blob: 3c2b24c49a84182b67d6a9c89e0baf5fc8f38a60 [file] [log] [blame]
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001/*
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
Catalin Marinas4698c1f2009-06-26 17:38:27 +010051 * 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
Catalin Marinas3c7b4e62009-06-11 13:22:39 +010055 *
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
Joe Perchesae281062009-06-23 14:40:26 +010064#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
65
Catalin Marinas3c7b4e62009-06-11 13:22:39 +010066#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 */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100106#define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100107#define SECS_FIRST_SCAN 60 /* delay before the first scan */
108#define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
109
110#define BYTES_PER_POINTER sizeof(void *)
111
Catalin Marinas216c04b2009-06-17 18:29:02 +0100112/* GFP bitmask for kmemleak internal allocations */
113#define GFP_KMEMLEAK_MASK (GFP_KERNEL | GFP_ATOMIC)
114
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100115/* scanning area inside a memory block */
116struct kmemleak_scan_area {
117 struct hlist_node node;
118 unsigned long offset;
119 size_t length;
120};
121
122/*
123 * Structure holding the metadata for each allocated memory block.
124 * Modifications to such objects should be made while holding the
125 * object->lock. Insertions or deletions from object_list, gray_list or
126 * tree_node are already protected by the corresponding locks or mutex (see
127 * the notes on locking above). These objects are reference-counted
128 * (use_count) and freed using the RCU mechanism.
129 */
130struct kmemleak_object {
131 spinlock_t lock;
132 unsigned long flags; /* object status flags */
133 struct list_head object_list;
134 struct list_head gray_list;
135 struct prio_tree_node tree_node;
136 struct rcu_head rcu; /* object_list lockless traversal */
137 /* object usage count; object freed when use_count == 0 */
138 atomic_t use_count;
139 unsigned long pointer;
140 size_t size;
141 /* minimum number of a pointers found before it is considered leak */
142 int min_count;
143 /* the total number of pointers found pointing to this object */
144 int count;
145 /* memory ranges to be scanned inside an object (empty for all) */
146 struct hlist_head area_list;
147 unsigned long trace[MAX_TRACE];
148 unsigned int trace_len;
149 unsigned long jiffies; /* creation timestamp */
150 pid_t pid; /* pid of the current task */
151 char comm[TASK_COMM_LEN]; /* executable name */
152};
153
154/* flag representing the memory block allocation status */
155#define OBJECT_ALLOCATED (1 << 0)
156/* flag set after the first reporting of an unreference object */
157#define OBJECT_REPORTED (1 << 1)
158/* flag set to not scan the object */
159#define OBJECT_NO_SCAN (1 << 2)
160
161/* the list of all allocated objects */
162static LIST_HEAD(object_list);
163/* the list of gray-colored objects (see color_gray comment below) */
164static LIST_HEAD(gray_list);
165/* prio search tree for object boundaries */
166static struct prio_tree_root object_tree_root;
167/* rw_lock protecting the access to object_list and prio_tree_root */
168static DEFINE_RWLOCK(kmemleak_lock);
169
170/* allocation caches for kmemleak internal data */
171static struct kmem_cache *object_cache;
172static struct kmem_cache *scan_area_cache;
173
174/* set if tracing memory operations is enabled */
175static atomic_t kmemleak_enabled = ATOMIC_INIT(0);
176/* set in the late_initcall if there were no errors */
177static atomic_t kmemleak_initialized = ATOMIC_INIT(0);
178/* enables or disables early logging of the memory operations */
179static atomic_t kmemleak_early_log = ATOMIC_INIT(1);
180/* set if a fata kmemleak error has occurred */
181static atomic_t kmemleak_error = ATOMIC_INIT(0);
182
183/* minimum and maximum address that may be valid pointers */
184static unsigned long min_addr = ULONG_MAX;
185static unsigned long max_addr;
186
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100187static struct task_struct *scan_thread;
Catalin Marinasacf49682009-06-26 17:38:29 +0100188/* used to avoid reporting of recently allocated objects */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100189static unsigned long jiffies_min_age;
Catalin Marinasacf49682009-06-26 17:38:29 +0100190static unsigned long jiffies_last_scan;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100191/* delay between automatic memory scannings */
192static signed long jiffies_scan_wait;
193/* enables or disables the task stacks scanning */
Catalin Marinase0a2a162009-06-26 17:38:25 +0100194static int kmemleak_stack_scan = 1;
Catalin Marinas4698c1f2009-06-26 17:38:27 +0100195/* protects the memory scanning, parameters and debug/kmemleak file access */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100196static DEFINE_MUTEX(scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100197
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100198/*
Catalin Marinas20301172009-06-17 18:29:04 +0100199 * Early object allocation/freeing logging. Kmemleak is initialized after the
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100200 * kernel allocator. However, both the kernel allocator and kmemleak may
Catalin Marinas20301172009-06-17 18:29:04 +0100201 * allocate memory blocks which need to be tracked. Kmemleak defines an
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100202 * arbitrary buffer to hold the allocation/freeing information before it is
203 * fully initialized.
