xref: /linux/mm/kmemleak.c (revision 9702969978695d9a699a1f34771580cdbb153b33)
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 __read_mostly = 1;
214 /* same as above but only for the kmemleak_free() callback */
215 static int kmemleak_free_enabled __read_mostly = 1;
216 /* set in the late_initcall if there were no errors */
217 static int kmemleak_late_initialized;
218 /* set if a fatal kmemleak error has occurred */
219 static int kmemleak_error;
220 
221 /* minimum and maximum address that may be valid pointers */
222 static unsigned long min_addr = ULONG_MAX;
223 static unsigned long max_addr;
224 
225 /* minimum and maximum address that may be valid per-CPU pointers */
226 static unsigned long min_percpu_addr = ULONG_MAX;
227 static unsigned long max_percpu_addr;
228 
229 static struct task_struct *scan_thread;
230 /* used to avoid reporting of recently allocated objects */
231 static unsigned long jiffies_min_age;
232 static unsigned long jiffies_last_scan;
233 /* delay between automatic memory scannings */
234 static unsigned long jiffies_scan_wait;
235 /* enables or disables the task stacks scanning */
236 static int kmemleak_stack_scan = 1;
237 /* protects the memory scanning, parameters and debug/kmemleak file access */
238 static DEFINE_MUTEX(scan_mutex);
239 /* setting kmemleak=on, will set this var, skipping the disable */
240 static int kmemleak_skip_disable;
241 /* If there are leaks that can be reported */
242 static bool kmemleak_found_leaks;
243 
244 static bool kmemleak_verbose;
245 module_param_named(verbose, kmemleak_verbose, bool, 0600);
246 
247 static void kmemleak_disable(void);
248 
249 /*
250  * Print a warning and dump the stack trace.
251  */
252 #define kmemleak_warn(x...)	do {		\
253 	pr_warn(x);				\
254 	dump_stack();				\
255 } while (0)
256 
257 /*
258  * Macro invoked when a serious kmemleak condition occurred and cannot be
259  * recovered from. Kmemleak will be disabled and further allocation/freeing
260  * tracing no longer available.
261  */
262 #define kmemleak_stop(x...)	do {	\
263 	kmemleak_warn(x);		\
264 	kmemleak_disable();		\
265 } while (0)
266 
267 #define warn_or_seq_printf(seq, fmt, ...)	do {	\
268 	if (seq)					\
269 		seq_printf(seq, fmt, ##__VA_ARGS__);	\
270 	else						\
271 		pr_warn(fmt, ##__VA_ARGS__);		\
272 } while (0)
273 
274 static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type,
275 				 int rowsize, int groupsize, const void *buf,
276 				 size_t len, bool ascii)
277 {
278 	if (seq)
279 		seq_hex_dump(seq, HEX_PREFIX, prefix_type, rowsize, groupsize,
280 			     buf, len, ascii);
281 	else
282 		print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type,
283 			       rowsize, groupsize, buf, len, ascii);
284 }
285 
286 /*
287  * Printing of the objects hex dump to the seq file. The number of lines to be
288  * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
289  * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
290  * with the object->lock held.
291  */
292 static void hex_dump_object(struct seq_file *seq,
293 			    struct kmemleak_object *object)
294 {
295 	const u8 *ptr = (const u8 *)object->pointer;
296 	size_t len;
297 
298 	if (WARN_ON_ONCE(object->flags & OBJECT_PHYS))
299 		return;
300 
301 	if (object->flags & OBJECT_PERCPU)
302 		ptr = (const u8 *)this_cpu_ptr((void __percpu *)object->pointer);
303 
304 	/* limit the number of lines to HEX_MAX_LINES */
305 	len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
306 
307 	if (object->flags & OBJECT_PERCPU)
308 		warn_or_seq_printf(seq, "  hex dump (first %zu bytes on cpu %d):\n",
309 				   len, raw_smp_processor_id());
310 	else
311 		warn_or_seq_printf(seq, "  hex dump (first %zu bytes):\n", len);
312 	kasan_disable_current();
313 	warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE,
314 			     HEX_GROUP_SIZE, kasan_reset_tag((void *)ptr), len, HEX_ASCII);
315 	kasan_enable_current();
316 }
317 
318 /*
319  * Object colors, encoded with count and min_count:
320  * - white - orphan object, not enough references to it (count < min_count)
321  * - gray  - not orphan, not marked as false positive (min_count == 0) or
322  *		sufficient references to it (count >= min_count)
323  * - black - ignore, it doesn't contain references (e.g. text section)
324  *		(min_count == -1). No function defined for this color.
325  */
326 static bool color_white(const struct kmemleak_object *object)
327 {
328 	return object->count != KMEMLEAK_BLACK &&
329 		object->count < object->min_count;
330 }
331 
332 static bool color_gray(const struct kmemleak_object *object)
333 {
334 	return object->min_count != KMEMLEAK_BLACK &&
335 		object->count >= object->min_count;
336 }
337 
338 /*
339  * Objects are considered unreferenced only if their color is white, they have
340  * not be deleted and have a minimum age to avoid false positives caused by
341  * pointers temporarily stored in CPU registers.
342  */
343 static bool unreferenced_object(struct kmemleak_object *object)
344 {
345 	return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
346 		time_before_eq(object->jiffies + jiffies_min_age,
347 			       jiffies_last_scan);
348 }
349 
350 static const char *__object_type_str(struct kmemleak_object *object)
351 {
352 	if (object->flags & OBJECT_PHYS)
353 		return " (phys)";
354 	if (object->flags & OBJECT_PERCPU)
355 		return " (percpu)";
356 	return "";
357 }
358 
359 /*
360  * Printing of the unreferenced objects information to the seq file. The
361  * print_unreferenced function must be called with the object->lock held.
362  */
363 static void print_unreferenced(struct seq_file *seq,
364 			       struct kmemleak_object *object)
365 {
366 	int i;
367 	unsigned long *entries;
368 	unsigned int nr_entries;
369 
370 	nr_entries = stack_depot_fetch(object->trace_handle, &entries);
371 	warn_or_seq_printf(seq, "unreferenced object%s 0x%08lx (size %zu):\n",
372 			   __object_type_str(object),
373 			   object->pointer, object->size);
374 	warn_or_seq_printf(seq, "  comm \"%s\", pid %d, jiffies %lu\n",
375 			   object->comm, object->pid, object->jiffies);
376 	hex_dump_object(seq, object);
377 	warn_or_seq_printf(seq, "  backtrace (crc %x):\n", object->checksum);
378 
379 	for (i = 0; i < nr_entries; i++) {
380 		void *ptr = (void *)entries[i];
381 		warn_or_seq_printf(seq, "    %pS\n", ptr);
382 	}
383 }
384 
385 /*
386  * Print the kmemleak_object information. This function is used mainly for
387  * debugging special cases when kmemleak operations. It must be called with
388  * the object->lock held.
389  */
390 static void dump_object_info(struct kmemleak_object *object)
391 {
392 	pr_notice("Object%s 0x%08lx (size %zu):\n",
393 		  __object_type_str(object), object->pointer, object->size);
394 	pr_notice("  comm \"%s\", pid %d, jiffies %lu\n",
395 		  object->comm, object->pid, object->jiffies);
396 	pr_notice("  min_count = %d\n", object->min_count);
397 	pr_notice("  count = %d\n", object->count);
398 	pr_notice("  flags = 0x%x\n", object->flags);
399 	pr_notice("  checksum = %u\n", object->checksum);
400 	pr_notice("  backtrace:\n");
401 	if (object->trace_handle)
402 		stack_depot_print(object->trace_handle);
403 }
404 
405 static struct rb_root *object_tree(unsigned long objflags)
406 {
407 	if (objflags & OBJECT_PHYS)
408 		return &object_phys_tree_root;
409 	if (objflags & OBJECT_PERCPU)
410 		return &object_percpu_tree_root;
411 	return &object_tree_root;
412 }
413 
414 /*
415  * Look-up a memory block metadata (kmemleak_object) in the object search
416  * tree based on a pointer value. If alias is 0, only values pointing to the
417  * beginning of the memory block are allowed. The kmemleak_lock must be held
418  * when calling this function.
419  */
420 static struct kmemleak_object *__lookup_object(unsigned long ptr, int alias,
421 					       unsigned int objflags)
422 {
423 	struct rb_node *rb = object_tree(objflags)->rb_node;
424 	unsigned long untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
425 
426 	while (rb) {
427 		struct kmemleak_object *object;
428 		unsigned long untagged_objp;
429 
430 		object = rb_entry(rb, struct kmemleak_object, rb_node);
431 		untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer);
432 
433 		if (untagged_ptr < untagged_objp)
434 			rb = object->rb_node.rb_left;
435 		else if (untagged_objp + object->size <= untagged_ptr)
436 			rb = object->rb_node.rb_right;
437 		else if (untagged_objp == untagged_ptr || alias)
438 			return object;
439 		else {
440 			/*
441 			 * Printk deferring due to the kmemleak_lock held.
442 			 * This is done to avoid deadlock.
443 			 */
444 			printk_deferred_enter();
445 			kmemleak_warn("Found object by alias at 0x%08lx\n",
446 				      ptr);
447 			dump_object_info(object);
448 			printk_deferred_exit();
449 			break;
450 		}
451 	}
452 	return NULL;
453 }
454 
455 /* Look-up a kmemleak object which allocated with virtual address. */
456 static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
457 {
458 	return __lookup_object(ptr, alias, 0);
459 }
460 
461 /*
462  * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
463  * that once an object's use_count reached 0, the RCU freeing was already
464  * registered and the object should no longer be used. This function must be
465  * called under the protection of rcu_read_lock().
