]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - mm/kmemleak.c
mm: kmemleak: take a full lowmem check in kmemleak_*_phys()
[mirror_ubuntu-jammy-kernel.git] / mm / kmemleak.c
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 modifications and
17 * accesses to the object_tree_root. The object_list is the main list
18 * holding the metadata (struct kmemleak_object) for the allocated memory
19 * blocks. The object_tree_root is a red black tree used to look-up
20 * metadata based on a pointer to the corresponding memory block. The
21 * kmemleak_object structures are added to the object_list and
22 * object_tree_root in the create_object() function called from the
23 * kmemleak_alloc() callback and removed in delete_object() called from the
24 * kmemleak_free() callback
25 * - kmemleak_object.lock (raw_spinlock_t): protects a kmemleak_object.
26 * Accesses to the metadata (e.g. count) are protected by this lock. Note
27 * that some members of this structure may be protected by other means
28 * (atomic or kmemleak_lock). This lock is also held when scanning the
29 * corresponding memory block to avoid the kernel freeing it via the
30 * kmemleak_free() callback. This is less heavyweight than holding a global
31 * lock like kmemleak_lock during scanning.
32 * - scan_mutex (mutex): ensures that only one thread may scan the memory for
33 * unreferenced objects at a time. The gray_list contains the objects which
34 * are already referenced or marked as false positives and need to be
35 * scanned. This list is only modified during a scanning episode when the
36 * scan_mutex is held. At the end of a scan, the gray_list is always empty.
37 * Note that the kmemleak_object.use_count is incremented when an object is
38 * added to the gray_list and therefore cannot be freed. This mutex also
39 * prevents multiple users of the "kmemleak" debugfs file together with
40 * modifications to the memory scanning parameters including the scan_thread
41 * pointer
42 *
43 * Locks and mutexes are acquired/nested in the following order:
44 *
45 * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING)
46 *
47 * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex
48 * regions.
49 *
50 * The kmemleak_object structures have a use_count incremented or decremented
51 * using the get_object()/put_object() functions. When the use_count becomes
52 * 0, this count can no longer be incremented and put_object() schedules the
53 * kmemleak_object freeing via an RCU callback. All calls to the get_object()
54 * function must be protected by rcu_read_lock() to avoid accessing a freed
55 * structure.
56 */
57
58 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
59
60 #include <linux/init.h>
61 #include <linux/kernel.h>
62 #include <linux/list.h>
63 #include <linux/sched/signal.h>
64 #include <linux/sched/task.h>
65 #include <linux/sched/task_stack.h>
66 #include <linux/jiffies.h>
67 #include <linux/delay.h>
68 #include <linux/export.h>
69 #include <linux/kthread.h>
70 #include <linux/rbtree.h>
71 #include <linux/fs.h>
72 #include <linux/debugfs.h>
73 #include <linux/seq_file.h>
74 #include <linux/cpumask.h>
75 #include <linux/spinlock.h>
76 #include <linux/module.h>
77 #include <linux/mutex.h>
78 #include <linux/rcupdate.h>
79 #include <linux/stacktrace.h>
80 #include <linux/cache.h>
81 #include <linux/percpu.h>
82 #include <linux/memblock.h>
83 #include <linux/pfn.h>
84 #include <linux/mmzone.h>
85 #include <linux/slab.h>
86 #include <linux/thread_info.h>
87 #include <linux/err.h>
88 #include <linux/uaccess.h>
89 #include <linux/string.h>
90 #include <linux/nodemask.h>
91 #include <linux/mm.h>
92 #include <linux/workqueue.h>
93 #include <linux/crc32.h>
94
95 #include <asm/sections.h>
96 #include <asm/processor.h>
97 #include <linux/atomic.h>
98
99 #include <linux/kasan.h>
100 #include <linux/kfence.h>
101 #include <linux/kmemleak.h>
102 #include <linux/memory_hotplug.h>
103
104 /*
105 * Kmemleak configuration and common defines.
106 */
107 #define MAX_TRACE 16 /* stack trace length */
108 #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
109 #define SECS_FIRST_SCAN 60 /* delay before the first scan */
110 #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
111 #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
112
113 #define BYTES_PER_POINTER sizeof(void *)
114
115 /* GFP bitmask for kmemleak internal allocations */
116 #define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC | \
117 __GFP_NOLOCKDEP)) | \
118 __GFP_NORETRY | __GFP_NOMEMALLOC | \
119 __GFP_NOWARN)
120
121 /* scanning area inside a memory block */
122 struct kmemleak_scan_area {
123 struct hlist_node node;
124 unsigned long start;
125 size_t size;
126 };
127
128 #define KMEMLEAK_GREY 0
129 #define KMEMLEAK_BLACK -1
130
131 /*
132 * Structure holding the metadata for each allocated memory block.
133 * Modifications to such objects should be made while holding the
134 * object->lock. Insertions or deletions from object_list, gray_list or
135 * rb_node are already protected by the corresponding locks or mutex (see
136 * the notes on locking above). These objects are reference-counted
137 * (use_count) and freed using the RCU mechanism.
138 */
139 struct kmemleak_object {
140 raw_spinlock_t lock;
141 unsigned int flags; /* object status flags */
142 struct list_head object_list;
143 struct list_head gray_list;
144 struct rb_node rb_node;
145 struct rcu_head rcu; /* object_list lockless traversal */
146 /* object usage count; object freed when use_count == 0 */
147 atomic_t use_count;
148 unsigned long pointer;
149 size_t size;
150 /* pass surplus references to this pointer */
151 unsigned long excess_ref;
152 /* minimum number of a pointers found before it is considered leak */
153 int min_count;
154 /* the total number of pointers found pointing to this object */
155 int count;
156 /* checksum for detecting modified objects */
157 u32 checksum;
158 /* memory ranges to be scanned inside an object (empty for all) */
159 struct hlist_head area_list;
160 unsigned long trace[MAX_TRACE];
161 unsigned int trace_len;
162 unsigned long jiffies; /* creation timestamp */
163 pid_t pid; /* pid of the current task */
164 char comm[TASK_COMM_LEN]; /* executable name */
165 };
166
167 /* flag representing the memory block allocation status */
168 #define OBJECT_ALLOCATED (1 << 0)
169 /* flag set after the first reporting of an unreference object */
170 #define OBJECT_REPORTED (1 << 1)
171 /* flag set to not scan the object */
172 #define OBJECT_NO_SCAN (1 << 2)
173 /* flag set to fully scan the object when scan_area allocation failed */
174 #define OBJECT_FULL_SCAN (1 << 3)
175
176 #define HEX_PREFIX " "
177 /* number of bytes to print per line; must be 16 or 32 */
178 #define HEX_ROW_SIZE 16
179 /* number of bytes to print at a time (1, 2, 4, 8) */
180 #define HEX_GROUP_SIZE 1
181 /* include ASCII after the hex output */
182 #define HEX_ASCII 1
183 /* max number of lines to be printed */
184 #define HEX_MAX_LINES 2
185
186 /* the list of all allocated objects */
187 static LIST_HEAD(object_list);
188 /* the list of gray-colored objects (see color_gray comment below) */
189 static LIST_HEAD(gray_list);
190 /* memory pool allocation */
191 static struct kmemleak_object mem_pool[CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE];
192 static int mem_pool_free_count = ARRAY_SIZE(mem_pool);
193 static LIST_HEAD(mem_pool_free_list);
194 /* search tree for object boundaries */
195 static struct rb_root object_tree_root = RB_ROOT;
196 /* protecting the access to object_list and object_tree_root */
197 static DEFINE_RAW_SPINLOCK(kmemleak_lock);
198
199 /* allocation caches for kmemleak internal data */
200 static struct kmem_cache *object_cache;
201 static struct kmem_cache *scan_area_cache;
202
203 /* set if tracing memory operations is enabled */
204 static int kmemleak_enabled = 1;
205 /* same as above but only for the kmemleak_free() callback */
206 static int kmemleak_free_enabled = 1;
207 /* set in the late_initcall if there were no errors */
208 static int kmemleak_initialized;
209 /* set if a kmemleak warning was issued */
210 static int kmemleak_warning;
211 /* set if a fatal kmemleak error has occurred */
212 static int kmemleak_error;
213
214 /* minimum and maximum address that may be valid pointers */
215 static unsigned long min_addr = ULONG_MAX;
216 static unsigned long max_addr;
217
218 static struct task_struct *scan_thread;
219 /* used to avoid reporting of recently allocated objects */
220 static unsigned long jiffies_min_age;
221 static unsigned long jiffies_last_scan;
222 /* delay between automatic memory scannings */
223 static unsigned long jiffies_scan_wait;
224 /* enables or disables the task stacks scanning */
225 static int kmemleak_stack_scan = 1;
226 /* protects the memory scanning, parameters and debug/kmemleak file access */
227 static DEFINE_MUTEX(scan_mutex);
228 /* setting kmemleak=on, will set this var, skipping the disable */
229 static int kmemleak_skip_disable;
230 /* If there are leaks that can be reported */
231 static bool kmemleak_found_leaks;
232
233 static bool kmemleak_verbose;
234 module_param_named(verbose, kmemleak_verbose, bool, 0600);
235
236 static void kmemleak_disable(void);
237
238 /*
239 * Print a warning and dump the stack trace.
