Loading...
1/*
2 * mm/kmemleak.c
3 *
4 * Copyright (C) 2008 ARM Limited
5 * Written by Catalin Marinas <catalin.marinas@arm.com>
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 *
20 *
21 * For more information on the algorithm and kmemleak usage, please see
22 * Documentation/dev-tools/kmemleak.rst.
23 *
24 * Notes on locking
25 * ----------------
26 *
27 * The following locks and mutexes are used by kmemleak:
28 *
29 * - kmemleak_lock (rwlock): protects the object_list modifications and
30 * accesses to the object_tree_root. The object_list is the main list
31 * holding the metadata (struct kmemleak_object) for the allocated memory
32 * blocks. The object_tree_root is a red black tree used to look-up
33 * metadata based on a pointer to the corresponding memory block. The
34 * kmemleak_object structures are added to the object_list and
35 * object_tree_root in the create_object() function called from the
36 * kmemleak_alloc() callback and removed in delete_object() called from the
37 * kmemleak_free() callback
38 * - kmemleak_object.lock (spinlock): protects a kmemleak_object. Accesses to
39 * the metadata (e.g. count) are protected by this lock. Note that some
40 * members of this structure may be protected by other means (atomic or
41 * kmemleak_lock). This lock is also held when scanning the corresponding
42 * memory block to avoid the kernel freeing it via the kmemleak_free()
43 * callback. This is less heavyweight than holding a global lock like
44 * kmemleak_lock during scanning
45 * - scan_mutex (mutex): ensures that only one thread may scan the memory for
46 * unreferenced objects at a time. The gray_list contains the objects which
47 * are already referenced or marked as false positives and need to be
48 * scanned. This list is only modified during a scanning episode when the
49 * scan_mutex is held. At the end of a scan, the gray_list is always empty.
50 * Note that the kmemleak_object.use_count is incremented when an object is
51 * added to the gray_list and therefore cannot be freed. This mutex also
52 * prevents multiple users of the "kmemleak" debugfs file together with
53 * modifications to the memory scanning parameters including the scan_thread
54 * pointer
55 *
56 * Locks and mutexes are acquired/nested in the following order:
57 *
58 * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING)
59 *
60 * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex
61 * regions.
62 *
63 * The kmemleak_object structures have a use_count incremented or decremented
64 * using the get_object()/put_object() functions. When the use_count becomes
65 * 0, this count can no longer be incremented and put_object() schedules the
66 * kmemleak_object freeing via an RCU callback. All calls to the get_object()
67 * function must be protected by rcu_read_lock() to avoid accessing a freed
68 * structure.
69 */
70
71#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
72
73#include <linux/init.h>
74#include <linux/kernel.h>
75#include <linux/list.h>
76#include <linux/sched/signal.h>
77#include <linux/sched/task.h>
78#include <linux/sched/task_stack.h>
79#include <linux/jiffies.h>
80#include <linux/delay.h>
81#include <linux/export.h>
82#include <linux/kthread.h>
83#include <linux/rbtree.h>
84#include <linux/fs.h>
85#include <linux/debugfs.h>
86#include <linux/seq_file.h>
87#include <linux/cpumask.h>
88#include <linux/spinlock.h>
89#include <linux/mutex.h>
90#include <linux/rcupdate.h>
91#include <linux/stacktrace.h>
92#include <linux/cache.h>
93#include <linux/percpu.h>
94#include <linux/bootmem.h>
95#include <linux/pfn.h>
96#include <linux/mmzone.h>
97#include <linux/slab.h>
98#include <linux/thread_info.h>
99#include <linux/err.h>
100#include <linux/uaccess.h>
101#include <linux/string.h>
102#include <linux/nodemask.h>
103#include <linux/mm.h>
104#include <linux/workqueue.h>
105#include <linux/crc32.h>
106
107#include <asm/sections.h>
108#include <asm/processor.h>
109#include <linux/atomic.h>
110
111#include <linux/kasan.h>
112#include <linux/kmemleak.h>
113#include <linux/memory_hotplug.h>
114
115/*
116 * Kmemleak configuration and common defines.
117 */
118#define MAX_TRACE 16 /* stack trace length */
119#define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
120#define SECS_FIRST_SCAN 60 /* delay before the first scan */
121#define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
122#define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
123
124#define BYTES_PER_POINTER sizeof(void *)
125
126/* GFP bitmask for kmemleak internal allocations */
127#define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC)) | \
128 __GFP_NORETRY | __GFP_NOMEMALLOC | \
129 __GFP_NOWARN | __GFP_NOFAIL)
130
131/* scanning area inside a memory block */
132struct kmemleak_scan_area {
133 struct hlist_node node;
134 unsigned long start;
135 size_t size;
136};
137
138#define KMEMLEAK_GREY 0
139#define KMEMLEAK_BLACK -1
140
141/*
142 * Structure holding the metadata for each allocated memory block.
143 * Modifications to such objects should be made while holding the
144 * object->lock. Insertions or deletions from object_list, gray_list or
145 * rb_node are already protected by the corresponding locks or mutex (see
146 * the notes on locking above). These objects are reference-counted
147 * (use_count) and freed using the RCU mechanism.
148 */
149struct kmemleak_object {
150 spinlock_t lock;
151 unsigned int flags; /* object status flags */
152 struct list_head object_list;
153 struct list_head gray_list;
154 struct rb_node rb_node;
155 struct rcu_head rcu; /* object_list lockless traversal */
156 /* object usage count; object freed when use_count == 0 */
157 atomic_t use_count;
158 unsigned long pointer;
159 size_t size;
160 /* pass surplus references to this pointer */
161 unsigned long excess_ref;
162 /* minimum number of a pointers found before it is considered leak */
163 int min_count;
164 /* the total number of pointers found pointing to this object */
165 int count;
166 /* checksum for detecting modified objects */
167 u32 checksum;
168 /* memory ranges to be scanned inside an object (empty for all) */
169 struct hlist_head area_list;
170 unsigned long trace[MAX_TRACE];
171 unsigned int trace_len;
172 unsigned long jiffies; /* creation timestamp */
173 pid_t pid; /* pid of the current task */
174 char comm[TASK_COMM_LEN]; /* executable name */
175};
176
177/* flag representing the memory block allocation status */
178#define OBJECT_ALLOCATED (1 << 0)
179/* flag set after the first reporting of an unreference object */
180#define OBJECT_REPORTED (1 << 1)
181/* flag set to not scan the object */
182#define OBJECT_NO_SCAN (1 << 2)
183
184/* number of bytes to print per line; must be 16 or 32 */
185#define HEX_ROW_SIZE 16
186/* number of bytes to print at a time (1, 2, 4, 8) */
187#define HEX_GROUP_SIZE 1
188/* include ASCII after the hex output */
189#define HEX_ASCII 1
190/* max number of lines to be printed */
191#define HEX_MAX_LINES 2
192
193/* the list of all allocated objects */
194static LIST_HEAD(object_list);
195/* the list of gray-colored objects (see color_gray comment below) */
196static LIST_HEAD(gray_list);
197/* search tree for object boundaries */
198static struct rb_root object_tree_root = RB_ROOT;
199/* rw_lock protecting the access to object_list and object_tree_root */
200static DEFINE_RWLOCK(kmemleak_lock);
201
202/* allocation caches for kmemleak internal data */
203static struct kmem_cache *object_cache;
204static struct kmem_cache *scan_area_cache;
205
206/* set if tracing memory operations is enabled */
207static int kmemleak_enabled;
208/* same as above but only for the kmemleak_free() callback */
209static int kmemleak_free_enabled;
210/* set in the late_initcall if there were no errors */
211static int kmemleak_initialized;
212/* enables or disables early logging of the memory operations */
213static int kmemleak_early_log = 1;
214/* set if a kmemleak warning was issued */
215static int kmemleak_warning;
216/* set if a fatal kmemleak error has occurred */
217static int kmemleak_error;
218
219/* minimum and maximum address that may be valid pointers */
220static unsigned long min_addr = ULONG_MAX;
221static unsigned long max_addr;
222
223static struct task_struct *scan_thread;
224/* used to avoid reporting of recently allocated objects */
225static unsigned long jiffies_min_age;
226static unsigned long jiffies_last_scan;
227/* delay between automatic memory scannings */
228static signed long jiffies_scan_wait;
229/* enables or disables the task stacks scanning */
230static int kmemleak_stack_scan = 1;
231/* protects the memory scanning, parameters and debug/kmemleak file access */
232static DEFINE_MUTEX(scan_mutex);
233/* setting kmemleak=on, will set this var, skipping the disable */
234static int kmemleak_skip_disable;
235/* If there are leaks that can be reported */
236static bool kmemleak_found_leaks;
237
238/*
239 * Early object allocation/freeing logging. Kmemleak is initialized after the
240 * kernel allocator. However, both the kernel allocator and kmemleak may
241 * allocate memory blocks which need to be tracked. Kmemleak defines an
242 * arbitrary buffer to hold the allocation/freeing information before it is
243 * fully initialized.
244 */
245
246/* kmemleak operation type for early logging */
247enum {
248 KMEMLEAK_ALLOC,
249 KMEMLEAK_ALLOC_PERCPU,
250 KMEMLEAK_FREE,
251 KMEMLEAK_FREE_PART,
252 KMEMLEAK_FREE_PERCPU,
253 KMEMLEAK_NOT_LEAK,
254 KMEMLEAK_IGNORE,
255 KMEMLEAK_SCAN_AREA,
256 KMEMLEAK_NO_SCAN,
257 KMEMLEAK_SET_EXCESS_REF
258};
259
260/*
261 * Structure holding the information passed to kmemleak callbacks during the
262 * early logging.
263 */
264struct early_log {
265 int op_type; /* kmemleak operation type */
266 int min_count; /* minimum reference count */
267 const void *ptr; /* allocated/freed memory block */
268 union {
269 size_t size; /* memory block size */
270 unsigned long excess_ref; /* surplus reference passing */
271 };
272 unsigned long trace[MAX_TRACE]; /* stack trace */
273 unsigned int trace_len; /* stack trace length */
274};
275
276/* early logging buffer and current position */
277static struct early_log
278 early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata;
279static int crt_early_log __initdata;
280
281static void kmemleak_disable(void);
282
283/*
284 * Print a warning and dump the stack trace.
285 */
286#define kmemleak_warn(x...) do { \
287 pr_warn(x); \
288 dump_stack(); \
289 kmemleak_warning = 1; \
290} while (0)
291
292/*
293 * Macro invoked when a serious kmemleak condition occurred and cannot be
294 * recovered from. Kmemleak will be disabled and further allocation/freeing
295 * tracing no longer available.
296 */
297#define kmemleak_stop(x...) do { \
298 kmemleak_warn(x); \
299 kmemleak_disable(); \
300} while (0)
301
302/*
303 * Printing of the objects hex dump to the seq file. The number of lines to be
304 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
305 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
306 * with the object->lock held.
307 */
308static void hex_dump_object(struct seq_file *seq,
309 struct kmemleak_object *object)
310{
311 const u8 *ptr = (const u8 *)object->pointer;
312 size_t len;
313
314 /* limit the number of lines to HEX_MAX_LINES */
315 len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
316
317 seq_printf(seq, " hex dump (first %zu bytes):\n", len);
318 kasan_disable_current();
319 seq_hex_dump(seq, " ", DUMP_PREFIX_NONE, HEX_ROW_SIZE,
320 HEX_GROUP_SIZE, ptr, len, HEX_ASCII);
321 kasan_enable_current();
322}
323
324/*
325 * Object colors, encoded with count and min_count:
326 * - white - orphan object, not enough references to it (count < min_count)
327 * - gray - not orphan, not marked as false positive (min_count == 0) or
328 * sufficient references to it (count >= min_count)
329 * - black - ignore, it doesn't contain references (e.g. text section)
330 * (min_count == -1). No function defined for this color.
331 * Newly created objects don't have any color assigned (object->count == -1)
332 * before the next memory scan when they become white.
333 */
334static bool color_white(const struct kmemleak_object *object)
335{
336 return object->count != KMEMLEAK_BLACK &&
337 object->count < object->min_count;
338}
339
340static bool color_gray(const struct kmemleak_object *object)
341{
342 return object->min_count != KMEMLEAK_BLACK &&
343 object->count >= object->min_count;
344}
345
346/*
347 * Objects are considered unreferenced only if their color is white, they have
348 * not be deleted and have a minimum age to avoid false positives caused by
349 * pointers temporarily stored in CPU registers.
350 */
351static bool unreferenced_object(struct kmemleak_object *object)
352{
353 return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
354 time_before_eq(object->jiffies + jiffies_min_age,
355 jiffies_last_scan);
356}
357
358/*
359 * Printing of the unreferenced objects information to the seq file. The
360 * print_unreferenced function must be called with the object->lock held.
361 */
362static void print_unreferenced(struct seq_file *seq,
363 struct kmemleak_object *object)
364{
365 int i;
366 unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
367
368 seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
369 object->pointer, object->size);
370 seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
371 object->comm, object->pid, object->jiffies,
372 msecs_age / 1000, msecs_age % 1000);
373 hex_dump_object(seq, object);
374 seq_printf(seq, " backtrace:\n");
375
376 for (i = 0; i < object->trace_len; i++) {
377 void *ptr = (void *)object->trace[i];
378 seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
379 }
380}
381
382/*
383 * Print the kmemleak_object information. This function is used mainly for
384 * debugging special cases when kmemleak operations. It must be called with
385 * the object->lock held.
386 */
387static void dump_object_info(struct kmemleak_object *object)
388{
389 struct stack_trace trace;
390
391 trace.nr_entries = object->trace_len;
392 trace.entries = object->trace;
393
394 pr_notice("Object 0x%08lx (size %zu):\n",
395 object->pointer, object->size);
396 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
397 object->comm, object->pid, object->jiffies);
398 pr_notice(" min_count = %d\n", object->min_count);
399 pr_notice(" count = %d\n", object->count);
400 pr_notice(" flags = 0x%x\n", object->flags);
401 pr_notice(" checksum = %u\n", object->checksum);
402 pr_notice(" backtrace:\n");
403 print_stack_trace(&trace, 4);
404}
405
406/*
407 * Look-up a memory block metadata (kmemleak_object) in the object search
408 * tree based on a pointer value. If alias is 0, only values pointing to the
409 * beginning of the memory block are allowed. The kmemleak_lock must be held
410 * when calling this function.
411 */
412static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
413{
414 struct rb_node *rb = object_tree_root.rb_node;
415
416 while (rb) {
417 struct kmemleak_object *object =
418 rb_entry(rb, struct kmemleak_object, rb_node);
419 if (ptr < object->pointer)
420 rb = object->rb_node.rb_left;
421 else if (object->pointer + object->size <= ptr)
422 rb = object->rb_node.rb_right;
423 else if (object->pointer == ptr || alias)
424 return object;
425 else {
426 kmemleak_warn("Found object by alias at 0x%08lx\n",
427 ptr);
428 dump_object_info(object);
429 break;
430 }
431 }
432 return NULL;
433}
434
435/*
436 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
437 * that once an object's use_count reached 0, the RCU freeing was already
438 * registered and the object should no longer be used. This function must be
439 * called under the protection of rcu_read_lock().
440 */
441static int get_object(struct kmemleak_object *object)
442{
443 return atomic_inc_not_zero(&object->use_count);
444}
445
446/*
447 * RCU callback to free a kmemleak_object.
448 */
449static void free_object_rcu(struct rcu_head *rcu)
450{
451 struct hlist_node *tmp;
452 struct kmemleak_scan_area *area;
453 struct kmemleak_object *object =
454 container_of(rcu, struct kmemleak_object, rcu);
455
456 /*
457 * Once use_count is 0 (guaranteed by put_object), there is no other
458 * code accessing this object, hence no need for locking.