466  */
467 static int get_object(struct kmemleak_object *object)
468 {
469 	return atomic_inc_not_zero(&object->use_count);
470 }
471 
472 /*
473  * Memory pool allocation and freeing. kmemleak_lock must not be held.
474  */
475 static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
476 {
477 	unsigned long flags;
478 	struct kmemleak_object *object;
479 	bool warn = false;
480 
481 	/* try the slab allocator first */
482 	if (object_cache) {
483 		object = kmem_cache_alloc_noprof(object_cache,
484 						 gfp_nested_mask(gfp));
485 		if (object)
486 			return object;
487 	}
488 
489 	/* slab allocation failed, try the memory pool */
490 	raw_spin_lock_irqsave(&kmemleak_lock, flags);
491 	object = list_first_entry_or_null(&mem_pool_free_list,
492 					  typeof(*object), object_list);
493 	if (object)
494 		list_del(&object->object_list);
495 	else if (mem_pool_free_count)
496 		object = &mem_pool[--mem_pool_free_count];
497 	else
498 		warn = true;
499 	raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
500 	if (warn)
501 		pr_warn_once("Memory pool empty, consider increasing CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE\n");
502 
503 	return object;
504 }
505 
506 /*
507  * Return the object to either the slab allocator or the memory pool.
508  */
509 static void mem_pool_free(struct kmemleak_object *object)
510 {
511 	unsigned long flags;
512 
513 	if (object < mem_pool || object >= ARRAY_END(mem_pool)) {
514 		kmem_cache_free(object_cache, object);
515 		return;
516 	}
517 
518 	/* add the object to the memory pool free list */
519 	raw_spin_lock_irqsave(&kmemleak_lock, flags);
520 	list_add(&object->object_list, &mem_pool_free_list);
521 	raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
522 }
523 
524 /*
525  * RCU callback to free a kmemleak_object.
526  */
527 static void free_object_rcu(struct rcu_head *rcu)
528 {
529 	struct hlist_node *tmp;
530 	struct kmemleak_scan_area *area;
531 	struct kmemleak_object *object =
532 		container_of(rcu, struct kmemleak_object, rcu);
533 
534 	/*
535 	 * Once use_count is 0 (guaranteed by put_object), there is no other
536 	 * code accessing this object, hence no need for locking.
537 	 */
538 	hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
539 		hlist_del(&area->node);
540 		kmem_cache_free(scan_area_cache, area);
541 	}
542 	mem_pool_free(object);
543 }
544 
545 /*
546  * Decrement the object use_count. Once the count is 0, free the object using
547  * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
548  * delete_object() path, the delayed RCU freeing ensures that there is no
549  * recursive call to the kernel allocator. Lock-less RCU object_list traversal
550  * is also possible.
551  */
552 static void put_object(struct kmemleak_object *object)
553 {
554 	if (!atomic_dec_and_test(&object->use_count))
555 		return;
556 
557 	/* should only get here after delete_object was called */
558 	WARN_ON(object->flags & OBJECT_ALLOCATED);
559 
560 	/*
561 	 * It may be too early for the RCU callbacks, however, there is no
562 	 * concurrent object_list traversal when !object_cache and all objects
563 	 * came from the memory pool. Free the object directly.
564 	 */
565 	if (object_cache)
566 		call_rcu(&object->rcu, free_object_rcu);
567 	else
568 		free_object_rcu(&object->rcu);
569 }
570 
571 /*
572  * Look up an object in the object search tree and increase its use_count.
573  */
574 static struct kmemleak_object *__find_and_get_object(unsigned long ptr, int alias,
575 						     unsigned int objflags)
576 {
577 	unsigned long flags;
578 	struct kmemleak_object *object;
579 
580 	rcu_read_lock();
581 	raw_spin_lock_irqsave(&kmemleak_lock, flags);
582 	object = __lookup_object(ptr, alias, objflags);
583 	raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
584 
585 	/* check whether the object is still available */
586 	if (object && !get_object(object))
587 		object = NULL;
588 	rcu_read_unlock();
589 
590 	return object;
591 }
592 
593 /* Look up and get an object which allocated with virtual address. */
594 static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
595 {
596 	return __find_and_get_object(ptr, alias, 0);
597 }
598 
599 /*
600  * Remove an object from its object tree and object_list. Must be called with
601  * the kmemleak_lock held _if_ kmemleak is still enabled.
602  */
603 static void __remove_object(struct kmemleak_object *object)
604 {
605 	rb_erase(&object->rb_node, object_tree(object->flags));
606 	if (!(object->del_state & DELSTATE_NO_DELETE))
607 		list_del_rcu(&object->object_list);
608 	object->del_state |= DELSTATE_REMOVED;
609 }
610 
611 static struct kmemleak_object *__find_and_remove_object(unsigned long ptr,
612 							int alias,
613 							unsigned int objflags)
614 {
615 	struct kmemleak_object *object;
616 
617 	object = __lookup_object(ptr, alias, objflags);
618 	if (object)
619 		__remove_object(object);
620 
621 	return object;
622 }
623 
624 /*
625  * Look up an object in the object search tree and remove it from both object
626  * tree root and object_list. The returned object's use_count should be at
627  * least 1, as initially set by create_object().
628  */
629 static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias,
630 						      unsigned int objflags)
631 {
632 	unsigned long flags;
633 	struct kmemleak_object *object;
634 
635 	raw_spin_lock_irqsave(&kmemleak_lock, flags);
636 	object = __find_and_remove_object(ptr, alias, objflags);
637 	raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
638 
639 	return object;
640 }
641 
642 static noinline depot_stack_handle_t set_track_prepare(void)
643 {
644 	depot_stack_handle_t trace_handle;
645 	unsigned long entries[MAX_TRACE];
646 	unsigned int nr_entries;
647 
648 	/*
649 	 * Use object_cache to determine whether kmemleak_init() has
650 	 * been invoked. stack_depot_early_init() is called before
651 	 * kmemleak_init() in mm_core_init().
652 	 */
653 	if (!object_cache)
654 		return 0;
655 	nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 3);
656 	trace_handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT);
657 
658 	return trace_handle;
659 }
660 
661 static struct kmemleak_object *__alloc_object(gfp_t gfp)
662 {
663 	struct kmemleak_object *object;
664 
665 	object = mem_pool_alloc(gfp);
666 	if (!object) {
667 		pr_warn("Cannot allocate a kmemleak_object structure\n");
668 		kmemleak_disable();
669 		return NULL;
670 	}
671 
672 	INIT_LIST_HEAD(&object->object_list);
673 	INIT_LIST_HEAD(&object->gray_list);
674 	INIT_HLIST_HEAD(&object->area_list);
675 	raw_spin_lock_init(&object->lock);
676 	atomic_set(&object->use_count, 1);
677 	object->excess_ref = 0;
678 	object->count = 0;			/* white color initially */
679 	object->checksum = 0;
680 	object->del_state = 0;
681 
682 	/* task information */
683 	if (in_hardirq()) {
684 		object->pid = 0;
685 		strscpy(object->comm, "hardirq");
686 	} else if (in_serving_softirq()) {
687 		object->pid = 0;
688 		strscpy(object->comm, "softirq");
689 	} else {
690 		object->pid = current->pid;
691 		/*
692 		 * There is a small chance of a race with set_task_comm(),
693 		 * however using get_task_comm() here may cause locking
694 		 * dependency issues with current->alloc_lock. In the worst
695 		 * case, the command line is not correct.
696 		 */
697 		strscpy(object->comm, current->comm);
698 	}
699 
700 	/* kernel backtrace */
701 	object->trace_handle = set_track_prepare();
702 
703 	return object;
704 }
705 
706 static int __link_object(struct kmemleak_object *object, unsigned long ptr,
707 			 size_t size, int min_count, unsigned int objflags)
708 {
709 
710 	struct kmemleak_object *parent;
711 	struct rb_node **link, *rb_parent;
712 	unsigned long untagged_ptr;
713 	unsigned long untagged_objp;
714 
715 	object->flags = OBJECT_ALLOCATED | objflags;
716 	object->pointer = ptr;
717 	object->size = kfence_ksize((void *)ptr) ?: size;
718 	object->min_count = min_count;
719 	object->jiffies = jiffies;
720 
721 	untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
722 	/*
723 	 * Only update min_addr and max_addr with object storing virtual
724 	 * address. And update min_percpu_addr max_percpu_addr for per-CPU
725 	 * objects.
726 	 */
727 	if (objflags & OBJECT_PERCPU) {
728 		min_percpu_addr = min(min_percpu_addr, untagged_ptr);
729 		max_percpu_addr = max(max_percpu_addr, untagged_ptr + size);
730 	} else if (!(objflags & OBJECT_PHYS)) {
731 		min_addr = min(min_addr, untagged_ptr);
732 		max_addr = max(max_addr, untagged_ptr + size);
733 	}
734 	link = &object_tree(objflags)->rb_node;
735 	rb_parent = NULL;
736 	while (*link) {
737 		rb_parent = *link;
738 		parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
739 		untagged_objp = (unsigned long)kasan_reset_tag((void *)parent->pointer);
740 		if (untagged_ptr + size <= untagged_objp)
741 			link = &parent->rb_node.rb_left;
742 		else if (untagged_objp + parent->size <= untagged_ptr)
743 			link = &parent->rb_node.rb_right;
744 		else {
745 			/*
746 			 * Printk deferring due to the kmemleak_lock held.