240 */
241 #define kmemleak_warn(x...) do { \
242 pr_warn(x); \
243 dump_stack(); \
244 kmemleak_warning = 1; \
245 } while (0)
246
247 /*
248 * Macro invoked when a serious kmemleak condition occurred and cannot be
249 * recovered from. Kmemleak will be disabled and further allocation/freeing
250 * tracing no longer available.
251 */
252 #define kmemleak_stop(x...) do { \
253 kmemleak_warn(x); \
254 kmemleak_disable(); \
255 } while (0)
256
257 #define warn_or_seq_printf(seq, fmt, ...) do { \
258 if (seq) \
259 seq_printf(seq, fmt, ##__VA_ARGS__); \
260 else \
261 pr_warn(fmt, ##__VA_ARGS__); \
262 } while (0)
263
264 static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type,
265 int rowsize, int groupsize, const void *buf,
266 size_t len, bool ascii)
267 {
268 if (seq)
269 seq_hex_dump(seq, HEX_PREFIX, prefix_type, rowsize, groupsize,
270 buf, len, ascii);
271 else
272 print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type,
273 rowsize, groupsize, buf, len, ascii);
274 }
275
276 /*
277 * Printing of the objects hex dump to the seq file. The number of lines to be
278 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
279 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
280 * with the object->lock held.
281 */
282 static void hex_dump_object(struct seq_file *seq,
283 struct kmemleak_object *object)
284 {
285 const u8 *ptr = (const u8 *)object->pointer;
286 size_t len;
287
288 /* limit the number of lines to HEX_MAX_LINES */
289 len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
290
291 warn_or_seq_printf(seq, " hex dump (first %zu bytes):\n", len);
292 kasan_disable_current();
293 warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE,
294 HEX_GROUP_SIZE, kasan_reset_tag((void *)ptr), len, HEX_ASCII);
295 kasan_enable_current();
296 }
297
298 /*
299 * Object colors, encoded with count and min_count:
300 * - white - orphan object, not enough references to it (count < min_count)
301 * - gray - not orphan, not marked as false positive (min_count == 0) or
302 * sufficient references to it (count >= min_count)
303 * - black - ignore, it doesn't contain references (e.g. text section)
304 * (min_count == -1). No function defined for this color.
305 * Newly created objects don't have any color assigned (object->count == -1)
306 * before the next memory scan when they become white.
307 */
308 static bool color_white(const struct kmemleak_object *object)
309 {
310 return object->count != KMEMLEAK_BLACK &&
311 object->count < object->min_count;
312 }
313
314 static bool color_gray(const struct kmemleak_object *object)
315 {
316 return object->min_count != KMEMLEAK_BLACK &&
317 object->count >= object->min_count;
318 }
319
320 /*
321 * Objects are considered unreferenced only if their color is white, they have
322 * not be deleted and have a minimum age to avoid false positives caused by
323 * pointers temporarily stored in CPU registers.
324 */
325 static bool unreferenced_object(struct kmemleak_object *object)
326 {
327 return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
328 time_before_eq(object->jiffies + jiffies_min_age,
329 jiffies_last_scan);
330 }
331
332 /*
333 * Printing of the unreferenced objects information to the seq file. The
334 * print_unreferenced function must be called with the object->lock held.
335 */
336 static void print_unreferenced(struct seq_file *seq,
337 struct kmemleak_object *object)
338 {
339 int i;
340 unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
341
342 warn_or_seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
343 object->pointer, object->size);
344 warn_or_seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
345 object->comm, object->pid, object->jiffies,
346 msecs_age / 1000, msecs_age % 1000);
347 hex_dump_object(seq, object);
348 warn_or_seq_printf(seq, " backtrace:\n");
349
350 for (i = 0; i < object->trace_len; i++) {
351 void *ptr = (void *)object->trace[i];
352 warn_or_seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
353 }
354 }
355
356 /*
357 * Print the kmemleak_object information. This function is used mainly for
358 * debugging special cases when kmemleak operations. It must be called with
359 * the object->lock held.
360 */
361 static void dump_object_info(struct kmemleak_object *object)
362 {
363 pr_notice("Object 0x%08lx (size %zu):\n",
364 object->pointer, object->size);
365 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
366 object->comm, object->pid, object->jiffies);
367 pr_notice(" min_count = %d\n", object->min_count);
368 pr_notice(" count = %d\n", object->count);
369 pr_notice(" flags = 0x%x\n", object->flags);
370 pr_notice(" checksum = %u\n", object->checksum);
371 pr_notice(" backtrace:\n");
372 stack_trace_print(object->trace, object->trace_len, 4);
373 }
374
375 /*
376 * Look-up a memory block metadata (kmemleak_object) in the object search
377 * tree based on a pointer value. If alias is 0, only values pointing to the
378 * beginning of the memory block are allowed. The kmemleak_lock must be held
379 * when calling this function.
380 */
381 static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
382 {
383 struct rb_node *rb = object_tree_root.rb_node;
384
385 while (rb) {
386 struct kmemleak_object *object =
387 rb_entry(rb, struct kmemleak_object, rb_node);
388 if (ptr < object->pointer)
389 rb = object->rb_node.rb_left;
390 else if (object->pointer + object->size <= ptr)
391 rb = object->rb_node.rb_right;
392 else if (object->pointer == ptr || alias)
393 return object;
394 else {
395 kmemleak_warn("Found object by alias at 0x%08lx\n",
396 ptr);
397 dump_object_info(object);
398 break;
399 }
400 }
401 return NULL;
402 }
403
404 /*
405 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
406 * that once an object's use_count reached 0, the RCU freeing was already
407 * registered and the object should no longer be used. This function must be
408 * called under the protection of rcu_read_lock().
409 */
410 static int get_object(struct kmemleak_object *object)
411 {
412 return atomic_inc_not_zero(&object->use_count);
413 }
414
415 /*
416 * Memory pool allocation and freeing. kmemleak_lock must not be held.
417 */
418 static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
419 {
420 unsigned long flags;
421 struct kmemleak_object *object;
422
423 /* try the slab allocator first */
424 if (object_cache) {
425 object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
426 if (object)
427 return object;
428 }
429
430 /* slab allocation failed, try the memory pool */
431 raw_spin_lock_irqsave(&kmemleak_lock, flags);
432 object = list_first_entry_or_null(&mem_pool_free_list,
433 typeof(*object), object_list);
434 if (object)
435 list_del(&object->object_list);
436 else if (mem_pool_free_count)
437 object = &mem_pool[--mem_pool_free_count];
438 else
439 pr_warn_once("Memory pool empty, consider increasing CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE\n");
440 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
441
442 return object;
443 }
444
445 /*
446 * Return the object to either the slab allocator or the memory pool.
447 */
448 static void mem_pool_free(struct kmemleak_object *object)
449 {
450 unsigned long flags;
451
452 if (object < mem_pool || object >= mem_pool + ARRAY_SIZE(mem_pool)) {
453 kmem_cache_free(object_cache, object);
454 return;
455 }
456
457 /* add the object to the memory pool free list */
458 raw_spin_lock_irqsave(&kmemleak_lock, flags);
459 list_add(&object->object_list, &mem_pool_free_list);
460 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
461 }
462
463 /*
464 * RCU callback to free a kmemleak_object.