459 */
460 hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
461 hlist_del(&area->node);
462 kmem_cache_free(scan_area_cache, area);
463 }
464 kmem_cache_free(object_cache, object);
465}
466
467/*
468 * Decrement the object use_count. Once the count is 0, free the object using
469 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
470 * delete_object() path, the delayed RCU freeing ensures that there is no
471 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
472 * is also possible.
473 */
474static void put_object(struct kmemleak_object *object)
475{
476 if (!atomic_dec_and_test(&object->use_count))
477 return;
478
479 /* should only get here after delete_object was called */
480 WARN_ON(object->flags & OBJECT_ALLOCATED);
481
482 call_rcu(&object->rcu, free_object_rcu);
483}
484
485/*
486 * Look up an object in the object search tree and increase its use_count.
487 */
488static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
489{
490 unsigned long flags;
491 struct kmemleak_object *object;
492
493 rcu_read_lock();
494 read_lock_irqsave(&kmemleak_lock, flags);
495 object = lookup_object(ptr, alias);
496 read_unlock_irqrestore(&kmemleak_lock, flags);
497
498 /* check whether the object is still available */
499 if (object && !get_object(object))
500 object = NULL;
501 rcu_read_unlock();
502
503 return object;
504}
505
506/*
507 * Look up an object in the object search tree and remove it from both
508 * object_tree_root and object_list. The returned object's use_count should be
509 * at least 1, as initially set by create_object().
510 */
511static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias)
512{
513 unsigned long flags;
514 struct kmemleak_object *object;
515
516 write_lock_irqsave(&kmemleak_lock, flags);
517 object = lookup_object(ptr, alias);
518 if (object) {
519 rb_erase(&object->rb_node, &object_tree_root);
520 list_del_rcu(&object->object_list);
521 }
522 write_unlock_irqrestore(&kmemleak_lock, flags);
523
524 return object;
525}
526
527/*
528 * Save stack trace to the given array of MAX_TRACE size.
529 */
530static int __save_stack_trace(unsigned long *trace)
531{
532 struct stack_trace stack_trace;
533
534 stack_trace.max_entries = MAX_TRACE;
535 stack_trace.nr_entries = 0;
536 stack_trace.entries = trace;
537 stack_trace.skip = 2;
538 save_stack_trace(&stack_trace);
539
540 return stack_trace.nr_entries;
541}
542
543/*
544 * Create the metadata (struct kmemleak_object) corresponding to an allocated
545 * memory block and add it to the object_list and object_tree_root.
546 */
547static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
548 int min_count, gfp_t gfp)
549{
550 unsigned long flags;
551 struct kmemleak_object *object, *parent;
552 struct rb_node **link, *rb_parent;
553
554 object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
555 if (!object) {
556 pr_warn("Cannot allocate a kmemleak_object structure\n");
557 kmemleak_disable();
558 return NULL;
559 }
560
561 INIT_LIST_HEAD(&object->object_list);
562 INIT_LIST_HEAD(&object->gray_list);
563 INIT_HLIST_HEAD(&object->area_list);
564 spin_lock_init(&object->lock);
565 atomic_set(&object->use_count, 1);
566 object->flags = OBJECT_ALLOCATED;
567 object->pointer = ptr;
568 object->size = size;
569 object->excess_ref = 0;
570 object->min_count = min_count;
571 object->count = 0; /* white color initially */
572 object->jiffies = jiffies;
573 object->checksum = 0;
574
575 /* task information */
576 if (in_irq()) {
577 object->pid = 0;
578 strncpy(object->comm, "hardirq", sizeof(object->comm));
579 } else if (in_softirq()) {
580 object->pid = 0;
581 strncpy(object->comm, "softirq", sizeof(object->comm));
582 } else {
583 object->pid = current->pid;
584 /*
585 * There is a small chance of a race with set_task_comm(),
586 * however using get_task_comm() here may cause locking
587 * dependency issues with current->alloc_lock. In the worst
588 * case, the command line is not correct.
589 */
590 strncpy(object->comm, current->comm, sizeof(object->comm));
591 }
592
593 /* kernel backtrace */
594 object->trace_len = __save_stack_trace(object->trace);
595
596 write_lock_irqsave(&kmemleak_lock, flags);
597
598 min_addr = min(min_addr, ptr);
599 max_addr = max(max_addr, ptr + size);
600 link = &object_tree_root.rb_node;
601 rb_parent = NULL;
602 while (*link) {
603 rb_parent = *link;
604 parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
605 if (ptr + size <= parent->pointer)
606 link = &parent->rb_node.rb_left;
607 else if (parent->pointer + parent->size <= ptr)
608 link = &parent->rb_node.rb_right;
609 else {
610 kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
611 ptr);
612 /*
613 * No need for parent->lock here since "parent" cannot
614 * be freed while the kmemleak_lock is held.
615 */
616 dump_object_info(parent);
617 kmem_cache_free(object_cache, object);
618 object = NULL;
619 goto out;
620 }
621 }
622 rb_link_node(&object->rb_node, rb_parent, link);
623 rb_insert_color(&object->rb_node, &object_tree_root);
624
625 list_add_tail_rcu(&object->object_list, &object_list);
626out:
627 write_unlock_irqrestore(&kmemleak_lock, flags);
628 return object;
629}
630
631/*
632 * Mark the object as not allocated and schedule RCU freeing via put_object().
633 */
634static void __delete_object(struct kmemleak_object *object)
635{
636 unsigned long flags;
637
638 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
639 WARN_ON(atomic_read(&object->use_count) < 1);
640
641 /*
642 * Locking here also ensures that the corresponding memory block
643 * cannot be freed when it is being scanned.
644 */
645 spin_lock_irqsave(&object->lock, flags);
646 object->flags &= ~OBJECT_ALLOCATED;
647 spin_unlock_irqrestore(&object->lock, flags);
648 put_object(object);
649}
650
651/*
652 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
653 * delete it.
654 */
655static void delete_object_full(unsigned long ptr)
656{
657 struct kmemleak_object *object;
658
659 object = find_and_remove_object(ptr, 0);
660 if (!object) {
661#ifdef DEBUG
662 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
663 ptr);
664#endif
665 return;
666 }
667 __delete_object(object);
668}
669
670/*
671 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
672 * delete it. If the memory block is partially freed, the function may create
673 * additional metadata for the remaining parts of the block.
674 */
675static void delete_object_part(unsigned long ptr, size_t size)
676{
677 struct kmemleak_object *object;
678 unsigned long start, end;
679
680 object = find_and_remove_object(ptr, 1);
681 if (!object) {
682#ifdef DEBUG
683 kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
684 ptr, size);
685#endif
686 return;
687 }
688
689 /*
690 * Create one or two objects that may result from the memory block
691 * split. Note that partial freeing is only done by free_bootmem() and
692 * this happens before kmemleak_init() is called. The path below is
693 * only executed during early log recording in kmemleak_init(), so
694 * GFP_KERNEL is enough.
695 */
696 start = object->pointer;
697 end = object->pointer + object->size;
698 if (ptr > start)
699 create_object(start, ptr - start, object->min_count,
700 GFP_KERNEL);
701 if (ptr + size < end)
702 create_object(ptr + size, end - ptr - size, object->min_count,
703 GFP_KERNEL);
704
705 __delete_object(object);
706}
707
708static void __paint_it(struct kmemleak_object *object, int color)
709{
710 object->min_count = color;
711 if (color == KMEMLEAK_BLACK)
712 object->flags |= OBJECT_NO_SCAN;
713}
714
715static void paint_it(struct kmemleak_object *object, int color)
716{
717 unsigned long flags;
718
719 spin_lock_irqsave(&object->lock, flags);
720 __paint_it(object, color);
721 spin_unlock_irqrestore(&object->lock, flags);
722}
723
724static void paint_ptr(unsigned long ptr, int color)
725{
726 struct kmemleak_object *object;
727
728 object = find_and_get_object(ptr, 0);
729 if (!object) {
730 kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n",
731 ptr,
732 (color == KMEMLEAK_GREY) ? "Grey" :
733 (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
734 return;
735 }
736 paint_it(object, color);
737 put_object(object);
738}
739
740/*
741 * Mark an object permanently as gray-colored so that it can no longer be
742 * reported as a leak. This is used in general to mark a false positive.
743 */
744static void make_gray_object(unsigned long ptr)
745{
746 paint_ptr(ptr, KMEMLEAK_GREY);
747}
748
749/*
750 * Mark the object as black-colored so that it is ignored from scans and
751 * reporting.
752 */
753static void make_black_object(unsigned long ptr)
754{
755 paint_ptr(ptr, KMEMLEAK_BLACK);
756}
757
758/*
759 * Add a scanning area to the object. If at least one such area is added,
760 * kmemleak will only scan these ranges rather than the whole memory block.
761 */
762static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
763{
764 unsigned long flags;
765 struct kmemleak_object *object;
766 struct kmemleak_scan_area *area;
767
768 object = find_and_get_object(ptr, 1);
769 if (!object) {
770 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
771 ptr);
772 return;
773 }
774
775 area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
776 if (!area) {
777 pr_warn("Cannot allocate a scan area\n");
778 goto out;
779 }
780
781 spin_lock_irqsave(&object->lock, flags);
782 if (size == SIZE_MAX) {
783 size = object->pointer + object->size - ptr;
784 } else if (ptr + size > object->pointer + object->size) {
785 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
786 dump_object_info(object);
787 kmem_cache_free(scan_area_cache, area);
788 goto out_unlock;
789 }
790
791 INIT_HLIST_NODE(&area->node);
792 area->start = ptr;
793 area->size = size;
794
795 hlist_add_head(&area->node, &object->area_list);
796out_unlock:
797 spin_unlock_irqrestore(&object->lock, flags);
798out:
799 put_object(object);
800}
801
802/*
803 * Any surplus references (object already gray) to 'ptr' are passed to
804 * 'excess_ref'. This is used in the vmalloc() case where a pointer to
805 * vm_struct may be used as an alternative reference to the vmalloc'ed object
806 * (see free_thread_stack()).
807 */
808static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref)
809{
810 unsigned long flags;
811 struct kmemleak_object *object;
812
813 object = find_and_get_object(ptr, 0);
814 if (!object) {
815 kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n",
816 ptr);
817 return;
818 }
819
820 spin_lock_irqsave(&object->lock, flags);
821 object->excess_ref = excess_ref;
822 spin_unlock_irqrestore(&object->lock, flags);
823 put_object(object);
824}
825
826/*
827 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
828 * pointer. Such object will not be scanned by kmemleak but references to it
829 * are searched.
830 */
831static void object_no_scan(unsigned long ptr)
832{
833 unsigned long flags;
834 struct kmemleak_object *object;
835
836 object = find_and_get_object(ptr, 0);
837 if (!object) {
838 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
839 return;
840 }
841
842 spin_lock_irqsave(&object->lock, flags);
843 object->flags |= OBJECT_NO_SCAN;
844 spin_unlock_irqrestore(&object->lock, flags);
845 put_object(object);
846}
847
848/*
849 * Log an early kmemleak_* call to the early_log buffer. These calls will be
850 * processed later once kmemleak is fully initialized.
851 */
852static void __init log_early(int op_type, const void *ptr, size_t size,
853 int min_count)
854{
855 unsigned long flags;
856 struct early_log *log;
857
858 if (kmemleak_error) {
859 /* kmemleak stopped recording, just count the requests */
860 crt_early_log++;
861 return;
862 }
863
864 if (crt_early_log >= ARRAY_SIZE(early_log)) {
865 crt_early_log++;
866 kmemleak_disable();
867 return;
868 }
869
870 /*
871 * There is no need for locking since the kernel is still in UP mode
872 * at this stage. Disabling the IRQs is enough.
873 */
874 local_irq_save(flags);
875 log = &early_log[crt_early_log];
876 log->op_type = op_type;
877 log->ptr = ptr;
878 log->size = size;
879 log->min_count = min_count;
880 log->trace_len = __save_stack_trace(log->trace);
881 crt_early_log++;
882 local_irq_restore(flags);
883}
884
885/*
886 * Log an early allocated block and populate the stack trace.
887 */
888static void early_alloc(struct early_log *log)
889{
890 struct kmemleak_object *object;
891 unsigned long flags;
892 int i;
893
894 if (!kmemleak_enabled || !log->ptr || IS_ERR(log->ptr))
895 return;
896
897 /*
898 * RCU locking needed to ensure object is not freed via put_object().
899 */
900 rcu_read_lock();
901 object = create_object((unsigned long)log->ptr, log->size,
902 log->min_count, GFP_ATOMIC);
903 if (!object)
904 goto out;
905 spin_lock_irqsave(&object->lock, flags);
906 for (i = 0; i < log->trace_len; i++)
907 object->trace[i] = log->trace[i];
908 object->trace_len = log->trace_len;
909 spin_unlock_irqrestore(&object->lock, flags);
910out:
911 rcu_read_unlock();
912}
913
914/*
915 * Log an early allocated block and populate the stack trace.
916 */
917static void early_alloc_percpu(struct early_log *log)
918{
919 unsigned int cpu;
920 const void __percpu *ptr = log->ptr;
921
922 for_each_possible_cpu(cpu) {
923 log->ptr = per_cpu_ptr(ptr, cpu);
924 early_alloc(log);
925 }
926}
927
928/**
929 * kmemleak_alloc - register a newly allocated object
930 * @ptr: pointer to beginning of the object
931 * @size: size of the object
932 * @min_count: minimum number of references to this object. If during memory
933 * scanning a number of references less than @min_count is found,
934 * the object is reported as a memory leak. If @min_count is 0,
935 * the object is never reported as a leak. If @min_count is -1,
936 * the object is ignored (not scanned and not reported as a leak)
937 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
938 *
939 * This function is called from the kernel allocators when a new object
940 * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.).
941 */
942void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
943 gfp_t gfp)
944{
945 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
946
947 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
948 create_object((unsigned long)ptr, size, min_count, gfp);
949 else if (kmemleak_early_log)
950 log_early(KMEMLEAK_ALLOC, ptr, size, min_count);
951}
952EXPORT_SYMBOL_GPL(kmemleak_alloc);
953
954/**
955 * kmemleak_alloc_percpu - register a newly allocated __percpu object
956 * @ptr: __percpu pointer to beginning of the object
957 * @size: size of the object
958 * @gfp: flags used for kmemleak internal memory allocations
959 *
960 * This function is called from the kernel percpu allocator when a new object
961 * (memory block) is allocated (alloc_percpu).
962 */
963void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
964 gfp_t gfp)
965{
966 unsigned int cpu;
967
968 pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size);
969
970 /*
971 * Percpu allocations are only scanned and not reported as leaks
972 * (min_count is set to 0).
973 */
974 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
975 for_each_possible_cpu(cpu)
976 create_object((unsigned long)per_cpu_ptr(ptr, cpu),
977 size, 0, gfp);
978 else if (kmemleak_early_log)
979 log_early(KMEMLEAK_ALLOC_PERCPU, ptr, size, 0);
980}
981EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
982
983/**
984 * kmemleak_vmalloc - register a newly vmalloc'ed object
985 * @area: pointer to vm_struct
986 * @size: size of the object
987 * @gfp: __vmalloc() flags used for kmemleak internal memory allocations
988 *
989 * This function is called from the vmalloc() kernel allocator when a new
990 * object (memory block) is allocated.
991 */
992void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp)
993{
994 pr_debug("%s(0x%p, %zu)\n", __func__, area, size);
995
996 /*
997 * A min_count = 2 is needed because vm_struct contains a reference to
998 * the virtual address of the vmalloc'ed block.
