentation/dev-tools/kmemleak.rst.
10 *
11 * Notes on locking
12 * ----------------
13 *
14 * The following locks and mutexes are used by kmemleak:
15 *
16 * - kmemleak_lock (raw_spinlock_t): protects the object_list as well as
17 * del_state modifications and accesses to the object trees
18 * (object_tree_root, object_phys_tree_root, object_percpu_tree_root). The
19 * object_list is the main list holding the metadata (struct
20 * kmemleak_object) for the allocated memory blocks. The object trees are
21 * red black trees used to look-up metadata based on a pointer to the
22 * corresponding memory block. The kmemleak_object structures are added to
23 * the object_list and the object tree root in the create_object() function
24 * called from the kmemleak_alloc{,_phys,_percpu}() callback and removed in
25 * delete_object() called from the kmemleak_free{,_phys,_percpu}() callback
26 * - kmemleak_object.lock (raw_spinlock_t): protects a kmemleak_object.
27 * Accesses to the metadata (e.g. count) are protected by this lock. Note
28 * that some members of this structure may be protected by other means
29 * (atomic or kmemleak_lock). This lock is also held when scanning the
30 * corresponding memory block to avoid the kernel freeing it via the
31 * kmemleak_free() callback. This is less heavyweight than holding a global
32 * lock like kmemleak_lock during scanning.
33 * - scan_mutex (mutex): ensures that only one thread may scan the memory for
34 * unreferenced objects at a time. The gray_list contains the objects which
35 * are already referenced or marked as false positives and need to be
36 * scanned. This list is only modified during a scanning episode when the
37 * scan_mutex is held. At the end of a scan, the gray_list is always empty.
38 * Note that the kmemleak_object.use_count is incremented when an object is
39 * added to the gray_list and therefore cannot be freed. This mutex also
40 * prevents multiple users of the "kmemleak" debugfs file together with
41 * modifications to the memory scanning parameters including the scan_thread
42 * pointer
43 *
44 * Locks and mutexes are acquired/nested in the following order:
45 *
46 * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING)
47 *
48 * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex
49 * regions.
50 *
51 * The kmemleak_object structures have a use_count incremented or decremented
52 * using the get_object()/put_object() functions. When the use_count becomes
53 * 0, this count can no longer be incremented and put_object() schedules the
54 * kmemleak_object freeing via an RCU callback. All calls to the get_object()
55 * function must be protected by rcu_read_lock() to avoid accessing a freed
56 * structure.
57 */
58
59 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
60
61 #include <linux/init.h>
62 #include <linux/kernel.h>
63 #include <linux/list.h>
64 #include <linux/sched/signal.h>
65 #include <linux/sched/task.h>
66 #include <linux/sched/task_stack.h>
67 #include <linux/jiffies.h>
68 #include <linux/delay.h>
69 #include <linux/export.h>
70 #include <linux/kthread.h>
71 #include <linux/rbtree.h>
72 #include <linux/fs.h>
73 #include <linux/debugfs.h>
74 #include <linux/seq_file.h>
75 #include <linux/cpumask.h>
76 #include <linux/spinlock.h>
77 #include <linux/module.h>
78 #include <linux/mutex.h>
79 #include <linux/rcupdate.h>
80 #include <linux/stacktrace.h>
81 #include <linux/stackdepot.h>
82 #include <linux/cache.h>
83 #include <linux/percpu.h>
84 #include <linux/memblock.h>
85 #include <linux/pfn.h>
86 #include <linux/mmzone.h>
87 #include <linux/slab.h>
88 #include <linux/thread_info.h>
89 #include <linux/err.h>
90 #include <linux/uaccess.h>
91 #include <linux/string.h>
92 #include <linux/nodemask.h>
93 #include <linux/mm.h>
94 #include <linux/workqueue.h>
95 #include <linux/crc32.h>
96
97 #include <asm/sections.h>
98 #include <asm/processor.h>
99 #include <linux/atomic.h>
100
101 #include <linux/kasan.h>
102 #include <linux/kfence.h>
103 #include <linux/kmemleak.h>
104 #include <linux/memory_hotplug.h>
105
106 /*
107 * Kmemleak configuration and common defines.
108 */
109 #define MAX_TRACE 16 /* stack trace length */
110 #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
111 #define SECS_FIRST_SCAN 60 /* delay before the first scan */
112 #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
113 #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
114
115 #define BYTES_PER_POINTER sizeof(void *)
116
117 /* scanning area inside a memory block */
118 struct kmemleak_scan_area {
119 struct hlist_node node;
120 unsigned long start;
121 size_t size;
122 };
123
124 #define KMEMLEAK_GREY 0
125 #define KMEMLEAK_BLACK -1
126
127 /*
128 * Structure holding the metadata for each allocated memory block.
129 * Modifications to such objects should be made while holding the
130 * object->lock. Insertions or deletions from object_list, gray_list or
131 * rb_node are already protected by the corresponding locks or mutex (see
132 * the notes on locking above). These objects are reference-counted
133 * (use_count) and freed using the RCU mechanism.
134 */
135 struct kmemleak_object {
136 raw_spinlock_t lock;
137 unsigned int flags; /* object status flags */
138 struct list_head object_list;
139 struct list_head gray_list;
140 struct rb_node rb_node;
141 struct rcu_head rcu; /* object_list lockless traversal */
142 /* object usage count; object freed when use_count == 0 */
143 atomic_t use_count;
144 unsigned int del_state; /* deletion state */
145 unsigned long pointer;
146 size_t size;
147 /* pass surplus references to this pointer */
148 unsigned long excess_ref;
149 /* minimum number of a pointers found before it is considered leak */
150 int min_count;
151 /* the total number of pointers found pointing to this object */
152 int count;
153 /* checksum for detecting modified objects */
154 u32 checksum;
155 depot_stack_handle_t trace_handle;
156 /* memory ranges to be scanned inside an object (empty for all) */
157 struct hlist_head area_list;
158 unsigned long jiffies; /* creation timestamp */
159 pid_t pid; /* pid of the current task */
160 char comm[TASK_COMM_LEN]; /* executable name */
161 };
162
163 /* flag representing the memory block allocation status */
164 #define OBJECT_ALLOCATED (1 << 0)
165 /* flag set after the first reporting of an unreference object */
166 #define OBJECT_REPORTED (1 << 1)
167 /* flag set to not scan the object */
168 #define OBJECT_NO_SCAN (1 << 2)
169 /* flag set to fully scan the object when scan_area allocation failed */
170 #define OBJECT_FULL_SCAN (1 << 3)
171 /* flag set for object allocated with physical address */
172 #define OBJECT_PHYS (1 << 4)
173 /* flag set for per-CPU pointers */
174 #define OBJECT_PERCPU (1 << 5)
175
176 /* set when __remove_object() called */
177 #define DELSTATE_REMOVED (1 << 0)
178 /* set to temporarily prevent deletion from object_list */
179 #define DELSTATE_NO_DELETE (1 << 1)
180
181 #define HEX_PREFIX " "
182 /* number of bytes to print per line; must be 16 or 32 */
183 #define HEX_ROW_SIZE 16
184 /* number of bytes to print at a time (1, 2, 4, 8) */
185 #define HEX_GROUP_SIZE 1
186 /* include ASCII after the hex output */
187 #define HEX_ASCII 1
188 /* max number of lines to be printed */
189 #define HEX_MAX_LINES 2
190
191 /* the list of all allocated objects */
192 static LIST_HEAD(object_list);
193 /* the list of gray-colored objects (see color_gray comment below) */
194 static LIST_HEAD(gray_list);
195 /* memory pool allocation */
196 static struct kmemleak_object mem_pool[CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE];
197 static int mem_pool_free_count = ARRAY_SIZE(mem_pool);
198 static LIST_HEAD(mem_pool_free_list);
199 /* search tree for object boundaries */
200 static struct rb_root object_tree_root = RB_ROOT;
201 /* search tree for object (with OBJECT_PHYS flag) boundaries */
202 static struct rb_root object_phys_tree_root = RB_ROOT;
203 /* search tree for object (with OBJECT_PERCPU flag) boundaries */
204 static struct rb_root object_percpu_tree_root = RB_ROOT;
205 /* protecting the access to object_list, object_tree_root (or object_phys_tree_root) */
206 static DEFINE_RAW_SPINLOCK(kmemleak_lock);
207
208 /* allocation caches for kmemleak internal data */
209 static struct kmem_cache *object_cache;
210 static struct kmem_cache *scan_area_cache;
211
212 /* set if tracing memory operations is enabled */
213 static int kmemleak_enabled = 1;
214 /* same as above but only for the kmemleak_free() callback */
215 static int kmemleak_free_enabled = 1;
216 /* set in the late_initcall if there were no errors */
217 static int kmemleak_late_initialized;
218 /* set if a kmemleak warning was issued */
219 static int kmemleak_warning;
220 /* set if a fatal kmemleak error has occurred */
221 static int kmemleak_error;
222
223 /* minimum and maximum address that may be valid pointers */
224 static unsigned long min_addr = ULONG_MAX;
225 static unsigned long max_addr;
226
227 static struct task_struct *scan_thread;
228 /* used to avoid reporting of recently allocated objects */
229 static unsigned long jiffies_min_age;
230 static unsigned long jiffies_last_scan;
231 /* delay between automatic memory scannings */
232 static unsigned long jiffies_scan_wait;
233 /* enables or disables the task stacks scanning */
234 static int kmemleak_stack_scan = 1;
235 /* protects the memory scanning, parameters and debug/kmemleak file access */
236 static DEFINE_MUTEX(scan_mutex);
237 /* setting kmemleak=on, will set this var, skipping the disable */
238 static int kmemleak_skip_disable;
239 /* If there are leaks that can be reported */
240 static bool kmemleak_found_leaks;
241
242 static bool kmemleak_verbose;
243 module_param_named(verbose, kmemleak_verbose, bool, 0600);
244
245 static void kmemleak_disable(void);
246
247 /*
248 * Print a warning and dump the stack trace.