204 */
205
206/* kmemleak operation type for early logging */
207enum {
208 KMEMLEAK_ALLOC,
209 KMEMLEAK_FREE,
210 KMEMLEAK_NOT_LEAK,
211 KMEMLEAK_IGNORE,
212 KMEMLEAK_SCAN_AREA,
213 KMEMLEAK_NO_SCAN
214};
215
216/*
217 * Structure holding the information passed to kmemleak callbacks during the
218 * early logging.
219 */
220struct early_log {
221 int op_type; /* kmemleak operation type */
222 const void *ptr; /* allocated/freed memory block */
223 size_t size; /* memory block size */
224 int min_count; /* minimum reference count */
225 unsigned long offset; /* scan area offset */
226 size_t length; /* scan area length */
227};
228
229/* early logging buffer and current position */
Catalin Marinasa9d90582009-06-25 10:16:11 +0100230static struct early_log early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE];
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100231static int crt_early_log;
232
233static void kmemleak_disable(void);
234
235/*
236 * Print a warning and dump the stack trace.
237 */
238#define kmemleak_warn(x...) do { \
239 pr_warning(x); \
240 dump_stack(); \
241} while (0)
242
243/*
244 * Macro invoked when a serious kmemleak condition occured and cannot be
Catalin Marinas20301172009-06-17 18:29:04 +0100245 * recovered from. Kmemleak will be disabled and further allocation/freeing
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100246 * tracing no longer available.
247 */
Catalin Marinas000814f2009-06-17 18:29:03 +0100248#define kmemleak_stop(x...) do { \
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100249 kmemleak_warn(x); \
250 kmemleak_disable(); \
251} while (0)
252
253/*
254 * Object colors, encoded with count and min_count:
255 * - white - orphan object, not enough references to it (count < min_count)
256 * - gray - not orphan, not marked as false positive (min_count == 0) or
257 * sufficient references to it (count >= min_count)
258 * - black - ignore, it doesn't contain references (e.g. text section)
259 * (min_count == -1). No function defined for this color.
260 * Newly created objects don't have any color assigned (object->count == -1)
261 * before the next memory scan when they become white.
262 */
263static int color_white(const struct kmemleak_object *object)
264{
265 return object->count != -1 && object->count < object->min_count;
266}
267
268static int color_gray(const struct kmemleak_object *object)
269{
270 return object->min_count != -1 && object->count >= object->min_count;
271}
272
273/*
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100274 * Objects are considered unreferenced only if their color is white, they have
275 * not be deleted and have a minimum age to avoid false positives caused by
276 * pointers temporarily stored in CPU registers.
277 */
278static int unreferenced_object(struct kmemleak_object *object)
279{
280 return (object->flags & OBJECT_ALLOCATED) && color_white(object) &&
Catalin Marinasacf49682009-06-26 17:38:29 +0100281 time_before_eq(object->jiffies + jiffies_min_age,
282 jiffies_last_scan);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100283}
284
285/*
Catalin Marinasbab4a342009-06-26 17:38:26 +0100286 * Printing of the unreferenced objects information to the seq file. The
287 * print_unreferenced function must be called with the object->lock held.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100288 */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100289static void print_unreferenced(struct seq_file *seq,
290 struct kmemleak_object *object)
291{
292 int i;
293
Catalin Marinasbab4a342009-06-26 17:38:26 +0100294 seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
295 object->pointer, object->size);
296 seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu\n",
297 object->comm, object->pid, object->jiffies);
298 seq_printf(seq, " backtrace:\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100299
300 for (i = 0; i < object->trace_len; i++) {
301 void *ptr = (void *)object->trace[i];
Catalin Marinasbab4a342009-06-26 17:38:26 +0100302 seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100303 }
304}
305
306/*
307 * Print the kmemleak_object information. This function is used mainly for
308 * debugging special cases when kmemleak operations. It must be called with
309 * the object->lock held.