465 */
466 static void free_object_rcu(struct rcu_head *rcu)
467 {
468 struct hlist_node *tmp;
469 struct kmemleak_scan_area *area;
470 struct kmemleak_object *object =
471 container_of(rcu, struct kmemleak_object, rcu);
472
473 /*
474 * Once use_count is 0 (guaranteed by put_object), there is no other
475 * code accessing this object, hence no need for locking.
476 */
477 hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
478 hlist_del(&area->node);
479 kmem_cache_free(scan_area_cache, area);
480 }
481 mem_pool_free(object);
482 }
483
484 /*
485 * Decrement the object use_count. Once the count is 0, free the object using
486 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
487 * delete_object() path, the delayed RCU freeing ensures that there is no
488 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
489 * is also possible.
490 */
491 static void put_object(struct kmemleak_object *object)
492 {
493 if (!atomic_dec_and_test(&object->use_count))
494 return;
495
496 /* should only get here after delete_object was called */
497 WARN_ON(object->flags & OBJECT_ALLOCATED);
498
499 /*
500 * It may be too early for the RCU callbacks, however, there is no
501 * concurrent object_list traversal when !object_cache and all objects
502 * came from the memory pool. Free the object directly.
503 */
504 if (object_cache)
505 call_rcu(&object->rcu, free_object_rcu);
506 else
507 free_object_rcu(&object->rcu);
508 }
509
510 /*
511 * Look up an object in the object search tree and increase its use_count.
512 */
513 static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
514 {
515 unsigned long flags;
516 struct kmemleak_object *object;
517
518 rcu_read_lock();
519 raw_spin_lock_irqsave(&kmemleak_lock, flags);
520 object = lookup_object(ptr, alias);
521 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
522
523 /* check whether the object is still available */
524 if (object && !get_object(object))
525 object = NULL;
526 rcu_read_unlock();
527
528 return object;
529 }
530
531 /*
532 * Remove an object from the object_tree_root and object_list. Must be called
533 * with the kmemleak_lock held _if_ kmemleak is still enabled.
534 */
535 static void __remove_object(struct kmemleak_object *object)
536 {
537 rb_erase(&object->rb_node, &object_tree_root);
538 list_del_rcu(&object->object_list);
539 }
540
541 /*
542 * Look up an object in the object search tree and remove it from both
543 * object_tree_root and object_list. The returned object's use_count should be
544 * at least 1, as initially set by create_object().
545 */
546 static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias)
547 {
548 unsigned long flags;
549 struct kmemleak_object *object;
550
551 raw_spin_lock_irqsave(&kmemleak_lock, flags);
552 object = lookup_object(ptr, alias);
553 if (object)
554 __remove_object(object);
555 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
556
557 return object;
558 }
559
560 /*
561 * Save stack trace to the given array of MAX_TRACE size.
562 */
563 static int __save_stack_trace(unsigned long *trace)
564 {
565 return stack_trace_save(trace, MAX_TRACE, 2);
566 }
567
568 /*
569 * Create the metadata (struct kmemleak_object) corresponding to an allocated
570 * memory block and add it to the object_list and object_tree_root.
571 */
572 static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
573 int min_count, gfp_t gfp)
574 {
575 unsigned long flags;
576 struct kmemleak_object *object, *parent;
577 struct rb_node **link, *rb_parent;
578 unsigned long untagged_ptr;
579
580 object = mem_pool_alloc(gfp);
581 if (!object) {
582 pr_warn("Cannot allocate a kmemleak_object structure\n");
583 kmemleak_disable();
584 return NULL;
585 }
586
587 INIT_LIST_HEAD(&object->object_list);
588 INIT_LIST_HEAD(&object->gray_list);
589 INIT_HLIST_HEAD(&object->area_list);
590 raw_spin_lock_init(&object->lock);
591 atomic_set(&object->use_count, 1);
592 object->flags = OBJECT_ALLOCATED;
593 object->pointer = ptr;
594 object->size = kfence_ksize((void *)ptr) ?: size;
595 object->excess_ref = 0;
596 object->min_count = min_count;
597 object->count = 0; /* white color initially */
598 object->jiffies = jiffies;
599 object->checksum = 0;
600
601 /* task information */
602 if (in_hardirq()) {
603 object->pid = 0;
604 strncpy(object->comm, "hardirq", sizeof(object->comm));
605 } else if (in_serving_softirq()) {
606 object->pid = 0;
607 strncpy(object->comm, "softirq", sizeof(object->comm));
608 } else {
609 object->pid = current->pid;
610 /*
611 * There is a small chance of a race with set_task_comm(),
612 * however using get_task_comm() here may cause locking
613 * dependency issues with current->alloc_lock. In the worst
614 * case, the command line is not correct.
615 */
616 strncpy(object->comm, current->comm, sizeof(object->comm));
617 }
618
619 /* kernel backtrace */
620 object->trace_len = __save_stack_trace(object->trace);
621
622 raw_spin_lock_irqsave(&kmemleak_lock, flags);
623
624 untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
625 min_addr = min(min_addr, untagged_ptr);
626 max_addr = max(max_addr, untagged_ptr + size);
627 link = &object_tree_root.rb_node;
628 rb_parent = NULL;
629 while (*link) {
630 rb_parent = *link;
631 parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
632 if (ptr + size <= parent->pointer)
633 link = &parent->rb_node.rb_left;
634 else if (parent->pointer + parent->size <= ptr)
635 link = &parent->rb_node.rb_right;
636 else {
637 kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
638 ptr);
639 /*
640 * No need for parent->lock here since "parent" cannot
641 * be freed while the kmemleak_lock is held.
642 */
643 dump_object_info(parent);
644 kmem_cache_free(object_cache, object);
645 object = NULL;
646 goto out;
647 }
648 }
649 rb_link_node(&object->rb_node, rb_parent, link);
650 rb_insert_color(&object->rb_node, &object_tree_root);
651
652 list_add_tail_rcu(&object->object_list, &object_list);
653 out:
654 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
655 return object;
656 }
657
658 /*
659 * Mark the object as not allocated and schedule RCU freeing via put_object().
660 */
661 static void __delete_object(struct kmemleak_object *object)
662 {
663 unsigned long flags;
664
665 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
666 WARN_ON(atomic_read(&object->use_count) < 1);
667
668 /*
669 * Locking here also ensures that the corresponding memory block
670 * cannot be freed when it is being scanned.
671 */
672 raw_spin_lock_irqsave(&object->lock, flags);
673 object->flags &= ~OBJECT_ALLOCATED;
674 raw_spin_unlock_irqrestore(&object->lock, flags);
675 put_object(object);
676 }
677
678 /*
679 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
680 * delete it.
681 */
682 static void delete_object_full(unsigned long ptr)
683 {
684 struct kmemleak_object *object;
685
686 object = find_and_remove_object(ptr, 0);
687 if (!object) {
688 #ifdef DEBUG
689 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
690 ptr);
691 #endif
692 return;
693 }
694 __delete_object(object);
695 }
696
697 /*
698 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
699 * delete it. If the memory block is partially freed, the function may create
700 * additional metadata for the remaining parts of the block.
701 */
702 static void delete_object_part(unsigned long ptr, size_t size)
703 {
704 struct kmemleak_object *object;
705 unsigned long start, end;
706
707 object = find_and_remove_object(ptr, 1);
708 if (!object) {
709 #ifdef DEBUG
710 kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
711 ptr, size);
712 #endif
713 return;
714 }
715
716 /*
717 * Create one or two objects that may result from the memory block
718 * split. Note that partial freeing is only done by free_bootmem() and
719 * this happens before kmemleak_init() is called.