747 			 * This is done to avoid deadlock.
748 			 */
749 			printk_deferred_enter();
750 			kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
751 				      ptr);
752 			/*
753 			 * No need for parent->lock here since "parent" cannot
754 			 * be freed while the kmemleak_lock is held.
755 			 */
756 			dump_object_info(parent);
757 			printk_deferred_exit();
758 			return -EEXIST;
759 		}
760 	}
761 	rb_link_node(&object->rb_node, rb_parent, link);
762 	rb_insert_color(&object->rb_node, object_tree(objflags));
763 	list_add_tail_rcu(&object->object_list, &object_list);
764 
765 	return 0;
766 }
767 
768 /*
769  * Create the metadata (struct kmemleak_object) corresponding to an allocated
770  * memory block and add it to the object_list and object tree.
771  */
772 static void __create_object(unsigned long ptr, size_t size,
773 				int min_count, gfp_t gfp, unsigned int objflags)
774 {
775 	struct kmemleak_object *object;
776 	unsigned long flags;
777 	int ret;
778 
779 	object = __alloc_object(gfp);
780 	if (!object)
781 		return;
782 
783 	raw_spin_lock_irqsave(&kmemleak_lock, flags);
784 	ret = __link_object(object, ptr, size, min_count, objflags);
785 	raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
786 	if (ret)
787 		mem_pool_free(object);
788 }
789 
790 /* Create kmemleak object which allocated with virtual address. */
791 static void create_object(unsigned long ptr, size_t size,
792 			  int min_count, gfp_t gfp)
793 {
794 	__create_object(ptr, size, min_count, gfp, 0);
795 }
796 
797 /* Create kmemleak object which allocated with physical address. */
798 static void create_object_phys(unsigned long ptr, size_t size,
799 			       int min_count, gfp_t gfp)
800 {
801 	__create_object(ptr, size, min_count, gfp, OBJECT_PHYS);
802 }
803 
804 /* Create kmemleak object corresponding to a per-CPU allocation. */
805 static void create_object_percpu(unsigned long ptr, size_t size,
806 				 int min_count, gfp_t gfp)
807 {
808 	__create_object(ptr, size, min_count, gfp, OBJECT_PERCPU);
809 }
810 
811 /*
812  * Mark the object as not allocated and schedule RCU freeing via put_object().
813  */
814 static void __delete_object(struct kmemleak_object *object)
815 {
816 	unsigned long flags;
817 
818 	WARN_ON(!(object->flags & OBJECT_ALLOCATED));
819 	WARN_ON(atomic_read(&object->use_count) < 1);
820 
821 	/*
822 	 * Locking here also ensures that the corresponding memory block
823 	 * cannot be freed when it is being scanned.
824 	 */
825 	raw_spin_lock_irqsave(&object->lock, flags);
826 	object->flags &= ~OBJECT_ALLOCATED;
827 	raw_spin_unlock_irqrestore(&object->lock, flags);
828 	put_object(object);
829 }
830 
831 /*
832  * Look up the metadata (struct kmemleak_object) corresponding to ptr and
833  * delete it.
834  */
835 static void delete_object_full(unsigned long ptr, unsigned int objflags)
836 {
837 	struct kmemleak_object *object;
838 
839 	object = find_and_remove_object(ptr, 0, objflags);
840 	if (!object)
841 		/*
842 		 * kmalloc_nolock() -> kfree() calls kmemleak_free()
843 		 * without kmemleak_alloc().
844 		 */
845 		return;
846 	__delete_object(object);
847 }
848 
849 /*
850  * Look up the metadata (struct kmemleak_object) corresponding to ptr and
851  * delete it. If the memory block is partially freed, the function may create
852  * additional metadata for the remaining parts of the block.
853  */
854 static void delete_object_part(unsigned long ptr, size_t size,
855 			       unsigned int objflags)
856 {
857 	struct kmemleak_object *object, *object_l, *object_r;
858 	unsigned long start, end, flags;
859 
860 	object_l = __alloc_object(GFP_KERNEL);
861 	if (!object_l)
862 		return;
863 
864 	object_r = __alloc_object(GFP_KERNEL);
865 	if (!object_r)
866 		goto out;
867 
868 	raw_spin_lock_irqsave(&kmemleak_lock, flags);
869 	object = __find_and_remove_object(ptr, 1, objflags);
870 	if (!object)
871 		goto unlock;
872 
873 	/*
874 	 * Create one or two objects that may result from the memory block
875 	 * split. Note that partial freeing is only done by free_bootmem() and
876 	 * this happens before kmemleak_init() is called.
877 	 */
878 	start = object->pointer;
879 	end = object->pointer + object->size;
880 	if ((ptr > start) &&
881 	    !__link_object(object_l, start, ptr - start,
882 			   object->min_count, objflags))
883 		object_l = NULL;
884 	if ((ptr + size < end) &&
885 	    !__link_object(object_r, ptr + size, end - ptr - size,
886 			   object->min_count, objflags))
887 		object_r = NULL;
888 
889 unlock:
890 	raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
891 	if (object) {
892 		__delete_object(object);
893 	} else {
894 #ifdef DEBUG
895 		kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
896 			      ptr, size);
897 #endif
898 	}
899 
900 out:
901 	if (object_l)
902 		mem_pool_free(object_l);
903 	if (object_r)
904 		mem_pool_free(object_r);
905 }
906 
907 static void __paint_it(struct kmemleak_object *object, int color)
908 {
909 	object->min_count = color;
910 	if (color == KMEMLEAK_BLACK)
911 		object->flags |= OBJECT_NO_SCAN;
912 }
913 
914 static void paint_it(struct kmemleak_object *object, int color)
915 {
916 	unsigned long flags;
917 
918 	raw_spin_lock_irqsave(&object->lock, flags);
919 	__paint_it(object, color);
920 	raw_spin_unlock_irqrestore(&object->lock, flags);
921 }
922 
923 static void paint_ptr(unsigned long ptr, int color, unsigned int objflags)
924 {
925 	struct kmemleak_object *object;
926 
927 	object = __find_and_get_object(ptr, 0, objflags);
928 	if (!object)
929 		/*
930 		 * kmalloc_nolock() -> kfree_rcu() calls kmemleak_ignore()
931 		 * without kmemleak_alloc().
932 		 */
933 		return;
934 	paint_it(object, color);
935 	put_object(object);
936 }
937 
938 /*
939  * Mark an object permanently as gray-colored so that it can no longer be
940  * reported as a leak. This is used in general to mark a false positive.
941  */
942 static void make_gray_object(unsigned long ptr)
943 {
944 	paint_ptr(ptr, KMEMLEAK_GREY, 0);
945 }
946 
947 /*
948  * Mark the object as black-colored so that it is ignored from scans and
949  * reporting.
950  */
951 static void make_black_object(unsigned long ptr, unsigned int objflags)
952 {
953 	paint_ptr(ptr, KMEMLEAK_BLACK, objflags);
954 }
955 
956 /*
957  * Reset the checksum of an object. The immediate effect is that it will not
958  * be reported as a leak during the next scan until its checksum is updated.
959  */
960 static void reset_checksum(unsigned long ptr)
961 {
962 	unsigned long flags;
963 	struct kmemleak_object *object;
964 
965 	object = find_and_get_object(ptr, 0);
966 	if (!object) {
967 		kmemleak_warn("Not resetting the checksum of an unknown object at 0x%08lx\n",
968 			      ptr);
969 		return;
970 	}
971 
972 	raw_spin_lock_irqsave(&object->lock, flags);
973 	object->checksum = 0;
974 	raw_spin_unlock_irqrestore(&object->lock, flags);
975 	put_object(object);
976 }
977 
978 /*
979  * Add a scanning area to the object. If at least one such area is added,
980  * kmemleak will only scan these ranges rather than the whole memory block.
981  */
982 static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
983 {
984 	unsigned long flags;
985 	struct kmemleak_object *object;
986 	struct kmemleak_scan_area *area = NULL;
987 	unsigned long untagged_ptr;
988 	unsigned long untagged_objp;
989 
990 	object = find_and_get_object(ptr, 1);
991 	if (!object) {
992 		kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
993 			      ptr);
994 		return;
995 	}
996 
997 	untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
998 	untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer);
999 
1000 	if (scan_area_cache)
1001 		area = kmem_cache_alloc_noprof(scan_area_cache,
1002 					       gfp_nested_mask(gfp));
1003 
1004 	raw_spin_lock_irqsave(&object->lock, flags);
1005 	if (!area) {
1006 		pr_warn_once("Cannot allocate a scan area, scanning the full object\n");
1007 		/* mark the object for full scan to avoid false positives */
1008 		object->flags |= OBJECT_FULL_SCAN;
1009 		goto out_unlock;
1010 	}
1011 	if (size == SIZE_MAX) {
1012 		size = untagged_objp + object->size - untagged_ptr;
1013 	} else if (untagged_ptr + size > untagged_objp + object->size) {
1014 		kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
1015 		dump_object_info(object);
1016 		kmem_cache_free(scan_area_cache, area);
1017 		goto out_unlock;
1018 	}
1019 
1020 	INIT_HLIST_NODE(&area->node);
1021 	area->start = ptr;
1022 	area->size = size;
1023 
1024 	hlist_add_head(&area->node, &object->area_list);
1025 out_unlock:
1026 	raw_spin_unlock_irqrestore(&object->lock, flags);
1027 	put_object(object);
1028 }
1029 
1030 /*
1031  * Any surplus references (object already gray) to 'ptr' are passed to
1032  * 'excess_ref'. This is used in the vmalloc() case where a pointer to
1033  * vm_struct may be used as an alternative reference to the vmalloc'ed object
1034  * (see free_thread_stack()).