310 */
311static void dump_object_info(struct kmemleak_object *object)
312{
313 struct stack_trace trace;
314
315 trace.nr_entries = object->trace_len;
316 trace.entries = object->trace;
317
Joe Perchesae281062009-06-23 14:40:26 +0100318 pr_notice("Object 0x%08lx (size %zu):\n",
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100319 object->tree_node.start, object->size);
320 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
321 object->comm, object->pid, object->jiffies);
322 pr_notice(" min_count = %d\n", object->min_count);
323 pr_notice(" count = %d\n", object->count);
324 pr_notice(" backtrace:\n");
325 print_stack_trace(&trace, 4);
326}
327
328/*
329 * Look-up a memory block metadata (kmemleak_object) in the priority search
330 * tree based on a pointer value. If alias is 0, only values pointing to the
331 * beginning of the memory block are allowed. The kmemleak_lock must be held
332 * when calling this function.
333 */
334static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
335{
336 struct prio_tree_node *node;
337 struct prio_tree_iter iter;
338 struct kmemleak_object *object;
339
340 prio_tree_iter_init(&iter, &object_tree_root, ptr, ptr);
341 node = prio_tree_next(&iter);
342 if (node) {
343 object = prio_tree_entry(node, struct kmemleak_object,
344 tree_node);
345 if (!alias && object->pointer != ptr) {
Joe Perchesae281062009-06-23 14:40:26 +0100346 kmemleak_warn("Found object by alias");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100347 object = NULL;
348 }
349 } else
350 object = NULL;
351
352 return object;
353}
354
355/*
356 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
357 * that once an object's use_count reached 0, the RCU freeing was already
358 * registered and the object should no longer be used. This function must be
359 * called under the protection of rcu_read_lock().
360 */
361static int get_object(struct kmemleak_object *object)
362{
363 return atomic_inc_not_zero(&object->use_count);
364}
365
366/*
367 * RCU callback to free a kmemleak_object.
368 */
369static void free_object_rcu(struct rcu_head *rcu)
370{
371 struct hlist_node *elem, *tmp;
372 struct kmemleak_scan_area *area;
373 struct kmemleak_object *object =
374 container_of(rcu, struct kmemleak_object, rcu);
375
376 /*
377 * Once use_count is 0 (guaranteed by put_object), there is no other
378 * code accessing this object, hence no need for locking.
379 */
380 hlist_for_each_entry_safe(area, elem, tmp, &object->area_list, node) {
381 hlist_del(elem);
382 kmem_cache_free(scan_area_cache, area);
383 }
384 kmem_cache_free(object_cache, object);
385}
386
387/*
388 * Decrement the object use_count. Once the count is 0, free the object using
389 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
390 * delete_object() path, the delayed RCU freeing ensures that there is no
391 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
392 * is also possible.
393 */
394static void put_object(struct kmemleak_object *object)
395{
396 if (!atomic_dec_and_test(&object->use_count))
397 return;
398
399 /* should only get here after delete_object was called */
400 WARN_ON(object->flags & OBJECT_ALLOCATED);
401
402 call_rcu(&object->rcu, free_object_rcu);
403}
404
405/*
406 * Look up an object in the prio search tree and increase its use_count.
407 */
408static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
409{
410 unsigned long flags;
411 struct kmemleak_object *object = NULL;
412
413 rcu_read_lock();
414 read_lock_irqsave(&kmemleak_lock, flags);
415 if (ptr >= min_addr && ptr < max_addr)
416 object = lookup_object(ptr, alias);
417 read_unlock_irqrestore(&kmemleak_lock, flags);
418
419 /* check whether the object is still available */
420 if (object && !get_object(object))
421 object = NULL;
422 rcu_read_unlock();
423
424 return object;
425}
426
427/*
428 * Create the metadata (struct kmemleak_object) corresponding to an allocated
429 * memory block and add it to the object_list and object_tree_root.
430 */
431static void create_object(unsigned long ptr, size_t size, int min_count,
432 gfp_t gfp)
433{
434 unsigned long flags;
435 struct kmemleak_object *object;
436 struct prio_tree_node *node;
437 struct stack_trace trace;
438
Catalin Marinas216c04b2009-06-17 18:29:02 +0100439 object = kmem_cache_alloc(object_cache, gfp & GFP_KMEMLEAK_MASK);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100440 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100441 kmemleak_stop("Cannot allocate a kmemleak_object structure\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100442 return;
443 }
444
445 INIT_LIST_HEAD(&object->object_list);
446 INIT_LIST_HEAD(&object->gray_list);
447 INIT_HLIST_HEAD(&object->area_list);
448 spin_lock_init(&object->lock);
449 atomic_set(&object->use_count, 1);
450 object->flags = OBJECT_ALLOCATED;
451 object->pointer = ptr;
452 object->size = size;
453 object->min_count = min_count;
454 object->count = -1; /* no color initially */
455 object->jiffies = jiffies;
456
457 /* task information */
458 if (in_irq()) {
459 object->pid = 0;
460 strncpy(object->comm, "hardirq", sizeof(object->comm));
461 } else if (in_softirq()) {
462 object->pid = 0;
463 strncpy(object->comm, "softirq", sizeof(object->comm));
464 } else {
465 object->pid = current->pid;
466 /*
467 * There is a small chance of a race with set_task_comm(),
468 * however using get_task_comm() here may cause locking
469 * dependency issues with current->alloc_lock. In the worst
470 * case, the command line is not correct.