999 */
1000 if (kmemleak_enabled) {
1001 create_object((unsigned long)area->addr, size, 2, gfp);
1002 object_set_excess_ref((unsigned long)area,
1003 (unsigned long)area->addr);
1004 } else if (kmemleak_early_log) {
1005 log_early(KMEMLEAK_ALLOC, area->addr, size, 2);
1006 /* reusing early_log.size for storing area->addr */
1007 log_early(KMEMLEAK_SET_EXCESS_REF,
1008 area, (unsigned long)area->addr, 0);
1009 }
1010}
1011EXPORT_SYMBOL_GPL(kmemleak_vmalloc);
1012
1013/**
1014 * kmemleak_free - unregister a previously registered object
1015 * @ptr: pointer to beginning of the object
1016 *
1017 * This function is called from the kernel allocators when an object (memory
1018 * block) is freed (kmem_cache_free, kfree, vfree etc.).
1019 */
1020void __ref kmemleak_free(const void *ptr)
1021{
1022 pr_debug("%s(0x%p)\n", __func__, ptr);
1023
1024 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
1025 delete_object_full((unsigned long)ptr);
1026 else if (kmemleak_early_log)
1027 log_early(KMEMLEAK_FREE, ptr, 0, 0);
1028}
1029EXPORT_SYMBOL_GPL(kmemleak_free);
1030
1031/**
1032 * kmemleak_free_part - partially unregister a previously registered object
1033 * @ptr: pointer to the beginning or inside the object. This also
1034 * represents the start of the range to be freed
1035 * @size: size to be unregistered
1036 *
1037 * This function is called when only a part of a memory block is freed
1038 * (usually from the bootmem allocator).
1039 */
1040void __ref kmemleak_free_part(const void *ptr, size_t size)
1041{
1042 pr_debug("%s(0x%p)\n", __func__, ptr);
1043
1044 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1045 delete_object_part((unsigned long)ptr, size);
1046 else if (kmemleak_early_log)
1047 log_early(KMEMLEAK_FREE_PART, ptr, size, 0);
1048}
1049EXPORT_SYMBOL_GPL(kmemleak_free_part);
1050
1051/**
1052 * kmemleak_free_percpu - unregister a previously registered __percpu object
1053 * @ptr: __percpu pointer to beginning of the object
1054 *
1055 * This function is called from the kernel percpu allocator when an object
1056 * (memory block) is freed (free_percpu).
1057 */
1058void __ref kmemleak_free_percpu(const void __percpu *ptr)
1059{
1060 unsigned int cpu;
1061
1062 pr_debug("%s(0x%p)\n", __func__, ptr);
1063
1064 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
1065 for_each_possible_cpu(cpu)
1066 delete_object_full((unsigned long)per_cpu_ptr(ptr,
1067 cpu));
1068 else if (kmemleak_early_log)
1069 log_early(KMEMLEAK_FREE_PERCPU, ptr, 0, 0);
1070}
1071EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
1072
1073/**
1074 * kmemleak_update_trace - update object allocation stack trace
1075 * @ptr: pointer to beginning of the object
1076 *
1077 * Override the object allocation stack trace for cases where the actual
1078 * allocation place is not always useful.
1079 */
1080void __ref kmemleak_update_trace(const void *ptr)
1081{
1082 struct kmemleak_object *object;
1083 unsigned long flags;
1084
1085 pr_debug("%s(0x%p)\n", __func__, ptr);
1086
1087 if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
1088 return;
1089
1090 object = find_and_get_object((unsigned long)ptr, 1);
1091 if (!object) {
1092#ifdef DEBUG
1093 kmemleak_warn("Updating stack trace for unknown object at %p\n",
1094 ptr);
1095#endif
1096 return;
1097 }
1098
1099 spin_lock_irqsave(&object->lock, flags);
1100 object->trace_len = __save_stack_trace(object->trace);
1101 spin_unlock_irqrestore(&object->lock, flags);
1102
1103 put_object(object);
1104}
1105EXPORT_SYMBOL(kmemleak_update_trace);
1106
1107/**
1108 * kmemleak_not_leak - mark an allocated object as false positive
1109 * @ptr: pointer to beginning of the object
1110 *
1111 * Calling this function on an object will cause the memory block to no longer
1112 * be reported as leak and always be scanned.
1113 */
1114void __ref kmemleak_not_leak(const void *ptr)
1115{
1116 pr_debug("%s(0x%p)\n", __func__, ptr);
1117
1118 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1119 make_gray_object((unsigned long)ptr);
1120 else if (kmemleak_early_log)
1121 log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0);
1122}
1123EXPORT_SYMBOL(kmemleak_not_leak);
1124
1125/**
1126 * kmemleak_ignore - ignore an allocated object
1127 * @ptr: pointer to beginning of the object
1128 *
1129 * Calling this function on an object will cause the memory block to be
1130 * ignored (not scanned and not reported as a leak). This is usually done when
1131 * it is known that the corresponding block is not a leak and does not contain
1132 * any references to other allocated memory blocks.
1133 */
1134void __ref kmemleak_ignore(const void *ptr)
1135{
1136 pr_debug("%s(0x%p)\n", __func__, ptr);
1137
1138 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1139 make_black_object((unsigned long)ptr);
1140 else if (kmemleak_early_log)
1141 log_early(KMEMLEAK_IGNORE, ptr, 0, 0);
1142}
1143EXPORT_SYMBOL(kmemleak_ignore);
1144
1145/**
1146 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1147 * @ptr: pointer to beginning or inside the object. This also
1148 * represents the start of the scan area
1149 * @size: size of the scan area
1150 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1151 *
1152 * This function is used when it is known that only certain parts of an object
1153 * contain references to other objects. Kmemleak will only scan these areas
1154 * reducing the number false negatives.
1155 */
1156void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1157{
1158 pr_debug("%s(0x%p)\n", __func__, ptr);
1159
1160 if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
1161 add_scan_area((unsigned long)ptr, size, gfp);
1162 else if (kmemleak_early_log)
1163 log_early(KMEMLEAK_SCAN_AREA, ptr, size, 0);
1164}
1165EXPORT_SYMBOL(kmemleak_scan_area);
1166
1167/**
1168 * kmemleak_no_scan - do not scan an allocated object
1169 * @ptr: pointer to beginning of the object
1170 *
1171 * This function notifies kmemleak not to scan the given memory block. Useful
1172 * in situations where it is known that the given object does not contain any
1173 * references to other objects. Kmemleak will not scan such objects reducing
1174 * the number of false negatives.
1175 */
1176void __ref kmemleak_no_scan(const void *ptr)
1177{
1178 pr_debug("%s(0x%p)\n", __func__, ptr);
1179
1180 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1181 object_no_scan((unsigned long)ptr);
1182 else if (kmemleak_early_log)
1183 log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0);
1184}
1185EXPORT_SYMBOL(kmemleak_no_scan);
1186
1187/**
1188 * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical
1189 * address argument
1190 * @phys: physical address of the object
1191 * @size: size of the object
1192 * @min_count: minimum number of references to this object.
1193 * See kmemleak_alloc()
1194 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1195 */
1196void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, int min_count,
1197 gfp_t gfp)
1198{
1199 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1200 kmemleak_alloc(__va(phys), size, min_count, gfp);
1201}
1202EXPORT_SYMBOL(kmemleak_alloc_phys);
1203
1204/**
1205 * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a
1206 * physical address argument
1207 * @phys: physical address if the beginning or inside an object. This
1208 * also represents the start of the range to be freed
1209 * @size: size to be unregistered
1210 */
1211void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size)
1212{
1213 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1214 kmemleak_free_part(__va(phys), size);
1215}
1216EXPORT_SYMBOL(kmemleak_free_part_phys);
1217
1218/**
1219 * kmemleak_not_leak_phys - similar to kmemleak_not_leak but taking a physical
1220 * address argument
1221 * @phys: physical address of the object
1222 */
1223void __ref kmemleak_not_leak_phys(phys_addr_t phys)
1224{
1225 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1226 kmemleak_not_leak(__va(phys));
1227}
1228EXPORT_SYMBOL(kmemleak_not_leak_phys);
1229
1230/**
1231 * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical
1232 * address argument
1233 * @phys: physical address of the object
1234 */
1235void __ref kmemleak_ignore_phys(phys_addr_t phys)
1236{
1237 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1238 kmemleak_ignore(__va(phys));
1239}
1240EXPORT_SYMBOL(kmemleak_ignore_phys);
1241
1242/*
1243 * Update an object's checksum and return true if it was modified.
1244 */
1245static bool update_checksum(struct kmemleak_object *object)
1246{
1247 u32 old_csum = object->checksum;
1248
1249 kasan_disable_current();
1250 object->checksum = crc32(0, (void *)object->pointer, object->size);
1251 kasan_enable_current();
1252
1253 return object->checksum != old_csum;
1254}
1255
1256/*
1257 * Update an object's references. object->lock must be held by the caller.
1258 */
1259static void update_refs(struct kmemleak_object *object)
1260{
1261 if (!color_white(object)) {
1262 /* non-orphan, ignored or new */
1263 return;
1264 }
1265
1266 /*
1267 * Increase the object's reference count (number of pointers to the
1268 * memory block). If this count reaches the required minimum, the
1269 * object's color will become gray and it will be added to the
1270 * gray_list.
1271 */
1272 object->count++;
1273 if (color_gray(object)) {
1274 /* put_object() called when removing from gray_list */
1275 WARN_ON(!get_object(object));
1276 list_add_tail(&object->gray_list, &gray_list);
1277 }
1278}
1279
1280/*
1281 * Memory scanning is a long process and it needs to be interruptable. This
1282 * function checks whether such interrupt condition occurred.
1283 */
1284static int scan_should_stop(void)
1285{
1286 if (!kmemleak_enabled)
1287 return 1;
1288
1289 /*
1290 * This function may be called from either process or kthread context,
1291 * hence the need to check for both stop conditions.
1292 */
1293 if (current->mm)
1294 return signal_pending(current);
1295 else
1296 return kthread_should_stop();
1297
1298 return 0;
1299}
1300
1301/*
1302 * Scan a memory block (exclusive range) for valid pointers and add those
1303 * found to the gray list.
1304 */
1305static void scan_block(void *_start, void *_end,
1306 struct kmemleak_object *scanned)
1307{
1308 unsigned long *ptr;
1309 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1310 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1311 unsigned long flags;
1312
1313 read_lock_irqsave(&kmemleak_lock, flags);
1314 for (ptr = start; ptr < end; ptr++) {
1315 struct kmemleak_object *object;
1316 unsigned long pointer;
1317 unsigned long excess_ref;
1318
1319 if (scan_should_stop())
1320 break;
1321
1322 kasan_disable_current();
1323 pointer = *ptr;
1324 kasan_enable_current();
1325
1326 if (pointer < min_addr || pointer >= max_addr)
1327 continue;
1328
1329 /*
1330 * No need for get_object() here since we hold kmemleak_lock.
1331 * object->use_count cannot be dropped to 0 while the object
1332 * is still present in object_tree_root and object_list
1333 * (with updates protected by kmemleak_lock).
1334 */
1335 object = lookup_object(pointer, 1);
1336 if (!object)
1337 continue;
1338 if (object == scanned)
1339 /* self referenced, ignore */
1340 continue;
1341
1342 /*
1343 * Avoid the lockdep recursive warning on object->lock being
1344 * previously acquired in scan_object(). These locks are
1345 * enclosed by scan_mutex.
1346 */
1347 spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1348 /* only pass surplus references (object already gray) */
1349 if (color_gray(object)) {
1350 excess_ref = object->excess_ref;
1351 /* no need for update_refs() if object already gray */
1352 } else {
1353 excess_ref = 0;
1354 update_refs(object);
1355 }
1356 spin_unlock(&object->lock);
1357
1358 if (excess_ref) {
1359 object = lookup_object(excess_ref, 0);
1360 if (!object)
1361 continue;
1362 if (object == scanned)
1363 /* circular reference, ignore */
1364 continue;
1365 spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1366 update_refs(object);
1367 spin_unlock(&object->lock);
1368 }
1369 }
1370 read_unlock_irqrestore(&kmemleak_lock, flags);
1371}
1372
1373/*
1374 * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
1375 */
1376static void scan_large_block(void *start, void *end)
1377{
1378 void *next;
1379
1380 while (start < end) {
1381 next = min(start + MAX_SCAN_SIZE, end);
1382 scan_block(start, next, NULL);
1383 start = next;
1384 cond_resched();
1385 }
1386}
1387
1388/*
1389 * Scan a memory block corresponding to a kmemleak_object. A condition is
1390 * that object->use_count >= 1.
1391 */
1392static void scan_object(struct kmemleak_object *object)
1393{
1394 struct kmemleak_scan_area *area;
1395 unsigned long flags;
1396
1397 /*
1398 * Once the object->lock is acquired, the corresponding memory block
1399 * cannot be freed (the same lock is acquired in delete_object).
1400 */
1401 spin_lock_irqsave(&object->lock, flags);
1402 if (object->flags & OBJECT_NO_SCAN)
1403 goto out;
1404 if (!(object->flags & OBJECT_ALLOCATED))
1405 /* already freed object */
1406 goto out;
1407 if (hlist_empty(&object->area_list)) {
1408 void *start = (void *)object->pointer;
1409 void *end = (void *)(object->pointer + object->size);
1410 void *next;
1411
1412 do {
1413 next = min(start + MAX_SCAN_SIZE, end);
1414 scan_block(start, next, object);
1415
1416 start = next;
1417 if (start >= end)
1418 break;
1419
1420 spin_unlock_irqrestore(&object->lock, flags);
1421 cond_resched();
1422 spin_lock_irqsave(&object->lock, flags);
1423 } while (object->flags & OBJECT_ALLOCATED);
1424 } else
1425 hlist_for_each_entry(area, &object->area_list, node)
1426 scan_block((void *)area->start,
1427 (void *)(area->start + area->size),
1428 object);
1429out:
1430 spin_unlock_irqrestore(&object->lock, flags);
1431}
1432
1433/*
1434 * Scan the objects already referenced (gray objects). More objects will be
1435 * referenced and, if there are no memory leaks, all the objects are scanned.
1436 */
1437static void scan_gray_list(void)
1438{
1439 struct kmemleak_object *object, *tmp;
1440
1441 /*
1442 * The list traversal is safe for both tail additions and removals
1443 * from inside the loop. The kmemleak objects cannot be freed from
1444 * outside the loop because their use_count was incremented.
1445 */
1446 object = list_entry(gray_list.next, typeof(*object), gray_list);
1447 while (&object->gray_list != &gray_list) {
1448 cond_resched();
1449
1450 /* may add new objects to the list */
1451 if (!scan_should_stop())
1452 scan_object(object);
1453
1454 tmp = list_entry(object->gray_list.next, typeof(*object),
1455 gray_list);
1456
1457 /* remove the object from the list and release it */
1458 list_del(&object->gray_list);
1459 put_object(object);
1460
1461 object = tmp;
1462 }
1463 WARN_ON(!list_empty(&gray_list));
1464}
1465
1466/*
1467 * Scan data sections and all the referenced memory blocks allocated via the
1468 * kernel's standard allocators. This function must be called with the
1469 * scan_mutex held.
1470 */
1471static void kmemleak_scan(void)
1472{
1473 unsigned long flags;
1474 struct kmemleak_object *object;
1475 int i;
1476 int new_leaks = 0;
1477
1478 jiffies_last_scan = jiffies;
1479
1480 /* prepare the kmemleak_object's */
1481 rcu_read_lock();
1482 list_for_each_entry_rcu(object, &object_list, object_list) {
1483 spin_lock_irqsave(&object->lock, flags);
1484#ifdef DEBUG
1485 /*
1486 * With a few exceptions there should be a maximum of
1487 * 1 reference to any object at this point.