720 */
721 start = object->pointer;
722 end = object->pointer + object->size;
723 if (ptr > start)
724 create_object(start, ptr - start, object->min_count,
725 GFP_KERNEL);
726 if (ptr + size < end)
727 create_object(ptr + size, end - ptr - size, object->min_count,
728 GFP_KERNEL);
729
730 __delete_object(object);
731 }
732
733 static void __paint_it(struct kmemleak_object *object, int color)
734 {
735 object->min_count = color;
736 if (color == KMEMLEAK_BLACK)
737 object->flags |= OBJECT_NO_SCAN;
738 }
739
740 static void paint_it(struct kmemleak_object *object, int color)
741 {
742 unsigned long flags;
743
744 raw_spin_lock_irqsave(&object->lock, flags);
745 __paint_it(object, color);
746 raw_spin_unlock_irqrestore(&object->lock, flags);
747 }
748
749 static void paint_ptr(unsigned long ptr, int color)
750 {
751 struct kmemleak_object *object;
752
753 object = find_and_get_object(ptr, 0);
754 if (!object) {
755 kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n",
756 ptr,
757 (color == KMEMLEAK_GREY) ? "Grey" :
758 (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
759 return;
760 }
761 paint_it(object, color);
762 put_object(object);
763 }
764
765 /*
766 * Mark an object permanently as gray-colored so that it can no longer be
767 * reported as a leak. This is used in general to mark a false positive.
768 */
769 static void make_gray_object(unsigned long ptr)
770 {
771 paint_ptr(ptr, KMEMLEAK_GREY);
772 }
773
774 /*
775 * Mark the object as black-colored so that it is ignored from scans and
776 * reporting.
777 */
778 static void make_black_object(unsigned long ptr)
779 {
780 paint_ptr(ptr, KMEMLEAK_BLACK);
781 }
782
783 /*
784 * Add a scanning area to the object. If at least one such area is added,
785 * kmemleak will only scan these ranges rather than the whole memory block.
786 */
787 static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
788 {
789 unsigned long flags;
790 struct kmemleak_object *object;
791 struct kmemleak_scan_area *area = NULL;
792 unsigned long untagged_ptr;
793 unsigned long untagged_objp;
794
795 object = find_and_get_object(ptr, 1);
796 if (!object) {
797 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
798 ptr);
799 return;
800 }
801
802 untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
803 untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer);
804
805 if (scan_area_cache)
806 area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
807
808 raw_spin_lock_irqsave(&object->lock, flags);
809 if (!area) {
810 pr_warn_once("Cannot allocate a scan area, scanning the full object\n");
811 /* mark the object for full scan to avoid false positives */
812 object->flags |= OBJECT_FULL_SCAN;
813 goto out_unlock;
814 }
815 if (size == SIZE_MAX) {
816 size = untagged_objp + object->size - untagged_ptr;
817 } else if (untagged_ptr + size > untagged_objp + object->size) {
818 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
819 dump_object_info(object);
820 kmem_cache_free(scan_area_cache, area);
821 goto out_unlock;
822 }
823
824 INIT_HLIST_NODE(&area->node);
825 area->start = ptr;
826 area->size = size;
827
828 hlist_add_head(&area->node, &object->area_list);
829 out_unlock:
830 raw_spin_unlock_irqrestore(&object->lock, flags);
831 put_object(object);
832 }
833
834 /*
835 * Any surplus references (object already gray) to 'ptr' are passed to
836 * 'excess_ref'. This is used in the vmalloc() case where a pointer to
837 * vm_struct may be used as an alternative reference to the vmalloc'ed object
838 * (see free_thread_stack()).
839 */
840 static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref)
841 {
842 unsigned long flags;
843 struct kmemleak_object *object;
844
845 object = find_and_get_object(ptr, 0);
846 if (!object) {
847 kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n",
848 ptr);
849 return;
850 }
851
852 raw_spin_lock_irqsave(&object->lock, flags);
853 object->excess_ref = excess_ref;
854 raw_spin_unlock_irqrestore(&object->lock, flags);
855 put_object(object);
856 }
857
858 /*
859 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
860 * pointer. Such object will not be scanned by kmemleak but references to it
861 * are searched.
862 */
863 static void object_no_scan(unsigned long ptr)
864 {
865 unsigned long flags;
866 struct kmemleak_object *object;
867
868 object = find_and_get_object(ptr, 0);
869 if (!object) {
870 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
871 return;
872 }
873
874 raw_spin_lock_irqsave(&object->lock, flags);
875 object->flags |= OBJECT_NO_SCAN;
876 raw_spin_unlock_irqrestore(&object->lock, flags);
877 put_object(object);
878 }
879
880 /**
881 * kmemleak_alloc - register a newly allocated object
882 * @ptr: pointer to beginning of the object
883 * @size: size of the object
884 * @min_count: minimum number of references to this object. If during memory
885 * scanning a number of references less than @min_count is found,
886 * the object is reported as a memory leak. If @min_count is 0,
887 * the object is never reported as a leak. If @min_count is -1,
888 * the object is ignored (not scanned and not reported as a leak)
889 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
890 *
891 * This function is called from the kernel allocators when a new object
892 * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.).
893 */
894 void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
895 gfp_t gfp)
896 {
897 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
898
899 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
900 create_object((unsigned long)ptr, size, min_count, gfp);
901 }
902 EXPORT_SYMBOL_GPL(kmemleak_alloc);
903
904 /**
905 * kmemleak_alloc_percpu - register a newly allocated __percpu object
906 * @ptr: __percpu pointer to beginning of the object
907 * @size: size of the object
908 * @gfp: flags used for kmemleak internal memory allocations
909 *
910 * This function is called from the kernel percpu allocator when a new object
911 * (memory block) is allocated (alloc_percpu).
912 */
913 void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
914 gfp_t gfp)
915 {
916 unsigned int cpu;
917
918 pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size);
919
920 /*
921 * Percpu allocations are only scanned and not reported as leaks
922 * (min_count is set to 0).
923 */
924 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
925 for_each_possible_cpu(cpu)
926 create_object((unsigned long)per_cpu_ptr(ptr, cpu),
927 size, 0, gfp);
928 }
929 EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
930
931 /**
932 * kmemleak_vmalloc - register a newly vmalloc'ed object
933 * @area: pointer to vm_struct
934 * @size: size of the object
935 * @gfp: __vmalloc() flags used for kmemleak internal memory allocations
936 *
937 * This function is called from the vmalloc() kernel allocator when a new
938 * object (memory block) is allocated.
939 */
940 void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp)
941 {
942 pr_debug("%s(0x%p, %zu)\n", __func__, area, size);
943
944 /*
945 * A min_count = 2 is needed because vm_struct contains a reference to
946 * the virtual address of the vmalloc'ed block.
947 */
948 if (kmemleak_enabled) {
949 create_object((unsigned long)area->addr, size, 2, gfp);
950 object_set_excess_ref((unsigned long)area,
951 (unsigned long)area->addr);
952 }
953 }
954 EXPORT_SYMBOL_GPL(kmemleak_vmalloc);
955
956 /**
957 * kmemleak_free - unregister a previously registered object
958 * @ptr: pointer to beginning of the object
959 *
960 * This function is called from the kernel allocators when an object (memory
961 * block) is freed (kmem_cache_free, kfree, vfree etc.).
962 */
963 void __ref kmemleak_free(const void *ptr)
964 {
965 pr_debug("%s(0x%p)\n", __func__, ptr);
966
967 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
968 delete_object_full((unsigned long)ptr);
969 }
970 EXPORT_SYMBOL_GPL(kmemleak_free);
971
972 /**
973 * kmemleak_free_part - partially unregister a previously registered object
974 * @ptr: pointer to the beginning or inside the object. This also
975 * represents the start of the range to be freed
976 * @size: size to be unregistered
977 *
978 * This function is called when only a part of a memory block is freed
979 * (usually from the bootmem allocator).
980 */
981 void __ref kmemleak_free_part(const void *ptr, size_t size)
982 {
983 pr_debug("%s(0x%p)\n", __func__, ptr);
984
985 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
986 delete_object_part((unsigned long)ptr, size);
987 }
988 EXPORT_SYMBOL_GPL(kmemleak_free_part);
989
990 /**
991 * kmemleak_free_percpu - unregister a previously registered __percpu object
992 * @ptr: __percpu pointer to beginning of the object
993 *
994 * This function is called from the kernel percpu allocator when an object
995 * (memory block) is freed (free_percpu).