249 */
250 #define kmemleak_warn(x...) do { \
251 pr_warn(x); \
252 dump_stack(); \
253 kmemleak_warning = 1; \
254 } while (0)
255
256 /*
257 * Macro invoked when a serious kmemleak condition occurred and cannot be
258 * recovered from. Kmemleak will be disabled and further allocation/freeing
259 * tracing no longer available.
260 */
261 #define kmemleak_stop(x...) do { \
262 kmemleak_warn(x); \
263 kmemleak_disable(); \
264 } while (0)
265
266 #define warn_or_seq_printf(seq, fmt, ...) do { \
267 if (seq) \
268 seq_printf(seq, fmt, ##__VA_ARGS__); \
269 else \
270 pr_warn(fmt, ##__VA_ARGS__); \
271 } while (0)
272
273 static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type,
274 int rowsize, int groupsize, const void *buf,
275 size_t len, bool ascii)
276 {
277 if (seq)
278 seq_hex_dump(seq, HEX_PREFIX, prefix_type, rowsize, groupsize,
279 buf, len, ascii);
280 else
281 print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type,
282 rowsize, groupsize, buf, len, ascii);
283 }
284
285 /*
286 * Printing of the objects hex dump to the seq file. The number of lines to be
287 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
288 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
289 * with the object->lock held.
290 */
291 static void hex_dump_object(struct seq_file *seq,
292 struct kmemleak_object *object)
293 {
294 const u8 *ptr = (const u8 *)object->pointer;
295 size_t len;
296
297 if (WARN_ON_ONCE(object->flags & (OBJECT_PHYS | OBJECT_PERCPU)))
298 return;
299
300 /* limit the number of lines to HEX_MAX_LINES */
301 len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
302
303 warn_or_seq_printf(seq, " hex dump (first %zu bytes):\n", len);
304 kasan_disable_current();
305 warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE,
306 HEX_GROUP_SIZE, kasan_reset_tag((void *)ptr), len, HEX_ASCII);
307 kasan_enable_current();
308 }
309
310 /*
311 * Object colors, encoded with count and min_count:
312 * - white - orphan object, not enough references to it (count < min_count)
313 * - gray - not orphan, not marked as false positive (min_count == 0) or
314 * sufficient references to it (count >= min_count)
315 * - black - ignore, it doesn't contain references (e.g. text section)
316 * (min_count == -1). No function defined for this color.
317 * Newly created objects don't have any color assigned (object->count == -1)
318 * before the next memory scan when they become white.
319 */
320 static bool color_white(const struct kmemleak_object *object)
321 {
322 return object->count != KMEMLEAK_BLACK &&
323 object->count < object->min_count;
324 }
325
326 static bool color_gray(const struct kmemleak_object *object)
327 {
328 return object->min_count != KMEMLEAK_BLACK &&
329 object->count >= object->min_count;
330 }
331
332 /*
333 * Objects are considered unreferenced only if their color is white, they have
334 * not be deleted and have a minimum age to avoid false positives caused by
335 * pointers temporarily stored in CPU registers.
336 */
337 static bool unreferenced_object(struct kmemleak_object *object)
338 {
339 return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
340 time_before_eq(object->jiffies + jiffies_min_age,
341 jiffies_last_scan);
342 }
343
344 /*
345 * Printing of the unreferenced objects information to the seq file. The
346 * print_unreferenced function must be called with the object->lock held.
347 */
348 static void print_unreferenced(struct seq_file *seq,
349 struct kmemleak_object *object)
350 {
351 int i;
352 unsigned long *entries;
353 unsigned int nr_entries;
354
355 nr_entries = stack_depot_fetch(object->trace_handle, &entries);
356 warn_or_seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
357 object->pointer, object->size);
358 warn_or_seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu\n",
359 object->comm, object->pid, object->jiffies);
360 hex_dump_object(seq, object);
361 warn_or_seq_printf(seq, " backtrace (crc %x):\n", object->checksum);
362
363 for (i = 0; i < nr_entries; i++) {
364 void *ptr = (void *)entries[i];
365 warn_or_seq_printf(seq, " [<%pK>] %pS\n", ptr, ptr);
366 }
367 }
368
369 /*
370 * Print the kmemleak_object information. This function is used mainly for
371 * debugging special cases when kmemleak operations. It must be called with
372 * the object->lock held.
373 */
374 static void dump_object_info(struct kmemleak_object *object)
375 {
376 pr_notice("Object 0x%08lx (size %zu):\n",
377 object->pointer, object->size);
378 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
379 object->comm, object->pid, object->jiffies);
380 pr_notice(" min_count = %d\n", object->min_count);
381 pr_notice(" count = %d\n", object->count);
382 pr_notice(" flags = 0x%x\n", object->flags);
383 pr_notice(" checksum = %u\n", object->checksum);
384 pr_notice(" backtrace:\n");
385 if (object->trace_handle)
386 stack_depot_print(object->trace_handle);
387 }
388
389 static struct rb_root *object_tree(unsigned long objflags)
390 {
391 if (objflags & OBJECT_PHYS)
392 return &object_phys_tree_root;
393 if (objflags & OBJECT_PERCPU)
394 return &object_percpu_tree_root;
395 return &object_tree_root;
396 }
397
398 /*
399 * Look-up a memory block metadata (kmemleak_object) in the object search
400 * tree based on a pointer value. If alias is 0, only values pointing to the
401 * beginning of the memory block are allowed. The kmemleak_lock must be held
402 * when calling this function.
403 */
404 static struct kmemleak_object *__lookup_object(unsigned long ptr, int alias,
405 unsigned int objflags)
406 {
407 struct rb_node *rb = object_tree(objflags)->rb_node;
408 unsigned long untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
409
410 while (rb) {
411 struct kmemleak_object *object;
412 unsigned long untagged_objp;
413
414 object = rb_entry(rb, struct kmemleak_object, rb_node);
415 untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer);
416
417 if (untagged_ptr < untagged_objp)
418 rb = object->rb_node.rb_left;
419 else if (untagged_objp + object->size <= untagged_ptr)
420 rb = object->rb_node.rb_right;
421 else if (untagged_objp == untagged_ptr || alias)
422 return object;
423 else {
424 kmemleak_warn("Found object by alias at 0x%08lx\n",
425 ptr);
426 dump_object_info(object);
427 break;
428 }
429 }
430 return NULL;
431 }
432
433 /* Look-up a kmemleak object which allocated with virtual address. */
434 static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
435 {
436 return __lookup_object(ptr, alias, 0);
437 }
438
439 /*
440 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
441 * that once an object's use_count reached 0, the RCU freeing was already
442 * registered and the object should no longer be used. This function must be
443 * called under the protection of rcu_read_lock().
444 */
445 static int get_object(struct kmemleak_object *object)
446 {
447 return atomic_inc_not_zero(&object->use_count);
448 }
449
450 /*
451 * Memory pool allocation and freeing. kmemleak_lock must not be held.
452 */
453 static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
454 {
455 unsigned long flags;
456 struct kmemleak_object *object;
457
458 /* try the slab allocator first */
459 if (object_cache) {
460 object = kmem_cache_alloc_noprof(object_cache,
461 gfp_nested_mask(gfp));
462 if (object)
463 return object;
464 }
465
466 /* slab allocation failed, try the memory pool */
467 raw_spin_lock_irqsave(&kmemleak_lock, flags);
468 object = list_first_entry_or_null(&mem_pool_free_list,
469 typeof(*object), object_list);
470 if (object)
471 list_del(&object->object_list);
472 else if (mem_pool_free_count)
473 object = &mem_pool[--mem_pool_free_count];
474 else
475 pr_warn_once("Memory pool empty, consider increasing CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE\n");
476 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
477
478 return object;
479 }
480
481 /*
482 * Return the object to either the slab allocator or the memory pool.
483 */
484 static void mem_pool_free(struct kmemleak_object *object)
485 {
486 unsigned long flags;
487
488 if (object < mem_pool || object >= mem_pool + ARRAY_SIZE(mem_pool)) {
489 kmem_cache_free(object_cache, object);
490 return;
491 }
492
493 /* add the object to the memory pool free list */
494 raw_spin_lock_irqsave(&kmemleak_lock, flags);
495 list_add(&object->object_list, &mem_pool_free_list);
496 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
497 }
498
499 /*
500 * RCU callback to free a kmemleak_object.