1035  */
1036 static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref)
1037 {
1038 	unsigned long flags;
1039 	struct kmemleak_object *object;
1040 
1041 	object = find_and_get_object(ptr, 0);
1042 	if (!object) {
1043 		kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n",
1044 			      ptr);
1045 		return;
1046 	}
1047 
1048 	raw_spin_lock_irqsave(&object->lock, flags);
1049 	object->excess_ref = excess_ref;
1050 	raw_spin_unlock_irqrestore(&object->lock, flags);
1051 	put_object(object);
1052 }
1053 
1054 /*
1055  * Set the OBJECT_NO_SCAN flag for the object corresponding to the given
1056  * pointer. Such object will not be scanned by kmemleak but references to it
1057  * are searched.
1058  */
1059 static void object_no_scan(unsigned long ptr)
1060 {
1061 	unsigned long flags;
1062 	struct kmemleak_object *object;
1063 
1064 	object = find_and_get_object(ptr, 0);
1065 	if (!object) {
1066 		kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
1067 		return;
1068 	}
1069 
1070 	raw_spin_lock_irqsave(&object->lock, flags);
1071 	object->flags |= OBJECT_NO_SCAN;
1072 	raw_spin_unlock_irqrestore(&object->lock, flags);
1073 	put_object(object);
1074 }
1075 
1076 /**
1077  * kmemleak_alloc - register a newly allocated object
1078  * @ptr:	pointer to beginning of the object
1079  * @size:	size of the object
1080  * @min_count:	minimum number of references to this object. If during memory
1081  *		scanning a number of references less than @min_count is found,
1082  *		the object is reported as a memory leak. If @min_count is 0,
1083  *		the object is never reported as a leak. If @min_count is -1,
1084  *		the object is ignored (not scanned and not reported as a leak)
1085  * @gfp:	kmalloc() flags used for kmemleak internal memory allocations
1086  *
1087  * This function is called from the kernel allocators when a new object
1088  * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.).
1089  */
1090 void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
1091 			  gfp_t gfp)
1092 {
1093 	pr_debug("%s(0x%px, %zu, %d)\n", __func__, ptr, size, min_count);
1094 
1095 	if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1096 		create_object((unsigned long)ptr, size, min_count, gfp);
1097 }
1098 EXPORT_SYMBOL_GPL(kmemleak_alloc);
1099 
1100 /**
1101  * kmemleak_alloc_percpu - register a newly allocated __percpu object
1102  * @ptr:	__percpu pointer to beginning of the object
1103  * @size:	size of the object
1104  * @gfp:	flags used for kmemleak internal memory allocations
1105  *
1106  * This function is called from the kernel percpu allocator when a new object
1107  * (memory block) is allocated (alloc_percpu).
1108  */
1109 void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
1110 				 gfp_t gfp)
1111 {
1112 	pr_debug("%s(0x%px, %zu)\n", __func__, ptr, size);
1113 
1114 	if (kmemleak_enabled && ptr && !IS_ERR_PCPU(ptr))
1115 		create_object_percpu((__force unsigned long)ptr, size, 1, gfp);
1116 }
1117 EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
1118 
1119 /**
1120  * kmemleak_vmalloc - register a newly vmalloc'ed object
1121  * @area:	pointer to vm_struct
1122  * @size:	size of the object
1123  * @gfp:	__vmalloc() flags used for kmemleak internal memory allocations
1124  *
1125  * This function is called from the vmalloc() kernel allocator when a new
1126  * object (memory block) is allocated.
1127  */
1128 void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp)
1129 {
1130 	pr_debug("%s(0x%px, %zu)\n", __func__, area, size);
1131 
1132 	/*
1133 	 * A min_count = 2 is needed because vm_struct contains a reference to
1134 	 * the virtual address of the vmalloc'ed block.
1135 	 */
1136 	if (kmemleak_enabled) {
1137 		create_object((unsigned long)area->addr, size, 2, gfp);
1138 		object_set_excess_ref((unsigned long)area,
1139 				      (unsigned long)area->addr);
1140 	}
1141 }
1142 EXPORT_SYMBOL_GPL(kmemleak_vmalloc);
1143 
1144 /**
1145  * kmemleak_free - unregister a previously registered object
1146  * @ptr:	pointer to beginning of the object
1147  *
1148  * This function is called from the kernel allocators when an object (memory
1149  * block) is freed (kmem_cache_free, kfree, vfree etc.).
1150  */
1151 void __ref kmemleak_free(const void *ptr)
1152 {
1153 	pr_debug("%s(0x%px)\n", __func__, ptr);
1154 
1155 	if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
1156 		delete_object_full((unsigned long)ptr, 0);
1157 }
1158 EXPORT_SYMBOL_GPL(kmemleak_free);
1159 
1160 /**
1161  * kmemleak_free_part - partially unregister a previously registered object
1162  * @ptr:	pointer to the beginning or inside the object. This also
1163  *		represents the start of the range to be freed
1164  * @size:	size to be unregistered
1165  *
1166  * This function is called when only a part of a memory block is freed
1167  * (usually from the bootmem allocator).
1168  */
1169 void __ref kmemleak_free_part(const void *ptr, size_t size)
1170 {
1171 	pr_debug("%s(0x%px)\n", __func__, ptr);
1172 
1173 	if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1174 		delete_object_part((unsigned long)ptr, size, 0);
1175 }
1176 EXPORT_SYMBOL_GPL(kmemleak_free_part);
1177 
1178 /**
1179  * kmemleak_free_percpu - unregister a previously registered __percpu object
1180  * @ptr:	__percpu pointer to beginning of the object
1181  *
1182  * This function is called from the kernel percpu allocator when an object
1183  * (memory block) is freed (free_percpu).
1184  */
1185 void __ref kmemleak_free_percpu(const void __percpu *ptr)
1186 {
1187 	pr_debug("%s(0x%px)\n", __func__, ptr);
1188 
1189 	if (kmemleak_free_enabled && ptr && !IS_ERR_PCPU(ptr))
1190 		delete_object_full((__force unsigned long)ptr, OBJECT_PERCPU);
1191 }
1192 EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
1193 
1194 /**
1195  * kmemleak_update_trace - update object allocation stack trace
1196  * @ptr:	pointer to beginning of the object
1197  *
1198  * Override the object allocation stack trace for cases where the actual
1199  * allocation place is not always useful.
1200  */
1201 void __ref kmemleak_update_trace(const void *ptr)
1202 {
1203 	struct kmemleak_object *object;
1204 	depot_stack_handle_t trace_handle;
1205 	unsigned long flags;
1206 
1207 	pr_debug("%s(0x%px)\n", __func__, ptr);
1208 
1209 	if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
1210 		return;
1211 
1212 	object = find_and_get_object((unsigned long)ptr, 1);
1213 	if (!object) {
1214 #ifdef DEBUG
1215 		kmemleak_warn("Updating stack trace for unknown object at %p\n",
1216 			      ptr);
1217 #endif
1218 		return;
1219 	}
1220 
1221 	trace_handle = set_track_prepare();
1222 	raw_spin_lock_irqsave(&object->lock, flags);
1223 	object->trace_handle = trace_handle;
1224 	raw_spin_unlock_irqrestore(&object->lock, flags);
1225 
1226 	put_object(object);
1227 }
1228 EXPORT_SYMBOL(kmemleak_update_trace);
1229 
1230 /**
1231  * kmemleak_not_leak - mark an allocated object as false positive
1232  * @ptr:	pointer to beginning of the object
1233  *
1234  * Calling this function on an object will cause the memory block to no longer
1235  * be reported as leak and always be scanned.
1236  */
1237 void __ref kmemleak_not_leak(const void *ptr)
1238 {
1239 	pr_debug("%s(0x%px)\n", __func__, ptr);
1240 
1241 	if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1242 		make_gray_object((unsigned long)ptr);
1243 }
1244 EXPORT_SYMBOL(kmemleak_not_leak);
1245 
1246 /**
1247  * kmemleak_transient_leak - mark an allocated object as transient false positive
1248  * @ptr:	pointer to beginning of the object
1249  *
1250  * Calling this function on an object will cause the memory block to not be
1251  * reported as a leak temporarily. This may happen, for example, if the object
1252  * is part of a singly linked list and the ->next reference to it is changed.