996 */
997 void __ref kmemleak_free_percpu(const void __percpu *ptr)
998 {
999 unsigned int cpu;
1000
1001 pr_debug("%s(0x%p)\n", __func__, ptr);
1002
1003 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
1004 for_each_possible_cpu(cpu)
1005 delete_object_full((unsigned long)per_cpu_ptr(ptr,
1006 cpu));
1007 }
1008 EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
1009
1010 /**
1011 * kmemleak_update_trace - update object allocation stack trace
1012 * @ptr: pointer to beginning of the object
1013 *
1014 * Override the object allocation stack trace for cases where the actual
1015 * allocation place is not always useful.
1016 */
1017 void __ref kmemleak_update_trace(const void *ptr)
1018 {
1019 struct kmemleak_object *object;
1020 unsigned long flags;
1021
1022 pr_debug("%s(0x%p)\n", __func__, ptr);
1023
1024 if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
1025 return;
1026
1027 object = find_and_get_object((unsigned long)ptr, 1);
1028 if (!object) {
1029 #ifdef DEBUG
1030 kmemleak_warn("Updating stack trace for unknown object at %p\n",
1031 ptr);
1032 #endif
1033 return;
1034 }
1035
1036 raw_spin_lock_irqsave(&object->lock, flags);
1037 object->trace_len = __save_stack_trace(object->trace);
1038 raw_spin_unlock_irqrestore(&object->lock, flags);
1039
1040 put_object(object);
1041 }
1042 EXPORT_SYMBOL(kmemleak_update_trace);
1043
1044 /**
1045 * kmemleak_not_leak - mark an allocated object as false positive
1046 * @ptr: pointer to beginning of the object
1047 *
1048 * Calling this function on an object will cause the memory block to no longer
1049 * be reported as leak and always be scanned.
1050 */
1051 void __ref kmemleak_not_leak(const void *ptr)
1052 {
1053 pr_debug("%s(0x%p)\n", __func__, ptr);
1054
1055 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1056 make_gray_object((unsigned long)ptr);
1057 }
1058 EXPORT_SYMBOL(kmemleak_not_leak);
1059
1060 /**
1061 * kmemleak_ignore - ignore an allocated object
1062 * @ptr: pointer to beginning of the object
1063 *
1064 * Calling this function on an object will cause the memory block to be
1065 * ignored (not scanned and not reported as a leak). This is usually done when
1066 * it is known that the corresponding block is not a leak and does not contain
1067 * any references to other allocated memory blocks.
1068 */
1069 void __ref kmemleak_ignore(const void *ptr)
1070 {
1071 pr_debug("%s(0x%p)\n", __func__, ptr);
1072
1073 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1074 make_black_object((unsigned long)ptr);
1075 }
1076 EXPORT_SYMBOL(kmemleak_ignore);
1077
1078 /**
1079 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1080 * @ptr: pointer to beginning or inside the object. This also
1081 * represents the start of the scan area
1082 * @size: size of the scan area
1083 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1084 *
1085 * This function is used when it is known that only certain parts of an object
1086 * contain references to other objects. Kmemleak will only scan these areas
1087 * reducing the number false negatives.
1088 */
1089 void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1090 {
1091 pr_debug("%s(0x%p)\n", __func__, ptr);
1092
1093 if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
1094 add_scan_area((unsigned long)ptr, size, gfp);
1095 }
1096 EXPORT_SYMBOL(kmemleak_scan_area);
1097
1098 /**
1099 * kmemleak_no_scan - do not scan an allocated object
1100 * @ptr: pointer to beginning of the object
1101 *
1102 * This function notifies kmemleak not to scan the given memory block. Useful
1103 * in situations where it is known that the given object does not contain any
1104 * references to other objects. Kmemleak will not scan such objects reducing
1105 * the number of false negatives.
1106 */
1107 void __ref kmemleak_no_scan(const void *ptr)
1108 {
1109 pr_debug("%s(0x%p)\n", __func__, ptr);
1110
1111 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1112 object_no_scan((unsigned long)ptr);
1113 }
1114 EXPORT_SYMBOL(kmemleak_no_scan);
1115
1116 /**
1117 * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical
1118 * address argument
1119 * @phys: physical address of the object
1120 * @size: size of the object
1121 * @min_count: minimum number of references to this object.
1122 * See kmemleak_alloc()
1123 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1124 */
1125 void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, int min_count,
1126 gfp_t gfp)
1127 {
1128 if (PHYS_PFN(phys) >= min_low_pfn && PHYS_PFN(phys) < max_low_pfn)
1129 kmemleak_alloc(__va(phys), size, min_count, gfp);
1130 }
1131 EXPORT_SYMBOL(kmemleak_alloc_phys);
1132
1133 /**
1134 * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a
1135 * physical address argument
1136 * @phys: physical address if the beginning or inside an object. This
1137 * also represents the start of the range to be freed
1138 * @size: size to be unregistered
1139 */
1140 void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size)
1141 {
1142 if (PHYS_PFN(phys) >= min_low_pfn && PHYS_PFN(phys) < max_low_pfn)
1143 kmemleak_free_part(__va(phys), size);
1144 }
1145 EXPORT_SYMBOL(kmemleak_free_part_phys);
1146
1147 /**
1148 * kmemleak_not_leak_phys - similar to kmemleak_not_leak but taking a physical
1149 * address argument
1150 * @phys: physical address of the object
1151 */
1152 void __ref kmemleak_not_leak_phys(phys_addr_t phys)
1153 {
1154 if (PHYS_PFN(phys) >= min_low_pfn && PHYS_PFN(phys) < max_low_pfn)
1155 kmemleak_not_leak(__va(phys));
1156 }
1157 EXPORT_SYMBOL(kmemleak_not_leak_phys);
1158
1159 /**
1160 * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical
1161 * address argument
1162 * @phys: physical address of the object
1163 */
1164 void __ref kmemleak_ignore_phys(phys_addr_t phys)
1165 {
1166 if (PHYS_PFN(phys) >= min_low_pfn && PHYS_PFN(phys) < max_low_pfn)
1167 kmemleak_ignore(__va(phys));
1168 }
1169 EXPORT_SYMBOL(kmemleak_ignore_phys);
1170
1171 /*
1172 * Update an object's checksum and return true if it was modified.
1173 */
1174 static bool update_checksum(struct kmemleak_object *object)
1175 {
1176 u32 old_csum = object->checksum;
1177
1178 kasan_disable_current();
1179 kcsan_disable_current();
1180 object->checksum = crc32(0, kasan_reset_tag((void *)object->pointer), object->size);
1181 kasan_enable_current();
1182 kcsan_enable_current();
1183
1184 return object->checksum != old_csum;
1185 }
1186
1187 /*
1188 * Update an object's references. object->lock must be held by the caller.
1189 */
1190 static void update_refs(struct kmemleak_object *object)
1191 {
1192 if (!color_white(object)) {
1193 /* non-orphan, ignored or new */
1194 return;
1195 }
1196
1197 /*
1198 * Increase the object's reference count (number of pointers to the
1199 * memory block). If this count reaches the required minimum, the
1200 * object's color will become gray and it will be added to the
1201 * gray_list.
1202 */
1203 object->count++;
1204 if (color_gray(object)) {
1205 /* put_object() called when removing from gray_list */
1206 WARN_ON(!get_object(object));
1207 list_add_tail(&object->gray_list, &gray_list);
1208 }
1209 }
1210
1211 /*
1212 * Memory scanning is a long process and it needs to be interruptible. This
1213 * function checks whether such interrupt condition occurred.
1214 */
1215 static int scan_should_stop(void)
1216 {
1217 if (!kmemleak_enabled)
1218 return 1;
1219
1220 /*
1221 * This function may be called from either process or kthread context,
1222 * hence the need to check for both stop conditions.
1223 */
1224 if (current->mm)
1225 return signal_pending(current);
1226 else
1227 return kthread_should_stop();
1228
1229 return 0;
1230 }
1231
1232 /*
1233 * Scan a memory block (exclusive range) for valid pointers and add those
1234 * found to the gray list.