501 */
502 static void free_object_rcu(struct rcu_head *rcu)
503 {
504 struct hlist_node *tmp;
505 struct kmemleak_scan_area *area;
506 struct kmemleak_object *object =
507 container_of(rcu, struct kmemleak_object, rcu);
508
509 /*
510 * Once use_count is 0 (guaranteed by put_object), there is no other
511 * code accessing this object, hence no need for locking.
512 */
513 hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
514 hlist_del(&area->node);
515 kmem_cache_free(scan_area_cache, area);
516 }
517 mem_pool_free(object);
518 }
519
520 /*
521 * Decrement the object use_count. Once the count is 0, free the object using
522 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
523 * delete_object() path, the delayed RCU freeing ensures that there is no
524 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
525 * is also possible.
526 */
527 static void put_object(struct kmemleak_object *object)
528 {
529 if (!atomic_dec_and_test(&object->use_count))
530 return;
531
532 /* should only get here after delete_object was called */
533 WARN_ON(object->flags & OBJECT_ALLOCATED);
534
535 /*
536 * It may be too early for the RCU callbacks, however, there is no
537 * concurrent object_list traversal when !object_cache and all objects
538 * came from the memory pool. Free the object directly.
539 */
540 if (object_cache)
541 call_rcu(&object->rcu, free_object_rcu);
542 else
543 free_object_rcu(&object->rcu);
544 }
545
546 /*
547 * Look up an object in the object search tree and increase its use_count.
548 */
549 static struct kmemleak_object *__find_and_get_object(unsigned long ptr, int alias,
550 unsigned int objflags)
551 {
552 unsigned long flags;
553 struct kmemleak_object *object;
554
555 rcu_read_lock();
556 raw_spin_lock_irqsave(&kmemleak_lock, flags);
557 object = __lookup_object(ptr, alias, objflags);
558 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
559
560 /* check whether the object is still available */
561 if (object && !get_object(object))
562 object = NULL;
563 rcu_read_unlock();
564
565 return object;
566 }
567
568 /* Look up and get an object which allocated with virtual address. */
569 static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
570 {
571 return __find_and_get_object(ptr, alias, 0);
572 }
573
574 /*
575 * Remove an object from its object tree and object_list. Must be called with
576 * the kmemleak_lock held _if_ kmemleak is still enabled.
577 */
578 static void __remove_object(struct kmemleak_object *object)
579 {
580 rb_erase(&object->rb_node, object_tree(object->flags));
581 if (!(object->del_state & DELSTATE_NO_DELETE))
582 list_del_rcu(&object->object_list);
583 object->del_state |= DELSTATE_REMOVED;
584 }
585
586 static struct kmemleak_object *__find_and_remove_object(unsigned long ptr,
587 int alias,
588 unsigned int objflags)
589 {
590 struct kmemleak_object *object;
591
592 object = __lookup_object(ptr, alias, objflags);
593 if (object)
594 __remove_object(object);
595
596 return object;
597 }
598
599 /*
600 * Look up an object in the object search tree and remove it from both object
601 * tree root and object_list. The returned object's use_count should be at
602 * least 1, as initially set by create_object().
603 */
604 static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias,
605 unsigned int objflags)
606 {
607 unsigned long flags;
608 struct kmemleak_object *object;
609
610 raw_spin_lock_irqsave(&kmemleak_lock, flags);
611 object = __find_and_remove_object(ptr, alias, objflags);
612 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
613
614 return object;
615 }
616
617 static noinline depot_stack_handle_t set_track_prepare(void)
618 {
619 depot_stack_handle_t trace_handle;
620 unsigned long entries[MAX_TRACE];
621 unsigned int nr_entries;
622
623 /*
624 * Use object_cache to determine whether kmemleak_init() has
625 * been invoked. stack_depot_early_init() is called before
626 * kmemleak_init() in mm_core_init().
627 */
628 if (!object_cache)
629 return 0;
630 nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 3);
631 trace_handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT);
632
633 return trace_handle;
634 }
635
636 static struct kmemleak_object *__alloc_object(gfp_t gfp)
637 {
638 struct kmemleak_object *object;
639
640 object = mem_pool_alloc(gfp);
641 if (!object) {
642 pr_warn("Cannot allocate a kmemleak_object structure\n");
643 kmemleak_disable();
644 return NULL;
645 }
646
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->excess_ref = 0;
653 object->count = 0; /* white color initially */
654 object->checksum = 0;
655 object->del_state = 0;
656
657 /* task information */
658 if (in_hardirq()) {
659 object->pid = 0;
660 strscpy(object->comm, "hardirq");
661 } else if (in_serving_softirq()) {
662 object->pid = 0;
663 strscpy(object->comm, "softirq");
664 } else {
665 object->pid = current->pid;
666 /*
667 * There is a small chance of a race with set_task_comm(),
668 * however using get_task_comm() here may cause locking
669 * dependency issues with current->alloc_lock. In the worst
670 * case, the command line is not correct.
671 */
672 strscpy(object->comm, current->comm);
673 }
674
675 /* kernel backtrace */
676 object->trace_handle = set_track_prepare();
677
678 return object;
679 }
680
681 static int __link_object(struct kmemleak_object *object, unsigned long ptr,
682 size_t size, int min_count, unsigned int objflags)
683 {
684
685 struct kmemleak_object *parent;
686 struct rb_node **link, *rb_parent;
687 unsigned long untagged_ptr;
688 unsigned long untagged_objp;
689
690 object->flags = OBJECT_ALLOCATED | objflags;
691 object->pointer = ptr;
692 object->size = kfence_ksize((void *)ptr) ?: size;
693 object->min_count = min_count;
694 object->jiffies = jiffies;
695
696 untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
697 /*
698 * Only update min_addr and max_addr with object
699 * storing virtual address.
700 */
701 if (!(objflags & (OBJECT_PHYS | OBJECT_PERCPU))) {
702 min_addr = min(min_addr, untagged_ptr);
703 max_addr = max(max_addr, untagged_ptr + size);
704 }
705 link = &object_tree(objflags)->rb_node;
706 rb_parent = NULL;
707 while (*link) {
708 rb_parent = *link;
709 parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
710 untagged_objp = (unsigned long)kasan_reset_tag((void *)parent->pointer);
711 if (untagged_ptr + size <= untagged_objp)
712 link = &parent->rb_node.rb_left;
713 else if (untagged_objp + parent->size <= untagged_ptr)
714 link = &parent->rb_node.rb_right;
715 else {
716 kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
717 ptr);
718 /*
719 * No need for parent->lock here since "parent" cannot
720 * be freed while the kmemleak_lock is held.
721 */
722 dump_object_info(parent);
723 return -EEXIST;
724 }
725 }
726 rb_link_node(&object->rb_node, rb_parent, link);
727 rb_insert_color(&object->rb_node, object_tree(objflags));
728 list_add_tail_rcu(&object->object_list, &object_list);
729
730 return 0;
731 }
732
733 /*
734 * Create the metadata (struct kmemleak_object) corresponding to an allocated
735 * memory block and add it to the object_list and object tree.
736 */
737 static void __create_object(unsigned long ptr, size_t size,
738 int min_count, gfp_t gfp, unsigned int objflags)
739 {
740 struct kmemleak_object *object;
741 unsigned long flags;
742 int ret;
743
744 object = __alloc_object(gfp);
745 if (!object)
746 return;
747
748 raw_spin_lock_irqsave(&kmemleak_lock, flags);
749 ret = __link_object(object, ptr, size, min_count, objflags);
750 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
751 if (ret)
752 mem_pool_free(object);
753 }
754
755 /* Create kmemleak object which allocated with virtual address. */
756 static void create_object(unsigned long ptr, size_t size,
757 int min_count, gfp_t gfp)
758 {
759 __create_object(ptr, size, min_count, gfp, 0);
760 }
761
762 /* Create kmemleak object which allocated with physical address. */
763 static void create_object_phys(unsigned long ptr, size_t size,
764 int min_count, gfp_t gfp)
765 {
766 __create_object(ptr, size, min_count, gfp, OBJECT_PHYS);
767 }
768
769 /* Create kmemleak object corresponding to a per-CPU allocation. */
770 static void create_object_percpu(unsigned long ptr, size_t size,
771 int min_count, gfp_t gfp)
772 {
773 __create_object(ptr, size, min_count, gfp, OBJECT_PERCPU);
774 }
775
776 /*
777 * Mark the object as not allocated and schedule RCU freeing via put_object().
778 */
779 static void __delete_object(struct kmemleak_object *object)
780 {
781 unsigned long flags;
782
783 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
784 WARN_ON(atomic_read(&object->use_count) < 1);
785
786 /*
787 * Locking here also ensures that the corresponding memory block
788 * cannot be freed when it is being scanned.
789 */
790 raw_spin_lock_irqsave(&object->lock, flags);
791 object->flags &= ~OBJECT_ALLOCATED;
792 raw_spin_unlock_irqrestore(&object->lock, flags);
793 put_object(object);
794 }
795
796 /*
797 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
798 * delete it.