471 */
472 strncpy(object->comm, current->comm, sizeof(object->comm));
473 }
474
475 /* kernel backtrace */
476 trace.max_entries = MAX_TRACE;
477 trace.nr_entries = 0;
478 trace.entries = object->trace;
479 trace.skip = 1;
480 save_stack_trace(&trace);
481 object->trace_len = trace.nr_entries;
482
483 INIT_PRIO_TREE_NODE(&object->tree_node);
484 object->tree_node.start = ptr;
485 object->tree_node.last = ptr + size - 1;
486
487 write_lock_irqsave(&kmemleak_lock, flags);
488 min_addr = min(min_addr, ptr);
489 max_addr = max(max_addr, ptr + size);
490 node = prio_tree_insert(&object_tree_root, &object->tree_node);
491 /*
492 * The code calling the kernel does not yet have the pointer to the
493 * memory block to be able to free it. However, we still hold the
494 * kmemleak_lock here in case parts of the kernel started freeing
495 * random memory blocks.
496 */
497 if (node != &object->tree_node) {
498 unsigned long flags;
499
Joe Perchesae281062009-06-23 14:40:26 +0100500 kmemleak_stop("Cannot insert 0x%lx into the object search tree "
501 "(already existing)\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100502 object = lookup_object(ptr, 1);
503 spin_lock_irqsave(&object->lock, flags);
504 dump_object_info(object);
505 spin_unlock_irqrestore(&object->lock, flags);
506
507 goto out;
508 }
509 list_add_tail_rcu(&object->object_list, &object_list);
510out:
511 write_unlock_irqrestore(&kmemleak_lock, flags);
512}
513
514/*
515 * Remove the metadata (struct kmemleak_object) for a memory block from the
516 * object_list and object_tree_root and decrement its use_count.
517 */
518static void delete_object(unsigned long ptr)
519{
520 unsigned long flags;
521 struct kmemleak_object *object;
522
523 write_lock_irqsave(&kmemleak_lock, flags);
524 object = lookup_object(ptr, 0);
525 if (!object) {
Catalin Marinasb6e68722009-06-29 17:13:57 +0100526#ifdef DEBUG
Joe Perchesae281062009-06-23 14:40:26 +0100527 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100528 ptr);
Catalin Marinasb6e68722009-06-29 17:13:57 +0100529#endif
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100530 write_unlock_irqrestore(&kmemleak_lock, flags);
531 return;
532 }
533 prio_tree_remove(&object_tree_root, &object->tree_node);
534 list_del_rcu(&object->object_list);
535 write_unlock_irqrestore(&kmemleak_lock, flags);
536
537 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
538 WARN_ON(atomic_read(&object->use_count) < 1);
539
540 /*
541 * Locking here also ensures that the corresponding memory block
542 * cannot be freed when it is being scanned.
543 */
544 spin_lock_irqsave(&object->lock, flags);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100545 object->flags &= ~OBJECT_ALLOCATED;
546 spin_unlock_irqrestore(&object->lock, flags);
547 put_object(object);
548}
549
550/*
551 * Make a object permanently as gray-colored so that it can no longer be
552 * reported as a leak. This is used in general to mark a false positive.
553 */
554static void make_gray_object(unsigned long ptr)
555{
556 unsigned long flags;
557 struct kmemleak_object *object;
558
559 object = find_and_get_object(ptr, 0);
560 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100561 kmemleak_warn("Graying unknown object at 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100562 return;
563 }
564
565 spin_lock_irqsave(&object->lock, flags);
566 object->min_count = 0;
567 spin_unlock_irqrestore(&object->lock, flags);
568 put_object(object);
569}
570
571/*
572 * Mark the object as black-colored so that it is ignored from scans and
573 * reporting.