1253  */
1254 void __ref kmemleak_transient_leak(const void *ptr)
1255 {
1256 	pr_debug("%s(0x%px)\n", __func__, ptr);
1257 
1258 	if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1259 		reset_checksum((unsigned long)ptr);
1260 }
1261 EXPORT_SYMBOL(kmemleak_transient_leak);
1262 
1263 /**
1264  * kmemleak_ignore_percpu - similar to kmemleak_ignore but taking a percpu
1265  *			    address argument
1266  * @ptr:	percpu address of the object
1267  */
1268 void __ref kmemleak_ignore_percpu(const void __percpu *ptr)
1269 {
1270 	pr_debug("%s(0x%px)\n", __func__, ptr);
1271 
1272 	if (kmemleak_enabled && ptr && !IS_ERR_PCPU(ptr))
1273 		make_black_object((unsigned long)ptr, OBJECT_PERCPU);
1274 }
1275 EXPORT_SYMBOL_GPL(kmemleak_ignore_percpu);
1276 
1277 /**
1278  * kmemleak_ignore - ignore an allocated object
1279  * @ptr:	pointer to beginning of the object
1280  *
1281  * Calling this function on an object will cause the memory block to be
1282  * ignored (not scanned and not reported as a leak). This is usually done when
1283  * it is known that the corresponding block is not a leak and does not contain
1284  * any references to other allocated memory blocks.
1285  */
1286 void __ref kmemleak_ignore(const void *ptr)
1287 {
1288 	pr_debug("%s(0x%px)\n", __func__, ptr);
1289 
1290 	if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1291 		make_black_object((unsigned long)ptr, 0);
1292 }
1293 EXPORT_SYMBOL(kmemleak_ignore);
1294 
1295 /**
1296  * kmemleak_scan_area - limit the range to be scanned in an allocated object
1297  * @ptr:	pointer to beginning or inside the object. This also
1298  *		represents the start of the scan area
1299  * @size:	size of the scan area
1300  * @gfp:	kmalloc() flags used for kmemleak internal memory allocations
1301  *
1302  * This function is used when it is known that only certain parts of an object
1303  * contain references to other objects. Kmemleak will only scan these areas
1304  * reducing the number false negatives.
1305  */
1306 void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1307 {
1308 	pr_debug("%s(0x%px)\n", __func__, ptr);
1309 
1310 	if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
1311 		add_scan_area((unsigned long)ptr, size, gfp);
1312 }
1313 EXPORT_SYMBOL(kmemleak_scan_area);
1314 
1315 /**
1316  * kmemleak_no_scan - do not scan an allocated object
1317  * @ptr:	pointer to beginning of the object
1318  *
1319  * This function notifies kmemleak not to scan the given memory block. Useful
1320  * in situations where it is known that the given object does not contain any
1321  * references to other objects. Kmemleak will not scan such objects reducing
1322  * the number of false negatives.
1323  */
1324 void __ref kmemleak_no_scan(const void *ptr)
1325 {
1326 	pr_debug("%s(0x%px)\n", __func__, ptr);
1327 
1328 	if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1329 		object_no_scan((unsigned long)ptr);
1330 }
1331 EXPORT_SYMBOL(kmemleak_no_scan);
1332 
1333 /**
1334  * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical
1335  *			 address argument
1336  * @phys:	physical address of the object
1337  * @size:	size of the object
1338  * @gfp:	kmalloc() flags used for kmemleak internal memory allocations
1339  */
1340 void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, gfp_t gfp)
1341 {
1342 	pr_debug("%s(0x%px, %zu)\n", __func__, &phys, size);
1343 
1344 	if (kmemleak_enabled)
1345 		/*
1346 		 * Create object with OBJECT_PHYS flag and
1347 		 * assume min_count 0.
1348 		 */
1349 		create_object_phys((unsigned long)phys, size, 0, gfp);
1350 }
1351 EXPORT_SYMBOL(kmemleak_alloc_phys);
1352 
1353 /**
1354  * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a
1355  *			     physical address argument
1356  * @phys:	physical address if the beginning or inside an object. This
1357  *		also represents the start of the range to be freed
1358  * @size:	size to be unregistered
1359  */
1360 void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size)
1361 {
1362 	pr_debug("%s(0x%px)\n", __func__, &phys);
1363 
1364 	if (kmemleak_enabled)
1365 		delete_object_part((unsigned long)phys, size, OBJECT_PHYS);
1366 }
1367 EXPORT_SYMBOL(kmemleak_free_part_phys);
1368 
1369 /**
1370  * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical
1371  *			  address argument
1372  * @phys:	physical address of the object
1373  */
1374 void __ref kmemleak_ignore_phys(phys_addr_t phys)
1375 {
1376 	pr_debug("%s(0x%px)\n", __func__, &phys);
1377 
1378 	if (kmemleak_enabled)
1379 		make_black_object((unsigned long)phys, OBJECT_PHYS);
1380 }
1381 EXPORT_SYMBOL(kmemleak_ignore_phys);
1382 
1383 /*
1384  * Update an object's checksum and return true if it was modified.
1385  */
1386 static bool update_checksum(struct kmemleak_object *object)
1387 {
1388 	u32 old_csum = object->checksum;
1389 
1390 	if (WARN_ON_ONCE(object->flags & OBJECT_PHYS))
1391 		return false;
1392 
1393 	kasan_disable_current();
1394 	kcsan_disable_current();
1395 	if (object->flags & OBJECT_PERCPU) {
1396 		unsigned int cpu;
1397 
1398 		object->checksum = 0;
1399 		for_each_possible_cpu(cpu) {
1400 			void *ptr = per_cpu_ptr((void __percpu *)object->pointer, cpu);
1401 
1402 			object->checksum ^= crc32(0, kasan_reset_tag((void *)ptr), object->size);
1403 		}
1404 	} else {
1405 		object->checksum = crc32(0, kasan_reset_tag((void *)object->pointer), object->size);
1406 	}
1407 	kasan_enable_current();
1408 	kcsan_enable_current();
1409 
1410 	return object->checksum != old_csum;
1411 }
1412 
1413 /*
1414  * Update an object's references. object->lock must be held by the caller.
1415  */
1416 static void update_refs(struct kmemleak_object *object)
1417 {
1418 	if (!color_white(object)) {
1419 		/* non-orphan, ignored or new */
1420 		return;
1421 	}
1422 
1423 	/*
1424 	 * Increase the object's reference count (number of pointers to the
1425 	 * memory block). If this count reaches the required minimum, the
1426 	 * object's color will become gray and it will be added to the
1427 	 * gray_list.
1428 	 */
1429 	object->count++;
1430 	if (color_gray(object)) {
1431 		/* put_object() called when removing from gray_list */
1432 		WARN_ON(!get_object(object));
1433 		list_add_tail(&object->gray_list, &gray_list);
1434 	}
1435 }
1436 
1437 static void pointer_update_refs(struct kmemleak_object *scanned,
1438 			 unsigned long pointer, unsigned int objflags)
1439 {
1440 	struct kmemleak_object *object;
1441 	unsigned long untagged_ptr;
1442 	unsigned long excess_ref;
1443 
1444 	untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer);
1445 	if (objflags & OBJECT_PERCPU) {
1446 		if (untagged_ptr < min_percpu_addr || untagged_ptr >= max_percpu_addr)
1447 			return;
1448 	} else {
1449 		if (untagged_ptr < min_addr || untagged_ptr >= max_addr)
1450 			return;
1451 	}
1452 
1453 	/*
1454 	 * No need for get_object() here since we hold kmemleak_lock.
1455 	 * object->use_count cannot be dropped to 0 while the object
1456 	 * is still present in object_tree_root and object_list
1457 	 * (with updates protected by kmemleak_lock).
1458 	 */
1459 	object = __lookup_object(pointer, 1, objflags);
1460 	if (!object)
1461 		return;
1462 	if (object == scanned)
1463 		/* self referenced, ignore */
1464 		return;
1465 
1466 	/*
1467 	 * Avoid the lockdep recursive warning on object->lock being
1468 	 * previously acquired in scan_object(). These locks are
1469 	 * enclosed by scan_mutex.
1470 	 */
1471 	raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1472 	/* only pass surplus references (object already gray) */
1473 	if (color_gray(object)) {
1474 		excess_ref = object->excess_ref;
1475 		/* no need for update_refs() if object already gray */
1476 	} else {
1477 		excess_ref = 0;
1478 		update_refs(object);
1479 	}
1480 	raw_spin_unlock(&object->lock);
1481 
1482 	if (excess_ref) {
1483 		object = lookup_object(excess_ref, 0);
1484 		if (!object)
1485 			return;
1486 		if (object == scanned)
1487 			/* circular reference, ignore */
1488 			return;
1489 		raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1490 		update_refs(object);
1491 		raw_spin_unlock(&object->lock);
1492 	}
1493 }
1494 
1495 /*
1496  * Memory scanning is a long process and it needs to be interruptible. This
1497  * function checks whether such interrupt condition occurred.
1498  */
1499 static int scan_should_stop(void)
1500 {
1501 	if (!kmemleak_enabled)
1502 		return 1;
1503 
1504 	/*
1505 	 * This function may be called from either process or kthread context,
1506 	 * hence the need to check for both stop conditions.
1507 	 */
1508 	if (current->mm)
1509 		return signal_pending(current);
1510 	else
1511 		return kthread_should_stop();
1512 
1513 	return 0;
1514 }
1515 
1516 /*
1517  * Scan a memory block (exclusive range) for valid pointers and add those
1518  * found to the gray list.