1235 */
1236 static void scan_block(void *_start, void *_end,
1237 struct kmemleak_object *scanned)
1238 {
1239 unsigned long *ptr;
1240 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1241 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1242 unsigned long flags;
1243 unsigned long untagged_ptr;
1244
1245 raw_spin_lock_irqsave(&kmemleak_lock, flags);
1246 for (ptr = start; ptr < end; ptr++) {
1247 struct kmemleak_object *object;
1248 unsigned long pointer;
1249 unsigned long excess_ref;
1250
1251 if (scan_should_stop())
1252 break;
1253
1254 kasan_disable_current();
1255 pointer = *(unsigned long *)kasan_reset_tag((void *)ptr);
1256 kasan_enable_current();
1257
1258 untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer);
1259 if (untagged_ptr < min_addr || untagged_ptr >= max_addr)
1260 continue;
1261
1262 /*
1263 * No need for get_object() here since we hold kmemleak_lock.
1264 * object->use_count cannot be dropped to 0 while the object
1265 * is still present in object_tree_root and object_list
1266 * (with updates protected by kmemleak_lock).
1267 */
1268 object = lookup_object(pointer, 1);
1269 if (!object)
1270 continue;
1271 if (object == scanned)
1272 /* self referenced, ignore */
1273 continue;
1274
1275 /*
1276 * Avoid the lockdep recursive warning on object->lock being
1277 * previously acquired in scan_object(). These locks are
1278 * enclosed by scan_mutex.
1279 */
1280 raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1281 /* only pass surplus references (object already gray) */
1282 if (color_gray(object)) {
1283 excess_ref = object->excess_ref;
1284 /* no need for update_refs() if object already gray */
1285 } else {
1286 excess_ref = 0;
1287 update_refs(object);
1288 }
1289 raw_spin_unlock(&object->lock);
1290
1291 if (excess_ref) {
1292 object = lookup_object(excess_ref, 0);
1293 if (!object)
1294 continue;
1295 if (object == scanned)
1296 /* circular reference, ignore */
1297 continue;
1298 raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1299 update_refs(object);
1300 raw_spin_unlock(&object->lock);
1301 }
1302 }
1303 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
1304 }
1305
1306 /*
1307 * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
1308 */
1309 #ifdef CONFIG_SMP
1310 static void scan_large_block(void *start, void *end)
1311 {
1312 void *next;
1313
1314 while (start < end) {
1315 next = min(start + MAX_SCAN_SIZE, end);
1316 scan_block(start, next, NULL);
1317 start = next;
1318 cond_resched();
1319 }
1320 }
1321 #endif
1322
1323 /*
1324 * Scan a memory block corresponding to a kmemleak_object. A condition is
1325 * that object->use_count >= 1.
1326 */
1327 static void scan_object(struct kmemleak_object *object)
1328 {
1329 struct kmemleak_scan_area *area;
1330 unsigned long flags;
1331
1332 /*
1333 * Once the object->lock is acquired, the corresponding memory block
1334 * cannot be freed (the same lock is acquired in delete_object).
1335 */
1336 raw_spin_lock_irqsave(&object->lock, flags);
1337 if (object->flags & OBJECT_NO_SCAN)
1338 goto out;
1339 if (!(object->flags & OBJECT_ALLOCATED))
1340 /* already freed object */
1341 goto out;
1342 if (hlist_empty(&object->area_list) ||
1343 object->flags & OBJECT_FULL_SCAN) {
1344 void *start = (void *)object->pointer;
1345 void *end = (void *)(object->pointer + object->size);
1346 void *next;
1347
1348 do {
1349 next = min(start + MAX_SCAN_SIZE, end);
1350 scan_block(start, next, object);
1351
1352 start = next;
1353 if (start >= end)
1354 break;
1355
1356 raw_spin_unlock_irqrestore(&object->lock, flags);
1357 cond_resched();
1358 raw_spin_lock_irqsave(&object->lock, flags);
1359 } while (object->flags & OBJECT_ALLOCATED);
1360 } else
1361 hlist_for_each_entry(area, &object->area_list, node)
1362 scan_block((void *)area->start,
1363 (void *)(area->start + area->size),
1364 object);
1365 out:
1366 raw_spin_unlock_irqrestore(&object->lock, flags);
1367 }
1368
1369 /*
1370 * Scan the objects already referenced (gray objects). More objects will be
1371 * referenced and, if there are no memory leaks, all the objects are scanned.
1372 */
1373 static void scan_gray_list(void)
1374 {
1375 struct kmemleak_object *object, *tmp;
1376
1377 /*
1378 * The list traversal is safe for both tail additions and removals
1379 * from inside the loop. The kmemleak objects cannot be freed from
1380 * outside the loop because their use_count was incremented.
1381 */
1382 object = list_entry(gray_list.next, typeof(*object), gray_list);
1383 while (&object->gray_list != &gray_list) {
1384 cond_resched();
1385
1386 /* may add new objects to the list */
1387 if (!scan_should_stop())
1388 scan_object(object);
1389
1390 tmp = list_entry(object->gray_list.next, typeof(*object),
1391 gray_list);
1392
1393 /* remove the object from the list and release it */
1394 list_del(&object->gray_list);
1395 put_object(object);
1396
1397 object = tmp;
1398 }
1399 WARN_ON(!list_empty(&gray_list));
1400 }
1401
1402 /*
1403 * Scan data sections and all the referenced memory blocks allocated via the
1404 * kernel's standard allocators. This function must be called with the
1405 * scan_mutex held.
1406 */
1407 static void kmemleak_scan(void)
1408 {
1409 unsigned long flags;
1410 struct kmemleak_object *object;
1411 struct zone *zone;
1412 int __maybe_unused i;
1413 int new_leaks = 0;
1414
1415 jiffies_last_scan = jiffies;
1416
1417 /* prepare the kmemleak_object's */
1418 rcu_read_lock();
1419 list_for_each_entry_rcu(object, &object_list, object_list) {
1420 raw_spin_lock_irqsave(&object->lock, flags);
1421 #ifdef DEBUG
1422 /*
1423 * With a few exceptions there should be a maximum of
1424 * 1 reference to any object at this point.
1425 */
1426 if (atomic_read(&object->use_count) > 1) {
1427 pr_debug("object->use_count = %d\n",
1428 atomic_read(&object->use_count));
1429 dump_object_info(object);
1430 }
1431 #endif
1432 /* reset the reference count (whiten the object) */
1433 object->count = 0;
1434 if (color_gray(object) && get_object(object))
1435 list_add_tail(&object->gray_list, &gray_list);
1436
1437 raw_spin_unlock_irqrestore(&object->lock, flags);
1438 }
1439 rcu_read_unlock();
1440
1441 #ifdef CONFIG_SMP
1442 /* per-cpu sections scanning */
1443 for_each_possible_cpu(i)
1444 scan_large_block(__per_cpu_start + per_cpu_offset(i),
1445 __per_cpu_end + per_cpu_offset(i));
1446 #endif
1447
1448 /*
1449 * Struct page scanning for each node.
1450 */
1451 get_online_mems();
1452 for_each_populated_zone(zone) {
1453 unsigned long start_pfn = zone->zone_start_pfn;
1454 unsigned long end_pfn = zone_end_pfn(zone);
1455 unsigned long pfn;
1456
1457 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1458 struct page *page = pfn_to_online_page(pfn);
1459
1460 if (!page)
1461 continue;
1462
1463 /* only scan pages belonging to this zone */
1464 if (page_zone(page) != zone)
1465 continue;
1466 /* only scan if page is in use */
1467 if (page_count(page) == 0)
1468 continue;
1469 scan_block(page, page + 1, NULL);
1470 if (!(pfn & 63))
1471 cond_resched();
1472 }
1473 }
1474 put_online_mems();
1475
1476 /*
1477 * Scanning the task stacks (may introduce false negatives).
1478 */
1479 if (kmemleak_stack_scan) {
1480 struct task_struct *p, *g;
1481
1482 rcu_read_lock();
1483 for_each_process_thread(g, p) {
1484 void *stack = try_get_task_stack(p);
1485 if (stack) {
1486 scan_block(stack, stack + THREAD_SIZE, NULL);
1487 put_task_stack(p);
1488 }
1489 }
1490 rcu_read_unlock();
1491 }
1492
1493 /*
1494 * Scan the objects already referenced from the sections scanned
1495 * above.
