]> bbs.cooldavid.org Git - net-next-2.6.git/blame - mm/kmemleak.c
kmemleak: Do not report alloc_bootmem blocks as leaks
[net-next-2.6.git] / mm / kmemleak.c
CommitLineData
3c7b4e6b
CM
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/kmemleak.txt.
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 priority search 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
4698c1f2
CM
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
3c7b4e6b
CM
55 *
56 * The kmemleak_object structures have a use_count incremented or decremented
57 * using the get_object()/put_object() functions. When the use_count becomes
58 * 0, this count can no longer be incremented and put_object() schedules the
59 * kmemleak_object freeing via an RCU callback. All calls to the get_object()
60 * function must be protected by rcu_read_lock() to avoid accessing a freed
61 * structure.
62 */
63
ae281064
JP
64#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
65
3c7b4e6b
CM
66#include <linux/init.h>
67#include <linux/kernel.h>
68#include <linux/list.h>
69#include <linux/sched.h>
70#include <linux/jiffies.h>
71#include <linux/delay.h>
72#include <linux/module.h>
73#include <linux/kthread.h>
74#include <linux/prio_tree.h>
75#include <linux/gfp.h>
76#include <linux/fs.h>
77#include <linux/debugfs.h>
78#include <linux/seq_file.h>
79#include <linux/cpumask.h>
80#include <linux/spinlock.h>
81#include <linux/mutex.h>
82#include <linux/rcupdate.h>
83#include <linux/stacktrace.h>
84#include <linux/cache.h>
85#include <linux/percpu.h>
86#include <linux/hardirq.h>
87#include <linux/mmzone.h>
88#include <linux/slab.h>
89#include <linux/thread_info.h>
90#include <linux/err.h>
91#include <linux/uaccess.h>
92#include <linux/string.h>
93#include <linux/nodemask.h>
94#include <linux/mm.h>
95
96#include <asm/sections.h>
97#include <asm/processor.h>
98#include <asm/atomic.h>
99
100#include <linux/kmemleak.h>
101
102/*
103 * Kmemleak configuration and common defines.
104 */
105#define MAX_TRACE 16 /* stack trace length */
3c7b4e6b 106#define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
3c7b4e6b
CM
107#define SECS_FIRST_SCAN 60 /* delay before the first scan */
108#define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
2587362e 109#define GRAY_LIST_PASSES 25 /* maximum number of gray list scans */
af98603d 110#define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
3c7b4e6b
CM
111
112#define BYTES_PER_POINTER sizeof(void *)
113
216c04b0
CM
114/* GFP bitmask for kmemleak internal allocations */
115#define GFP_KMEMLEAK_MASK (GFP_KERNEL | GFP_ATOMIC)
116
3c7b4e6b
CM
117/* scanning area inside a memory block */
118struct kmemleak_scan_area {
119 struct hlist_node node;
120 unsigned long offset;
121 size_t length;
122};
123
124/*
125 * Structure holding the metadata for each allocated memory block.
126 * Modifications to such objects should be made while holding the
127 * object->lock. Insertions or deletions from object_list, gray_list or
128 * tree_node are already protected by the corresponding locks or mutex (see
129 * the notes on locking above). These objects are reference-counted
130 * (use_count) and freed using the RCU mechanism.
131 */
132struct kmemleak_object {
133 spinlock_t lock;
134 unsigned long flags; /* object status flags */
135 struct list_head object_list;
136 struct list_head gray_list;
137 struct prio_tree_node tree_node;
138 struct rcu_head rcu; /* object_list lockless traversal */
139 /* object usage count; object freed when use_count == 0 */
140 atomic_t use_count;
141 unsigned long pointer;
142 size_t size;
143 /* minimum number of a pointers found before it is considered leak */
144 int min_count;
145 /* the total number of pointers found pointing to this object */
146 int count;
147 /* memory ranges to be scanned inside an object (empty for all) */
148 struct hlist_head area_list;
149 unsigned long trace[MAX_TRACE];
150 unsigned int trace_len;
151 unsigned long jiffies; /* creation timestamp */
152 pid_t pid; /* pid of the current task */
153 char comm[TASK_COMM_LEN]; /* executable name */
154};
155
156/* flag representing the memory block allocation status */
157#define OBJECT_ALLOCATED (1 << 0)
158/* flag set after the first reporting of an unreference object */
159#define OBJECT_REPORTED (1 << 1)
160/* flag set to not scan the object */
161#define OBJECT_NO_SCAN (1 << 2)
2587362e
CM
162/* flag set on newly allocated objects */
163#define OBJECT_NEW (1 << 3)
3c7b4e6b
CM
164
165/* the list of all allocated objects */
166static LIST_HEAD(object_list);
167/* the list of gray-colored objects (see color_gray comment below) */
168static LIST_HEAD(gray_list);
169/* prio search tree for object boundaries */
170static struct prio_tree_root object_tree_root;
171/* rw_lock protecting the access to object_list and prio_tree_root */
172static DEFINE_RWLOCK(kmemleak_lock);
173
174/* allocation caches for kmemleak internal data */
175static struct kmem_cache *object_cache;
176static struct kmem_cache *scan_area_cache;
177
178/* set if tracing memory operations is enabled */
179static atomic_t kmemleak_enabled = ATOMIC_INIT(0);
180/* set in the late_initcall if there were no errors */
181static atomic_t kmemleak_initialized = ATOMIC_INIT(0);
182/* enables or disables early logging of the memory operations */
183static atomic_t kmemleak_early_log = ATOMIC_INIT(1);
184/* set if a fata kmemleak error has occurred */
185static atomic_t kmemleak_error = ATOMIC_INIT(0);
186
187/* minimum and maximum address that may be valid pointers */
188static unsigned long min_addr = ULONG_MAX;
189static unsigned long max_addr;
190
3c7b4e6b 191static struct task_struct *scan_thread;
acf4968e 192/* used to avoid reporting of recently allocated objects */
3c7b4e6b 193static unsigned long jiffies_min_age;
acf4968e 194static unsigned long jiffies_last_scan;
3c7b4e6b
CM
195/* delay between automatic memory scannings */
196static signed long jiffies_scan_wait;
197/* enables or disables the task stacks scanning */
e0a2a160 198static int kmemleak_stack_scan = 1;
4698c1f2 199/* protects the memory scanning, parameters and debug/kmemleak file access */
3c7b4e6b 200static DEFINE_MUTEX(scan_mutex);
3c7b4e6b 201
3c7b4e6b 202/*
2030117d 203 * Early object allocation/freeing logging. Kmemleak is initialized after the
3c7b4e6b 204 * kernel allocator. However, both the kernel allocator and kmemleak may
2030117d 205 * allocate memory blocks which need to be tracked. Kmemleak defines an
3c7b4e6b
CM
206 * arbitrary buffer to hold the allocation/freeing information before it is
207 * fully initialized.