1519  */
1520 static void scan_block(void *_start, void *_end,
1521 		       struct kmemleak_object *scanned)
1522 {
1523 	unsigned long *ptr;
1524 	unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1525 	unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1526 	unsigned long flags;
1527 
1528 	raw_spin_lock_irqsave(&kmemleak_lock, flags);
1529 	for (ptr = start; ptr < end; ptr++) {
1530 		unsigned long pointer;
1531 
1532 		if (scan_should_stop())
1533 			break;
1534 
1535 		kasan_disable_current();
1536 		pointer = *(unsigned long *)kasan_reset_tag((void *)ptr);
1537 		kasan_enable_current();
1538 
1539 		pointer_update_refs(scanned, pointer, 0);
1540 		pointer_update_refs(scanned, pointer, OBJECT_PERCPU);
1541 	}
1542 	raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
1543 }
1544 
1545 /*
1546  * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
1547  */
1548 #ifdef CONFIG_SMP
1549 static void scan_large_block(void *start, void *end)
1550 {
1551 	void *next;
1552 
1553 	while (start < end) {
1554 		next = min(start + MAX_SCAN_SIZE, end);
1555 		scan_block(start, next, NULL);
1556 		start = next;
1557 		cond_resched();
1558 	}
1559 }
1560 #endif
1561 
1562 /*
1563  * Scan a memory block corresponding to a kmemleak_object. A condition is
1564  * that object->use_count >= 1.
1565  */
1566 static void scan_object(struct kmemleak_object *object)
1567 {
1568 	struct kmemleak_scan_area *area;
1569 	unsigned long flags;
1570 
1571 	/*
1572 	 * Once the object->lock is acquired, the corresponding memory block
1573 	 * cannot be freed (the same lock is acquired in delete_object).
1574 	 */
1575 	raw_spin_lock_irqsave(&object->lock, flags);
1576 	if (object->flags & OBJECT_NO_SCAN)
1577 		goto out;
1578 	if (!(object->flags & OBJECT_ALLOCATED))
1579 		/* already freed object */
1580 		goto out;
1581 
1582 	if (object->flags & OBJECT_PERCPU) {
1583 		unsigned int cpu;
1584 
1585 		for_each_possible_cpu(cpu) {
1586 			void *start = per_cpu_ptr((void __percpu *)object->pointer, cpu);
1587 			void *end = start + object->size;
1588 
1589 			scan_block(start, end, object);
1590 
1591 			raw_spin_unlock_irqrestore(&object->lock, flags);
1592 			cond_resched();
1593 			raw_spin_lock_irqsave(&object->lock, flags);
1594 			if (!(object->flags & OBJECT_ALLOCATED))
1595 				break;
1596 		}
1597 	} else if (hlist_empty(&object->area_list) ||
1598 	    object->flags & OBJECT_FULL_SCAN) {
1599 		void *start = object->flags & OBJECT_PHYS ?
1600 				__va((phys_addr_t)object->pointer) :
1601 				(void *)object->pointer;
1602 		void *end = start + object->size;
1603 		void *next;
1604 
1605 		do {
1606 			next = min(start + MAX_SCAN_SIZE, end);
1607 			scan_block(start, next, object);
1608 
1609 			start = next;
1610 			if (start >= end)
1611 				break;
1612 
1613 			raw_spin_unlock_irqrestore(&object->lock, flags);
1614 			cond_resched();
1615 			raw_spin_lock_irqsave(&object->lock, flags);
1616 		} while (object->flags & OBJECT_ALLOCATED);
1617 	} else {
1618 		hlist_for_each_entry(area, &object->area_list, node)
1619 			scan_block((void *)area->start,
1620 				   (void *)(area->start + area->size),
1621 				   object);
1622 	}
1623 out:
1624 	raw_spin_unlock_irqrestore(&object->lock, flags);
1625 }
1626 
1627 /*
1628  * Scan the objects already referenced (gray objects). More objects will be
1629  * referenced and, if there are no memory leaks, all the objects are scanned.
1630  */
1631 static void scan_gray_list(void)
1632 {
1633 	struct kmemleak_object *object, *tmp;
1634 
1635 	/*
1636 	 * The list traversal is safe for both tail additions and removals
1637 	 * from inside the loop. The kmemleak objects cannot be freed from
1638 	 * outside the loop because their use_count was incremented.
1639 	 */
1640 	object = list_entry(gray_list.next, typeof(*object), gray_list);
1641 	while (&object->gray_list != &gray_list) {
1642 		cond_resched();
1643 
1644 		/* may add new objects to the list */
1645 		if (!scan_should_stop())
1646 			scan_object(object);
1647 
1648 		tmp = list_entry(object->gray_list.next, typeof(*object),
1649 				 gray_list);
1650 
1651 		/* remove the object from the list and release it */
1652 		list_del(&object->gray_list);
1653 		put_object(object);
1654 
1655 		object = tmp;
1656 	}
1657 	WARN_ON(!list_empty(&gray_list));
1658 }
1659 
1660 /*
1661  * Conditionally call resched() in an object iteration loop while making sure
1662  * that the given object won't go away without RCU read lock by performing a
1663  * get_object() if necessaary.
1664  */
1665 static void kmemleak_cond_resched(struct kmemleak_object *object)
1666 {
1667 	if (!get_object(object))
1668 		return;	/* Try next object */
1669 
1670 	raw_spin_lock_irq(&kmemleak_lock);
1671 	if (object->del_state & DELSTATE_REMOVED)
1672 		goto unlock_put;	/* Object removed */
1673 	object->del_state |= DELSTATE_NO_DELETE;
1674 	raw_spin_unlock_irq(&kmemleak_lock);
1675 
1676 	rcu_read_unlock();
1677 	cond_resched();
1678 	rcu_read_lock();
1679 
1680 	raw_spin_lock_irq(&kmemleak_lock);
1681 	if (object->del_state & DELSTATE_REMOVED)
1682 		list_del_rcu(&object->object_list);
1683 	object->del_state &= ~DELSTATE_NO_DELETE;
1684 unlock_put:
1685 	raw_spin_unlock_irq(&kmemleak_lock);
1686 	put_object(object);
1687 }
1688 
1689 /*
1690  * Scan data sections and all the referenced memory blocks allocated via the
1691  * kernel's standard allocators. This function must be called with the
1692  * scan_mutex held.
1693  */
1694 static void kmemleak_scan(void)
1695 {
1696 	struct kmemleak_object *object;
1697 	struct zone *zone;
1698 	int __maybe_unused i;
1699 	int new_leaks = 0;
1700 
1701 	jiffies_last_scan = jiffies;
1702 
1703 	/* prepare the kmemleak_object's */
1704 	rcu_read_lock();
1705 	list_for_each_entry_rcu(object, &object_list, object_list) {
1706 		raw_spin_lock_irq(&object->lock);
1707 #ifdef DEBUG
1708 		/*
1709 		 * With a few exceptions there should be a maximum of
1710 		 * 1 reference to any object at this point.
1711 		 */
1712 		if (atomic_read(&object->use_count) > 1) {
1713 			pr_debug("object->use_count = %d\n",
1714 				 atomic_read(&object->use_count));
1715 			dump_object_info(object);
1716 		}
1717 #endif
1718 
1719 		/* ignore objects outside lowmem (paint them black) */
1720 		if ((object->flags & OBJECT_PHYS) &&
1721 		   !(object->flags & OBJECT_NO_SCAN)) {
1722 			unsigned long phys = object->pointer;
1723 
1724 			if (PHYS_PFN(phys) < min_low_pfn ||
1725 			    PHYS_PFN(phys + object->size) > max_low_pfn)
1726 				__paint_it(object, KMEMLEAK_BLACK);
1727 		}
1728 
1729 		/* reset the reference count (whiten the object) */
1730 		object->count = 0;
1731 		if (color_gray(object) && get_object(object))
1732 			list_add_tail(&object->gray_list, &gray_list);
1733 
1734 		raw_spin_unlock_irq(&object->lock);
1735 
1736 		if (need_resched())
1737 			kmemleak_cond_resched(object);
1738 	}
1739 	rcu_read_unlock();
1740 
1741 #ifdef CONFIG_SMP
1742 	/* per-cpu sections scanning */
1743 	for_each_possible_cpu(i)
1744 		scan_large_block(__per_cpu_start + per_cpu_offset(i),
1745 				 __per_cpu_end + per_cpu_offset(i));
1746 #endif
1747 
1748 	/*
1749 	 * Struct page scanning for each node.
1750 	 */
1751 	get_online_mems();
1752 	for_each_populated_zone(zone) {
1753 		unsigned long start_pfn = zone->zone_start_pfn;
1754 		unsigned long end_pfn = zone_end_pfn(zone);
1755 		unsigned long pfn;
1756 
1757 		for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1758 			struct page *page = pfn_to_online_page(pfn);
1759 
1760 			if (!(pfn & 63))
1761 				cond_resched();
1762 
1763 			if (!page)
1764 				continue;
1765 
1766 			/* only scan pages belonging to this zone */
1767 			if (page_zone(page) != zone)
1768 				continue;
1769 			/* only scan if page is in use */
1770 			if (page_count(page) == 0)
1771 				continue;
1772 			scan_block(page, page + 1, NULL);
1773 		}
1774 	}
1775 	put_online_mems();
1776 
1777 	/*
1778 	 * Scanning the task stacks (may introduce false negatives).
1779 	 */
1780 	if (kmemleak_stack_scan) {
1781 		struct task_struct *p, *g;
1782 
1783 		rcu_read_lock();
1784 		for_each_process_thread(g, p) {
1785 			void *stack = try_get_task_stack(p);
1786 			if (stack) {
1787 				scan_block(stack, stack + THREAD_SIZE, NULL);
1788 				put_task_stack(p);
1789 			}
1790 		}
1791 		rcu_read_unlock();
1792 	}
1793 
1794 	/*
1795 	 * Scan the objects already referenced from the sections scanned
1796 	 * above.