1496 */
1497 scan_gray_list();
1498
1499 /*
1500 * Check for new or unreferenced objects modified since the previous
1501 * scan and color them gray until the next scan.
1502 */
1503 rcu_read_lock();
1504 list_for_each_entry_rcu(object, &object_list, object_list) {
1505 raw_spin_lock_irqsave(&object->lock, flags);
1506 if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1507 && update_checksum(object) && get_object(object)) {
1508 /* color it gray temporarily */
1509 object->count = object->min_count;
1510 list_add_tail(&object->gray_list, &gray_list);
1511 }
1512 raw_spin_unlock_irqrestore(&object->lock, flags);
1513 }
1514 rcu_read_unlock();
1515
1516 /*
1517 * Re-scan the gray list for modified unreferenced objects.
1518 */
1519 scan_gray_list();
1520
1521 /*
1522 * If scanning was stopped do not report any new unreferenced objects.
1523 */
1524 if (scan_should_stop())
1525 return;
1526
1527 /*
1528 * Scanning result reporting.
1529 */
1530 rcu_read_lock();
1531 list_for_each_entry_rcu(object, &object_list, object_list) {
1532 raw_spin_lock_irqsave(&object->lock, flags);
1533 if (unreferenced_object(object) &&
1534 !(object->flags & OBJECT_REPORTED)) {
1535 object->flags |= OBJECT_REPORTED;
1536
1537 if (kmemleak_verbose)
1538 print_unreferenced(NULL, object);
1539
1540 new_leaks++;
1541 }
1542 raw_spin_unlock_irqrestore(&object->lock, flags);
1543 }
1544 rcu_read_unlock();
1545
1546 if (new_leaks) {
1547 kmemleak_found_leaks = true;
1548
1549 pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
1550 new_leaks);
1551 }
1552
1553 }
1554
1555 /*
1556 * Thread function performing automatic memory scanning. Unreferenced objects
1557 * at the end of a memory scan are reported but only the first time.
1558 */
1559 static int kmemleak_scan_thread(void *arg)
1560 {
1561 static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN);
1562
1563 pr_info("Automatic memory scanning thread started\n");
1564 set_user_nice(current, 10);
1565
1566 /*
1567 * Wait before the first scan to allow the system to fully initialize.
1568 */
1569 if (first_run) {
1570 signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000);
1571 first_run = 0;
1572 while (timeout && !kthread_should_stop())
1573 timeout = schedule_timeout_interruptible(timeout);
1574 }
1575
1576 while (!kthread_should_stop()) {
1577 signed long timeout = READ_ONCE(jiffies_scan_wait);
1578
1579 mutex_lock(&scan_mutex);
1580 kmemleak_scan();
1581 mutex_unlock(&scan_mutex);
1582
1583 /* wait before the next scan */
1584 while (timeout && !kthread_should_stop())
1585 timeout = schedule_timeout_interruptible(timeout);
1586 }
1587
1588 pr_info("Automatic memory scanning thread ended\n");
1589
1590 return 0;
1591 }
1592
1593 /*
1594 * Start the automatic memory scanning thread. This function must be called
1595 * with the scan_mutex held.
1596 */
1597 static void start_scan_thread(void)
1598 {
1599 if (scan_thread)
1600 return;
1601 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1602 if (IS_ERR(scan_thread)) {
1603 pr_warn("Failed to create the scan thread\n");
1604 scan_thread = NULL;
1605 }
1606 }
1607
1608 /*
1609 * Stop the automatic memory scanning thread.
1610 */
1611 static void stop_scan_thread(void)
1612 {
1613 if (scan_thread) {
1614 kthread_stop(scan_thread);
1615 scan_thread = NULL;
1616 }
1617 }
1618
1619 /*
1620 * Iterate over the object_list and return the first valid object at or after
1621 * the required position with its use_count incremented. The function triggers
1622 * a memory scanning when the pos argument points to the first position.
1623 */
1624 static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1625 {
1626 struct kmemleak_object *object;
1627 loff_t n = *pos;
1628 int err;
1629
1630 err = mutex_lock_interruptible(&scan_mutex);
1631 if (err < 0)
1632 return ERR_PTR(err);
1633
1634 rcu_read_lock();
1635 list_for_each_entry_rcu(object, &object_list, object_list) {
1636 if (n-- > 0)
1637 continue;
1638 if (get_object(object))
1639 goto out;
1640 }
1641 object = NULL;
1642 out:
1643 return object;
1644 }
1645
1646 /*
1647 * Return the next object in the object_list. The function decrements the
1648 * use_count of the previous object and increases that of the next one.
1649 */
1650 static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1651 {
1652 struct kmemleak_object *prev_obj = v;
1653 struct kmemleak_object *next_obj = NULL;
1654 struct kmemleak_object *obj = prev_obj;
1655
1656 ++(*pos);
1657
1658 list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
1659 if (get_object(obj)) {
1660 next_obj = obj;
1661 break;
1662 }
1663 }
1664
1665 put_object(prev_obj);
1666 return next_obj;
1667 }
1668
1669 /*
1670 * Decrement the use_count of the last object required, if any.
1671 */
1672 static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1673 {
1674 if (!IS_ERR(v)) {
1675 /*
1676 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1677 * waiting was interrupted, so only release it if !IS_ERR.
1678 */
1679 rcu_read_unlock();
1680 mutex_unlock(&scan_mutex);
1681 if (v)
1682 put_object(v);
1683 }
1684 }
1685
1686 /*
1687 * Print the information for an unreferenced object to the seq file.
1688 */
1689 static int kmemleak_seq_show(struct seq_file *seq, void *v)
1690 {
1691 struct kmemleak_object *object = v;
1692 unsigned long flags;
1693
1694 raw_spin_lock_irqsave(&object->lock, flags);
1695 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1696 print_unreferenced(seq, object);
1697 raw_spin_unlock_irqrestore(&object->lock, flags);
1698 return 0;
1699 }
1700
1701 static const struct seq_operations kmemleak_seq_ops = {
1702 .start = kmemleak_seq_start,
1703 .next = kmemleak_seq_next,
1704 .stop = kmemleak_seq_stop,
1705 .show = kmemleak_seq_show,
1706 };
1707
1708 static int kmemleak_open(struct inode *inode, struct file *file)
1709 {
1710 return seq_open(file, &kmemleak_seq_ops);
1711 }
1712
1713 static int dump_str_object_info(const char *str)
1714 {
1715 unsigned long flags;
1716 struct kmemleak_object *object;
1717 unsigned long addr;
1718
1719 if (kstrtoul(str, 0, &addr))
1720 return -EINVAL;
1721 object = find_and_get_object(addr, 0);
1722 if (!object) {
1723 pr_info("Unknown object at 0x%08lx\n", addr);
1724 return -EINVAL;
1725 }
1726
1727 raw_spin_lock_irqsave(&object->lock, flags);
1728 dump_object_info(object);
1729 raw_spin_unlock_irqrestore(&object->lock, flags);
1730
1731 put_object(object);
1732 return 0;
1733 }
1734
1735 /*
1736 * We use grey instead of black to ensure we can do future scans on the same
1737 * objects. If we did not do future scans these black objects could
1738 * potentially contain references to newly allocated objects in the future and
1739 * we'd end up with false positives.
1740 */
1741 static void kmemleak_clear(void)
1742 {
1743 struct kmemleak_object *object;
1744 unsigned long flags;
1745
1746 rcu_read_lock();
1747 list_for_each_entry_rcu(object, &object_list, object_list) {
1748 raw_spin_lock_irqsave(&object->lock, flags);
1749 if ((object->flags & OBJECT_REPORTED) &&
1750 unreferenced_object(object))
1751 __paint_it(object, KMEMLEAK_GREY);
1752 raw_spin_unlock_irqrestore(&object->lock, flags);
1753 }
1754 rcu_read_unlock();
1755
1756 kmemleak_found_leaks = false;
1757 }
1758
1759 static void __kmemleak_do_cleanup(void);
1760
1761 /*
1762 * File write operation to configure kmemleak at run-time. The following
1763 * commands can be written to the /sys/kernel/debug/kmemleak file:
1764 * off - disable kmemleak (irreversible)
1765 * stack=on - enable the task stacks scanning
1766 * stack=off - disable the tasks stacks scanning
1767 * scan=on - start the automatic memory scanning thread
1768 * scan=off - stop the automatic memory scanning thread
1769 * scan=... - set the automatic memory scanning period in seconds (0 to
1770 * disable it)
1771 * scan - trigger a memory scan
1772 * clear - mark all current reported unreferenced kmemleak objects as
1773 * grey to ignore printing them, or free all kmemleak objects
1774 * if kmemleak has been disabled.