1488 */
1489 if (atomic_read(&object->use_count) > 1) {
1490 pr_debug("object->use_count = %d\n",
1491 atomic_read(&object->use_count));
1492 dump_object_info(object);
1493 }
1494#endif
1495 /* reset the reference count (whiten the object) */
1496 object->count = 0;
1497 if (color_gray(object) && get_object(object))
1498 list_add_tail(&object->gray_list, &gray_list);
1499
1500 spin_unlock_irqrestore(&object->lock, flags);
1501 }
1502 rcu_read_unlock();
1503
1504 /* data/bss scanning */
1505 scan_large_block(_sdata, _edata);
1506 scan_large_block(__bss_start, __bss_stop);
1507 scan_large_block(__start_ro_after_init, __end_ro_after_init);
1508
1509#ifdef CONFIG_SMP
1510 /* per-cpu sections scanning */
1511 for_each_possible_cpu(i)
1512 scan_large_block(__per_cpu_start + per_cpu_offset(i),
1513 __per_cpu_end + per_cpu_offset(i));
1514#endif
1515
1516 /*
1517 * Struct page scanning for each node.
1518 */
1519 get_online_mems();
1520 for_each_online_node(i) {
1521 unsigned long start_pfn = node_start_pfn(i);
1522 unsigned long end_pfn = node_end_pfn(i);
1523 unsigned long pfn;
1524
1525 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1526 struct page *page;
1527
1528 if (!pfn_valid(pfn))
1529 continue;
1530 page = pfn_to_page(pfn);
1531 /* only scan if page is in use */
1532 if (page_count(page) == 0)
1533 continue;
1534 scan_block(page, page + 1, NULL);
1535 if (!(pfn & 63))
1536 cond_resched();
1537 }
1538 }
1539 put_online_mems();
1540
1541 /*
1542 * Scanning the task stacks (may introduce false negatives).
1543 */
1544 if (kmemleak_stack_scan) {
1545 struct task_struct *p, *g;
1546
1547 read_lock(&tasklist_lock);
1548 do_each_thread(g, p) {
1549 void *stack = try_get_task_stack(p);
1550 if (stack) {
1551 scan_block(stack, stack + THREAD_SIZE, NULL);
1552 put_task_stack(p);
1553 }
1554 } while_each_thread(g, p);
1555 read_unlock(&tasklist_lock);
1556 }
1557
1558 /*
1559 * Scan the objects already referenced from the sections scanned
1560 * above.
1561 */
1562 scan_gray_list();
1563
1564 /*
1565 * Check for new or unreferenced objects modified since the previous
1566 * scan and color them gray until the next scan.
1567 */
1568 rcu_read_lock();
1569 list_for_each_entry_rcu(object, &object_list, object_list) {
1570 spin_lock_irqsave(&object->lock, flags);
1571 if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1572 && update_checksum(object) && get_object(object)) {
1573 /* color it gray temporarily */
1574 object->count = object->min_count;
1575 list_add_tail(&object->gray_list, &gray_list);
1576 }
1577 spin_unlock_irqrestore(&object->lock, flags);
1578 }
1579 rcu_read_unlock();
1580
1581 /*
1582 * Re-scan the gray list for modified unreferenced objects.
1583 */
1584 scan_gray_list();
1585
1586 /*
1587 * If scanning was stopped do not report any new unreferenced objects.
1588 */
1589 if (scan_should_stop())
1590 return;
1591
1592 /*
1593 * Scanning result reporting.
1594 */
1595 rcu_read_lock();
1596 list_for_each_entry_rcu(object, &object_list, object_list) {
1597 spin_lock_irqsave(&object->lock, flags);
1598 if (unreferenced_object(object) &&
1599 !(object->flags & OBJECT_REPORTED)) {
1600 object->flags |= OBJECT_REPORTED;
1601 new_leaks++;
1602 }
1603 spin_unlock_irqrestore(&object->lock, flags);
1604 }
1605 rcu_read_unlock();
1606
1607 if (new_leaks) {
1608 kmemleak_found_leaks = true;
1609
1610 pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
1611 new_leaks);
1612 }
1613
1614}
1615
1616/*
1617 * Thread function performing automatic memory scanning. Unreferenced objects
1618 * at the end of a memory scan are reported but only the first time.
1619 */
1620static int kmemleak_scan_thread(void *arg)
1621{
1622 static int first_run = 1;
1623
1624 pr_info("Automatic memory scanning thread started\n");
1625 set_user_nice(current, 10);
1626
1627 /*
1628 * Wait before the first scan to allow the system to fully initialize.
1629 */
1630 if (first_run) {
1631 signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000);
1632 first_run = 0;
1633 while (timeout && !kthread_should_stop())
1634 timeout = schedule_timeout_interruptible(timeout);
1635 }
1636
1637 while (!kthread_should_stop()) {
1638 signed long timeout = jiffies_scan_wait;
1639
1640 mutex_lock(&scan_mutex);
1641 kmemleak_scan();
1642 mutex_unlock(&scan_mutex);
1643
1644 /* wait before the next scan */
1645 while (timeout && !kthread_should_stop())
1646 timeout = schedule_timeout_interruptible(timeout);
1647 }
1648
1649 pr_info("Automatic memory scanning thread ended\n");
1650
1651 return 0;
1652}
1653
1654/*
1655 * Start the automatic memory scanning thread. This function must be called
1656 * with the scan_mutex held.
1657 */
1658static void start_scan_thread(void)
1659{
1660 if (scan_thread)
1661 return;
1662 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1663 if (IS_ERR(scan_thread)) {
1664 pr_warn("Failed to create the scan thread\n");
1665 scan_thread = NULL;
1666 }
1667}
1668
1669/*
1670 * Stop the automatic memory scanning thread.
1671 */
1672static void stop_scan_thread(void)
1673{
1674 if (scan_thread) {
1675 kthread_stop(scan_thread);
1676 scan_thread = NULL;
1677 }
1678}
1679
1680/*
1681 * Iterate over the object_list and return the first valid object at or after
1682 * the required position with its use_count incremented. The function triggers
1683 * a memory scanning when the pos argument points to the first position.
1684 */
1685static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1686{
1687 struct kmemleak_object *object;
1688 loff_t n = *pos;
1689 int err;
1690
1691 err = mutex_lock_interruptible(&scan_mutex);
1692 if (err < 0)
1693 return ERR_PTR(err);
1694
1695 rcu_read_lock();
1696 list_for_each_entry_rcu(object, &object_list, object_list) {
1697 if (n-- > 0)
1698 continue;
1699 if (get_object(object))
1700 goto out;
1701 }
1702 object = NULL;
1703out:
1704 return object;
1705}
1706
1707/*
1708 * Return the next object in the object_list. The function decrements the
1709 * use_count of the previous object and increases that of the next one.
1710 */
1711static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1712{
1713 struct kmemleak_object *prev_obj = v;
1714 struct kmemleak_object *next_obj = NULL;
1715 struct kmemleak_object *obj = prev_obj;
1716
1717 ++(*pos);
1718
1719 list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
1720 if (get_object(obj)) {
1721 next_obj = obj;
1722 break;
1723 }
1724 }
1725
1726 put_object(prev_obj);
1727 return next_obj;
1728}
1729
1730/*
1731 * Decrement the use_count of the last object required, if any.
1732 */
1733static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1734{
1735 if (!IS_ERR(v)) {
1736 /*
1737 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1738 * waiting was interrupted, so only release it if !IS_ERR.
1739 */
1740 rcu_read_unlock();
1741 mutex_unlock(&scan_mutex);
1742 if (v)
1743 put_object(v);
1744 }
1745}
1746
1747/*
1748 * Print the information for an unreferenced object to the seq file.
1749 */
1750static int kmemleak_seq_show(struct seq_file *seq, void *v)
1751{
1752 struct kmemleak_object *object = v;
1753 unsigned long flags;
1754
1755 spin_lock_irqsave(&object->lock, flags);
1756 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1757 print_unreferenced(seq, object);
1758 spin_unlock_irqrestore(&object->lock, flags);
1759 return 0;
1760}
1761
1762static const struct seq_operations kmemleak_seq_ops = {
1763 .start = kmemleak_seq_start,
1764 .next = kmemleak_seq_next,
1765 .stop = kmemleak_seq_stop,
1766 .show = kmemleak_seq_show,
1767};
1768
1769static int kmemleak_open(struct inode *inode, struct file *file)
1770{
1771 return seq_open(file, &kmemleak_seq_ops);
1772}
1773
1774static int dump_str_object_info(const char *str)
1775{
1776 unsigned long flags;
1777 struct kmemleak_object *object;
1778 unsigned long addr;
1779
1780 if (kstrtoul(str, 0, &addr))
1781 return -EINVAL;
1782 object = find_and_get_object(addr, 0);
1783 if (!object) {
1784 pr_info("Unknown object at 0x%08lx\n", addr);
1785 return -EINVAL;
1786 }
1787
1788 spin_lock_irqsave(&object->lock, flags);
1789 dump_object_info(object);
1790 spin_unlock_irqrestore(&object->lock, flags);
1791
1792 put_object(object);
1793 return 0;
1794}
1795
1796/*
1797 * We use grey instead of black to ensure we can do future scans on the same
1798 * objects. If we did not do future scans these black objects could
1799 * potentially contain references to newly allocated objects in the future and
1800 * we'd end up with false positives.
1801 */
1802static void kmemleak_clear(void)
1803{
1804 struct kmemleak_object *object;
1805 unsigned long flags;
1806
1807 rcu_read_lock();
1808 list_for_each_entry_rcu(object, &object_list, object_list) {
1809 spin_lock_irqsave(&object->lock, flags);
1810 if ((object->flags & OBJECT_REPORTED) &&
1811 unreferenced_object(object))
1812 __paint_it(object, KMEMLEAK_GREY);
1813 spin_unlock_irqrestore(&object->lock, flags);
1814 }
1815 rcu_read_unlock();
1816
1817 kmemleak_found_leaks = false;
1818}
1819
1820static void __kmemleak_do_cleanup(void);
1821
1822/*
1823 * File write operation to configure kmemleak at run-time. The following
1824 * commands can be written to the /sys/kernel/debug/kmemleak file:
1825 * off - disable kmemleak (irreversible)
1826 * stack=on - enable the task stacks scanning
1827 * stack=off - disable the tasks stacks scanning
1828 * scan=on - start the automatic memory scanning thread
1829 * scan=off - stop the automatic memory scanning thread
1830 * scan=... - set the automatic memory scanning period in seconds (0 to
1831 * disable it)
1832 * scan - trigger a memory scan
1833 * clear - mark all current reported unreferenced kmemleak objects as
1834 * grey to ignore printing them, or free all kmemleak objects
1835 * if kmemleak has been disabled.
1836 * dump=... - dump information about the object found at the given address
1837 */
1838static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1839 size_t size, loff_t *ppos)
1840{
1841 char buf[64];
1842 int buf_size;
1843 int ret;
1844
1845 buf_size = min(size, (sizeof(buf) - 1));
1846 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1847 return -EFAULT;
1848 buf[buf_size] = 0;
1849
1850 ret = mutex_lock_interruptible(&scan_mutex);
1851 if (ret < 0)
1852 return ret;
1853
1854 if (strncmp(buf, "clear", 5) == 0) {
1855 if (kmemleak_enabled)
1856 kmemleak_clear();
1857 else
1858 __kmemleak_do_cleanup();
1859 goto out;
1860 }
1861
1862 if (!kmemleak_enabled) {
1863 ret = -EBUSY;
1864 goto out;
1865 }
1866
1867 if (strncmp(buf, "off", 3) == 0)
1868 kmemleak_disable();
1869 else if (strncmp(buf, "stack=on", 8) == 0)
1870 kmemleak_stack_scan = 1;
1871 else if (strncmp(buf, "stack=off", 9) == 0)
1872 kmemleak_stack_scan = 0;
1873 else if (strncmp(buf, "scan=on", 7) == 0)
1874 start_scan_thread();
1875 else if (strncmp(buf, "scan=off", 8) == 0)
1876 stop_scan_thread();
1877 else if (strncmp(buf, "scan=", 5) == 0) {
1878 unsigned long secs;
1879
1880 ret = kstrtoul(buf + 5, 0, &secs);
1881 if (ret < 0)
1882 goto out;
1883 stop_scan_thread();
1884 if (secs) {
1885 jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1886 start_scan_thread();
1887 }
1888 } else if (strncmp(buf, "scan", 4) == 0)
1889 kmemleak_scan();
1890 else if (strncmp(buf, "dump=", 5) == 0)
1891 ret = dump_str_object_info(buf + 5);
1892 else
1893 ret = -EINVAL;
1894
1895out:
1896 mutex_unlock(&scan_mutex);
1897 if (ret < 0)
1898 return ret;
1899
1900 /* ignore the rest of the buffer, only one command at a time */
1901 *ppos += size;
1902 return size;
1903}
1904
1905static const struct file_operations kmemleak_fops = {
1906 .owner = THIS_MODULE,
1907 .open = kmemleak_open,
1908 .read = seq_read,
1909 .write = kmemleak_write,
1910 .llseek = seq_lseek,
1911 .release = seq_release,
1912};
1913
1914static void __kmemleak_do_cleanup(void)
1915{
1916 struct kmemleak_object *object;
1917
1918 rcu_read_lock();
1919 list_for_each_entry_rcu(object, &object_list, object_list)
1920 delete_object_full(object->pointer);
1921 rcu_read_unlock();
1922}
1923
1924/*
1925 * Stop the memory scanning thread and free the kmemleak internal objects if
1926 * no previous scan thread (otherwise, kmemleak may still have some useful
1927 * information on memory leaks).
1928 */
1929static void kmemleak_do_cleanup(struct work_struct *work)
1930{
1931 stop_scan_thread();
1932
1933 mutex_lock(&scan_mutex);
1934 /*
1935 * Once it is made sure that kmemleak_scan has stopped, it is safe to no
1936 * longer track object freeing. Ordering of the scan thread stopping and
1937 * the memory accesses below is guaranteed by the kthread_stop()
1938 * function.
1939 */
1940 kmemleak_free_enabled = 0;
1941 mutex_unlock(&scan_mutex);
1942
1943 if (!kmemleak_found_leaks)
1944 __kmemleak_do_cleanup();
1945 else
1946 pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
1947}
1948
1949static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
1950
1951/*
1952 * Disable kmemleak. No memory allocation/freeing will be traced once this
1953 * function is called. Disabling kmemleak is an irreversible operation.
1954 */
1955static void kmemleak_disable(void)
1956{
1957 /* atomically check whether it was already invoked */
1958 if (cmpxchg(&kmemleak_error, 0, 1))
1959 return;
1960
1961 /* stop any memory operation tracing */
1962 kmemleak_enabled = 0;
1963
1964 /* check whether it is too early for a kernel thread */
1965 if (kmemleak_initialized)
1966 schedule_work(&cleanup_work);
1967 else
1968 kmemleak_free_enabled = 0;
1969
1970 pr_info("Kernel memory leak detector disabled\n");
1971}
1972
1973/*
1974 * Allow boot-time kmemleak disabling (enabled by default).
1975 */
1976static int __init kmemleak_boot_config(char *str)
1977{
1978 if (!str)
1979 return -EINVAL;
1980 if (strcmp(str, "off") == 0)
1981 kmemleak_disable();
1982 else if (strcmp(str, "on") == 0)
1983 kmemleak_skip_disable = 1;
1984 else
1985 return -EINVAL;
1986 return 0;
1987}
1988early_param("kmemleak", kmemleak_boot_config);
1989
1990static void __init print_log_trace(struct early_log *log)
1991{
1992 struct stack_trace trace;
1993
1994 trace.nr_entries = log->trace_len;
1995 trace.entries = log->trace;
1996
1997 pr_notice("Early log backtrace:\n");
1998 print_stack_trace(&trace, 2);
1999}
2000
2001/*
2002 * Kmemleak initialization.