799 */
800 static void delete_object_full(unsigned long ptr, unsigned int objflags)
801 {
802 struct kmemleak_object *object;
803
804 object = find_and_remove_object(ptr, 0, objflags);
805 if (!object) {
806 #ifdef DEBUG
807 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
808 ptr);
809 #endif
810 return;
811 }
812 __delete_object(object);
813 }
814
815 /*
816 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
817 * delete it. If the memory block is partially freed, the function may create
818 * additional metadata for the remaining parts of the block.
819 */
820 static void delete_object_part(unsigned long ptr, size_t size,
821 unsigned int objflags)
822 {
823 struct kmemleak_object *object, *object_l, *object_r;
824 unsigned long start, end, flags;
825
826 object_l = __alloc_object(GFP_KERNEL);
827 if (!object_l)
828 return;
829
830 object_r = __alloc_object(GFP_KERNEL);
831 if (!object_r)
832 goto out;
833
834 raw_spin_lock_irqsave(&kmemleak_lock, flags);
835 object = __find_and_remove_object(ptr, 1, objflags);
836 if (!object) {
837 #ifdef DEBUG
838 kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
839 ptr, size);
840 #endif
841 goto unlock;
842 }
843
844 /*
845 * Create one or two objects that may result from the memory block
846 * split. Note that partial freeing is only done by free_bootmem() and
847 * this happens before kmemleak_init() is called.
848 */
849 start = object->pointer;
850 end = object->pointer + object->size;
851 if ((ptr > start) &&
852 !__link_object(object_l, start, ptr - start,
853 object->min_count, objflags))
854 object_l = NULL;
855 if ((ptr + size < end) &&
856 !__link_object(object_r, ptr + size, end - ptr - size,
857 object->min_count, objflags))
858 object_r = NULL;
859
860 unlock:
861 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
862 if (object)
863 __delete_object(object);
864
865 out:
866 if (object_l)
867 mem_pool_free(object_l);
868 if (object_r)
869 mem_pool_free(object_r);
870 }
871
872 static void __paint_it(struct kmemleak_object *object, int color)
873 {
874 object->min_count = color;
875 if (color == KMEMLEAK_BLACK)
876 object->flags |= OBJECT_NO_SCAN;
877 }
878
879 static void paint_it(struct kmemleak_object *object, int color)
880 {
881 unsigned long flags;
882
883 raw_spin_lock_irqsave(&object->lock, flags);
884 __paint_it(object, color);
885 raw_spin_unlock_irqrestore(&object->lock, flags);
886 }
887
888 static void paint_ptr(unsigned long ptr, int color, unsigned int objflags)
889 {
890 struct kmemleak_object *object;
891
892 object = __find_and_get_object(ptr, 0, objflags);
893 if (!object) {
894 kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n",
895 ptr,
896 (color == KMEMLEAK_GREY) ? "Grey" :
897 (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
898 return;
899 }
900 paint_it(object, color);
901 put_object(object);
902 }
903
904 /*
905 * Mark an object permanently as gray-colored so that it can no longer be
906 * reported as a leak. This is used in general to mark a false positive.
907 */
908 static void make_gray_object(unsigned long ptr)
909 {
910 paint_ptr(ptr, KMEMLEAK_GREY, 0);
911 }
912
913 /*
914 * Mark the object as black-colored so that it is ignored from scans and
915 * reporting.
916 */
917 static void make_black_object(unsigned long ptr, unsigned int objflags)
918 {
919 paint_ptr(ptr, KMEMLEAK_BLACK, objflags);
920 }
921
922 /*
923 * Add a scanning area to the object. If at least one such area is added,
924 * kmemleak will only scan these ranges rather than the whole memory block.
925 */
926 static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
927 {
928 unsigned long flags;
929 struct kmemleak_object *object;
930 struct kmemleak_scan_area *area = NULL;
931 unsigned long untagged_ptr;
932 unsigned long untagged_objp;
933
934 object = find_and_get_object(ptr, 1);
935 if (!object) {
936 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
937 ptr);
938 return;
939 }
940
941 untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
942 untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer);
943
944 if (scan_area_cache)
945 area = kmem_cache_alloc_noprof(scan_area_cache,
946 gfp_nested_mask(gfp));
947
948 raw_spin_lock_irqsave(&object->lock, flags);
949 if (!area) {
950 pr_warn_once("Cannot allocate a scan area, scanning the full object\n");
951 /* mark the object for full scan to avoid false positives */
952 object->flags |= OBJECT_FULL_SCAN;
953 goto out_unlock;
954 }
955 if (size == SIZE_MAX) {
956 size = untagged_objp + object->size - untagged_ptr;
957 } else if (untagged_ptr + size > untagged_objp + object->size) {
958 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
959 dump_object_info(object);
960 kmem_cache_free(scan_area_cache, area);
961 goto out_unlock;
962 }
963
964 INIT_HLIST_NODE(&area->node);
965 area->start = ptr;
966 area->size = size;
967
968 hlist_add_head(&area->node, &object->area_list);
969 out_unlock:
970 raw_spin_unlock_irqrestore(&object->lock, flags);
971 put_object(object);
972 }
973
974 /*
975 * Any surplus references (object already gray) to 'ptr' are passed to
976 * 'excess_ref'. This is used in the vmalloc() case where a pointer to
977 * vm_struct may be used as an alternative reference to the vmalloc'ed object
978 * (see free_thread_stack()).
979 */
980 static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref)
981 {
982 unsigned long flags;
983 struct kmemleak_object *object;
984
985 object = find_and_get_object(ptr, 0);
986 if (!object) {
987 kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n",
988 ptr);
989 return;
990 }
991
992 raw_spin_lock_irqsave(&object->lock, flags);
993 object->excess_ref = excess_ref;
994 raw_spin_unlock_irqrestore(&object->lock, flags);
995 put_object(object);
996 }
997
998 /*
999 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
1000 * pointer. Such object will not be scanned by kmemleak but references to it
1001 * are searched.
1002 */
1003 static void object_no_scan(unsigned long ptr)
1004 {
1005 unsigned long flags;
1006 struct kmemleak_object *object;
1007
1008 object = find_and_get_object(ptr, 0);
1009 if (!object) {
1010 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
1011 return;
1012 }
1013
1014 raw_spin_lock_irqsave(&object->lock, flags);
1015 object->flags |= OBJECT_NO_SCAN;
1016 raw_spin_unlock_irqrestore(&object->lock, flags);
1017 put_object(object);
1018 }
1019
1020 /**
1021 * kmemleak_alloc - register a newly allocated object
1022 * @ptr: pointer to beginning of the object
1023 * @size: size of the object
1024 * @min_count: minimum number of references to this object. If during memory
1025 * scanning a number of references less than @min_count is found,
1026 * the object is reported as a memory leak. If @min_count is 0,
1027 * the object is never reported as a leak. If @min_count is -1,
1028 * the object is ignored (not scanned and not reported as a leak)
1029 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1030 *
1031 * This function is called from the kernel allocators when a new object
1032 * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.).
1033 */
1034 void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
1035 gfp_t gfp)
1036 {
1037 pr_debug("%s(0x%px, %zu, %d)\n", __func__, ptr, size, min_count);
1038
1039 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1040 create_object((unsigned long)ptr, size, min_count, gfp);
1041 }
1042 EXPORT_SYMBOL_GPL(kmemleak_alloc);
1043
1044 /**
1045 * kmemleak_alloc_percpu - register a newly allocated __percpu object
1046 * @ptr: __percpu pointer to beginning of the object
1047 * @size: size of the object
1048 * @gfp: flags used for kmemleak internal memory allocations
1049 *
1050 * This function is called from the kernel percpu allocator when a new object
1051 * (memory block) is allocated (alloc_percpu).
1052 */
1053 void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
1054 gfp_t gfp)
1055 {
1056 pr_debug("%s(0x%px, %zu)\n", __func__, ptr, size);
1057
1058 /*
1059 * Percpu allocations are only scanned and not reported as leaks
1060 * (min_count is set to 0).
1061 */
1062 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1063 create_object_percpu((unsigned long)ptr, size, 0, gfp);
1064 }
1065 EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
1066
1067 /**
1068 * kmemleak_vmalloc - register a newly vmalloc'ed object
1069 * @area: pointer to vm_struct
1070 * @size: size of the object
1071 * @gfp: __vmalloc() flags used for kmemleak internal memory allocations
1072 *
1073 * This function is called from the vmalloc() kernel allocator when a new
1074 * object (memory block) is allocated.
1075 */
1076 void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp)
1077 {
1078 pr_debug("%s(0x%px, %zu)\n", __func__, area, size);
1079
1080 /*
1081 * A min_count = 2 is needed because vm_struct contains a reference to
1082 * the virtual address of the vmalloc'ed block.
1083 */
1084 if (kmemleak_enabled) {
1085 create_object((unsigned long)area->addr, size, 2, gfp);
1086 object_set_excess_ref((unsigned long)area,
1087 (unsigned long)area->addr);
1088 }
1089 }
1090 EXPORT_SYMBOL_GPL(kmemleak_vmalloc);
1091
1092 /**
1093 * kmemleak_free - unregister a previously registered object
1094 * @ptr: pointer to beginning of the object
1095 *
1096 * This function is called from the kernel allocators when an object (memory
1097 * block) is freed (kmem_cache_free, kfree, vfree etc.).