1775 * dump=... - dump information about the object found at the given address
1776 */
1777 static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1778 size_t size, loff_t *ppos)
1779 {
1780 char buf[64];
1781 int buf_size;
1782 int ret;
1783
1784 buf_size = min(size, (sizeof(buf) - 1));
1785 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1786 return -EFAULT;
1787 buf[buf_size] = 0;
1788
1789 ret = mutex_lock_interruptible(&scan_mutex);
1790 if (ret < 0)
1791 return ret;
1792
1793 if (strncmp(buf, "clear", 5) == 0) {
1794 if (kmemleak_enabled)
1795 kmemleak_clear();
1796 else
1797 __kmemleak_do_cleanup();
1798 goto out;
1799 }
1800
1801 if (!kmemleak_enabled) {
1802 ret = -EPERM;
1803 goto out;
1804 }
1805
1806 if (strncmp(buf, "off", 3) == 0)
1807 kmemleak_disable();
1808 else if (strncmp(buf, "stack=on", 8) == 0)
1809 kmemleak_stack_scan = 1;
1810 else if (strncmp(buf, "stack=off", 9) == 0)
1811 kmemleak_stack_scan = 0;
1812 else if (strncmp(buf, "scan=on", 7) == 0)
1813 start_scan_thread();
1814 else if (strncmp(buf, "scan=off", 8) == 0)
1815 stop_scan_thread();
1816 else if (strncmp(buf, "scan=", 5) == 0) {
1817 unsigned secs;
1818 unsigned long msecs;
1819
1820 ret = kstrtouint(buf + 5, 0, &secs);
1821 if (ret < 0)
1822 goto out;
1823
1824 msecs = secs * MSEC_PER_SEC;
1825 if (msecs > UINT_MAX)
1826 msecs = UINT_MAX;
1827
1828 stop_scan_thread();
1829 if (msecs) {
1830 WRITE_ONCE(jiffies_scan_wait, msecs_to_jiffies(msecs));
1831 start_scan_thread();
1832 }
1833 } else if (strncmp(buf, "scan", 4) == 0)
1834 kmemleak_scan();
1835 else if (strncmp(buf, "dump=", 5) == 0)
1836 ret = dump_str_object_info(buf + 5);
1837 else
1838 ret = -EINVAL;
1839
1840 out:
1841 mutex_unlock(&scan_mutex);
1842 if (ret < 0)
1843 return ret;
1844
1845 /* ignore the rest of the buffer, only one command at a time */
1846 *ppos += size;
1847 return size;
1848 }
1849
1850 static const struct file_operations kmemleak_fops = {
1851 .owner = THIS_MODULE,
1852 .open = kmemleak_open,
1853 .read = seq_read,
1854 .write = kmemleak_write,
1855 .llseek = seq_lseek,
1856 .release = seq_release,
1857 };
1858
1859 static void __kmemleak_do_cleanup(void)
1860 {
1861 struct kmemleak_object *object, *tmp;
1862
1863 /*
1864 * Kmemleak has already been disabled, no need for RCU list traversal
1865 * or kmemleak_lock held.
1866 */
1867 list_for_each_entry_safe(object, tmp, &object_list, object_list) {
1868 __remove_object(object);
1869 __delete_object(object);
1870 }
1871 }
1872
1873 /*
1874 * Stop the memory scanning thread and free the kmemleak internal objects if
1875 * no previous scan thread (otherwise, kmemleak may still have some useful
1876 * information on memory leaks).
1877 */
1878 static void kmemleak_do_cleanup(struct work_struct *work)
1879 {
1880 stop_scan_thread();
1881
1882 mutex_lock(&scan_mutex);
1883 /*
1884 * Once it is made sure that kmemleak_scan has stopped, it is safe to no
1885 * longer track object freeing. Ordering of the scan thread stopping and
1886 * the memory accesses below is guaranteed by the kthread_stop()
1887 * function.
1888 */
1889 kmemleak_free_enabled = 0;
1890 mutex_unlock(&scan_mutex);
1891
1892 if (!kmemleak_found_leaks)
1893 __kmemleak_do_cleanup();
1894 else
1895 pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
1896 }
1897
1898 static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
1899
1900 /*
1901 * Disable kmemleak. No memory allocation/freeing will be traced once this
1902 * function is called. Disabling kmemleak is an irreversible operation.
1903 */
1904 static void kmemleak_disable(void)
1905 {
1906 /* atomically check whether it was already invoked */
1907 if (cmpxchg(&kmemleak_error, 0, 1))
1908 return;
1909
1910 /* stop any memory operation tracing */
1911 kmemleak_enabled = 0;
1912
1913 /* check whether it is too early for a kernel thread */
1914 if (kmemleak_initialized)
1915 schedule_work(&cleanup_work);
1916 else
1917 kmemleak_free_enabled = 0;
1918
1919 pr_info("Kernel memory leak detector disabled\n");
1920 }
1921
1922 /*
1923 * Allow boot-time kmemleak disabling (enabled by default).
1924 */
1925 static int __init kmemleak_boot_config(char *str)
1926 {
1927 if (!str)
1928 return -EINVAL;
1929 if (strcmp(str, "off") == 0)
1930 kmemleak_disable();
1931 else if (strcmp(str, "on") == 0)
1932 kmemleak_skip_disable = 1;
1933 else
1934 return -EINVAL;
1935 return 0;
1936 }
1937 early_param("kmemleak", kmemleak_boot_config);
1938
1939 /*
1940 * Kmemleak initialization.
1941 */
1942 void __init kmemleak_init(void)
1943 {
1944 #ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
1945 if (!kmemleak_skip_disable) {
1946 kmemleak_disable();
1947 return;
1948 }
1949 #endif
1950
1951 if (kmemleak_error)
1952 return;
1953
1954 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1955 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1956
1957 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1958 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1959
1960 /* register the data/bss sections */
1961 create_object((unsigned long)_sdata, _edata - _sdata,
1962 KMEMLEAK_GREY, GFP_ATOMIC);
1963 create_object((unsigned long)__bss_start, __bss_stop - __bss_start,
1964 KMEMLEAK_GREY, GFP_ATOMIC);
1965 /* only register .data..ro_after_init if not within .data */
1966 if (&__start_ro_after_init < &_sdata || &__end_ro_after_init > &_edata)
1967 create_object((unsigned long)__start_ro_after_init,
1968 __end_ro_after_init - __start_ro_after_init,
1969 KMEMLEAK_GREY, GFP_ATOMIC);
1970 }
1971
1972 /*
1973 * Late initialization function.
1974 */
1975 static int __init kmemleak_late_init(void)
1976 {
1977 kmemleak_initialized = 1;
1978
1979 debugfs_create_file("kmemleak", 0644, NULL, NULL, &kmemleak_fops);
1980
1981 if (kmemleak_error) {
1982 /*
1983 * Some error occurred and kmemleak was disabled. There is a
1984 * small chance that kmemleak_disable() was called immediately
1985 * after setting kmemleak_initialized and we may end up with
1986 * two clean-up threads but serialized by scan_mutex.
1987 */
1988 schedule_work(&cleanup_work);
1989 return -ENOMEM;
1990 }
1991
1992 if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) {
1993 mutex_lock(&scan_mutex);
1994 start_scan_thread();
1995 mutex_unlock(&scan_mutex);
1996 }
1997
1998 pr_info("Kernel memory leak detector initialized (mem pool available: %d)\n",
1999 mem_pool_free_count);
2000
2001 return 0;
2002 }
2003 late_initcall(kmemleak_late_init);