208 */
209
210/* kmemleak operation type for early logging */
211enum {
212 KMEMLEAK_ALLOC,
213 KMEMLEAK_FREE,
53238a60 214 KMEMLEAK_FREE_PART,
3c7b4e6b
CM
215 KMEMLEAK_NOT_LEAK,
216 KMEMLEAK_IGNORE,
217 KMEMLEAK_SCAN_AREA,
218 KMEMLEAK_NO_SCAN
219};
220
221/*
222 * Structure holding the information passed to kmemleak callbacks during the
223 * early logging.
224 */
225struct early_log {
226 int op_type; /* kmemleak operation type */
227 const void *ptr; /* allocated/freed memory block */
228 size_t size; /* memory block size */
229 int min_count; /* minimum reference count */
230 unsigned long offset; /* scan area offset */
231 size_t length; /* scan area length */
fd678967
CM
232 unsigned long trace[MAX_TRACE]; /* stack trace */
233 unsigned int trace_len; /* stack trace length */
3c7b4e6b
CM
234};
235
236/* early logging buffer and current position */
a6186d89
CM
237static struct early_log
238 early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata;
239static int crt_early_log __initdata;
3c7b4e6b
CM
240
241static void kmemleak_disable(void);
242
243/*
244 * Print a warning and dump the stack trace.
245 */
246#define kmemleak_warn(x...) do { \
247 pr_warning(x); \
248 dump_stack(); \
249} while (0)
250
251/*
252 * Macro invoked when a serious kmemleak condition occured and cannot be
2030117d 253 * recovered from. Kmemleak will be disabled and further allocation/freeing
3c7b4e6b
CM
254 * tracing no longer available.
255 */
000814f4 256#define kmemleak_stop(x...) do { \
3c7b4e6b
CM
257 kmemleak_warn(x); \
258 kmemleak_disable(); \
259} while (0)
260
261/*
262 * Object colors, encoded with count and min_count:
263 * - white - orphan object, not enough references to it (count < min_count)
264 * - gray - not orphan, not marked as false positive (min_count == 0) or
265 * sufficient references to it (count >= min_count)
266 * - black - ignore, it doesn't contain references (e.g. text section)
267 * (min_count == -1). No function defined for this color.
268 * Newly created objects don't have any color assigned (object->count == -1)
269 * before the next memory scan when they become white.
270 */
271static int color_white(const struct kmemleak_object *object)
272{
273 return object->count != -1 && object->count < object->min_count;
274}
275
276static int color_gray(const struct kmemleak_object *object)
277{
278 return object->min_count != -1 && object->count >= object->min_count;
279}
280
2587362e
CM
281static int color_black(const struct kmemleak_object *object)
282{
283 return object->min_count == -1;
284}
285
3c7b4e6b
CM
286/*
287 * Objects are considered unreferenced only if their color is white, they have
288 * not be deleted and have a minimum age to avoid false positives caused by
289 * pointers temporarily stored in CPU registers.
290 */
291static int unreferenced_object(struct kmemleak_object *object)
292{
293 return (object->flags & OBJECT_ALLOCATED) && color_white(object) &&
acf4968e
CM
294 time_before_eq(object->jiffies + jiffies_min_age,
295 jiffies_last_scan);
3c7b4e6b
CM
296}
297
298/*
bab4a34a
CM
299 * Printing of the unreferenced objects information to the seq file. The
300 * print_unreferenced function must be called with the object->lock held.
3c7b4e6b 301 */
3c7b4e6b
CM
302static void print_unreferenced(struct seq_file *seq,
303 struct kmemleak_object *object)
304{
305 int i;
306
bab4a34a
CM
307 seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
308 object->pointer, object->size);
309 seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu\n",
310 object->comm, object->pid, object->jiffies);
311 seq_printf(seq, " backtrace:\n");
3c7b4e6b
CM
312
313 for (i = 0; i < object->trace_len; i++) {
314 void *ptr = (void *)object->trace[i];
bab4a34a 315 seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
3c7b4e6b
CM
316 }
317}
318
319/*
320 * Print the kmemleak_object information. This function is used mainly for
321 * debugging special cases when kmemleak operations. It must be called with
322 * the object->lock held.
323 */
324static void dump_object_info(struct kmemleak_object *object)
325{
326 struct stack_trace trace;
327
328 trace.nr_entries = object->trace_len;
329 trace.entries = object->trace;
330
ae281064 331 pr_notice("Object 0x%08lx (size %zu):\n",
3c7b4e6b
CM
332 object->tree_node.start, object->size);
333 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
334 object->comm, object->pid, object->jiffies);
335 pr_notice(" min_count = %d\n", object->min_count);
336 pr_notice(" count = %d\n", object->count);
189d84ed 337 pr_notice(" flags = 0x%lx\n", object->flags);
3c7b4e6b
CM
338 pr_notice(" backtrace:\n");
339 print_stack_trace(&trace, 4);
340}
341
342/*
343 * Look-up a memory block metadata (kmemleak_object) in the priority search
344 * tree based on a pointer value. If alias is 0, only values pointing to the
345 * beginning of the memory block are allowed. The kmemleak_lock must be held
346 * when calling this function.
347 */
348static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
349{
350 struct prio_tree_node *node;
351 struct prio_tree_iter iter;
352 struct kmemleak_object *object;
353
354 prio_tree_iter_init(&iter, &object_tree_root, ptr, ptr);
355 node = prio_tree_next(&iter);
356 if (node) {
357 object = prio_tree_entry(node, struct kmemleak_object,
358 tree_node);
359 if (!alias && object->pointer != ptr) {
ae281064 360 kmemleak_warn("Found object by alias");
3c7b4e6b
CM
361 object = NULL;
362 }
363 } else
364 object = NULL;
365
366 return object;
367}
368
369/*
370 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
371 * that once an object's use_count reached 0, the RCU freeing was already
372 * registered and the object should no longer be used. This function must be
373 * called under the protection of rcu_read_lock().
374 */
375static int get_object(struct kmemleak_object *object)
376{
377 return atomic_inc_not_zero(&object->use_count);
378}
379
380/*
381 * RCU callback to free a kmemleak_object.
382 */
383static void free_object_rcu(struct rcu_head *rcu)
384{
385 struct hlist_node *elem, *tmp;
386 struct kmemleak_scan_area *area;
387 struct kmemleak_object *object =
388 container_of(rcu, struct kmemleak_object, rcu);
389
390 /*
391 * Once use_count is 0 (guaranteed by put_object), there is no other
392 * code accessing this object, hence no need for locking.
393 */
394 hlist_for_each_entry_safe(area, elem, tmp, &object->area_list, node) {
395 hlist_del(elem);
396 kmem_cache_free(scan_area_cache, area);
397 }
398 kmem_cache_free(object_cache, object);
399}
400
401/*
402 * Decrement the object use_count. Once the count is 0, free the object using
403 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
404 * delete_object() path, the delayed RCU freeing ensures that there is no
405 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
406 * is also possible.
407 */
408static void put_object(struct kmemleak_object *object)
409{
410 if (!atomic_dec_and_test(&object->use_count))
411 return;
412
413 /* should only get here after delete_object was called */
414 WARN_ON(object->flags & OBJECT_ALLOCATED);
415
416 call_rcu(&object->rcu, free_object_rcu);
417}
418
419/*
420 * Look up an object in the prio search tree and increase its use_count.