2003 */
2004void __init kmemleak_init(void)
2005{
2006 int i;
2007 unsigned long flags;
2008
2009#ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
2010 if (!kmemleak_skip_disable) {
2011 kmemleak_early_log = 0;
2012 kmemleak_disable();
2013 return;
2014 }
2015#endif
2016
2017 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
2018 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
2019
2020 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
2021 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
2022
2023 if (crt_early_log > ARRAY_SIZE(early_log))
2024 pr_warn("Early log buffer exceeded (%d), please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n",
2025 crt_early_log);
2026
2027 /* the kernel is still in UP mode, so disabling the IRQs is enough */
2028 local_irq_save(flags);
2029 kmemleak_early_log = 0;
2030 if (kmemleak_error) {
2031 local_irq_restore(flags);
2032 return;
2033 } else {
2034 kmemleak_enabled = 1;
2035 kmemleak_free_enabled = 1;
2036 }
2037 local_irq_restore(flags);
2038
2039 /*
2040 * This is the point where tracking allocations is safe. Automatic
2041 * scanning is started during the late initcall. Add the early logged
2042 * callbacks to the kmemleak infrastructure.
2043 */
2044 for (i = 0; i < crt_early_log; i++) {
2045 struct early_log *log = &early_log[i];
2046
2047 switch (log->op_type) {
2048 case KMEMLEAK_ALLOC:
2049 early_alloc(log);
2050 break;
2051 case KMEMLEAK_ALLOC_PERCPU:
2052 early_alloc_percpu(log);
2053 break;
2054 case KMEMLEAK_FREE:
2055 kmemleak_free(log->ptr);
2056 break;
2057 case KMEMLEAK_FREE_PART:
2058 kmemleak_free_part(log->ptr, log->size);
2059 break;
2060 case KMEMLEAK_FREE_PERCPU:
2061 kmemleak_free_percpu(log->ptr);
2062 break;
2063 case KMEMLEAK_NOT_LEAK:
2064 kmemleak_not_leak(log->ptr);
2065 break;
2066 case KMEMLEAK_IGNORE:
2067 kmemleak_ignore(log->ptr);
2068 break;
2069 case KMEMLEAK_SCAN_AREA:
2070 kmemleak_scan_area(log->ptr, log->size, GFP_KERNEL);
2071 break;
2072 case KMEMLEAK_NO_SCAN:
2073 kmemleak_no_scan(log->ptr);
2074 break;
2075 case KMEMLEAK_SET_EXCESS_REF:
2076 object_set_excess_ref((unsigned long)log->ptr,
2077 log->excess_ref);
2078 break;
2079 default:
2080 kmemleak_warn("Unknown early log operation: %d\n",
2081 log->op_type);
2082 }
2083
2084 if (kmemleak_warning) {
2085 print_log_trace(log);
2086 kmemleak_warning = 0;
2087 }
2088 }
2089}
2090
2091/*
2092 * Late initialization function.
2093 */
2094static int __init kmemleak_late_init(void)
2095{
2096 struct dentry *dentry;
2097
2098 kmemleak_initialized = 1;
2099
2100 if (kmemleak_error) {
2101 /*
2102 * Some error occurred and kmemleak was disabled. There is a
2103 * small chance that kmemleak_disable() was called immediately
2104 * after setting kmemleak_initialized and we may end up with
2105 * two clean-up threads but serialized by scan_mutex.
2106 */
2107 schedule_work(&cleanup_work);
2108 return -ENOMEM;
2109 }
2110
2111 dentry = debugfs_create_file("kmemleak", 0644, NULL, NULL,
2112 &kmemleak_fops);
2113 if (!dentry)
2114 pr_warn("Failed to create the debugfs kmemleak file\n");
2115 mutex_lock(&scan_mutex);
2116 start_scan_thread();
2117 mutex_unlock(&scan_mutex);
2118
2119 pr_info("Kernel memory leak detector initialized\n");
2120
2121 return 0;
2122}
2123late_initcall(kmemleak_late_init);
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_NORETRY | __GFP_NOMEMALLOC | \
118 __GFP_NOWARN)
119
120/* scanning area inside a memory block */
121struct kmemleak_scan_area {
122 struct hlist_node node;
123 unsigned long start;
124 size_t size;
125};
126
127#define KMEMLEAK_GREY 0
128#define KMEMLEAK_BLACK -1
129
130/*
131 * Structure holding the metadata for each allocated memory block.
132 * Modifications to such objects should be made while holding the
133 * object->lock. Insertions or deletions from object_list, gray_list or
134 * rb_node are already protected by the corresponding locks or mutex (see
135 * the notes on locking above). These objects are reference-counted
136 * (use_count) and freed using the RCU mechanism.
137 */
138struct kmemleak_object {
139 raw_spinlock_t lock;
140 unsigned int flags; /* object status flags */
141 struct list_head object_list;
142 struct list_head gray_list;
143 struct rb_node rb_node;
144 struct rcu_head rcu; /* object_list lockless traversal */
145 /* object usage count; object freed when use_count == 0 */
146 atomic_t use_count;
147 unsigned long pointer;
148 size_t size;
149 /* pass surplus references to this pointer */
150 unsigned long excess_ref;
151 /* minimum number of a pointers found before it is considered leak */
152 int min_count;
153 /* the total number of pointers found pointing to this object */
154 int count;
155 /* checksum for detecting modified objects */
156 u32 checksum;
157 /* memory ranges to be scanned inside an object (empty for all) */
158 struct hlist_head area_list;
159 unsigned long trace[MAX_TRACE];
160 unsigned int trace_len;
161 unsigned long jiffies; /* creation timestamp */
162 pid_t pid; /* pid of the current task */
163 char comm[TASK_COMM_LEN]; /* executable name */
164};
165
166/* flag representing the memory block allocation status */
167#define OBJECT_ALLOCATED (1 << 0)
168/* flag set after the first reporting of an unreference object */
169#define OBJECT_REPORTED (1 << 1)
170/* flag set to not scan the object */
171#define OBJECT_NO_SCAN (1 << 2)
172/* flag set to fully scan the object when scan_area allocation failed */
173#define OBJECT_FULL_SCAN (1 << 3)
174
175#define HEX_PREFIX " "
176/* number of bytes to print per line; must be 16 or 32 */
177#define HEX_ROW_SIZE 16
178/* number of bytes to print at a time (1, 2, 4, 8) */
179#define HEX_GROUP_SIZE 1
180/* include ASCII after the hex output */
181#define HEX_ASCII 1
182/* max number of lines to be printed */
183#define HEX_MAX_LINES 2
184
185/* the list of all allocated objects */
186static LIST_HEAD(object_list);
187/* the list of gray-colored objects (see color_gray comment below) */
188static LIST_HEAD(gray_list);
189/* memory pool allocation */
190static struct kmemleak_object mem_pool[CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE];
191static int mem_pool_free_count = ARRAY_SIZE(mem_pool);
192static LIST_HEAD(mem_pool_free_list);
193/* search tree for object boundaries */
194static struct rb_root object_tree_root = RB_ROOT;
195/* protecting the access to object_list and object_tree_root */
196static DEFINE_RAW_SPINLOCK(kmemleak_lock);
197
198/* allocation caches for kmemleak internal data */
199static struct kmem_cache *object_cache;
200static struct kmem_cache *scan_area_cache;
201
202/* set if tracing memory operations is enabled */
203static int kmemleak_enabled = 1;
204/* same as above but only for the kmemleak_free() callback */
205static int kmemleak_free_enabled = 1;
206/* set in the late_initcall if there were no errors */
207static int kmemleak_initialized;
208/* set if a kmemleak warning was issued */
209static int kmemleak_warning;
210/* set if a fatal kmemleak error has occurred */
211static int kmemleak_error;
212
213/* minimum and maximum address that may be valid pointers */
214static unsigned long min_addr = ULONG_MAX;
215static unsigned long max_addr;
216
217static struct task_struct *scan_thread;
218/* used to avoid reporting of recently allocated objects */
219static unsigned long jiffies_min_age;
220static unsigned long jiffies_last_scan;
221/* delay between automatic memory scannings */
222static unsigned long jiffies_scan_wait;
223/* enables or disables the task stacks scanning */
224static int kmemleak_stack_scan = 1;
225/* protects the memory scanning, parameters and debug/kmemleak file access */
226static DEFINE_MUTEX(scan_mutex);
227/* setting kmemleak=on, will set this var, skipping the disable */
228static int kmemleak_skip_disable;
229/* If there are leaks that can be reported */
230static bool kmemleak_found_leaks;
231
232static bool kmemleak_verbose;
233module_param_named(verbose, kmemleak_verbose, bool, 0600);
234
235static void kmemleak_disable(void);
236
237/*
238 * Print a warning and dump the stack trace.
239 */
240#define kmemleak_warn(x...) do { \
241 pr_warn(x); \
242 dump_stack(); \
243 kmemleak_warning = 1; \
244} while (0)
245
246/*
247 * Macro invoked when a serious kmemleak condition occurred and cannot be
248 * recovered from. Kmemleak will be disabled and further allocation/freeing
249 * tracing no longer available.
250 */
251#define kmemleak_stop(x...) do { \
252 kmemleak_warn(x); \
253 kmemleak_disable(); \
254} while (0)
255
256#define warn_or_seq_printf(seq, fmt, ...) do { \
257 if (seq) \
258 seq_printf(seq, fmt, ##__VA_ARGS__); \
259 else \
260 pr_warn(fmt, ##__VA_ARGS__); \
261} while (0)
262
263static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type,
264 int rowsize, int groupsize, const void *buf,
265 size_t len, bool ascii)
266{
267 if (seq)
268 seq_hex_dump(seq, HEX_PREFIX, prefix_type, rowsize, groupsize,
269 buf, len, ascii);
270 else
271 print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type,
272 rowsize, groupsize, buf, len, ascii);
273}
274
275/*
276 * Printing of the objects hex dump to the seq file. The number of lines to be
277 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
278 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
279 * with the object->lock held.
280 */
281static void hex_dump_object(struct seq_file *seq,
282 struct kmemleak_object *object)
283{
284 const u8 *ptr = (const u8 *)object->pointer;
285 size_t len;
286
287 /* limit the number of lines to HEX_MAX_LINES */
288 len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
289
290 warn_or_seq_printf(seq, " hex dump (first %zu bytes):\n", len);
291 kasan_disable_current();
292 warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE,
293 HEX_GROUP_SIZE, kasan_reset_tag((void *)ptr), len, HEX_ASCII);
294 kasan_enable_current();
295}
296
297/*
298 * Object colors, encoded with count and min_count:
299 * - white - orphan object, not enough references to it (count < min_count)
300 * - gray - not orphan, not marked as false positive (min_count == 0) or
301 * sufficient references to it (count >= min_count)
302 * - black - ignore, it doesn't contain references (e.g. text section)
303 * (min_count == -1). No function defined for this color.
304 * Newly created objects don't have any color assigned (object->count == -1)
305 * before the next memory scan when they become white.
306 */
307static bool color_white(const struct kmemleak_object *object)
308{
309 return object->count != KMEMLEAK_BLACK &&
310 object->count < object->min_count;
311}
312
313static bool color_gray(const struct kmemleak_object *object)
314{
315 return object->min_count != KMEMLEAK_BLACK &&
316 object->count >= object->min_count;
317}
318
319/*
320 * Objects are considered unreferenced only if their color is white, they have
321 * not be deleted and have a minimum age to avoid false positives caused by
322 * pointers temporarily stored in CPU registers.
323 */
324static bool unreferenced_object(struct kmemleak_object *object)
325{
326 return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
327 time_before_eq(object->jiffies + jiffies_min_age,
328 jiffies_last_scan);
329}
330
331/*
332 * Printing of the unreferenced objects information to the seq file. The
333 * print_unreferenced function must be called with the object->lock held.
334 */
335static void print_unreferenced(struct seq_file *seq,
336 struct kmemleak_object *object)
337{
338 int i;
339 unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
340
341 warn_or_seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
342 object->pointer, object->size);
343 warn_or_seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
344 object->comm, object->pid, object->jiffies,
345 msecs_age / 1000, msecs_age % 1000);
346 hex_dump_object(seq, object);
347 warn_or_seq_printf(seq, " backtrace:\n");
348
349 for (i = 0; i < object->trace_len; i++) {
350 void *ptr = (void *)object->trace[i];
351 warn_or_seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
352 }
353}
354
355/*
356 * Print the kmemleak_object information. This function is used mainly for
357 * debugging special cases when kmemleak operations. It must be called with
358 * the object->lock held.
359 */
360static void dump_object_info(struct kmemleak_object *object)
361{
362 pr_notice("Object 0x%08lx (size %zu):\n",
363 object->pointer, object->size);
364 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
365 object->comm, object->pid, object->jiffies);
366 pr_notice(" min_count = %d\n", object->min_count);
367 pr_notice(" count = %d\n", object->count);
368 pr_notice(" flags = 0x%x\n", object->flags);
369 pr_notice(" checksum = %u\n", object->checksum);
370 pr_notice(" backtrace:\n");
371 stack_trace_print(object->trace, object->trace_len, 4);
372}
373
374/*
375 * Look-up a memory block metadata (kmemleak_object) in the object search
376 * tree based on a pointer value. If alias is 0, only values pointing to the
377 * beginning of the memory block are allowed. The kmemleak_lock must be held
378 * when calling this function.
379 */
380static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
381{
382 struct rb_node *rb = object_tree_root.rb_node;
383
384 while (rb) {
385 struct kmemleak_object *object =
386 rb_entry(rb, struct kmemleak_object, rb_node);
387 if (ptr < object->pointer)
388 rb = object->rb_node.rb_left;
389 else if (object->pointer + object->size <= ptr)
390 rb = object->rb_node.rb_right;
391 else if (object->pointer == ptr || alias)
392 return object;
393 else {
394 kmemleak_warn("Found object by alias at 0x%08lx\n",
395 ptr);
396 dump_object_info(object);
397 break;
398 }
399 }
400 return NULL;
401}
402
403/*
404 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
405 * that once an object's use_count reached 0, the RCU freeing was already
406 * registered and the object should no longer be used. This function must be
407 * called under the protection of rcu_read_lock().
408 */
409static int get_object(struct kmemleak_object *object)
410{
411 return atomic_inc_not_zero(&object->use_count);
412}
413
414/*
415 * Memory pool allocation and freeing. kmemleak_lock must not be held.
416 */
417static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
418{
419 unsigned long flags;
420 struct kmemleak_object *object;
421
422 /* try the slab allocator first */
423 if (object_cache) {
424 object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
425 if (object)
426 return object;
427 }
428
429 /* slab allocation failed, try the memory pool */
430 raw_spin_lock_irqsave(&kmemleak_lock, flags);
431 object = list_first_entry_or_null(&mem_pool_free_list,
432 typeof(*object), object_list);
433 if (object)
434 list_del(&object->object_list);
435 else if (mem_pool_free_count)
436 object = &mem_pool[--mem_pool_free_count];
437 else
438 pr_warn_once("Memory pool empty, consider increasing CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE\n");
439 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
440
441 return object;
442}
443
444/*
445 * Return the object to either the slab allocator or the memory pool.