1797 	 */
1798 	scan_gray_list();
1799 
1800 	/*
1801 	 * Check for new or unreferenced objects modified since the previous
1802 	 * scan and color them gray until the next scan.
1803 	 */
1804 	rcu_read_lock();
1805 	list_for_each_entry_rcu(object, &object_list, object_list) {
1806 		if (need_resched())
1807 			kmemleak_cond_resched(object);
1808 
1809 		/*
1810 		 * This is racy but we can save the overhead of lock/unlock
1811 		 * calls. The missed objects, if any, should be caught in
1812 		 * the next scan.
1813 		 */
1814 		if (!color_white(object))
1815 			continue;
1816 		raw_spin_lock_irq(&object->lock);
1817 		if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1818 		    && update_checksum(object) && get_object(object)) {
1819 			/* color it gray temporarily */
1820 			object->count = object->min_count;
1821 			list_add_tail(&object->gray_list, &gray_list);
1822 		}
1823 		raw_spin_unlock_irq(&object->lock);
1824 	}
1825 	rcu_read_unlock();
1826 
1827 	/*
1828 	 * Re-scan the gray list for modified unreferenced objects.
1829 	 */
1830 	scan_gray_list();
1831 
1832 	/*
1833 	 * If scanning was stopped do not report any new unreferenced objects.
1834 	 */
1835 	if (scan_should_stop())
1836 		return;
1837 
1838 	/*
1839 	 * Scanning result reporting.
1840 	 */
1841 	rcu_read_lock();
1842 	list_for_each_entry_rcu(object, &object_list, object_list) {
1843 		if (need_resched())
1844 			kmemleak_cond_resched(object);
1845 
1846 		/*
1847 		 * This is racy but we can save the overhead of lock/unlock
1848 		 * calls. The missed objects, if any, should be caught in
1849 		 * the next scan.
1850 		 */
1851 		if (!color_white(object))
1852 			continue;
1853 		raw_spin_lock_irq(&object->lock);
1854 		if (unreferenced_object(object) &&
1855 		    !(object->flags & OBJECT_REPORTED)) {
1856 			object->flags |= OBJECT_REPORTED;
1857 
1858 			if (kmemleak_verbose)
1859 				print_unreferenced(NULL, object);
1860 
1861 			new_leaks++;
1862 		}
1863 		raw_spin_unlock_irq(&object->lock);
1864 	}
1865 	rcu_read_unlock();
1866 
1867 	if (new_leaks) {
1868 		kmemleak_found_leaks = true;
1869 
1870 		pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
1871 			new_leaks);
1872 	}
1873 
1874 }
1875 
1876 /*
1877  * Thread function performing automatic memory scanning. Unreferenced objects
1878  * at the end of a memory scan are reported but only the first time.
1879  */
1880 static int kmemleak_scan_thread(void *arg)
1881 {
1882 	static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN);
1883 
1884 	pr_info("Automatic memory scanning thread started\n");
1885 	set_user_nice(current, 10);
1886 
1887 	/*
1888 	 * Wait before the first scan to allow the system to fully initialize.
1889 	 */
1890 	if (first_run) {
1891 		signed long timeout = secs_to_jiffies(SECS_FIRST_SCAN);
1892 		first_run = 0;
1893 		while (timeout && !kthread_should_stop())
1894 			timeout = schedule_timeout_interruptible(timeout);
1895 	}
1896 
1897 	while (!kthread_should_stop()) {
1898 		signed long timeout = READ_ONCE(jiffies_scan_wait);
1899 
1900 		mutex_lock(&scan_mutex);
1901 		kmemleak_scan();
1902 		mutex_unlock(&scan_mutex);
1903 
1904 		/* wait before the next scan */
1905 		while (timeout && !kthread_should_stop())
1906 			timeout = schedule_timeout_interruptible(timeout);
1907 	}
1908 
1909 	pr_info("Automatic memory scanning thread ended\n");
1910 
1911 	return 0;
1912 }
1913 
1914 /*
1915  * Start the automatic memory scanning thread. This function must be called
1916  * with the scan_mutex held.
1917  */
1918 static void start_scan_thread(void)
1919 {
1920 	if (scan_thread)
1921 		return;
1922 	scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1923 	if (IS_ERR(scan_thread)) {
1924 		pr_warn("Failed to create the scan thread\n");
1925 		scan_thread = NULL;
1926 	}
1927 }
1928 
1929 /*
1930  * Stop the automatic memory scanning thread.
1931  */
1932 static void stop_scan_thread(void)
1933 {
1934 	if (scan_thread) {
1935 		kthread_stop(scan_thread);
1936 		scan_thread = NULL;
1937 	}
1938 }
1939 
1940 /*
1941  * Iterate over the object_list and return the first valid object at or after
1942  * the required position with its use_count incremented. The function triggers
1943  * a memory scanning when the pos argument points to the first position.
1944  */
1945 static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1946 {
1947 	struct kmemleak_object *object;
1948 	loff_t n = *pos;
1949 	int err;
1950 
1951 	err = mutex_lock_interruptible(&scan_mutex);
1952 	if (err < 0)
1953 		return ERR_PTR(err);
1954 
1955 	rcu_read_lock();
1956 	list_for_each_entry_rcu(object, &object_list, object_list) {
1957 		if (n-- > 0)
1958 			continue;
1959 		if (get_object(object))
1960 			goto out;
1961 	}
1962 	object = NULL;
1963 out:
1964 	return object;
1965 }
1966 
1967 /*
1968  * Return the next object in the object_list. The function decrements the
1969  * use_count of the previous object and increases that of the next one.
1970  */
1971 static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1972 {
1973 	struct kmemleak_object *prev_obj = v;
1974 	struct kmemleak_object *next_obj = NULL;
1975 	struct kmemleak_object *obj = prev_obj;
1976 
1977 	++(*pos);
1978 
1979 	list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
1980 		if (get_object(obj)) {
1981 			next_obj = obj;
1982 			break;
1983 		}
1984 	}
1985 
1986 	put_object(prev_obj);
1987 	return next_obj;
1988 }
1989 
1990 /*
1991  * Decrement the use_count of the last object required, if any.
1992  */
1993 static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1994 {
1995 	if (!IS_ERR(v)) {
1996 		/*
1997 		 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1998 		 * waiting was interrupted, so only release it if !IS_ERR.
1999 		 */
2000 		rcu_read_unlock();
2001 		mutex_unlock(&scan_mutex);
2002 		if (v)
2003 			put_object(v);
2004 	}
2005 }
2006 
2007 /*
2008  * Print the information for an unreferenced object to the seq file.
2009  */
2010 static int kmemleak_seq_show(struct seq_file *seq, void *v)
2011 {
2012 	struct kmemleak_object *object = v;
2013 	unsigned long flags;
2014 
2015 	raw_spin_lock_irqsave(&object->lock, flags);
2016 	if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
2017 		print_unreferenced(seq, object);
2018 	raw_spin_unlock_irqrestore(&object->lock, flags);
2019 	return 0;
2020 }
2021 
2022 static const struct seq_operations kmemleak_seq_ops = {
2023 	.start = kmemleak_seq_start,
2024 	.next  = kmemleak_seq_next,
2025 	.stop  = kmemleak_seq_stop,
2026 	.show  = kmemleak_seq_show,
2027 };
2028 
2029 static int kmemleak_open(struct inode *inode, struct file *file)
2030 {
2031 	return seq_open(file, &kmemleak_seq_ops);
2032 }
2033 
2034 static bool __dump_str_object_info(unsigned long addr, unsigned int objflags)
2035 {
2036 	unsigned long flags;
2037 	struct kmemleak_object *object;
2038 
2039 	object = __find_and_get_object(addr, 1, objflags);
2040 	if (!object)
2041 		return false;
2042 
2043 	raw_spin_lock_irqsave(&object->lock, flags);
2044 	dump_object_info(object);
2045 	raw_spin_unlock_irqrestore(&object->lock, flags);
2046 
2047 	put_object(object);
2048 
2049 	return true;
2050 }
2051 
2052 static int dump_str_object_info(const char *str)
2053 {
2054 	unsigned long addr;
2055 	bool found = false;
2056 
2057 	if (kstrtoul(str, 0, &addr))
2058 		return -EINVAL;
2059 
2060 	found |= __dump_str_object_info(addr, 0);
2061 	found |= __dump_str_object_info(addr, OBJECT_PHYS);
2062 	found |= __dump_str_object_info(addr, OBJECT_PERCPU);
2063 
2064 	if (!found) {
2065 		pr_info("Unknown object at 0x%08lx\n", addr);
2066 		return -EINVAL;
2067 	}
2068 
2069 	return 0;
2070 }
2071 
2072 /*
2073  * We use grey instead of black to ensure we can do future scans on the same
2074  * objects. If we did not do future scans these black objects could
2075  * potentially contain references to newly allocated objects in the future and
2076  * we'd end up with false positives.