421 */
422static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
423{
424 unsigned long flags;
425 struct kmemleak_object *object = NULL;
426
427 rcu_read_lock();
428 read_lock_irqsave(&kmemleak_lock, flags);
429 if (ptr >= min_addr && ptr < max_addr)
430 object = lookup_object(ptr, alias);
431 read_unlock_irqrestore(&kmemleak_lock, flags);
432
433 /* check whether the object is still available */
434 if (object && !get_object(object))
435 object = NULL;
436 rcu_read_unlock();
437
438 return object;
439}
440
fd678967
CM
441/*
442 * Save stack trace to the given array of MAX_TRACE size.
443 */
444static int __save_stack_trace(unsigned long *trace)
445{
446 struct stack_trace stack_trace;
447
448 stack_trace.max_entries = MAX_TRACE;
449 stack_trace.nr_entries = 0;
450 stack_trace.entries = trace;
451 stack_trace.skip = 2;
452 save_stack_trace(&stack_trace);
453
454 return stack_trace.nr_entries;
455}
456
3c7b4e6b
CM
457/*
458 * Create the metadata (struct kmemleak_object) corresponding to an allocated
459 * memory block and add it to the object_list and object_tree_root.
460 */
fd678967
CM
461static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
462 int min_count, gfp_t gfp)
3c7b4e6b
CM
463{
464 unsigned long flags;
465 struct kmemleak_object *object;
466 struct prio_tree_node *node;
3c7b4e6b 467
216c04b0 468 object = kmem_cache_alloc(object_cache, gfp & GFP_KMEMLEAK_MASK);
3c7b4e6b 469 if (!object) {
ae281064 470 kmemleak_stop("Cannot allocate a kmemleak_object structure\n");
fd678967 471 return NULL;
3c7b4e6b
CM
472 }
473
474 INIT_LIST_HEAD(&object->object_list);
475 INIT_LIST_HEAD(&object->gray_list);
476 INIT_HLIST_HEAD(&object->area_list);
477 spin_lock_init(&object->lock);
478 atomic_set(&object->use_count, 1);
2587362e 479 object->flags = OBJECT_ALLOCATED | OBJECT_NEW;
3c7b4e6b
CM
480 object->pointer = ptr;
481 object->size = size;
482 object->min_count = min_count;
483 object->count = -1; /* no color initially */
484 object->jiffies = jiffies;
485
486 /* task information */
487 if (in_irq()) {
488 object->pid = 0;
489 strncpy(object->comm, "hardirq", sizeof(object->comm));
490 } else if (in_softirq()) {
491 object->pid = 0;
492 strncpy(object->comm, "softirq", sizeof(object->comm));
493 } else {
494 object->pid = current->pid;
495 /*
496 * There is a small chance of a race with set_task_comm(),
497 * however using get_task_comm() here may cause locking
498 * dependency issues with current->alloc_lock. In the worst
499 * case, the command line is not correct.
500 */
501 strncpy(object->comm, current->comm, sizeof(object->comm));
502 }
503
504 /* kernel backtrace */
fd678967 505 object->trace_len = __save_stack_trace(object->trace);
3c7b4e6b
CM
506
507 INIT_PRIO_TREE_NODE(&object->tree_node);
508 object->tree_node.start = ptr;
509 object->tree_node.last = ptr + size - 1;
510
511 write_lock_irqsave(&kmemleak_lock, flags);
512 min_addr = min(min_addr, ptr);
513 max_addr = max(max_addr, ptr + size);
514 node = prio_tree_insert(&object_tree_root, &object->tree_node);
515 /*
516 * The code calling the kernel does not yet have the pointer to the
517 * memory block to be able to free it. However, we still hold the
518 * kmemleak_lock here in case parts of the kernel started freeing
519 * random memory blocks.
520 */
521 if (node != &object->tree_node) {
522 unsigned long flags;
523
ae281064
JP
524 kmemleak_stop("Cannot insert 0x%lx into the object search tree "
525 "(already existing)\n", ptr);
3c7b4e6b
CM
526 object = lookup_object(ptr, 1);
527 spin_lock_irqsave(&object->lock, flags);
528 dump_object_info(object);
529 spin_unlock_irqrestore(&object->lock, flags);
530
531 goto out;
532 }
533 list_add_tail_rcu(&object->object_list, &object_list);
534out:
535 write_unlock_irqrestore(&kmemleak_lock, flags);
fd678967 536 return object;
3c7b4e6b
CM
537}
538
539/*
540 * Remove the metadata (struct kmemleak_object) for a memory block from the
541 * object_list and object_tree_root and decrement its use_count.
542 */
53238a60 543static void __delete_object(struct kmemleak_object *object)
3c7b4e6b
CM
544{
545 unsigned long flags;
3c7b4e6b
CM
546
547 write_lock_irqsave(&kmemleak_lock, flags);
3c7b4e6b
CM
548 prio_tree_remove(&object_tree_root, &object->tree_node);
549 list_del_rcu(&object->object_list);
550 write_unlock_irqrestore(&kmemleak_lock, flags);
551
552 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
53238a60 553 WARN_ON(atomic_read(&object->use_count) < 2);
3c7b4e6b
CM
554
555 /*
556 * Locking here also ensures that the corresponding memory block
557 * cannot be freed when it is being scanned.
558 */
559 spin_lock_irqsave(&object->lock, flags);
3c7b4e6b
CM
560 object->flags &= ~OBJECT_ALLOCATED;
561 spin_unlock_irqrestore(&object->lock, flags);
562 put_object(object);
563}
564
53238a60
CM
565/*
566 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
567 * delete it.
568 */
569static void delete_object_full(unsigned long ptr)
570{
571 struct kmemleak_object *object;
572
573 object = find_and_get_object(ptr, 0);
574 if (!object) {
575#ifdef DEBUG
576 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
577 ptr);
578#endif
579 return;
580 }
581 __delete_object(object);
582 put_object(object);
583}
584
585/*
586 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
587 * delete it. If the memory block is partially freed, the function may create
588 * additional metadata for the remaining parts of the block.
589 */
590static void delete_object_part(unsigned long ptr, size_t size)
591{
592 struct kmemleak_object *object;
593 unsigned long start, end;
594
595 object = find_and_get_object(ptr, 1);
596 if (!object) {
597#ifdef DEBUG
598 kmemleak_warn("Partially freeing unknown object at 0x%08lx "
599 "(size %zu)\n", ptr, size);
600#endif
601 return;
602 }
603 __delete_object(object);
604
605 /*
606 * Create one or two objects that may result from the memory block
607 * split. Note that partial freeing is only done by free_bootmem() and
608 * this happens before kmemleak_init() is called. The path below is
609 * only executed during early log recording in kmemleak_init(), so
610 * GFP_KERNEL is enough.
611 */
612 start = object->pointer;
613 end = object->pointer + object->size;
614 if (ptr > start)
615 create_object(start, ptr - start, object->min_count,
616 GFP_KERNEL);
617 if (ptr + size < end)
618 create_object(ptr + size, end - ptr - size, object->min_count,
619 GFP_KERNEL);
620
621 put_object(object);
622}
3c7b4e6b
CM
623/*
624 * Make a object permanently as gray-colored so that it can no longer be
625 * reported as a leak. This is used in general to mark a false positive.