446 */
447static void mem_pool_free(struct kmemleak_object *object)
448{
449 unsigned long flags;
450
451 if (object < mem_pool || object >= mem_pool + ARRAY_SIZE(mem_pool)) {
452 kmem_cache_free(object_cache, object);
453 return;
454 }
455
456 /* add the object to the memory pool free list */
457 raw_spin_lock_irqsave(&kmemleak_lock, flags);
458 list_add(&object->object_list, &mem_pool_free_list);
459 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
460}
461
462/*
463 * RCU callback to free a kmemleak_object.
464 */
465static void free_object_rcu(struct rcu_head *rcu)
466{
467 struct hlist_node *tmp;
468 struct kmemleak_scan_area *area;
469 struct kmemleak_object *object =
470 container_of(rcu, struct kmemleak_object, rcu);
471
472 /*
473 * Once use_count is 0 (guaranteed by put_object), there is no other
474 * code accessing this object, hence no need for locking.
475 */
476 hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
477 hlist_del(&area->node);
478 kmem_cache_free(scan_area_cache, area);
479 }
480 mem_pool_free(object);
481}
482
483/*
484 * Decrement the object use_count. Once the count is 0, free the object using
485 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
486 * delete_object() path, the delayed RCU freeing ensures that there is no
487 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
488 * is also possible.
489 */
490static void put_object(struct kmemleak_object *object)
491{
492 if (!atomic_dec_and_test(&object->use_count))
493 return;
494
495 /* should only get here after delete_object was called */
496 WARN_ON(object->flags & OBJECT_ALLOCATED);
497
498 /*
499 * It may be too early for the RCU callbacks, however, there is no
500 * concurrent object_list traversal when !object_cache and all objects
501 * came from the memory pool. Free the object directly.
502 */
503 if (object_cache)
504 call_rcu(&object->rcu, free_object_rcu);
505 else
506 free_object_rcu(&object->rcu);
507}
508
509/*
510 * Look up an object in the object search tree and increase its use_count.
511 */
512static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
513{
514 unsigned long flags;
515 struct kmemleak_object *object;
516
517 rcu_read_lock();
518 raw_spin_lock_irqsave(&kmemleak_lock, flags);
519 object = lookup_object(ptr, alias);
520 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
521
522 /* check whether the object is still available */
523 if (object && !get_object(object))
524 object = NULL;
525 rcu_read_unlock();
526
527 return object;
528}
529
530/*
531 * Remove an object from the object_tree_root and object_list. Must be called
532 * with the kmemleak_lock held _if_ kmemleak is still enabled.
533 */
534static void __remove_object(struct kmemleak_object *object)
535{
536 rb_erase(&object->rb_node, &object_tree_root);
537 list_del_rcu(&object->object_list);
538}
539
540/*
541 * Look up an object in the object search tree and remove it from both
542 * object_tree_root and object_list. The returned object's use_count should be
543 * at least 1, as initially set by create_object().
544 */
545static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias)
546{
547 unsigned long flags;
548 struct kmemleak_object *object;
549
550 raw_spin_lock_irqsave(&kmemleak_lock, flags);
551 object = lookup_object(ptr, alias);
552 if (object)
553 __remove_object(object);
554 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
555
556 return object;
557}
558
559/*
560 * Save stack trace to the given array of MAX_TRACE size.
561 */
562static int __save_stack_trace(unsigned long *trace)
563{
564 return stack_trace_save(trace, MAX_TRACE, 2);
565}
566
567/*
568 * Create the metadata (struct kmemleak_object) corresponding to an allocated
569 * memory block and add it to the object_list and object_tree_root.
570 */
571static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
572 int min_count, gfp_t gfp)
573{
574 unsigned long flags;
575 struct kmemleak_object *object, *parent;
576 struct rb_node **link, *rb_parent;
577 unsigned long untagged_ptr;
578
579 object = mem_pool_alloc(gfp);
580 if (!object) {
581 pr_warn("Cannot allocate a kmemleak_object structure\n");
582 kmemleak_disable();
583 return NULL;
584 }
585
586 INIT_LIST_HEAD(&object->object_list);
587 INIT_LIST_HEAD(&object->gray_list);
588 INIT_HLIST_HEAD(&object->area_list);
589 raw_spin_lock_init(&object->lock);
590 atomic_set(&object->use_count, 1);
591 object->flags = OBJECT_ALLOCATED;
592 object->pointer = ptr;
593 object->size = kfence_ksize((void *)ptr) ?: size;
594 object->excess_ref = 0;
595 object->min_count = min_count;
596 object->count = 0; /* white color initially */
597 object->jiffies = jiffies;
598 object->checksum = 0;
599
600 /* task information */
601 if (in_irq()) {
602 object->pid = 0;
603 strncpy(object->comm, "hardirq", sizeof(object->comm));
604 } else if (in_serving_softirq()) {
605 object->pid = 0;
606 strncpy(object->comm, "softirq", sizeof(object->comm));
607 } else {
608 object->pid = current->pid;
609 /*
610 * There is a small chance of a race with set_task_comm(),
611 * however using get_task_comm() here may cause locking
612 * dependency issues with current->alloc_lock. In the worst
613 * case, the command line is not correct.
614 */
615 strncpy(object->comm, current->comm, sizeof(object->comm));
616 }
617
618 /* kernel backtrace */
619 object->trace_len = __save_stack_trace(object->trace);
620
621 raw_spin_lock_irqsave(&kmemleak_lock, flags);
622
623 untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
624 min_addr = min(min_addr, untagged_ptr);
625 max_addr = max(max_addr, untagged_ptr + size);
626 link = &object_tree_root.rb_node;
627 rb_parent = NULL;
628 while (*link) {
629 rb_parent = *link;
630 parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
631 if (ptr + size <= parent->pointer)
632 link = &parent->rb_node.rb_left;
633 else if (parent->pointer + parent->size <= ptr)
634 link = &parent->rb_node.rb_right;
635 else {
636 kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
637 ptr);
638 /*
639 * No need for parent->lock here since "parent" cannot
640 * be freed while the kmemleak_lock is held.
641 */
642 dump_object_info(parent);
643 kmem_cache_free(object_cache, object);
644 object = NULL;
645 goto out;
646 }
647 }
648 rb_link_node(&object->rb_node, rb_parent, link);
649 rb_insert_color(&object->rb_node, &object_tree_root);
650
651 list_add_tail_rcu(&object->object_list, &object_list);
652out:
653 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
654 return object;
655}
656
657/*
658 * Mark the object as not allocated and schedule RCU freeing via put_object().
659 */
660static void __delete_object(struct kmemleak_object *object)
661{
662 unsigned long flags;
663
664 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
665 WARN_ON(atomic_read(&object->use_count) < 1);
666
667 /*
668 * Locking here also ensures that the corresponding memory block
669 * cannot be freed when it is being scanned.
670 */
671 raw_spin_lock_irqsave(&object->lock, flags);
672 object->flags &= ~OBJECT_ALLOCATED;
673 raw_spin_unlock_irqrestore(&object->lock, flags);
674 put_object(object);
675}
676
677/*
678 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
679 * delete it.
680 */
681static void delete_object_full(unsigned long ptr)
682{
683 struct kmemleak_object *object;
684
685 object = find_and_remove_object(ptr, 0);
686 if (!object) {
687#ifdef DEBUG
688 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
689 ptr);
690#endif
691 return;
692 }
693 __delete_object(object);
694}
695
696/*
697 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
698 * delete it. If the memory block is partially freed, the function may create
699 * additional metadata for the remaining parts of the block.
700 */
701static void delete_object_part(unsigned long ptr, size_t size)
702{
703 struct kmemleak_object *object;
704 unsigned long start, end;
705
706 object = find_and_remove_object(ptr, 1);
707 if (!object) {
708#ifdef DEBUG
709 kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
710 ptr, size);
711#endif
712 return;
713 }
714
715 /*
716 * Create one or two objects that may result from the memory block
717 * split. Note that partial freeing is only done by free_bootmem() and
718 * this happens before kmemleak_init() is called.
719 */
720 start = object->pointer;
721 end = object->pointer + object->size;
722 if (ptr > start)
723 create_object(start, ptr - start, object->min_count,
724 GFP_KERNEL);
725 if (ptr + size < end)
726 create_object(ptr + size, end - ptr - size, object->min_count,
727 GFP_KERNEL);
728
729 __delete_object(object);
730}
731
732static void __paint_it(struct kmemleak_object *object, int color)
733{
734 object->min_count = color;
735 if (color == KMEMLEAK_BLACK)
736 object->flags |= OBJECT_NO_SCAN;
737}
738
739static void paint_it(struct kmemleak_object *object, int color)
740{
741 unsigned long flags;
742
743 raw_spin_lock_irqsave(&object->lock, flags);
744 __paint_it(object, color);
745 raw_spin_unlock_irqrestore(&object->lock, flags);
746}
747
748static void paint_ptr(unsigned long ptr, int color)
749{
750 struct kmemleak_object *object;
751
752 object = find_and_get_object(ptr, 0);
753 if (!object) {
754 kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n",
755 ptr,
756 (color == KMEMLEAK_GREY) ? "Grey" :
757 (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
758 return;
759 }
760 paint_it(object, color);
761 put_object(object);
762}
763
764/*
765 * Mark an object permanently as gray-colored so that it can no longer be
766 * reported as a leak. This is used in general to mark a false positive.
767 */
768static void make_gray_object(unsigned long ptr)
769{
770 paint_ptr(ptr, KMEMLEAK_GREY);
771}
772
773/*
774 * Mark the object as black-colored so that it is ignored from scans and
775 * reporting.
776 */
777static void make_black_object(unsigned long ptr)
778{
779 paint_ptr(ptr, KMEMLEAK_BLACK);
780}
781
782/*
783 * Add a scanning area to the object. If at least one such area is added,
784 * kmemleak will only scan these ranges rather than the whole memory block.
785 */
786static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
787{
788 unsigned long flags;
789 struct kmemleak_object *object;
790 struct kmemleak_scan_area *area = NULL;
791
792 object = find_and_get_object(ptr, 1);
793 if (!object) {
794 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
795 ptr);
796 return;
797 }
798
799 if (scan_area_cache)
800 area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
801
802 raw_spin_lock_irqsave(&object->lock, flags);
803 if (!area) {
804 pr_warn_once("Cannot allocate a scan area, scanning the full object\n");
805 /* mark the object for full scan to avoid false positives */
806 object->flags |= OBJECT_FULL_SCAN;
807 goto out_unlock;
808 }
809 if (size == SIZE_MAX) {
810 size = object->pointer + object->size - ptr;
811 } else if (ptr + size > object->pointer + object->size) {
812 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
813 dump_object_info(object);
814 kmem_cache_free(scan_area_cache, area);
815 goto out_unlock;
816 }
817
818 INIT_HLIST_NODE(&area->node);
819 area->start = ptr;
820 area->size = size;
821
822 hlist_add_head(&area->node, &object->area_list);
823out_unlock:
824 raw_spin_unlock_irqrestore(&object->lock, flags);
825 put_object(object);
826}
827
828/*
829 * Any surplus references (object already gray) to 'ptr' are passed to
830 * 'excess_ref'. This is used in the vmalloc() case where a pointer to
831 * vm_struct may be used as an alternative reference to the vmalloc'ed object
832 * (see free_thread_stack()).
833 */
834static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref)
835{
836 unsigned long flags;
837 struct kmemleak_object *object;
838
839 object = find_and_get_object(ptr, 0);
840 if (!object) {
841 kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n",
842 ptr);
843 return;
844 }
845
846 raw_spin_lock_irqsave(&object->lock, flags);
847 object->excess_ref = excess_ref;
848 raw_spin_unlock_irqrestore(&object->lock, flags);
849 put_object(object);
850}
851
852/*
853 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
854 * pointer. Such object will not be scanned by kmemleak but references to it
855 * are searched.
856 */
857static void object_no_scan(unsigned long ptr)
858{
859 unsigned long flags;
860 struct kmemleak_object *object;
861
862 object = find_and_get_object(ptr, 0);
863 if (!object) {
864 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
865 return;
866 }
867
868 raw_spin_lock_irqsave(&object->lock, flags);
869 object->flags |= OBJECT_NO_SCAN;
870 raw_spin_unlock_irqrestore(&object->lock, flags);
871 put_object(object);
872}
873
874/**
875 * kmemleak_alloc - register a newly allocated object
876 * @ptr: pointer to beginning of the object
877 * @size: size of the object
878 * @min_count: minimum number of references to this object. If during memory
879 * scanning a number of references less than @min_count is found,
880 * the object is reported as a memory leak. If @min_count is 0,
881 * the object is never reported as a leak. If @min_count is -1,
882 * the object is ignored (not scanned and not reported as a leak)
883 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
884 *
885 * This function is called from the kernel allocators when a new object
886 * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.).
887 */
888void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
889 gfp_t gfp)
890{
891 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
892
893 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
894 create_object((unsigned long)ptr, size, min_count, gfp);
895}
896EXPORT_SYMBOL_GPL(kmemleak_alloc);
897
898/**
899 * kmemleak_alloc_percpu - register a newly allocated __percpu object
900 * @ptr: __percpu pointer to beginning of the object
901 * @size: size of the object
902 * @gfp: flags used for kmemleak internal memory allocations
903 *
904 * This function is called from the kernel percpu allocator when a new object
905 * (memory block) is allocated (alloc_percpu).
906 */
907void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
908 gfp_t gfp)
909{
910 unsigned int cpu;
911
912 pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size);
913
914 /*
915 * Percpu allocations are only scanned and not reported as leaks
916 * (min_count is set to 0).
917 */
918 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
919 for_each_possible_cpu(cpu)
920 create_object((unsigned long)per_cpu_ptr(ptr, cpu),
921 size, 0, gfp);
922}
923EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
924
925/**
926 * kmemleak_vmalloc - register a newly vmalloc'ed object
927 * @area: pointer to vm_struct
928 * @size: size of the object
929 * @gfp: __vmalloc() flags used for kmemleak internal memory allocations
930 *
931 * This function is called from the vmalloc() kernel allocator when a new
932 * object (memory block) is allocated.
933 */
934void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp)
935{
936 pr_debug("%s(0x%p, %zu)\n", __func__, area, size);
937
938 /*
939 * A min_count = 2 is needed because vm_struct contains a reference to
940 * the virtual address of the vmalloc'ed block.
941 */
942 if (kmemleak_enabled) {
943 create_object((unsigned long)area->addr, size, 2, gfp);
944 object_set_excess_ref((unsigned long)area,
945 (unsigned long)area->addr);
946 }
947}
948EXPORT_SYMBOL_GPL(kmemleak_vmalloc);
949
950/**
951 * kmemleak_free - unregister a previously registered object
952 * @ptr: pointer to beginning of the object
953 *
954 * This function is called from the kernel allocators when an object (memory
955 * block) is freed (kmem_cache_free, kfree, vfree etc.).
956 */
957void __ref kmemleak_free(const void *ptr)
958{
959 pr_debug("%s(0x%p)\n", __func__, ptr);
960
961 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
962 delete_object_full((unsigned long)ptr);
963}
964EXPORT_SYMBOL_GPL(kmemleak_free);
965
966/**
967 * kmemleak_free_part - partially unregister a previously registered object
968 * @ptr: pointer to the beginning or inside the object. This also
969 * represents the start of the range to be freed
970 * @size: size to be unregistered
971 *
972 * This function is called when only a part of a memory block is freed
973 * (usually from the bootmem allocator).
974 */
975void __ref kmemleak_free_part(const void *ptr, size_t size)
976{
977 pr_debug("%s(0x%p)\n", __func__, ptr);
978
979 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
980 delete_object_part((unsigned long)ptr, size);
981}
982EXPORT_SYMBOL_GPL(kmemleak_free_part);
983
984/**
985 * kmemleak_free_percpu - unregister a previously registered __percpu object
986 * @ptr: __percpu pointer to beginning of the object
987 *
988 * This function is called from the kernel percpu allocator when an object
989 * (memory block) is freed (free_percpu).