1098 */
1099 void __ref kmemleak_free(const void *ptr)
1100 {
1101 pr_debug("%s(0x%px)\n", __func__, ptr);
1102
1103 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
1104 delete_object_full((unsigned long)ptr, 0);
1105 }
1106 EXPORT_SYMBOL_GPL(kmemleak_free);
1107
1108 /**
1109 * kmemleak_free_part - partially unregister a previously registered object
1110 * @ptr: pointer to the beginning or inside the object. This also
1111 * represents the start of the range to be freed
1112 * @size: size to be unregistered
1113 *
1114 * This function is called when only a part of a memory block is freed
1115 * (usually from the bootmem allocator).
1116 */
1117 void __ref kmemleak_free_part(const void *ptr, size_t size)
1118 {
1119 pr_debug("%s(0x%px)\n", __func__, ptr);
1120
1121 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1122 delete_object_part((unsigned long)ptr, size, 0);
1123 }
1124 EXPORT_SYMBOL_GPL(kmemleak_free_part);
1125
1126 /**
1127 * kmemleak_free_percpu - unregister a previously registered __percpu object
1128 * @ptr: __percpu pointer to beginning of the object
1129 *
1130 * This function is called from the kernel percpu allocator when an object
1131 * (memory block) is freed (free_percpu).
1132 */
1133 void __ref kmemleak_free_percpu(const void __percpu *ptr)
1134 {
1135 pr_debug("%s(0x%px)\n", __func__, ptr);
1136
1137 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
1138 delete_object_full((unsigned long)ptr, OBJECT_PERCPU);
1139 }
1140 EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
1141
1142 /**
1143 * kmemleak_update_trace - update object allocation stack trace
1144 * @ptr: pointer to beginning of the object
1145 *
1146 * Override the object allocation stack trace for cases where the actual
1147 * allocation place is not always useful.
1148 */
1149 void __ref kmemleak_update_trace(const void *ptr)
1150 {
1151 struct kmemleak_object *object;
1152 depot_stack_handle_t trace_handle;
1153 unsigned long flags;
1154
1155 pr_debug("%s(0x%px)\n", __func__, ptr);
1156
1157 if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
1158 return;
1159
1160 object = find_and_get_object((unsigned long)ptr, 1);
1161 if (!object) {
1162 #ifdef DEBUG
1163 kmemleak_warn("Updating stack trace for unknown object at %p\n",
1164 ptr);
1165 #endif
1166 return;
1167 }
1168
1169 trace_handle = set_track_prepare();
1170 raw_spin_lock_irqsave(&object->lock, flags);
1171 object->trace_handle = trace_handle;
1172 raw_spin_unlock_irqrestore(&object->lock, flags);
1173
1174 put_object(object);
1175 }
1176 EXPORT_SYMBOL(kmemleak_update_trace);
1177
1178 /**
1179 * kmemleak_not_leak - mark an allocated object as false positive
1180 * @ptr: pointer to beginning of the object
1181 *
1182 * Calling this function on an object will cause the memory block to no longer
1183 * be reported as leak and always be scanned.
1184 */
1185 void __ref kmemleak_not_leak(const void *ptr)
1186 {
1187 pr_debug("%s(0x%px)\n", __func__, ptr);
1188
1189 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1190 make_gray_object((unsigned long)ptr);
1191 }
1192 EXPORT_SYMBOL(kmemleak_not_leak);
1193
1194 /**
1195 * kmemleak_ignore - ignore an allocated object
1196 * @ptr: pointer to beginning of the object
1197 *
1198 * Calling this function on an object will cause the memory block to be
1199 * ignored (not scanned and not reported as a leak). This is usually done when
1200 * it is known that the corresponding block is not a leak and does not contain
1201 * any references to other allocated memory blocks.
1202 */
1203 void __ref kmemleak_ignore(const void *ptr)
1204 {
1205 pr_debug("%s(0x%px)\n", __func__, ptr);
1206
1207 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1208 make_black_object((unsigned long)ptr, 0);
1209 }
1210 EXPORT_SYMBOL(kmemleak_ignore);
1211
1212 /**
1213 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1214 * @ptr: pointer to beginning or inside the object. This also
1215 * represents the start of the scan area
1216 * @size: size of the scan area
1217 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1218 *
1219 * This function is used when it is known that only certain parts of an object
1220 * contain references to other objects. Kmemleak will only scan these areas
1221 * reducing the number false negatives.
1222 */
1223 void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1224 {
1225 pr_debug("%s(0x%px)\n", __func__, ptr);
1226
1227 if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
1228 add_scan_area((unsigned long)ptr, size, gfp);
1229 }
1230 EXPORT_SYMBOL(kmemleak_scan_area);
1231
1232 /**
1233 * kmemleak_no_scan - do not scan an allocated object
1234 * @ptr: pointer to beginning of the object
1235 *
1236 * This function notifies kmemleak not to scan the given memory block. Useful
1237 * in situations where it is known that the given object does not contain any
1238 * references to other objects. Kmemleak will not scan such objects reducing
1239 * the number of false negatives.
1240 */
1241 void __ref kmemleak_no_scan(const void *ptr)
1242 {
1243 pr_debug("%s(0x%px)\n", __func__, ptr);
1244
1245 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1246 object_no_scan((unsigned long)ptr);
1247 }
1248 EXPORT_SYMBOL(kmemleak_no_scan);
1249
1250 /**
1251 * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical
1252 * address argument
1253 * @phys: physical address of the object
1254 * @size: size of the object
1255 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1256 */
1257 void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, gfp_t gfp)
1258 {
1259 pr_debug("%s(0x%px, %zu)\n", __func__, &phys, size);
1260
1261 if (kmemleak_enabled)
1262 /*
1263 * Create object with OBJECT_PHYS flag and
1264 * assume min_count 0.
1265 */
1266 create_object_phys((unsigned long)phys, size, 0, gfp);
1267 }
1268 EXPORT_SYMBOL(kmemleak_alloc_phys);
1269
1270 /**
1271 * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a
1272 * physical address argument
1273 * @phys: physical address if the beginning or inside an object. This
1274 * also represents the start of the range to be freed
1275 * @size: size to be unregistered
1276 */
1277 void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size)
1278 {
1279 pr_debug("%s(0x%px)\n", __func__, &phys);
1280
1281 if (kmemleak_enabled)
1282 delete_object_part((unsigned long)phys, size, OBJECT_PHYS);
1283 }
1284 EXPORT_SYMBOL(kmemleak_free_part_phys);
1285
1286 /**
1287 * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical
1288 * address argument
1289 * @phys: physical address of the object
1290 */
1291 void __ref kmemleak_ignore_phys(phys_addr_t phys)
1292 {
1293 pr_debug("%s(0x%px)\n", __func__, &phys);
1294
1295 if (kmemleak_enabled)
1296 make_black_object((unsigned long)phys, OBJECT_PHYS);
1297 }
1298 EXPORT_SYMBOL(kmemleak_ignore_phys);
1299
1300 /*
1301 * Update an object's checksum and return true if it was modified.
1302 */
1303 static bool update_checksum(struct kmemleak_object *object)
1304 {
1305 u32 old_csum = object->checksum;
1306
1307 if (WARN_ON_ONCE(object->flags & (OBJECT_PHYS | OBJECT_PERCPU)))
1308 return false;
1309
1310 kasan_disable_current();
1311 kcsan_disable_current();
1312 object->checksum = crc32(0, kasan_reset_tag((void *)object->pointer), object->size);
1313 kasan_enable_current();
1314 kcsan_enable_current();
1315
1316 return object->checksum != old_csum;
1317 }
1318
1319 /*
1320 * Update an object's references. object->lock must be held by the caller.
1321 */
1322 static void update_refs(struct kmemleak_object *object)
1323 {
1324 if (!color_white(object)) {
1325 /* non-orphan, ignored or new */
1326 return;
1327 }
1328
1329 /*
1330 * Increase the object's reference count (number of pointers to the
1331 * memory block). If this count reaches the required minimum, the
1332 * object's color will become gray and it will be added to the
1333 * gray_list.
1334 */
1335 object->count++;
1336 if (color_gray(object)) {
1337 /* put_object() called when removing from gray_list */
1338 WARN_ON(!get_object(object));
1339 list_add_tail(&object->gray_list, &gray_list);
1340 }
1341 }
1342
1343 /*
1344 * Memory scanning is a long process and it needs to be interruptible. This
1345 * function checks whether such interrupt condition occurred.
1346 */
1347 static int scan_should_stop(void)
1348 {
1349 if (!kmemleak_enabled)
1350 return 1;
1351
1352 /*
1353 * This function may be called from either process or kthread context,
1354 * hence the need to check for both stop conditions.
1355 */
1356 if (current->mm)
1357 return signal_pending(current);
1358 else
1359 return kthread_should_stop();
1360
1361 return 0;
1362 }
1363
1364 /*
1365 * Scan a memory block (exclusive range) for valid pointers and add those
1366 * found to the gray list.