626 */
627static void make_gray_object(unsigned long ptr)
628{
629 unsigned long flags;
630 struct kmemleak_object *object;
631
632 object = find_and_get_object(ptr, 0);
633 if (!object) {
ae281064 634 kmemleak_warn("Graying unknown object at 0x%08lx\n", ptr);
3c7b4e6b
CM
635 return;
636 }
637
638 spin_lock_irqsave(&object->lock, flags);
639 object->min_count = 0;
640 spin_unlock_irqrestore(&object->lock, flags);
641 put_object(object);
642}
643
644/*
645 * Mark the object as black-colored so that it is ignored from scans and
646 * reporting.
647 */
648static void make_black_object(unsigned long ptr)
649{
650 unsigned long flags;
651 struct kmemleak_object *object;
652
653 object = find_and_get_object(ptr, 0);
654 if (!object) {
ae281064 655 kmemleak_warn("Blacking unknown object at 0x%08lx\n", ptr);
3c7b4e6b
CM
656 return;
657 }
658
659 spin_lock_irqsave(&object->lock, flags);
660 object->min_count = -1;
af98603d 661 object->flags |= OBJECT_NO_SCAN;
3c7b4e6b
CM
662 spin_unlock_irqrestore(&object->lock, flags);
663 put_object(object);
664}
665
666/*
667 * Add a scanning area to the object. If at least one such area is added,
668 * kmemleak will only scan these ranges rather than the whole memory block.
669 */
670static void add_scan_area(unsigned long ptr, unsigned long offset,
671 size_t length, gfp_t gfp)
672{
673 unsigned long flags;
674 struct kmemleak_object *object;
675 struct kmemleak_scan_area *area;
676
677 object = find_and_get_object(ptr, 0);
678 if (!object) {
ae281064
JP
679 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
680 ptr);
3c7b4e6b
CM
681 return;
682 }
683
216c04b0 684 area = kmem_cache_alloc(scan_area_cache, gfp & GFP_KMEMLEAK_MASK);
3c7b4e6b 685 if (!area) {
ae281064 686 kmemleak_warn("Cannot allocate a scan area\n");
3c7b4e6b
CM
687 goto out;
688 }
689
690 spin_lock_irqsave(&object->lock, flags);
691 if (offset + length > object->size) {
ae281064 692 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
3c7b4e6b
CM
693 dump_object_info(object);
694 kmem_cache_free(scan_area_cache, area);
695 goto out_unlock;
696 }
697
698 INIT_HLIST_NODE(&area->node);
699 area->offset = offset;
700 area->length = length;
701
702 hlist_add_head(&area->node, &object->area_list);
703out_unlock:
704 spin_unlock_irqrestore(&object->lock, flags);
705out:
706 put_object(object);
707}
708
709/*
710 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
711 * pointer. Such object will not be scanned by kmemleak but references to it
712 * are searched.
713 */
714static void object_no_scan(unsigned long ptr)
715{
716 unsigned long flags;
717 struct kmemleak_object *object;
718
719 object = find_and_get_object(ptr, 0);
720 if (!object) {
ae281064 721 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
3c7b4e6b
CM
722 return;
723 }
724
725 spin_lock_irqsave(&object->lock, flags);
726 object->flags |= OBJECT_NO_SCAN;
727 spin_unlock_irqrestore(&object->lock, flags);
728 put_object(object);
729}
730
731/*
732 * Log an early kmemleak_* call to the early_log buffer. These calls will be
733 * processed later once kmemleak is fully initialized.
734 */
a6186d89
CM
735static void __init log_early(int op_type, const void *ptr, size_t size,
736 int min_count, unsigned long offset, size_t length)
3c7b4e6b
CM
737{
738 unsigned long flags;
739 struct early_log *log;
740
741 if (crt_early_log >= ARRAY_SIZE(early_log)) {
a9d9058a
CM
742 pr_warning("Early log buffer exceeded\n");
743 kmemleak_disable();
3c7b4e6b
CM
744 return;
745 }
746
747 /*
748 * There is no need for locking since the kernel is still in UP mode
749 * at this stage. Disabling the IRQs is enough.
750 */
751 local_irq_save(flags);
752 log = &early_log[crt_early_log];
753 log->op_type = op_type;
754 log->ptr = ptr;
755 log->size = size;
756 log->min_count = min_count;
757 log->offset = offset;
758 log->length = length;
fd678967
CM
759 if (op_type == KMEMLEAK_ALLOC)
760 log->trace_len = __save_stack_trace(log->trace);
3c7b4e6b
CM
761 crt_early_log++;
762 local_irq_restore(flags);
763}
764
fd678967
CM
765/*
766 * Log an early allocated block and populate the stack trace.
767 */
768static void early_alloc(struct early_log *log)
769{
770 struct kmemleak_object *object;
771 unsigned long flags;
772 int i;
773
774 if (!atomic_read(&kmemleak_enabled) || !log->ptr || IS_ERR(log->ptr))
775 return;
776
777 /*
778 * RCU locking needed to ensure object is not freed via put_object().
779 */
780 rcu_read_lock();
781 object = create_object((unsigned long)log->ptr, log->size,
782 log->min_count, GFP_KERNEL);
783 spin_lock_irqsave(&object->lock, flags);
784 for (i = 0; i < log->trace_len; i++)
785 object->trace[i] = log->trace[i];
786 object->trace_len = log->trace_len;
787 spin_unlock_irqrestore(&object->lock, flags);
788 rcu_read_unlock();
789}
790
3c7b4e6b
CM
791/*
792 * Memory allocation function callback. This function is called from the
793 * kernel allocators when a new block is allocated (kmem_cache_alloc, kmalloc,
794 * vmalloc etc.).
795 */
a6186d89
CM
796void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
797 gfp_t gfp)
3c7b4e6b
CM
798{
799 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
800
801 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
802 create_object((unsigned long)ptr, size, min_count, gfp);
803 else if (atomic_read(&kmemleak_early_log))
804 log_early(KMEMLEAK_ALLOC, ptr, size, min_count, 0, 0);
805}
806EXPORT_SYMBOL_GPL(kmemleak_alloc);
807
808/*
809 * Memory freeing function callback. This function is called from the kernel
810 * allocators when a block is freed (kmem_cache_free, kfree, vfree etc.).
811 */
a6186d89 812void __ref kmemleak_free(const void *ptr)
3c7b4e6b
CM
813{
814 pr_debug("%s(0x%p)\n", __func__, ptr);
815
816 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
53238a60 817 delete_object_full((unsigned long)ptr);
3c7b4e6b
CM
818 else if (atomic_read(&kmemleak_early_log))
819 log_early(KMEMLEAK_FREE, ptr, 0, 0, 0, 0);
820}
821EXPORT_SYMBOL_GPL(kmemleak_free);
822
53238a60
CM
823/*
824 * Partial memory freeing function callback. This function is usually called
825 * from bootmem allocator when (part of) a memory block is freed.