2077  */
2078 static void kmemleak_clear(void)
2079 {
2080 	struct kmemleak_object *object;
2081 
2082 	rcu_read_lock();
2083 	list_for_each_entry_rcu(object, &object_list, object_list) {
2084 		raw_spin_lock_irq(&object->lock);
2085 		if ((object->flags & OBJECT_REPORTED) &&
2086 		    unreferenced_object(object))
2087 			__paint_it(object, KMEMLEAK_GREY);
2088 		raw_spin_unlock_irq(&object->lock);
2089 	}
2090 	rcu_read_unlock();
2091 
2092 	kmemleak_found_leaks = false;
2093 }
2094 
2095 static void __kmemleak_do_cleanup(void);
2096 
2097 /*
2098  * File write operation to configure kmemleak at run-time. The following
2099  * commands can be written to the /sys/kernel/debug/kmemleak file:
2100  *   off	- disable kmemleak (irreversible)
2101  *   stack=on	- enable the task stacks scanning
2102  *   stack=off	- disable the tasks stacks scanning
2103  *   scan=on	- start the automatic memory scanning thread
2104  *   scan=off	- stop the automatic memory scanning thread
2105  *   scan=...	- set the automatic memory scanning period in seconds (0 to
2106  *		  disable it)
2107  *   scan	- trigger a memory scan
2108  *   clear	- mark all current reported unreferenced kmemleak objects as
2109  *		  grey to ignore printing them, or free all kmemleak objects
2110  *		  if kmemleak has been disabled.
2111  *   dump=...	- dump information about the object found at the given address
2112  */
2113 static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
2114 			      size_t size, loff_t *ppos)
2115 {
2116 	char buf[64];
2117 	int buf_size;
2118 	int ret;
2119 
2120 	buf_size = min(size, (sizeof(buf) - 1));
2121 	if (strncpy_from_user(buf, user_buf, buf_size) < 0)
2122 		return -EFAULT;
2123 	buf[buf_size] = 0;
2124 
2125 	ret = mutex_lock_interruptible(&scan_mutex);
2126 	if (ret < 0)
2127 		return ret;
2128 
2129 	if (strncmp(buf, "clear", 5) == 0) {
2130 		if (kmemleak_enabled)
2131 			kmemleak_clear();
2132 		else
2133 			__kmemleak_do_cleanup();
2134 		goto out;
2135 	}
2136 
2137 	if (!kmemleak_enabled) {
2138 		ret = -EPERM;
2139 		goto out;
2140 	}
2141 
2142 	if (strncmp(buf, "off", 3) == 0)
2143 		kmemleak_disable();
2144 	else if (strncmp(buf, "stack=on", 8) == 0)
2145 		kmemleak_stack_scan = 1;
2146 	else if (strncmp(buf, "stack=off", 9) == 0)
2147 		kmemleak_stack_scan = 0;
2148 	else if (strncmp(buf, "scan=on", 7) == 0)
2149 		start_scan_thread();
2150 	else if (strncmp(buf, "scan=off", 8) == 0)
2151 		stop_scan_thread();
2152 	else if (strncmp(buf, "scan=", 5) == 0) {
2153 		unsigned secs;
2154 		unsigned long msecs;
2155 
2156 		ret = kstrtouint(buf + 5, 0, &secs);
2157 		if (ret < 0)
2158 			goto out;
2159 
2160 		msecs = secs * MSEC_PER_SEC;
2161 		if (msecs > UINT_MAX)
2162 			msecs = UINT_MAX;
2163 
2164 		stop_scan_thread();
2165 		if (msecs) {
2166 			WRITE_ONCE(jiffies_scan_wait, msecs_to_jiffies(msecs));
2167 			start_scan_thread();
2168 		}
2169 	} else if (strncmp(buf, "scan", 4) == 0)
2170 		kmemleak_scan();
2171 	else if (strncmp(buf, "dump=", 5) == 0)
2172 		ret = dump_str_object_info(buf + 5);
2173 	else
2174 		ret = -EINVAL;
2175 
2176 out:
2177 	mutex_unlock(&scan_mutex);
2178 	if (ret < 0)
2179 		return ret;
2180 
2181 	/* ignore the rest of the buffer, only one command at a time */
2182 	*ppos += size;
2183 	return size;
2184 }
2185 
2186 static const struct file_operations kmemleak_fops = {
2187 	.owner		= THIS_MODULE,
2188 	.open		= kmemleak_open,
2189 	.read		= seq_read,
2190 	.write		= kmemleak_write,
2191 	.llseek		= seq_lseek,
2192 	.release	= seq_release,
2193 };
2194 
2195 static void __kmemleak_do_cleanup(void)
2196 {
2197 	struct kmemleak_object *object, *tmp;
2198 	unsigned int cnt = 0;
2199 
2200 	/*
2201 	 * Kmemleak has already been disabled, no need for RCU list traversal
2202 	 * or kmemleak_lock held.
2203 	 */
2204 	list_for_each_entry_safe(object, tmp, &object_list, object_list) {
2205 		__remove_object(object);
2206 		__delete_object(object);
2207 
2208 		/* Call cond_resched() once per 64 iterations to avoid soft lockup */
2209 		if (!(++cnt & 0x3f))
2210 			cond_resched();
2211 	}
2212 }
2213 
2214 /*
2215  * Stop the memory scanning thread and free the kmemleak internal objects if
2216  * no previous scan thread (otherwise, kmemleak may still have some useful
2217  * information on memory leaks).
2218  */
2219 static void kmemleak_do_cleanup(struct work_struct *work)
2220 {
2221 	stop_scan_thread();
2222 
2223 	mutex_lock(&scan_mutex);
2224 	/*
2225 	 * Once it is made sure that kmemleak_scan has stopped, it is safe to no
2226 	 * longer track object freeing. Ordering of the scan thread stopping and
2227 	 * the memory accesses below is guaranteed by the kthread_stop()
2228 	 * function.
2229 	 */
2230 	kmemleak_free_enabled = 0;
2231 	mutex_unlock(&scan_mutex);
2232 
2233 	if (!kmemleak_found_leaks)
2234 		__kmemleak_do_cleanup();
2235 	else
2236 		pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
2237 }
2238 
2239 static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
2240 
2241 /*
2242  * Disable kmemleak. No memory allocation/freeing will be traced once this
2243  * function is called. Disabling kmemleak is an irreversible operation.
2244  */
2245 static void kmemleak_disable(void)
2246 {
2247 	/* atomically check whether it was already invoked */
2248 	if (cmpxchg(&kmemleak_error, 0, 1))
2249 		return;
2250 
2251 	/* stop any memory operation tracing */
2252 	kmemleak_enabled = 0;
2253 
2254 	/* check whether it is too early for a kernel thread */
2255 	if (kmemleak_late_initialized)
2256 		schedule_work(&cleanup_work);
2257 	else
2258 		kmemleak_free_enabled = 0;
2259 
2260 	pr_info("Kernel memory leak detector disabled\n");
2261 }
2262 
2263 /*
2264  * Allow boot-time kmemleak disabling (enabled by default).
2265  */
2266 static int __init kmemleak_boot_config(char *str)
2267 {
2268 	if (!str)
2269 		return -EINVAL;
2270 	if (strcmp(str, "off") == 0)
2271 		kmemleak_disable();
2272 	else if (strcmp(str, "on") == 0) {
2273 		kmemleak_skip_disable = 1;
2274 		stack_depot_request_early_init();
2275 	}
2276 	else
2277 		return -EINVAL;
2278 	return 0;
2279 }
2280 early_param("kmemleak", kmemleak_boot_config);
2281 
2282 /*
2283  * Kmemleak initialization.
2284  */
2285 void __init kmemleak_init(void)
2286 {
2287 #ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
2288 	if (!kmemleak_skip_disable) {
2289 		kmemleak_disable();
2290 		return;
2291 	}
2292 #endif
2293 
2294 	if (kmemleak_error)
2295 		return;
2296 
2297 	jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
2298 	jiffies_scan_wait = secs_to_jiffies(SECS_SCAN_WAIT);
2299 
2300 	object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
2301 	scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
2302 
2303 	/* register the data/bss sections */
2304 	create_object((unsigned long)_sdata, _edata - _sdata,
2305 		      KMEMLEAK_GREY, GFP_ATOMIC);
2306 	create_object((unsigned long)__bss_start, __bss_stop - __bss_start,
2307 		      KMEMLEAK_GREY, GFP_ATOMIC);
2308 	/* only register .data..ro_after_init if not within .data */
2309 	if (&__start_ro_after_init < &_sdata || &__end_ro_after_init > &_edata)
2310 		create_object((unsigned long)__start_ro_after_init,
2311 			      __end_ro_after_init - __start_ro_after_init,
2312 			      KMEMLEAK_GREY, GFP_ATOMIC);
2313 }
2314 
2315 /*
2316  * Late initialization function.
2317  */
2318 static int __init kmemleak_late_init(void)
2319 {
2320 	kmemleak_late_initialized = 1;
2321 
2322 	debugfs_create_file("kmemleak", 0644, NULL, NULL, &kmemleak_fops);
2323 
2324 	if (kmemleak_error) {
2325 		/*
2326 		 * Some error occurred and kmemleak was disabled. There is a
2327 		 * small chance that kmemleak_disable() was called immediately
2328 		 * after setting kmemleak_late_initialized and we may end up with
2329 		 * two clean-up threads but serialized by scan_mutex.
2330 		 */
2331 		schedule_work(&cleanup_work);
2332 		return -ENOMEM;
2333 	}
2334 
2335 	if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) {
2336 		mutex_lock(&scan_mutex);
2337 		start_scan_thread();
2338 		mutex_unlock(&scan_mutex);
2339 	}
2340 
2341 	pr_info("Kernel memory leak detector initialized (mem pool available: %d)\n",
2342 		mem_pool_free_count);
2343 
2344 	return 0;
2345 }
2346 late_initcall(kmemleak_late_init);
2347