990 */
991void __ref kmemleak_free_percpu(const void __percpu *ptr)
992{
993 unsigned int cpu;
994
995 pr_debug("%s(0x%p)\n", __func__, ptr);
996
997 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
998 for_each_possible_cpu(cpu)
999 delete_object_full((unsigned long)per_cpu_ptr(ptr,
1000 cpu));
1001}
1002EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
1003
1004/**
1005 * kmemleak_update_trace - update object allocation stack trace
1006 * @ptr: pointer to beginning of the object
1007 *
1008 * Override the object allocation stack trace for cases where the actual
1009 * allocation place is not always useful.
1010 */
1011void __ref kmemleak_update_trace(const void *ptr)
1012{
1013 struct kmemleak_object *object;
1014 unsigned long flags;
1015
1016 pr_debug("%s(0x%p)\n", __func__, ptr);
1017
1018 if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
1019 return;
1020
1021 object = find_and_get_object((unsigned long)ptr, 1);
1022 if (!object) {
1023#ifdef DEBUG
1024 kmemleak_warn("Updating stack trace for unknown object at %p\n",
1025 ptr);
1026#endif
1027 return;
1028 }
1029
1030 raw_spin_lock_irqsave(&object->lock, flags);
1031 object->trace_len = __save_stack_trace(object->trace);
1032 raw_spin_unlock_irqrestore(&object->lock, flags);
1033
1034 put_object(object);
1035}
1036EXPORT_SYMBOL(kmemleak_update_trace);
1037
1038/**
1039 * kmemleak_not_leak - mark an allocated object as false positive
1040 * @ptr: pointer to beginning of the object
1041 *
1042 * Calling this function on an object will cause the memory block to no longer
1043 * be reported as leak and always be scanned.
1044 */
1045void __ref kmemleak_not_leak(const void *ptr)
1046{
1047 pr_debug("%s(0x%p)\n", __func__, ptr);
1048
1049 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1050 make_gray_object((unsigned long)ptr);
1051}
1052EXPORT_SYMBOL(kmemleak_not_leak);
1053
1054/**
1055 * kmemleak_ignore - ignore an allocated object
1056 * @ptr: pointer to beginning of the object
1057 *
1058 * Calling this function on an object will cause the memory block to be
1059 * ignored (not scanned and not reported as a leak). This is usually done when
1060 * it is known that the corresponding block is not a leak and does not contain
1061 * any references to other allocated memory blocks.
1062 */
1063void __ref kmemleak_ignore(const void *ptr)
1064{
1065 pr_debug("%s(0x%p)\n", __func__, ptr);
1066
1067 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1068 make_black_object((unsigned long)ptr);
1069}
1070EXPORT_SYMBOL(kmemleak_ignore);
1071
1072/**
1073 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1074 * @ptr: pointer to beginning or inside the object. This also
1075 * represents the start of the scan area
1076 * @size: size of the scan area
1077 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1078 *
1079 * This function is used when it is known that only certain parts of an object
1080 * contain references to other objects. Kmemleak will only scan these areas
1081 * reducing the number false negatives.
1082 */
1083void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1084{
1085 pr_debug("%s(0x%p)\n", __func__, ptr);
1086
1087 if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
1088 add_scan_area((unsigned long)ptr, size, gfp);
1089}
1090EXPORT_SYMBOL(kmemleak_scan_area);
1091
1092/**
1093 * kmemleak_no_scan - do not scan an allocated object
1094 * @ptr: pointer to beginning of the object
1095 *
1096 * This function notifies kmemleak not to scan the given memory block. Useful
1097 * in situations where it is known that the given object does not contain any
1098 * references to other objects. Kmemleak will not scan such objects reducing
1099 * the number of false negatives.
1100 */
1101void __ref kmemleak_no_scan(const void *ptr)
1102{
1103 pr_debug("%s(0x%p)\n", __func__, ptr);
1104
1105 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1106 object_no_scan((unsigned long)ptr);
1107}
1108EXPORT_SYMBOL(kmemleak_no_scan);
1109
1110/**
1111 * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical
1112 * address argument
1113 * @phys: physical address of the object
1114 * @size: size of the object
1115 * @min_count: minimum number of references to this object.
1116 * See kmemleak_alloc()
1117 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1118 */
1119void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, int min_count,
1120 gfp_t gfp)
1121{
1122 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1123 kmemleak_alloc(__va(phys), size, min_count, gfp);
1124}
1125EXPORT_SYMBOL(kmemleak_alloc_phys);
1126
1127/**
1128 * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a
1129 * physical address argument
1130 * @phys: physical address if the beginning or inside an object. This
1131 * also represents the start of the range to be freed
1132 * @size: size to be unregistered
1133 */
1134void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size)
1135{
1136 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1137 kmemleak_free_part(__va(phys), size);
1138}
1139EXPORT_SYMBOL(kmemleak_free_part_phys);
1140
1141/**
1142 * kmemleak_not_leak_phys - similar to kmemleak_not_leak but taking a physical
1143 * address argument
1144 * @phys: physical address of the object
1145 */
1146void __ref kmemleak_not_leak_phys(phys_addr_t phys)
1147{
1148 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1149 kmemleak_not_leak(__va(phys));
1150}
1151EXPORT_SYMBOL(kmemleak_not_leak_phys);
1152
1153/**
1154 * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical
1155 * address argument
1156 * @phys: physical address of the object
1157 */
1158void __ref kmemleak_ignore_phys(phys_addr_t phys)
1159{
1160 if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
1161 kmemleak_ignore(__va(phys));
1162}
1163EXPORT_SYMBOL(kmemleak_ignore_phys);
1164
1165/*
1166 * Update an object's checksum and return true if it was modified.
1167 */
1168static bool update_checksum(struct kmemleak_object *object)
1169{
1170 u32 old_csum = object->checksum;
1171
1172 kasan_disable_current();
1173 kcsan_disable_current();
1174 object->checksum = crc32(0, kasan_reset_tag((void *)object->pointer), object->size);
1175 kasan_enable_current();
1176 kcsan_enable_current();
1177
1178 return object->checksum != old_csum;
1179}
1180
1181/*
1182 * Update an object's references. object->lock must be held by the caller.
1183 */
1184static void update_refs(struct kmemleak_object *object)
1185{
1186 if (!color_white(object)) {
1187 /* non-orphan, ignored or new */
1188 return;
1189 }
1190
1191 /*
1192 * Increase the object's reference count (number of pointers to the
1193 * memory block). If this count reaches the required minimum, the
1194 * object's color will become gray and it will be added to the
1195 * gray_list.
1196 */
1197 object->count++;
1198 if (color_gray(object)) {
1199 /* put_object() called when removing from gray_list */
1200 WARN_ON(!get_object(object));
1201 list_add_tail(&object->gray_list, &gray_list);
1202 }
1203}
1204
1205/*
1206 * Memory scanning is a long process and it needs to be interruptible. This
1207 * function checks whether such interrupt condition occurred.
1208 */
1209static int scan_should_stop(void)
1210{
1211 if (!kmemleak_enabled)
1212 return 1;
1213
1214 /*
1215 * This function may be called from either process or kthread context,
1216 * hence the need to check for both stop conditions.
1217 */
1218 if (current->mm)
1219 return signal_pending(current);
1220 else
1221 return kthread_should_stop();
1222
1223 return 0;
1224}
1225
1226/*
1227 * Scan a memory block (exclusive range) for valid pointers and add those
1228 * found to the gray list.
1229 */
1230static void scan_block(void *_start, void *_end,
1231 struct kmemleak_object *scanned)
1232{
1233 unsigned long *ptr;
1234 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1235 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1236 unsigned long flags;
1237 unsigned long untagged_ptr;
1238
1239 raw_spin_lock_irqsave(&kmemleak_lock, flags);
1240 for (ptr = start; ptr < end; ptr++) {
1241 struct kmemleak_object *object;
1242 unsigned long pointer;
1243 unsigned long excess_ref;
1244
1245 if (scan_should_stop())
1246 break;
1247
1248 kasan_disable_current();
1249 pointer = *(unsigned long *)kasan_reset_tag((void *)ptr);
1250 kasan_enable_current();
1251
1252 untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer);
1253 if (untagged_ptr < min_addr || untagged_ptr >= max_addr)
1254 continue;
1255
1256 /*
1257 * No need for get_object() here since we hold kmemleak_lock.
1258 * object->use_count cannot be dropped to 0 while the object
1259 * is still present in object_tree_root and object_list
1260 * (with updates protected by kmemleak_lock).
1261 */
1262 object = lookup_object(pointer, 1);
1263 if (!object)
1264 continue;
1265 if (object == scanned)
1266 /* self referenced, ignore */
1267 continue;
1268
1269 /*
1270 * Avoid the lockdep recursive warning on object->lock being
1271 * previously acquired in scan_object(). These locks are
1272 * enclosed by scan_mutex.
1273 */
1274 raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1275 /* only pass surplus references (object already gray) */
1276 if (color_gray(object)) {
1277 excess_ref = object->excess_ref;
1278 /* no need for update_refs() if object already gray */
1279 } else {
1280 excess_ref = 0;
1281 update_refs(object);
1282 }
1283 raw_spin_unlock(&object->lock);
1284
1285 if (excess_ref) {
1286 object = lookup_object(excess_ref, 0);
1287 if (!object)
1288 continue;
1289 if (object == scanned)
1290 /* circular reference, ignore */
1291 continue;
1292 raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1293 update_refs(object);
1294 raw_spin_unlock(&object->lock);
1295 }
1296 }
1297 raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
1298}
1299
1300/*
1301 * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
1302 */
1303#ifdef CONFIG_SMP
1304static void scan_large_block(void *start, void *end)
1305{
1306 void *next;
1307
1308 while (start < end) {
1309 next = min(start + MAX_SCAN_SIZE, end);
1310 scan_block(start, next, NULL);
1311 start = next;
1312 cond_resched();
1313 }
1314}
1315#endif
1316
1317/*
1318 * Scan a memory block corresponding to a kmemleak_object. A condition is
1319 * that object->use_count >= 1.
1320 */
1321static void scan_object(struct kmemleak_object *object)
1322{
1323 struct kmemleak_scan_area *area;
1324 unsigned long flags;
1325
1326 /*
1327 * Once the object->lock is acquired, the corresponding memory block
1328 * cannot be freed (the same lock is acquired in delete_object).
1329 */
1330 raw_spin_lock_irqsave(&object->lock, flags);
1331 if (object->flags & OBJECT_NO_SCAN)
1332 goto out;
1333 if (!(object->flags & OBJECT_ALLOCATED))
1334 /* already freed object */
1335 goto out;
1336 if (hlist_empty(&object->area_list) ||
1337 object->flags & OBJECT_FULL_SCAN) {
1338 void *start = (void *)object->pointer;
1339 void *end = (void *)(object->pointer + object->size);
1340 void *next;
1341
1342 do {
1343 next = min(start + MAX_SCAN_SIZE, end);
1344 scan_block(start, next, object);
1345
1346 start = next;
1347 if (start >= end)
1348 break;
1349
1350 raw_spin_unlock_irqrestore(&object->lock, flags);
1351 cond_resched();
1352 raw_spin_lock_irqsave(&object->lock, flags);
1353 } while (object->flags & OBJECT_ALLOCATED);
1354 } else
1355 hlist_for_each_entry(area, &object->area_list, node)
1356 scan_block((void *)area->start,
1357 (void *)(area->start + area->size),
1358 object);
1359out:
1360 raw_spin_unlock_irqrestore(&object->lock, flags);
1361}
1362
1363/*
1364 * Scan the objects already referenced (gray objects). More objects will be
1365 * referenced and, if there are no memory leaks, all the objects are scanned.
1366 */
1367static void scan_gray_list(void)
1368{
1369 struct kmemleak_object *object, *tmp;
1370
1371 /*
1372 * The list traversal is safe for both tail additions and removals
1373 * from inside the loop. The kmemleak objects cannot be freed from
1374 * outside the loop because their use_count was incremented.
1375 */
1376 object = list_entry(gray_list.next, typeof(*object), gray_list);
1377 while (&object->gray_list != &gray_list) {
1378 cond_resched();
1379
1380 /* may add new objects to the list */
1381 if (!scan_should_stop())
1382 scan_object(object);
1383
1384 tmp = list_entry(object->gray_list.next, typeof(*object),
1385 gray_list);
1386
1387 /* remove the object from the list and release it */
1388 list_del(&object->gray_list);
1389 put_object(object);
1390
1391 object = tmp;
1392 }
1393 WARN_ON(!list_empty(&gray_list));
1394}
1395
1396/*
1397 * Scan data sections and all the referenced memory blocks allocated via the
1398 * kernel's standard allocators. This function must be called with the
1399 * scan_mutex held.
1400 */
1401static void kmemleak_scan(void)
1402{
1403 unsigned long flags;
1404 struct kmemleak_object *object;
1405 int i;
1406 int new_leaks = 0;
1407
1408 jiffies_last_scan = jiffies;
1409
1410 /* prepare the kmemleak_object's */
1411 rcu_read_lock();
1412 list_for_each_entry_rcu(object, &object_list, object_list) {
1413 raw_spin_lock_irqsave(&object->lock, flags);
1414#ifdef DEBUG
1415 /*
1416 * With a few exceptions there should be a maximum of
1417 * 1 reference to any object at this point.
1418 */
1419 if (atomic_read(&object->use_count) > 1) {
1420 pr_debug("object->use_count = %d\n",
1421 atomic_read(&object->use_count));
1422 dump_object_info(object);
1423 }
1424#endif
1425 /* reset the reference count (whiten the object) */
1426 object->count = 0;
1427 if (color_gray(object) && get_object(object))
1428 list_add_tail(&object->gray_list, &gray_list);
1429
1430 raw_spin_unlock_irqrestore(&object->lock, flags);
1431 }
1432 rcu_read_unlock();
1433
1434#ifdef CONFIG_SMP
1435 /* per-cpu sections scanning */
1436 for_each_possible_cpu(i)
1437 scan_large_block(__per_cpu_start + per_cpu_offset(i),
1438 __per_cpu_end + per_cpu_offset(i));
1439#endif
1440
1441 /*
1442 * Struct page scanning for each node.
1443 */
1444 get_online_mems();
1445 for_each_online_node(i) {
1446 unsigned long start_pfn = node_start_pfn(i);
1447 unsigned long end_pfn = node_end_pfn(i);
1448 unsigned long pfn;
1449
1450 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1451 struct page *page = pfn_to_online_page(pfn);
1452
1453 if (!page)
1454 continue;
1455
1456 /* only scan pages belonging to this node */
1457 if (page_to_nid(page) != i)
1458 continue;
1459 /* only scan if page is in use */
1460 if (page_count(page) == 0)
1461 continue;
1462 scan_block(page, page + 1, NULL);
1463 if (!(pfn & 63))
1464 cond_resched();
1465 }
1466 }
1467 put_online_mems();
1468
1469 /*
1470 * Scanning the task stacks (may introduce false negatives).
1471 */
1472 if (kmemleak_stack_scan) {
1473 struct task_struct *p, *g;
1474
1475 rcu_read_lock();
1476 for_each_process_thread(g, p) {
1477 void *stack = try_get_task_stack(p);
1478 if (stack) {
1479 scan_block(stack, stack + THREAD_SIZE, NULL);
1480 put_task_stack(p);
1481 }
1482 }
1483 rcu_read_unlock();
1484 }
1485
1486 /*
1487 * Scan the objects already referenced from the sections scanned
1488 * above.
1489 */
1490 scan_gray_list();
1491
1492 /*
1493 * Check for new or unreferenced objects modified since the previous
1494 * scan and color them gray until the next scan.