826 */
a6186d89 827void __ref kmemleak_free_part(const void *ptr, size_t size)
53238a60
CM
828{
829 pr_debug("%s(0x%p)\n", __func__, ptr);
830
831 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
832 delete_object_part((unsigned long)ptr, size);
833 else if (atomic_read(&kmemleak_early_log))
834 log_early(KMEMLEAK_FREE_PART, ptr, size, 0, 0, 0);
835}
836EXPORT_SYMBOL_GPL(kmemleak_free_part);
837
3c7b4e6b
CM
838/*
839 * Mark an already allocated memory block as a false positive. This will cause
840 * the block to no longer be reported as leak and always be scanned.
841 */
a6186d89 842void __ref kmemleak_not_leak(const void *ptr)
3c7b4e6b
CM
843{
844 pr_debug("%s(0x%p)\n", __func__, ptr);
845
846 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
847 make_gray_object((unsigned long)ptr);
848 else if (atomic_read(&kmemleak_early_log))
849 log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0, 0, 0);
850}
851EXPORT_SYMBOL(kmemleak_not_leak);
852
853/*
854 * Ignore a memory block. This is usually done when it is known that the
855 * corresponding block is not a leak and does not contain any references to
856 * other allocated memory blocks.
857 */
a6186d89 858void __ref kmemleak_ignore(const void *ptr)
3c7b4e6b
CM
859{
860 pr_debug("%s(0x%p)\n", __func__, ptr);
861
862 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
863 make_black_object((unsigned long)ptr);
864 else if (atomic_read(&kmemleak_early_log))
865 log_early(KMEMLEAK_IGNORE, ptr, 0, 0, 0, 0);
866}
867EXPORT_SYMBOL(kmemleak_ignore);
868
869/*
870 * Limit the range to be scanned in an allocated memory block.
871 */
a6186d89
CM
872void __ref kmemleak_scan_area(const void *ptr, unsigned long offset,
873 size_t length, gfp_t gfp)
3c7b4e6b
CM
874{
875 pr_debug("%s(0x%p)\n", __func__, ptr);
876
877 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
878 add_scan_area((unsigned long)ptr, offset, length, gfp);
879 else if (atomic_read(&kmemleak_early_log))
880 log_early(KMEMLEAK_SCAN_AREA, ptr, 0, 0, offset, length);
881}
882EXPORT_SYMBOL(kmemleak_scan_area);
883
884/*
885 * Inform kmemleak not to scan the given memory block.
886 */
a6186d89 887void __ref kmemleak_no_scan(const void *ptr)
3c7b4e6b
CM
888{
889 pr_debug("%s(0x%p)\n", __func__, ptr);
890
891 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
892 object_no_scan((unsigned long)ptr);
893 else if (atomic_read(&kmemleak_early_log))
894 log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0, 0, 0);
895}
896EXPORT_SYMBOL(kmemleak_no_scan);
897
3c7b4e6b
CM
898/*
899 * Memory scanning is a long process and it needs to be interruptable. This
900 * function checks whether such interrupt condition occured.
901 */
902static int scan_should_stop(void)
903{
904 if (!atomic_read(&kmemleak_enabled))
905 return 1;
906
907 /*
908 * This function may be called from either process or kthread context,
909 * hence the need to check for both stop conditions.
910 */
911 if (current->mm)
912 return signal_pending(current);
913 else
914 return kthread_should_stop();
915
916 return 0;
917}
918
919/*
920 * Scan a memory block (exclusive range) for valid pointers and add those
921 * found to the gray list.
922 */
923static void scan_block(void *_start, void *_end,
4b8a9674 924 struct kmemleak_object *scanned, int allow_resched)
3c7b4e6b
CM
925{
926 unsigned long *ptr;
927 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
928 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
929
930 for (ptr = start; ptr < end; ptr++) {
931 unsigned long flags;
932 unsigned long pointer = *ptr;
933 struct kmemleak_object *object;
934
4b8a9674
CM
935 if (allow_resched)
936 cond_resched();
3c7b4e6b
CM
937 if (scan_should_stop())
938 break;
939
3c7b4e6b
CM
940 object = find_and_get_object(pointer, 1);
941 if (!object)
942 continue;
943 if (object == scanned) {
944 /* self referenced, ignore */
945 put_object(object);
946 continue;
947 }
948
949 /*
950 * Avoid the lockdep recursive warning on object->lock being
951 * previously acquired in scan_object(). These locks are
952 * enclosed by scan_mutex.
953 */
954 spin_lock_irqsave_nested(&object->lock, flags,
955 SINGLE_DEPTH_NESTING);
956 if (!color_white(object)) {
957 /* non-orphan, ignored or new */
958 spin_unlock_irqrestore(&object->lock, flags);
959 put_object(object);
960 continue;
961 }
962
963 /*
964 * Increase the object's reference count (number of pointers
965 * to the memory block). If this count reaches the required
966 * minimum, the object's color will become gray and it will be
967 * added to the gray_list.
968 */
969 object->count++;
970 if (color_gray(object))
971 list_add_tail(&object->gray_list, &gray_list);
972 else
973 put_object(object);
974 spin_unlock_irqrestore(&object->lock, flags);
975 }
976}
977
978/*
979 * Scan a memory block corresponding to a kmemleak_object. A condition is
980 * that object->use_count >= 1.
981 */
982static void scan_object(struct kmemleak_object *object)
983{
984 struct kmemleak_scan_area *area;
985 struct hlist_node *elem;
986 unsigned long flags;
987
988 /*
989 * Once the object->lock is aquired, the corresponding memory block
990 * cannot be freed (the same lock is aquired in delete_object).
991 */
992 spin_lock_irqsave(&object->lock, flags);
993 if (object->flags & OBJECT_NO_SCAN)
994 goto out;
995 if (!(object->flags & OBJECT_ALLOCATED))
996 /* already freed object */
997 goto out;
af98603d
CM
998 if (hlist_empty(&object->area_list)) {
999 void *start = (void *)object->pointer;
1000 void *end = (void *)(object->pointer + object->size);
1001
1002 while (start < end && (object->flags & OBJECT_ALLOCATED) &&
1003 !(object->flags & OBJECT_NO_SCAN)) {
1004 scan_block(start, min(start + MAX_SCAN_SIZE, end),
1005 object, 0);
1006 start += MAX_SCAN_SIZE;
1007
1008 spin_unlock_irqrestore(&object->lock, flags);
1009 cond_resched();
1010 spin_lock_irqsave(&object->lock, flags);
1011 }
1012 } else
3c7b4e6b
CM
1013 hlist_for_each_entry(area, elem, &object->area_list, node)
1014 scan_block((void *)(object->pointer + area->offset),
1015 (void *)(object->pointer + area->offset
4b8a9674 1016 + area->length), object, 0);
3c7b4e6b
CM
1017out:
1018 spin_unlock_irqrestore(&object->lock, flags);
1019}
1020
1021/*
1022 * Scan data sections and all the referenced memory blocks allocated via the
1023 * kernel's standard allocators. This function must be called with the
1024 * scan_mutex held.