1367 */
1368 static void scan_block(void *_start, void *_end,
1369 struct kmemleak_object *scanned)
1370 {
1371 unsigned long *ptr;
1372 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1373 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1374 unsigned long flags;
1375 unsigned long untagged_ptr;
1376
1377 raw_spin_lock_irqsave(&kmemleak_lock, flags);
1378 for (ptr = start; ptr < end; ptr++) {
1379 struct kmemleak_object *object;
1380 unsigned long pointer;
1381 unsigned long excess_ref;
1382
1383 if (scan_should_stop())
1384 break;
1385
1386 kasan_disable_current();
1387 pointer = *(unsigned long *)kasan_reset_tag((void *)ptr);
1388 kasan_enable_current();
1389
1390 untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer);
1391 if (untagged_ptr < min_addr || untagged_ptr >= max_addr)
1392 continue;
1393
1394 /*
1395 * No need for get_object() here since we hold kmemleak_lock.
1396 * object->use_count cannot be dropped to 0 while the object
1397 * is still present in object_tree_root and object_list
1398 * (with updates protected by kmemleak_lock).
1399 */
1400 object = lookup_object(pointer, 1);
1401 if (!object)
1402 continue;
1403 if (object == scanned)
1404 /* self referenced, ignore */
1405 continue;
1406
1407 /*
1408 * Avoid the lockdep recursive warning on object->lock being
1409 * previously acquired in scan_object(). These locks are
1410 * enclosed by scan_mutex.
1411 */
1412 raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1413 /* only pass surplus references (object already gray) */
1414 if (color_gray(object)) {
1415 excess_ref = object->excess_ref;
1416 /* no need for update_refs() if object already gray */
1417 } else {
1418 excess_ref = 0;
1419 update_refs(object);
1420 }
1421 raw_spin_unlock(&object->lock);
1422
1423 if (excess_ref) {
1424 object = lookup_object(excess_ref, 0);
1425 if (!object)
1426 continue;
1427 if (object == scanned)
1428 /* circular reference, ignore */
1429 continue;
1430 raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1431 update_refs(object);
1432 raw_spin_unlock(&object->lock);
1433 }
1434 }
1435 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
1436 }
1437
1438 /*
1439 * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
1440 */
1441 #ifdef CONFIG_SMP
1442 static void scan_large_block(void *start, void *end)
1443 {
1444 void *next;
1445
1446 while (start < end) {
1447 next = min(start + MAX_SCAN_SIZE, end);
1448 scan_block(start, next, NULL);
1449 start = next;
1450 cond_resched();
1451 }
1452 }
1453 #endif
1454
1455 /*
1456 * Scan a memory block corresponding to a kmemleak_object. A condition is
1457 * that object->use_count >= 1.
1458 */
1459 static void scan_object(struct kmemleak_object *object)
1460 {
1461 struct kmemleak_scan_area *area;
1462 unsigned long flags;
1463
1464 /*
1465 * Once the object->lock is acquired, the corresponding memory block
1466 * cannot be freed (the same lock is acquired in delete_object).
1467 */
1468 raw_spin_lock_irqsave(&object->lock, flags);
1469 if (object->flags & OBJECT_NO_SCAN)
1470 goto out;
1471 if (!(object->flags & OBJECT_ALLOCATED))
1472 /* already freed object */
1473 goto out;
1474
1475 if (object->flags & OBJECT_PERCPU) {
1476 unsigned int cpu;
1477
1478 for_each_possible_cpu(cpu) {
1479 void *start = per_cpu_ptr((void __percpu *)object->pointer, cpu);
1480 void *end = start + object->size;
1481
1482 scan_block(start, end, object);
1483
1484 raw_spin_unlock_irqrestore(&object->lock, flags);
1485 cond_resched();
1486 raw_spin_lock_irqsave(&object->lock, flags);
1487 if (!(object->flags & OBJECT_ALLOCATED))
1488 break;
1489 }
1490 } else if (hlist_empty(&object->area_list) ||
1491 object->flags & OBJECT_FULL_SCAN) {
1492 void *start = object->flags & OBJECT_PHYS ?
1493 __va((phys_addr_t)object->pointer) :
1494 (void *)object->pointer;
1495 void *end = start + object->size;
1496 void *next;
1497
1498 do {
1499 next = min(start + MAX_SCAN_SIZE, end);
1500 scan_block(start, next, object);
1501
1502 start = next;
1503 if (start >= end)
1504 break;
1505
1506 raw_spin_unlock_irqrestore(&object->lock, flags);
1507 cond_resched();
1508 raw_spin_lock_irqsave(&object->lock, flags);
1509 } while (object->flags & OBJECT_ALLOCATED);
1510 } else {
1511 hlist_for_each_entry(area, &object->area_list, node)
1512 scan_block((void *)area->start,
1513 (void *)(area->start + area->size),
1514 object);
1515 }
1516 out:
1517 raw_spin_unlock_irqrestore(&object->lock, flags);
1518 }
1519
1520 /*
1521 * Scan the objects already referenced (gray objects). More objects will be
1522 * referenced and, if there are no memory leaks, all the objects are scanned.
1523 */
1524 static void scan_gray_list(void)
1525 {
1526 struct kmemleak_object *object, *tmp;
1527
1528 /*
1529 * The list traversal is safe for both tail additions and removals
1530 * from inside the loop. The kmemleak objects cannot be freed from
1531 * outside the loop because their use_count was incremented.
1532 */
1533 object = list_entry(gray_list.next, typeof(*object), gray_list);
1534 while (&object->gray_list != &gray_list) {
1535 cond_resched();
1536
1537 /* may add new objects to the list */
1538 if (!scan_should_stop())
1539 scan_object(object);
1540
1541 tmp = list_entry(object->gray_list.next, typeof(*object),
1542 gray_list);
1543
1544 /* remove the object from the list and release it */
1545 list_del(&object->gray_list);
1546 put_object(object);
1547
1548 object = tmp;
1549 }
1550 WARN_ON(!list_empty(&gray_list));
1551 }
1552
1553 /*
1554 * Conditionally call resched() in an object iteration loop while making sure
1555 * that the given object won't go away without RCU read lock by performing a
1556 * get_object() if necessaary.
1557 */
1558 static void kmemleak_cond_resched(struct kmemleak_object *object)
1559 {
1560 if (!get_object(object))
1561 return; /* Try next object */
1562
1563 raw_spin_lock_irq(&kmemleak_lock);
1564 if (object->del_state & DELSTATE_REMOVED)
1565 goto unlock_put; /* Object removed */
1566 object->del_state |= DELSTATE_NO_DELETE;
1567 raw_spin_unlock_irq(&kmemleak_lock);
1568
1569 rcu_read_unlock();
1570 cond_resched();
1571 rcu_read_lock();
1572
1573 raw_spin_lock_irq(&kmemleak_lock);
1574 if (object->del_state & DELSTATE_REMOVED)
1575 list_del_rcu(&object->object_list);
1576 object->del_state &= ~DELSTATE_NO_DELETE;
1577 unlock_put:
1578 raw_spin_unlock_irq(&kmemleak_lock);
1579 put_object(object);
1580 }
1581
1582 /*
1583 * Scan data sections and all the referenced memory blocks allocated via the
1584 * kernel's standard allocators. This function must be called with the
1585 * scan_mutex held.
1586 */
1587 static void kmemleak_scan(void)
1588 {
1589 struct kmemleak_object *object;
1590 struct zone *zone;
1591 int __maybe_unused i;
1592 int new_leaks = 0;
1593
1594 jiffies_last_scan = jiffies;
1595
1596 /* prepare the kmemleak_object's */
1597 rcu_read_lock();
1598 list_for_each_entry_rcu(object, &object_list, object_list) {
1599 raw_spin_lock_irq(&object->lock);
1600 #ifdef DEBUG
1601 /*
1602 * With a few exceptions there should be a maximum of
1603 * 1 reference to any object at this point.
1604 */
1605 if (atomic_read(&object->use_count) > 1) {
1606 pr_debug("object->use_count = %d\n",
1607 atomic_read(&object->use_count));
1608 dump_object_info(object);
1609 }
1610 #endif
1611
1612 /* ignore objects outside lowmem (paint them black) */
1613 if ((object->flags & OBJECT_PHYS) &&
1614 !(object->flags & OBJECT_NO_SCAN)) {
1615 unsigned long phys = object->pointer;
1616
1617 if (PHYS_PFN(phys) < min_low_pfn ||
1618 PHYS_PFN(phys + object->size) >= max_low_pfn)
1619 __paint_it(object, KMEMLEAK_BLACK);
1620 }
1621
1622 /* reset the reference count (whiten the object) */
1623 object->count = 0;
1624 if (color_gray(object) && get_object(object))
1625 list_add_tail(&object->gray_list, &gray_list);
1626
1627 raw_spin_unlock_irq(&object->lock);
1628
1629 if (need_resched())
1630 kmemleak_cond_resched(object);
1631 }
1632 rcu_read_unlock();
1633
1634 #ifdef CONFIG_SMP
1635 /* per-cpu sections scanning */
1636 for_each_possible_cpu(i)
1637 scan_large_block(__per_cpu_start + per_cpu_offset(i),
1638 __per_cpu_end + per_cpu_offset(i));
1639 #endif
1640
1641 /*
1642 * Struct page scanning for each node.