1495 */
1496 rcu_read_lock();
1497 list_for_each_entry_rcu(object, &object_list, object_list) {
1498 raw_spin_lock_irqsave(&object->lock, flags);
1499 if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1500 && update_checksum(object) && get_object(object)) {
1501 /* color it gray temporarily */
1502 object->count = object->min_count;
1503 list_add_tail(&object->gray_list, &gray_list);
1504 }
1505 raw_spin_unlock_irqrestore(&object->lock, flags);
1506 }
1507 rcu_read_unlock();
1508
1509 /*
1510 * Re-scan the gray list for modified unreferenced objects.
1511 */
1512 scan_gray_list();
1513
1514 /*
1515 * If scanning was stopped do not report any new unreferenced objects.
1516 */
1517 if (scan_should_stop())
1518 return;
1519
1520 /*
1521 * Scanning result reporting.
1522 */
1523 rcu_read_lock();
1524 list_for_each_entry_rcu(object, &object_list, object_list) {
1525 raw_spin_lock_irqsave(&object->lock, flags);
1526 if (unreferenced_object(object) &&
1527 !(object->flags & OBJECT_REPORTED)) {
1528 object->flags |= OBJECT_REPORTED;
1529
1530 if (kmemleak_verbose)
1531 print_unreferenced(NULL, object);
1532
1533 new_leaks++;
1534 }
1535 raw_spin_unlock_irqrestore(&object->lock, flags);
1536 }
1537 rcu_read_unlock();
1538
1539 if (new_leaks) {
1540 kmemleak_found_leaks = true;
1541
1542 pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
1543 new_leaks);
1544 }
1545
1546}
1547
1548/*
1549 * Thread function performing automatic memory scanning. Unreferenced objects
1550 * at the end of a memory scan are reported but only the first time.
1551 */
1552static int kmemleak_scan_thread(void *arg)
1553{
1554 static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN);
1555
1556 pr_info("Automatic memory scanning thread started\n");
1557 set_user_nice(current, 10);
1558
1559 /*
1560 * Wait before the first scan to allow the system to fully initialize.
1561 */
1562 if (first_run) {
1563 signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000);
1564 first_run = 0;
1565 while (timeout && !kthread_should_stop())
1566 timeout = schedule_timeout_interruptible(timeout);
1567 }
1568
1569 while (!kthread_should_stop()) {
1570 signed long timeout = READ_ONCE(jiffies_scan_wait);
1571
1572 mutex_lock(&scan_mutex);
1573 kmemleak_scan();
1574 mutex_unlock(&scan_mutex);
1575
1576 /* wait before the next scan */
1577 while (timeout && !kthread_should_stop())
1578 timeout = schedule_timeout_interruptible(timeout);
1579 }
1580
1581 pr_info("Automatic memory scanning thread ended\n");
1582
1583 return 0;
1584}
1585
1586/*
1587 * Start the automatic memory scanning thread. This function must be called
1588 * with the scan_mutex held.
1589 */
1590static void start_scan_thread(void)
1591{
1592 if (scan_thread)
1593 return;
1594 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1595 if (IS_ERR(scan_thread)) {
1596 pr_warn("Failed to create the scan thread\n");
1597 scan_thread = NULL;
1598 }
1599}
1600
1601/*
1602 * Stop the automatic memory scanning thread.
1603 */
1604static void stop_scan_thread(void)
1605{
1606 if (scan_thread) {
1607 kthread_stop(scan_thread);
1608 scan_thread = NULL;
1609 }
1610}
1611
1612/*
1613 * Iterate over the object_list and return the first valid object at or after
1614 * the required position with its use_count incremented. The function triggers
1615 * a memory scanning when the pos argument points to the first position.
1616 */
1617static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1618{
1619 struct kmemleak_object *object;
1620 loff_t n = *pos;
1621 int err;
1622
1623 err = mutex_lock_interruptible(&scan_mutex);
1624 if (err < 0)
1625 return ERR_PTR(err);
1626
1627 rcu_read_lock();
1628 list_for_each_entry_rcu(object, &object_list, object_list) {
1629 if (n-- > 0)
1630 continue;
1631 if (get_object(object))
1632 goto out;
1633 }
1634 object = NULL;
1635out:
1636 return object;
1637}
1638
1639/*
1640 * Return the next object in the object_list. The function decrements the
1641 * use_count of the previous object and increases that of the next one.
1642 */
1643static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1644{
1645 struct kmemleak_object *prev_obj = v;
1646 struct kmemleak_object *next_obj = NULL;
1647 struct kmemleak_object *obj = prev_obj;
1648
1649 ++(*pos);
1650
1651 list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
1652 if (get_object(obj)) {
1653 next_obj = obj;
1654 break;
1655 }
1656 }
1657
1658 put_object(prev_obj);
1659 return next_obj;
1660}
1661
1662/*
1663 * Decrement the use_count of the last object required, if any.
1664 */
1665static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1666{
1667 if (!IS_ERR(v)) {
1668 /*
1669 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1670 * waiting was interrupted, so only release it if !IS_ERR.
1671 */
1672 rcu_read_unlock();
1673 mutex_unlock(&scan_mutex);
1674 if (v)
1675 put_object(v);
1676 }
1677}
1678
1679/*
1680 * Print the information for an unreferenced object to the seq file.
1681 */
1682static int kmemleak_seq_show(struct seq_file *seq, void *v)
1683{
1684 struct kmemleak_object *object = v;
1685 unsigned long flags;
1686
1687 raw_spin_lock_irqsave(&object->lock, flags);
1688 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1689 print_unreferenced(seq, object);
1690 raw_spin_unlock_irqrestore(&object->lock, flags);
1691 return 0;
1692}
1693
1694static const struct seq_operations kmemleak_seq_ops = {
1695 .start = kmemleak_seq_start,
1696 .next = kmemleak_seq_next,
1697 .stop = kmemleak_seq_stop,
1698 .show = kmemleak_seq_show,
1699};
1700
1701static int kmemleak_open(struct inode *inode, struct file *file)
1702{
1703 return seq_open(file, &kmemleak_seq_ops);
1704}
1705
1706static int dump_str_object_info(const char *str)
1707{
1708 unsigned long flags;
1709 struct kmemleak_object *object;
1710 unsigned long addr;
1711
1712 if (kstrtoul(str, 0, &addr))
1713 return -EINVAL;
1714 object = find_and_get_object(addr, 0);
1715 if (!object) {
1716 pr_info("Unknown object at 0x%08lx\n", addr);
1717 return -EINVAL;
1718 }
1719
1720 raw_spin_lock_irqsave(&object->lock, flags);
1721 dump_object_info(object);
1722 raw_spin_unlock_irqrestore(&object->lock, flags);
1723
1724 put_object(object);
1725 return 0;
1726}
1727
1728/*
1729 * We use grey instead of black to ensure we can do future scans on the same
1730 * objects. If we did not do future scans these black objects could
1731 * potentially contain references to newly allocated objects in the future and
1732 * we'd end up with false positives.
1733 */
1734static void kmemleak_clear(void)
1735{
1736 struct kmemleak_object *object;
1737 unsigned long flags;
1738
1739 rcu_read_lock();
1740 list_for_each_entry_rcu(object, &object_list, object_list) {
1741 raw_spin_lock_irqsave(&object->lock, flags);
1742 if ((object->flags & OBJECT_REPORTED) &&
1743 unreferenced_object(object))
1744 __paint_it(object, KMEMLEAK_GREY);
1745 raw_spin_unlock_irqrestore(&object->lock, flags);
1746 }
1747 rcu_read_unlock();
1748
1749 kmemleak_found_leaks = false;
1750}
1751
1752static void __kmemleak_do_cleanup(void);
1753
1754/*
1755 * File write operation to configure kmemleak at run-time. The following
1756 * commands can be written to the /sys/kernel/debug/kmemleak file:
1757 * off - disable kmemleak (irreversible)
1758 * stack=on - enable the task stacks scanning
1759 * stack=off - disable the tasks stacks scanning
1760 * scan=on - start the automatic memory scanning thread
1761 * scan=off - stop the automatic memory scanning thread
1762 * scan=... - set the automatic memory scanning period in seconds (0 to
1763 * disable it)
1764 * scan - trigger a memory scan
1765 * clear - mark all current reported unreferenced kmemleak objects as
1766 * grey to ignore printing them, or free all kmemleak objects
1767 * if kmemleak has been disabled.
1768 * dump=... - dump information about the object found at the given address
1769 */
1770static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1771 size_t size, loff_t *ppos)
1772{
1773 char buf[64];
1774 int buf_size;
1775 int ret;
1776
1777 buf_size = min(size, (sizeof(buf) - 1));
1778 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1779 return -EFAULT;
1780 buf[buf_size] = 0;
1781
1782 ret = mutex_lock_interruptible(&scan_mutex);
1783 if (ret < 0)
1784 return ret;
1785
1786 if (strncmp(buf, "clear", 5) == 0) {
1787 if (kmemleak_enabled)
1788 kmemleak_clear();
1789 else
1790 __kmemleak_do_cleanup();
1791 goto out;
1792 }
1793
1794 if (!kmemleak_enabled) {
1795 ret = -EPERM;
1796 goto out;
1797 }
1798
1799 if (strncmp(buf, "off", 3) == 0)
1800 kmemleak_disable();
1801 else if (strncmp(buf, "stack=on", 8) == 0)
1802 kmemleak_stack_scan = 1;
1803 else if (strncmp(buf, "stack=off", 9) == 0)
1804 kmemleak_stack_scan = 0;
1805 else if (strncmp(buf, "scan=on", 7) == 0)
1806 start_scan_thread();
1807 else if (strncmp(buf, "scan=off", 8) == 0)
1808 stop_scan_thread();
1809 else if (strncmp(buf, "scan=", 5) == 0) {
1810 unsigned secs;
1811 unsigned long msecs;
1812
1813 ret = kstrtouint(buf + 5, 0, &secs);
1814 if (ret < 0)
1815 goto out;
1816
1817 msecs = secs * MSEC_PER_SEC;
1818 if (msecs > UINT_MAX)
1819 msecs = UINT_MAX;
1820
1821 stop_scan_thread();
1822 if (msecs) {
1823 WRITE_ONCE(jiffies_scan_wait, msecs_to_jiffies(msecs));
1824 start_scan_thread();
1825 }
1826 } else if (strncmp(buf, "scan", 4) == 0)
1827 kmemleak_scan();
1828 else if (strncmp(buf, "dump=", 5) == 0)
1829 ret = dump_str_object_info(buf + 5);
1830 else
1831 ret = -EINVAL;
1832
1833out:
1834 mutex_unlock(&scan_mutex);
1835 if (ret < 0)
1836 return ret;
1837
1838 /* ignore the rest of the buffer, only one command at a time */
1839 *ppos += size;
1840 return size;
1841}
1842
1843static const struct file_operations kmemleak_fops = {
1844 .owner = THIS_MODULE,
1845 .open = kmemleak_open,
1846 .read = seq_read,
1847 .write = kmemleak_write,
1848 .llseek = seq_lseek,
1849 .release = seq_release,
1850};
1851
1852static void __kmemleak_do_cleanup(void)
1853{
1854 struct kmemleak_object *object, *tmp;
1855
1856 /*
1857 * Kmemleak has already been disabled, no need for RCU list traversal
1858 * or kmemleak_lock held.
1859 */
1860 list_for_each_entry_safe(object, tmp, &object_list, object_list) {
1861 __remove_object(object);
1862 __delete_object(object);
1863 }
1864}
1865
1866/*
1867 * Stop the memory scanning thread and free the kmemleak internal objects if
1868 * no previous scan thread (otherwise, kmemleak may still have some useful
1869 * information on memory leaks).
1870 */
1871static void kmemleak_do_cleanup(struct work_struct *work)
1872{
1873 stop_scan_thread();
1874
1875 mutex_lock(&scan_mutex);
1876 /*
1877 * Once it is made sure that kmemleak_scan has stopped, it is safe to no
1878 * longer track object freeing. Ordering of the scan thread stopping and
1879 * the memory accesses below is guaranteed by the kthread_stop()
1880 * function.
1881 */
1882 kmemleak_free_enabled = 0;
1883 mutex_unlock(&scan_mutex);
1884
1885 if (!kmemleak_found_leaks)
1886 __kmemleak_do_cleanup();
1887 else
1888 pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
1889}
1890
1891static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
1892
1893/*
1894 * Disable kmemleak. No memory allocation/freeing will be traced once this
1895 * function is called. Disabling kmemleak is an irreversible operation.
1896 */
1897static void kmemleak_disable(void)
1898{
1899 /* atomically check whether it was already invoked */
1900 if (cmpxchg(&kmemleak_error, 0, 1))
1901 return;
1902
1903 /* stop any memory operation tracing */
1904 kmemleak_enabled = 0;
1905
1906 /* check whether it is too early for a kernel thread */
1907 if (kmemleak_initialized)
1908 schedule_work(&cleanup_work);
1909 else
1910 kmemleak_free_enabled = 0;
1911
1912 pr_info("Kernel memory leak detector disabled\n");
1913}
1914
1915/*
1916 * Allow boot-time kmemleak disabling (enabled by default).
1917 */
1918static int __init kmemleak_boot_config(char *str)
1919{
1920 if (!str)
1921 return -EINVAL;
1922 if (strcmp(str, "off") == 0)
1923 kmemleak_disable();
1924 else if (strcmp(str, "on") == 0)
1925 kmemleak_skip_disable = 1;
1926 else
1927 return -EINVAL;
1928 return 0;
1929}
1930early_param("kmemleak", kmemleak_boot_config);
1931
1932/*
1933 * Kmemleak initialization.
1934 */
1935void __init kmemleak_init(void)
1936{
1937#ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
1938 if (!kmemleak_skip_disable) {
1939 kmemleak_disable();
1940 return;
1941 }
1942#endif
1943
1944 if (kmemleak_error)
1945 return;
1946
1947 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1948 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1949
1950 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1951 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1952
1953 /* register the data/bss sections */
1954 create_object((unsigned long)_sdata, _edata - _sdata,
1955 KMEMLEAK_GREY, GFP_ATOMIC);
1956 create_object((unsigned long)__bss_start, __bss_stop - __bss_start,
1957 KMEMLEAK_GREY, GFP_ATOMIC);
1958 /* only register .data..ro_after_init if not within .data */
1959 if (&__start_ro_after_init < &_sdata || &__end_ro_after_init > &_edata)
1960 create_object((unsigned long)__start_ro_after_init,
1961 __end_ro_after_init - __start_ro_after_init,
1962 KMEMLEAK_GREY, GFP_ATOMIC);
1963}
1964
1965/*
1966 * Late initialization function.
1967 */
1968static int __init kmemleak_late_init(void)
1969{
1970 kmemleak_initialized = 1;
1971
1972 debugfs_create_file("kmemleak", 0644, NULL, NULL, &kmemleak_fops);
1973
1974 if (kmemleak_error) {
1975 /*
1976 * Some error occurred and kmemleak was disabled. There is a
1977 * small chance that kmemleak_disable() was called immediately
1978 * after setting kmemleak_initialized and we may end up with
1979 * two clean-up threads but serialized by scan_mutex.
1980 */
1981 schedule_work(&cleanup_work);
1982 return -ENOMEM;
1983 }
1984
1985 if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) {
1986 mutex_lock(&scan_mutex);
1987 start_scan_thread();
1988 mutex_unlock(&scan_mutex);
1989 }
1990
1991 pr_info("Kernel memory leak detector initialized (mem pool available: %d)\n",
1992 mem_pool_free_count);
1993
1994 return 0;
1995}
1996late_initcall(kmemleak_late_init);