1025 */
1026static void kmemleak_scan(void)
1027{
1028 unsigned long flags;
1029 struct kmemleak_object *object, *tmp;
1030 struct task_struct *task;
1031 int i;
4698c1f2 1032 int new_leaks = 0;
2587362e 1033 int gray_list_pass = 0;
3c7b4e6b 1034
acf4968e
CM
1035 jiffies_last_scan = jiffies;
1036
3c7b4e6b
CM
1037 /* prepare the kmemleak_object's */
1038 rcu_read_lock();
1039 list_for_each_entry_rcu(object, &object_list, object_list) {
1040 spin_lock_irqsave(&object->lock, flags);
1041#ifdef DEBUG
1042 /*
1043 * With a few exceptions there should be a maximum of
1044 * 1 reference to any object at this point.
1045 */
1046 if (atomic_read(&object->use_count) > 1) {
ae281064 1047 pr_debug("object->use_count = %d\n",
3c7b4e6b
CM
1048 atomic_read(&object->use_count));
1049 dump_object_info(object);
1050 }
1051#endif
1052 /* reset the reference count (whiten the object) */
1053 object->count = 0;
2587362e 1054 object->flags &= ~OBJECT_NEW;
3c7b4e6b
CM
1055 if (color_gray(object) && get_object(object))
1056 list_add_tail(&object->gray_list, &gray_list);
1057
1058 spin_unlock_irqrestore(&object->lock, flags);
1059 }
1060 rcu_read_unlock();
1061
1062 /* data/bss scanning */
4b8a9674
CM
1063 scan_block(_sdata, _edata, NULL, 1);
1064 scan_block(__bss_start, __bss_stop, NULL, 1);
3c7b4e6b
CM
1065
1066#ifdef CONFIG_SMP
1067 /* per-cpu sections scanning */
1068 for_each_possible_cpu(i)
1069 scan_block(__per_cpu_start + per_cpu_offset(i),
4b8a9674 1070 __per_cpu_end + per_cpu_offset(i), NULL, 1);
3c7b4e6b
CM
1071#endif
1072
1073 /*
1074 * Struct page scanning for each node. The code below is not yet safe
1075 * with MEMORY_HOTPLUG.
1076 */
1077 for_each_online_node(i) {
1078 pg_data_t *pgdat = NODE_DATA(i);
1079 unsigned long start_pfn = pgdat->node_start_pfn;
1080 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1081 unsigned long pfn;
1082
1083 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1084 struct page *page;
1085
1086 if (!pfn_valid(pfn))
1087 continue;
1088 page = pfn_to_page(pfn);
1089 /* only scan if page is in use */
1090 if (page_count(page) == 0)
1091 continue;
4b8a9674 1092 scan_block(page, page + 1, NULL, 1);
3c7b4e6b
CM
1093 }
1094 }
1095
1096 /*
1097 * Scanning the task stacks may introduce false negatives and it is
1098 * not enabled by default.
1099 */
1100 if (kmemleak_stack_scan) {
1101 read_lock(&tasklist_lock);
1102 for_each_process(task)
1103 scan_block(task_stack_page(task),
4b8a9674
CM
1104 task_stack_page(task) + THREAD_SIZE,
1105 NULL, 0);
3c7b4e6b
CM
1106 read_unlock(&tasklist_lock);
1107 }
1108
1109 /*
1110 * Scan the objects already referenced from the sections scanned
1111 * above. More objects will be referenced and, if there are no memory
1112 * leaks, all the objects will be scanned. The list traversal is safe
1113 * for both tail additions and removals from inside the loop. The
1114 * kmemleak objects cannot be freed from outside the loop because their
1115 * use_count was increased.
1116 */
2587362e 1117repeat:
3c7b4e6b
CM
1118 object = list_entry(gray_list.next, typeof(*object), gray_list);
1119 while (&object->gray_list != &gray_list) {
57d81f6f 1120 cond_resched();
3c7b4e6b
CM
1121
1122 /* may add new objects to the list */
1123 if (!scan_should_stop())
1124 scan_object(object);
1125
1126 tmp = list_entry(object->gray_list.next, typeof(*object),
1127 gray_list);
1128
1129 /* remove the object from the list and release it */
1130 list_del(&object->gray_list);
1131 put_object(object);
1132
1133 object = tmp;
1134 }
2587362e
CM
1135
1136 if (scan_should_stop() || ++gray_list_pass >= GRAY_LIST_PASSES)
1137 goto scan_end;
1138
1139 /*
1140 * Check for new objects allocated during this scanning and add them
1141 * to the gray list.
1142 */
1143 rcu_read_lock();
1144 list_for_each_entry_rcu(object, &object_list, object_list) {
1145 spin_lock_irqsave(&object->lock, flags);
1146 if ((object->flags & OBJECT_NEW) && !color_black(object) &&
1147 get_object(object)) {
1148 object->flags &= ~OBJECT_NEW;
1149 list_add_tail(&object->gray_list, &gray_list);
1150 }
1151 spin_unlock_irqrestore(&object->lock, flags);
1152 }
1153 rcu_read_unlock();
1154
1155 if (!list_empty(&gray_list))
1156 goto repeat;
1157
1158scan_end:
3c7b4e6b 1159 WARN_ON(!list_empty(&gray_list));
4698c1f2 1160
17bb9e0d 1161 /*
2587362e
CM
1162 * If scanning was stopped or new objects were being allocated at a
1163 * higher rate than gray list scanning, do not report any new
1164 * unreferenced objects.
17bb9e0d 1165 */
2587362e 1166 if (scan_should_stop() || gray_list_pass >= GRAY_LIST_PASSES)
17bb9e0d
CM
1167 return;
1168
4698c1f2
CM
1169 /*
1170 * Scanning result reporting.
1171 */
1172 rcu_read_lock();
1173 list_for_each_entry_rcu(object, &object_list, object_list) {
1174 spin_lock_irqsave(&object->lock, flags);
1175 if (unreferenced_object(object) &&
1176 !(object->flags & OBJECT_REPORTED)) {
1177 object->flags |= OBJECT_REPORTED;
1178 new_leaks++;
1179 }
1180 spin_unlock_irqrestore(&object->lock, flags);
1181 }
1182 rcu_read_unlock();
1183
1184 if (new_leaks)
1185 pr_info("%d new suspected memory leaks (see "
1186 "/sys/kernel/debug/kmemleak)\n", new_leaks);
1187
3c7b4e6b
CM
1188}
1189
1190/*
1191 * Thread function performing automatic memory scanning. Unreferenced objects
1192 * at the end of a memory scan are reported but only the first time.
1193 */
1194static int kmemleak_scan_thread(void *arg)
1195{
1196 static int first_run = 1;
1197
ae281064 1198 pr_info("Automatic memory scanning thread started\n");
bf2a76b3 1199 set_user_nice(current, 10);
3c7b4e6b
CM
1200
1201 /*
1202 * Wait before the first scan to allow the system to fully initialize.