1643 */
1644 get_online_mems();
1645 for_each_populated_zone(zone) {
1646 unsigned long start_pfn = zone->zone_start_pfn;
1647 unsigned long end_pfn = zone_end_pfn(zone);
1648 unsigned long pfn;
1649
1650 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1651 struct page *page = pfn_to_online_page(pfn);
1652
1653 if (!(pfn & 63))
1654 cond_resched();
1655
1656 if (!page)
1657 continue;
1658
1659 /* only scan pages belonging to this zone */
1660 if (page_zone(page) != zone)
1661 continue;
1662 /* only scan if page is in use */
1663 if (page_count(page) == 0)
1664 continue;
1665 scan_block(page, page + 1, NULL);
1666 }
1667 }
1668 put_online_mems();
1669
1670 /*
1671 * Scanning the task stacks (may introduce false negatives).
1672 */
1673 if (kmemleak_stack_scan) {
1674 struct task_struct *p, *g;
1675
1676 rcu_read_lock();
1677 for_each_process_thread(g, p) {
1678 void *stack = try_get_task_stack(p);
1679 if (stack) {
1680 scan_block(stack, stack + THREAD_SIZE, NULL);
1681 put_task_stack(p);
1682 }
1683 }
1684 rcu_read_unlock();
1685 }
1686
1687 /*
1688 * Scan the objects already referenced from the sections scanned
1689 * above.
1690 */
1691 scan_gray_list();
1692
1693 /*
1694 * Check for new or unreferenced objects modified since the previous
1695 * scan and color them gray until the next scan.
1696 */
1697 rcu_read_lock();
1698 list_for_each_entry_rcu(object, &object_list, object_list) {
1699 if (need_resched())
1700 kmemleak_cond_resched(object);
1701
1702 /*
1703 * This is racy but we can save the overhead of lock/unlock
1704 * calls. The missed objects, if any, should be caught in
1705 * the next scan.
1706 */
1707 if (!color_white(object))
1708 continue;
1709 raw_spin_lock_irq(&object->lock);
1710 if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1711 && update_checksum(object) && get_object(object)) {
1712 /* color it gray temporarily */
1713 object->count = object->min_count;
1714 list_add_tail(&object->gray_list, &gray_list);
1715 }
1716 raw_spin_unlock_irq(&object->lock);
1717 }
1718 rcu_read_unlock();
1719
1720 /*
1721 * Re-scan the gray list for modified unreferenced objects.
1722 */
1723 scan_gray_list();
1724
1725 /*
1726 * If scanning was stopped do not report any new unreferenced objects.
1727 */
1728 if (scan_should_stop())
1729 return;
1730
1731 /*
1732 * Scanning result reporting.
1733 */
1734 rcu_read_lock();
1735 list_for_each_entry_rcu(object, &object_list, object_list) {
1736 if (need_resched())
1737 kmemleak_cond_resched(object);
1738
1739 /*
1740 * This is racy but we can save the overhead of lock/unlock
1741 * calls. The missed objects, if any, should be caught in
1742 * the next scan.
1743 */
1744 if (!color_white(object))
1745 continue;
1746 raw_spin_lock_irq(&object->lock);
1747 if (unreferenced_object(object) &&
1748 !(object->flags & OBJECT_REPORTED)) {
1749 object->flags |= OBJECT_REPORTED;
1750
1751 if (kmemleak_verbose)
1752 print_unreferenced(NULL, object);
1753
1754 new_leaks++;
1755 }
1756 raw_spin_unlock_irq(&object->lock);
1757 }
1758 rcu_read_unlock();
1759
1760 if (new_leaks) {
1761 kmemleak_found_leaks = true;
1762
1763 pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
1764 new_leaks);
1765 }
1766
1767 }
1768
1769 /*
1770 * Thread function performing automatic memory scanning. Unreferenced objects
1771 * at the end of a memory scan are reported but only the first time.
1772 */
1773 static int kmemleak_scan_thread(void *arg)
1774 {
1775 static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN);
1776
1777 pr_info("Automatic memory scanning thread started\n");
1778 set_user_nice(current, 10);
1779
1780 /*
1781 * Wait before the first scan to allow the system to fully initialize.
1782 */
1783 if (first_run) {
1784 signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000);
1785 first_run = 0;
1786 while (timeout && !kthread_should_stop())
1787 timeout = schedule_timeout_interruptible(timeout);
1788 }
1789
1790 while (!kthread_should_stop()) {
1791 signed long timeout = READ_ONCE(jiffies_scan_wait);
1792
1793 mutex_lock(&scan_mutex);
1794 kmemleak_scan();
1795 mutex_unlock(&scan_mutex);
1796
1797 /* wait before the next scan */
1798 while (timeout && !kthread_should_stop())
1799 timeout = schedule_timeout_interruptible(timeout);
1800 }
1801
1802 pr_info("Automatic memory scanning thread ended\n");
1803
1804 return 0;
1805 }
1806
1807 /*
1808 * Start the automatic memory scanning thread. This function must be called
1809 * with the scan_mutex held.
1810 */
1811 static void start_scan_thread(void)
1812 {
1813 if (scan_thread)
1814 return;
1815 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1816 if (IS_ERR(scan_thread)) {
1817 pr_warn("Failed to create the scan thread\n");
1818 scan_thread = NULL;
1819 }
1820 }
1821
1822 /*
1823 * Stop the automatic memory scanning thread.
1824 */
1825 static void stop_scan_thread(void)
1826 {
1827 if (scan_thread) {
1828 kthread_stop(scan_thread);
1829 scan_thread = NULL;
1830 }
1831 }
1832
1833 /*
1834 * Iterate over the object_list and return the first valid object at or after
1835 * the required position with its use_count incremented. The function triggers
1836 * a memory scanning when the pos argument points to the first position.
1837 */
1838 static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1839 {
1840 struct kmemleak_object *object;
1841 loff_t n = *pos;
1842 int err;
1843
1844 err = mutex_lock_interruptible(&scan_mutex);
1845 if (err < 0)
1846 return ERR_PTR(err);
1847
1848 rcu_read_lock();
1849 list_for_each_entry_rcu(object, &object_list, object_list) {
1850 if (n-- > 0)
1851 continue;
1852 if (get_object(object))
1853 goto out;
1854 }
1855 object = NULL;
1856 out:
1857 return object;
1858 }
1859
1860 /*
1861 * Return the next object in the object_list. The function decrements the
1862 * use_count of the previous object and increases that of the next one.
1863 */
1864 static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1865 {
1866 struct kmemleak_object *prev_obj = v;
1867 struct kmemleak_object *next_obj = NULL;
1868 struct kmemleak_object *obj = prev_obj;
1869
1870 ++(*pos);
1871
1872 list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
1873 if (get_object(obj)) {
1874 next_obj = obj;
1875 break;
1876 }
1877 }
1878
1879 put_object(prev_obj);
1880 return next_obj;
1881 }
1882
1883 /*
1884 * Decrement the use_count of the last object required, if any.
1885 */
1886 static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1887 {
1888 if (!IS_ERR(v)) {
1889 /*
1890 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1891 * waiting was interrupted, so only release it if !IS_ERR.
1892 */
1893 rcu_read_unlock();
1894 mutex_unlock(&scan_mutex);
1895 if (v)
1896 put_object(v);
1897 }
1898 }
1899
1900 /*
1901 * Print the information for an unreferenced object to the seq file.
1902 */
1903 static int kmemleak_seq_show(struct seq_file *seq, void *v)
1904 {
1905 struct kmemleak_object *object = v;
1906 unsigned long flags;
1907
1908 raw_spin_lock_irqsave(&object->lock, flags);
1909 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1910 print_unreferenced(seq, object);
1911 raw_spin_unlock_irqrestore(&object->lock, flags);
1912 return 0;
1913 }
1914
1915 static const struct seq_operations kmemleak_seq_ops = {
1916 .start = kmemleak_seq_start,
1917 .next = kmemleak_seq_next,
1918 .stop = kmemleak_seq_stop,
1919 .show = kmemleak_seq_show,
1920 };
1921
1922 static int kmemleak_open(struct inode *inode, struct file *file)
1923 {
1924 return seq_open(file, &kmemleak_seq_ops);
1925 }
1926
1927 static int dump_str_object_info(const char *str)
1928 {
1929 unsigned long flags;
1930 struct kmemleak_object *object;
1931 unsigned long addr;
1932
1933 if (kstrtoul(str, 0, &addr))
1934 return -EINVAL;
1935 object = find_and_get_object(addr, 0);
1936 if (!object) {
1937 pr_info("Unknown object at 0x%08lx\n", addr);
1938 return -EINVAL;
1939 }
1940
1941 raw_spin_lock_irqsave(&object->lock, flags);
1942 dump_object_info(object);
1943 raw_spin_unlock_irqrestore(&object->lock, flags);
1944
1945 put_object(object);
1946 return 0;
1947 }
1948
1949 /*
1950 * We use grey instead of black to ensure we can do future scans on the same
1951 * objects. If we did not do future scans these black objects could
1952 * potentially contain references to newly allocated objects in the future and
1953 * we'd end up with false positives.
1954 */
1955 static void kmemleak_clear(void)
1956 {
1957 struct kmemleak_object *object;
1958
1959 rcu_read_lock();
1960 list_for_each_entry_rcu(object, &object_list, object_list) {
1961 raw_spin_lock_irq(&object->lock);
1962 if ((object->flags & OBJECT_REPORTED) &&
1963 unreferenced_object(object))
1964 __paint_it(object, KMEMLEAK_GREY);
1965 raw_spin_unlock_irq(&object->lock);
1966 }
1967 rcu_read_unlock();
1968
1969 kmemleak_found_leaks = false;
1970 }
1971
1972 static void __kmemleak_do_cleanup(void);
1973
1974 /*
1975 * File write operation to configure kmemleak at run-time. The following
1976 * commands can be written to the /sys/kernel/debug/kmemleak file:
1977 * off - disable kmemleak (irreversible)
1978 * stack=on - enable the task stacks scanning
1979 * stack=off - disable the tasks stacks scanning
1980 * scan=on - start the automatic memory scanning thread
1981 * scan=off - stop the automatic memory scanning thread
1982 * scan=... - set the automatic memory scanning period in seconds (0 to
1983 * disable it)
1984 * scan - trigger a memory scan
1985 * clear - mark all current reported unreferenced kmemleak objects as
1986 * grey to ignore printing them, or free all kmemleak objects
1987 * if kmemleak has been disabled.
1988 * dump=... - dump information about the object found at the given address
1989 */
1990 static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1991 size_t size, loff_t *ppos)
1992 {
1993 char buf[64];
1994 int buf_size;
1995 int ret;
1996
1997 buf_size = min(size, (sizeof(buf) - 1));
1998 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1999 return -EFAULT;
2000 buf[buf_size] = 0;
2001
2002 ret = mutex_lock_interruptible(&scan_mutex);
2003 if (ret < 0)
2004 return ret;
2005
2006 if (strncmp(buf, "clear", 5) == 0) {
2007 if (kmemleak_enabled)
2008 kmemleak_clear();
2009 else
2010 __kmemleak_do_cleanup();
2011 goto out;
2012 }
2013
2014 if (!kmemleak_enabled) {
2015 ret = -EPERM;
2016 goto out;
2017 }
2018
2019 if (strncmp(buf, "off", 3) == 0)
2020 kmemleak_disable();
2021 else if (strncmp(buf, "stack=on", 8) == 0)
2022 kmemleak_stack_scan = 1;
2023 else if (strncmp(buf, "stack=off", 9) == 0)
2024 kmemleak_stack_scan = 0;
2025 else if (strncmp(buf, "scan=on", 7) == 0)
2026 start_scan_thread();
2027 else if (strncmp(buf, "scan=off", 8) == 0)
2028 stop_scan_thread();
2029 else if (strncmp(buf, "scan=", 5) == 0) {
2030 unsigned secs;
2031 unsigned long msecs;
2032
2033 ret = kstrtouint(buf + 5, 0, &secs);
2034 if (ret < 0)
2035 goto out;
2036
2037 msecs = secs * MSEC_PER_SEC;
2038 if (msecs > UINT_MAX)
2039 msecs = UINT_MAX;
2040
2041 stop_scan_thread();
2042 if (msecs) {
2043 WRITE_ONCE(jiffies_scan_wait, msecs_to_jiffies(msecs));
2044 start_scan_thread();
2045 }
2046 } else if (strncmp(buf, "scan", 4) == 0)
2047 kmemleak_scan();
2048 else if (strncmp(buf, "dump=", 5) == 0)
2049 ret = dump_str_object_info(buf + 5);
2050 else
2051 ret = -EINVAL;
2052
2053 out:
2054 mutex_unlock(&scan_mutex);
2055 if (ret < 0)
2056 return ret;
2057
2058 /* ignore the rest of the buffer, only one command at a time */
2059 *ppos += size;
2060 return size;
2061 }
2062
2063 static const struct file_operations kmemleak_fops = {
2064 .owner = THIS_MODULE,
2065 .open = kmemleak_open,
2066 .read = seq_read,
2067 .write = kmemleak_write,
2068 .llseek = seq_lseek,
2069 .release = seq_release,
2070 };
2071
2072 static void __kmemleak_do_cleanup(void)
2073 {
2074 struct kmemleak_object *object, *tmp;
2075
2076 /*
2077 * Kmemleak has already been disabled, no need for RCU list traversal
2078 * or kmemleak_lock held.
2079 */
2080 list_for_each_entry_safe(object, tmp, &object_list, object_list) {
2081 __remove_object(object);
2082 __delete_object(object);
2083 }
2084 }
2085
2086 /*
2087 * Stop the memory scanning thread and free the kmemleak internal objects if
2088 * no previous scan thread (otherwise, kmemleak may still have some useful
2089 * information on memory leaks).
2090 */
2091 static void kmemleak_do_cleanup(struct work_struct *work)
2092 {
2093 stop_scan_thread();
2094
2095 mutex_lock(&scan_mutex);
2096 /*
2097 * Once it is made sure that kmemleak_scan has stopped, it is safe to no
2098 * longer track object freeing. Ordering of the scan thread stopping and
2099 * the memory accesses below is guaranteed by the kthread_stop()
2100 * function.
2101 */
2102 kmemleak_free_enabled = 0;
2103 mutex_unlock(&scan_mutex);
2104
2105 if (!kmemleak_found_leaks)
2106 __kmemleak_do_cleanup();
2107 else
2108 pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
2109 }
2110
2111 static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
2112
2113 /*
2114 * Disable kmemleak. No memory allocation/freeing will be traced once this
2115 * function is called. Disabling kmemleak is an irreversible operation.
2116 */
2117 static void kmemleak_disable(void)
2118 {
2119 /* atomically check whether it was already invoked */
2120 if (cmpxchg(&kmemleak_error, 0, 1))
2121 return;
2122
2123 /* stop any memory operation tracing */
2124 kmemleak_enabled = 0;
2125
2126 /* check whether it is too early for a kernel thread */
2127 if (kmemleak_late_initialized)
2128 schedule_work(&cleanup_work);
2129 else
2130 kmemleak_free_enabled = 0;
2131
2132 pr_info("Kernel memory leak detector disabled\n");
2133 }
2134
2135 /*
2136 * Allow boot-time kmemleak disabling (enabled by default).
2137 */
2138 static int __init kmemleak_boot_config(char *str)
2139 {
2140 if (!str)
2141 return -EINVAL;
2142 if (strcmp(str, "off") == 0)
2143 kmemleak_disable();
2144 else if (strcmp(str, "on") == 0) {
2145 kmemleak_skip_disable = 1;
2146 stack_depot_request_early_init();
2147 }
2148 else
2149 return -EINVAL;
2150 return 0;
2151 }
2152 early_param("kmemleak", kmemleak_boot_config);
2153
2154 /*
2155 * Kmemleak initialization.
2156 */
2157 void __init kmemleak_init(void)
2158 {
2159 #ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
2160 if (!kmemleak_skip_disable) {
2161 kmemleak_disable();
2162 return;
2163 }
2164 #endif
2165
2166 if (kmemleak_error)
2167 return;
2168
2169 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
2170 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
2171
2172 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
2173 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
2174
2175 /* register the data/bss sections */
2176 create_object((unsigned long)_sdata, _edata - _sdata,
2177 KMEMLEAK_GREY, GFP_ATOMIC);
2178 create_object((unsigned long)__bss_start, __bss_stop - __bss_start,
2179 KMEMLEAK_GREY, GFP_ATOMIC);
2180 /* only register .data..ro_after_init if not within .data */
2181 if (&__start_ro_after_init < &_sdata || &__end_ro_after_init > &_edata)
2182 create_object((unsigned long)__start_ro_after_init,
2183 __end_ro_after_init - __start_ro_after_init,
2184 KMEMLEAK_GREY, GFP_ATOMIC);
2185 }
2186
2187 /*
2188 * Late initialization function.
2189 */
2190 static int __init kmemleak_late_init(void)
2191 {
2192 kmemleak_late_initialized = 1;
2193
2194 debugfs_create_file("kmemleak", 0644, NULL, NULL, &kmemleak_fops);
2195
2196 if (kmemleak_error) {
2197 /*
2198 * Some error occurred and kmemleak was disabled. There is a
2199 * small chance that kmemleak_disable() was called immediately
2200 * after setting kmemleak_late_initialized and we may end up with
2201 * two clean-up threads but serialized by scan_mutex.
2202 */
2203 schedule_work(&cleanup_work);
2204 return -ENOMEM;
2205 }
2206
2207 if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) {
2208 mutex_lock(&scan_mutex);
2209 start_scan_thread();
2210 mutex_unlock(&scan_mutex);
2211 }
2212
2213 pr_info("Kernel memory leak detector initialized (mem pool available: %d)\n",
2214 mem_pool_free_count);
2215
2216 return 0;
2217 }
2218 late_initcall(kmemleak_late_init);
2219
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.