1203 */
1204 if (first_run) {
1205 first_run = 0;
1206 ssleep(SECS_FIRST_SCAN);
1207 }
1208
1209 while (!kthread_should_stop()) {
3c7b4e6b
CM
1210 signed long timeout = jiffies_scan_wait;
1211
1212 mutex_lock(&scan_mutex);
3c7b4e6b 1213 kmemleak_scan();
3c7b4e6b 1214 mutex_unlock(&scan_mutex);
4698c1f2 1215
3c7b4e6b
CM
1216 /* wait before the next scan */
1217 while (timeout && !kthread_should_stop())
1218 timeout = schedule_timeout_interruptible(timeout);
1219 }
1220
ae281064 1221 pr_info("Automatic memory scanning thread ended\n");
3c7b4e6b
CM
1222
1223 return 0;
1224}
1225
1226/*
1227 * Start the automatic memory scanning thread. This function must be called
4698c1f2 1228 * with the scan_mutex held.
3c7b4e6b
CM
1229 */
1230void start_scan_thread(void)
1231{
1232 if (scan_thread)
1233 return;
1234 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1235 if (IS_ERR(scan_thread)) {
ae281064 1236 pr_warning("Failed to create the scan thread\n");
3c7b4e6b
CM
1237 scan_thread = NULL;
1238 }
1239}
1240
1241/*
1242 * Stop the automatic memory scanning thread. This function must be called
4698c1f2 1243 * with the scan_mutex held.
3c7b4e6b
CM
1244 */
1245void stop_scan_thread(void)
1246{
1247 if (scan_thread) {
1248 kthread_stop(scan_thread);
1249 scan_thread = NULL;
1250 }
1251}
1252
1253/*
1254 * Iterate over the object_list and return the first valid object at or after
1255 * the required position with its use_count incremented. The function triggers
1256 * a memory scanning when the pos argument points to the first position.
1257 */
1258static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1259{
1260 struct kmemleak_object *object;
1261 loff_t n = *pos;
b87324d0
CM
1262 int err;
1263
1264 err = mutex_lock_interruptible(&scan_mutex);
1265 if (err < 0)
1266 return ERR_PTR(err);
3c7b4e6b 1267
3c7b4e6b
CM
1268 rcu_read_lock();
1269 list_for_each_entry_rcu(object, &object_list, object_list) {
1270 if (n-- > 0)
1271 continue;
1272 if (get_object(object))
1273 goto out;
1274 }
1275 object = NULL;
1276out:
3c7b4e6b
CM
1277 return object;
1278}
1279
1280/*
1281 * Return the next object in the object_list. The function decrements the
1282 * use_count of the previous object and increases that of the next one.
1283 */
1284static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1285{
1286 struct kmemleak_object *prev_obj = v;
1287 struct kmemleak_object *next_obj = NULL;
1288 struct list_head *n = &prev_obj->object_list;
1289
1290 ++(*pos);
3c7b4e6b 1291
3c7b4e6b
CM
1292 list_for_each_continue_rcu(n, &object_list) {
1293 next_obj = list_entry(n, struct kmemleak_object, object_list);
1294 if (get_object(next_obj))
1295 break;
1296 }
288c857d 1297
3c7b4e6b
CM
1298 put_object(prev_obj);
1299 return next_obj;
1300}
1301
1302/*
1303 * Decrement the use_count of the last object required, if any.
1304 */
1305static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1306{
b87324d0
CM
1307 if (!IS_ERR(v)) {
1308 /*
1309 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1310 * waiting was interrupted, so only release it if !IS_ERR.
1311 */
f5886c7f 1312 rcu_read_unlock();
b87324d0
CM
1313 mutex_unlock(&scan_mutex);
1314 if (v)
1315 put_object(v);
1316 }
3c7b4e6b
CM
1317}
1318
1319/*
1320 * Print the information for an unreferenced object to the seq file.
1321 */
1322static int kmemleak_seq_show(struct seq_file *seq, void *v)
1323{
1324 struct kmemleak_object *object = v;
1325 unsigned long flags;
1326
1327 spin_lock_irqsave(&object->lock, flags);
288c857d 1328 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
17bb9e0d 1329 print_unreferenced(seq, object);
3c7b4e6b
CM
1330 spin_unlock_irqrestore(&object->lock, flags);
1331 return 0;
1332}
1333
1334static const struct seq_operations kmemleak_seq_ops = {
1335 .start = kmemleak_seq_start,
1336 .next = kmemleak_seq_next,
1337 .stop = kmemleak_seq_stop,
1338 .show = kmemleak_seq_show,
1339};
1340
1341static int kmemleak_open(struct inode *inode, struct file *file)
1342{
3c7b4e6b
CM
1343 if (!atomic_read(&kmemleak_enabled))
1344 return -EBUSY;
1345
b87324d0 1346 return seq_open(file, &kmemleak_seq_ops);
3c7b4e6b
CM
1347}
1348
1349static int kmemleak_release(struct inode *inode, struct file *file)
1350{
b87324d0 1351 return seq_release(inode, file);
3c7b4e6b
CM
1352}
1353
189d84ed
CM
1354static int dump_str_object_info(const char *str)
1355{
1356 unsigned long flags;
1357 struct kmemleak_object *object;
1358 unsigned long addr;
1359
1360 addr= simple_strtoul(str, NULL, 0);
1361 object = find_and_get_object(addr, 0);
1362 if (!object) {
1363 pr_info("Unknown object at 0x%08lx\n", addr);
1364 return -EINVAL;
1365 }
1366
1367 spin_lock_irqsave(&object->lock, flags);
1368 dump_object_info(object);
1369 spin_unlock_irqrestore(&object->lock, flags);
1370
1371 put_object(object);
1372 return 0;
1373}
1374
3c7b4e6b
CM
1375/*
1376 * File write operation to configure kmemleak at run-time. The following
1377 * commands can be written to the /sys/kernel/debug/kmemleak file:
1378 * off - disable kmemleak (irreversible)
1379 * stack=on - enable the task stacks scanning
1380 * stack=off - disable the tasks stacks scanning
1381 * scan=on - start the automatic memory scanning thread
1382 * scan=off - stop the automatic memory scanning thread
1383 * scan=... - set the automatic memory scanning period in seconds (0 to
1384 * disable it)
4698c1f2 1385 * scan - trigger a memory scan
189d84ed 1386 * dump=... - dump information about the object found at the given address
3c7b4e6b
CM
1387 */
1388static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1389 size_t size, loff_t *ppos)
1390{
1391 char buf[64];
1392 int buf_size;
b87324d0 1393 int ret;
3c7b4e6b
CM
1394
1395 buf_size = min(size, (sizeof(buf) - 1));
1396 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1397 return -EFAULT;
1398 buf[buf_size] = 0;
1399
b87324d0
CM
1400 ret = mutex_lock_interruptible(&scan_mutex);
1401 if (ret < 0)
1402 return ret;
1403
3c7b4e6b
CM
1404 if (strncmp(buf, "off", 3) == 0)
1405 kmemleak_disable();
1406 else if (strncmp(buf, "stack=on", 8) == 0)
1407 kmemleak_stack_scan = 1;
1408 else if (strncmp(buf, "stack=off", 9) == 0)
1409 kmemleak_stack_scan = 0;
1410 else if (strncmp(buf, "scan=on", 7) == 0)
1411 start_scan_thread();
1412 else if (strncmp(buf, "scan=off", 8) == 0)
1413 stop_scan_thread();
1414 else if (strncmp(buf, "scan=", 5) == 0) {
1415 unsigned long secs;
3c7b4e6b 1416
b87324d0
CM
1417 ret = strict_strtoul(buf + 5, 0, &secs);
1418 if (ret < 0)
1419 goto out;
3c7b4e6b
CM
1420 stop_scan_thread();
1421 if (secs) {
1422 jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1423 start_scan_thread();
1424 }
4698c1f2
CM
1425 } else if (strncmp(buf, "scan", 4) == 0)
1426 kmemleak_scan();
189d84ed
CM
1427 else if (strncmp(buf, "dump=", 5) == 0)
1428 ret = dump_str_object_info(buf + 5);
4698c1f2 1429 else
b87324d0
CM
1430 ret = -EINVAL;
1431
1432out:
1433 mutex_unlock(&scan_mutex);
1434 if (ret < 0)
1435 return ret;
3c7b4e6b
CM
1436
1437 /* ignore the rest of the buffer, only one command at a time */
1438 *ppos += size;
1439 return size;
1440}
1441
1442static const struct file_operations kmemleak_fops = {
1443 .owner = THIS_MODULE,
1444 .open = kmemleak_open,
1445 .read = seq_read,
1446 .write = kmemleak_write,
1447 .llseek = seq_lseek,
1448 .release = kmemleak_release,
1449};
1450
1451/*
1452 * Perform the freeing of the kmemleak internal objects after waiting for any
1453 * current memory scan to complete.
1454 */
1455static int kmemleak_cleanup_thread(void *arg)
1456{
1457 struct kmemleak_object *object;
1458
4698c1f2 1459 mutex_lock(&scan_mutex);
3c7b4e6b 1460 stop_scan_thread();
3c7b4e6b 1461
3c7b4e6b
CM
1462 rcu_read_lock();
1463 list_for_each_entry_rcu(object, &object_list, object_list)
53238a60 1464 delete_object_full(object->pointer);
3c7b4e6b
CM
1465 rcu_read_unlock();
1466 mutex_unlock(&scan_mutex);
1467
1468 return 0;
1469}
1470
1471/*
1472 * Start the clean-up thread.
1473 */
1474static void kmemleak_cleanup(void)
1475{
1476 struct task_struct *cleanup_thread;
1477
1478 cleanup_thread = kthread_run(kmemleak_cleanup_thread, NULL,
1479 "kmemleak-clean");
1480 if (IS_ERR(cleanup_thread))
ae281064 1481 pr_warning("Failed to create the clean-up thread\n");
3c7b4e6b
CM
1482}
1483
1484/*
1485 * Disable kmemleak. No memory allocation/freeing will be traced once this
1486 * function is called. Disabling kmemleak is an irreversible operation.
1487 */
1488static void kmemleak_disable(void)
1489{
1490 /* atomically check whether it was already invoked */
1491 if (atomic_cmpxchg(&kmemleak_error, 0, 1))
1492 return;
1493
1494 /* stop any memory operation tracing */
1495 atomic_set(&kmemleak_early_log, 0);
1496 atomic_set(&kmemleak_enabled, 0);
1497
1498 /* check whether it is too early for a kernel thread */
1499 if (atomic_read(&kmemleak_initialized))
1500 kmemleak_cleanup();
1501
1502 pr_info("Kernel memory leak detector disabled\n");
1503}
1504
1505/*
1506 * Allow boot-time kmemleak disabling (enabled by default).
1507 */
1508static int kmemleak_boot_config(char *str)
1509{
1510 if (!str)
1511 return -EINVAL;
1512 if (strcmp(str, "off") == 0)
1513 kmemleak_disable();
1514 else if (strcmp(str, "on") != 0)
1515 return -EINVAL;
1516 return 0;
1517}
1518early_param("kmemleak", kmemleak_boot_config);
1519
1520/*
2030117d 1521 * Kmemleak initialization.
3c7b4e6b
CM
1522 */
1523void __init kmemleak_init(void)
1524{
1525 int i;
1526 unsigned long flags;
1527
3c7b4e6b
CM
1528 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1529 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1530
1531 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1532 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1533 INIT_PRIO_TREE_ROOT(&object_tree_root);
1534
1535 /* the kernel is still in UP mode, so disabling the IRQs is enough */
1536 local_irq_save(flags);
1537 if (!atomic_read(&kmemleak_error)) {
1538 atomic_set(&kmemleak_enabled, 1);
1539 atomic_set(&kmemleak_early_log, 0);
1540 }
1541 local_irq_restore(flags);
1542
1543 /*
1544 * This is the point where tracking allocations is safe. Automatic
1545 * scanning is started during the late initcall. Add the early logged
1546 * callbacks to the kmemleak infrastructure.
1547 */
1548 for (i = 0; i < crt_early_log; i++) {
1549 struct early_log *log = &early_log[i];
1550
1551 switch (log->op_type) {
1552 case KMEMLEAK_ALLOC:
fd678967 1553 early_alloc(log);
3c7b4e6b
CM
1554 break;
1555 case KMEMLEAK_FREE:
1556 kmemleak_free(log->ptr);
1557 break;
53238a60
CM
1558 case KMEMLEAK_FREE_PART:
1559 kmemleak_free_part(log->ptr, log->size);
1560 break;
3c7b4e6b
CM
1561 case KMEMLEAK_NOT_LEAK:
1562 kmemleak_not_leak(log->ptr);
1563 break;
1564 case KMEMLEAK_IGNORE:
1565 kmemleak_ignore(log->ptr);
1566 break;
1567 case KMEMLEAK_SCAN_AREA:
1568 kmemleak_scan_area(log->ptr, log->offset, log->length,
1569 GFP_KERNEL);
1570 break;
1571 case KMEMLEAK_NO_SCAN:
1572 kmemleak_no_scan(log->ptr);
1573 break;
1574 default:
1575 WARN_ON(1);
1576 }
1577 }
1578}
1579
1580/*
1581 * Late initialization function.
1582 */
1583static int __init kmemleak_late_init(void)
1584{
1585 struct dentry *dentry;
1586
1587 atomic_set(&kmemleak_initialized, 1);
1588
1589 if (atomic_read(&kmemleak_error)) {
1590 /*
1591 * Some error occured and kmemleak was disabled. There is a
1592 * small chance that kmemleak_disable() was called immediately
1593 * after setting kmemleak_initialized and we may end up with
1594 * two clean-up threads but serialized by scan_mutex.
1595 */
1596 kmemleak_cleanup();
1597 return -ENOMEM;
1598 }
1599
1600 dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL,
1601 &kmemleak_fops);
1602 if (!dentry)
ae281064 1603 pr_warning("Failed to create the debugfs kmemleak file\n");
4698c1f2 1604 mutex_lock(&scan_mutex);
3c7b4e6b 1605 start_scan_thread();
4698c1f2 1606 mutex_unlock(&scan_mutex);
3c7b4e6b
CM
1607
1608 pr_info("Kernel memory leak detector initialized\n");
1609
1610 return 0;
1611}
1612late_initcall(kmemleak_late_init);