574 */
575static void make_black_object(unsigned long ptr)
576{
577 unsigned long flags;
578 struct kmemleak_object *object;
579
580 object = find_and_get_object(ptr, 0);
581 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100582 kmemleak_warn("Blacking unknown object at 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100583 return;
584 }
585
586 spin_lock_irqsave(&object->lock, flags);
587 object->min_count = -1;
588 spin_unlock_irqrestore(&object->lock, flags);
589 put_object(object);
590}
591
592/*
593 * Add a scanning area to the object. If at least one such area is added,
594 * kmemleak will only scan these ranges rather than the whole memory block.
595 */
596static void add_scan_area(unsigned long ptr, unsigned long offset,
597 size_t length, gfp_t gfp)
598{
599 unsigned long flags;
600 struct kmemleak_object *object;
601 struct kmemleak_scan_area *area;
602
603 object = find_and_get_object(ptr, 0);
604 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100605 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
606 ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100607 return;
608 }
609
Catalin Marinas216c04b2009-06-17 18:29:02 +0100610 area = kmem_cache_alloc(scan_area_cache, gfp & GFP_KMEMLEAK_MASK);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100611 if (!area) {
Joe Perchesae281062009-06-23 14:40:26 +0100612 kmemleak_warn("Cannot allocate a scan area\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100613 goto out;
614 }
615
616 spin_lock_irqsave(&object->lock, flags);
617 if (offset + length > object->size) {
Joe Perchesae281062009-06-23 14:40:26 +0100618 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100619 dump_object_info(object);
620 kmem_cache_free(scan_area_cache, area);
621 goto out_unlock;
622 }
623
624 INIT_HLIST_NODE(&area->node);
625 area->offset = offset;
626 area->length = length;
627
628 hlist_add_head(&area->node, &object->area_list);
629out_unlock:
630 spin_unlock_irqrestore(&object->lock, flags);
631out:
632 put_object(object);
633}
634
635/*
636 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
637 * pointer. Such object will not be scanned by kmemleak but references to it
638 * are searched.
639 */
640static void object_no_scan(unsigned long ptr)
641{
642 unsigned long flags;
643 struct kmemleak_object *object;
644
645 object = find_and_get_object(ptr, 0);
646 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100647 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100648 return;
649 }
650
651 spin_lock_irqsave(&object->lock, flags);
652 object->flags |= OBJECT_NO_SCAN;
653 spin_unlock_irqrestore(&object->lock, flags);
654 put_object(object);
655}
656
657/*
658 * Log an early kmemleak_* call to the early_log buffer. These calls will be
659 * processed later once kmemleak is fully initialized.
660 */
661static void log_early(int op_type, const void *ptr, size_t size,
662 int min_count, unsigned long offset, size_t length)
663{
664 unsigned long flags;
665 struct early_log *log;
666
667 if (crt_early_log >= ARRAY_SIZE(early_log)) {
Catalin Marinasa9d90582009-06-25 10:16:11 +0100668 pr_warning("Early log buffer exceeded\n");
669 kmemleak_disable();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100670 return;
671 }
672
673 /*
674 * There is no need for locking since the kernel is still in UP mode
675 * at this stage. Disabling the IRQs is enough.
676 */
677 local_irq_save(flags);
678 log = &early_log[crt_early_log];
679 log->op_type = op_type;
680 log->ptr = ptr;
681 log->size = size;
682 log->min_count = min_count;
683 log->offset = offset;
684 log->length = length;
685 crt_early_log++;
686 local_irq_restore(flags);
687}
688
689/*
690 * Memory allocation function callback. This function is called from the
691 * kernel allocators when a new block is allocated (kmem_cache_alloc, kmalloc,
692 * vmalloc etc.).
693 */
694void kmemleak_alloc(const void *ptr, size_t size, int min_count, gfp_t gfp)
695{
696 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
697
698 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
699 create_object((unsigned long)ptr, size, min_count, gfp);
700 else if (atomic_read(&kmemleak_early_log))
701 log_early(KMEMLEAK_ALLOC, ptr, size, min_count, 0, 0);
702}
703EXPORT_SYMBOL_GPL(kmemleak_alloc);
704
705/*
706 * Memory freeing function callback. This function is called from the kernel
707 * allocators when a block is freed (kmem_cache_free, kfree, vfree etc.).
708 */
709void kmemleak_free(const void *ptr)
710{
711 pr_debug("%s(0x%p)\n", __func__, ptr);
712
713 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
714 delete_object((unsigned long)ptr);
715 else if (atomic_read(&kmemleak_early_log))
716 log_early(KMEMLEAK_FREE, ptr, 0, 0, 0, 0);
717}
718EXPORT_SYMBOL_GPL(kmemleak_free);
719
720/*
721 * Mark an already allocated memory block as a false positive. This will cause
722 * the block to no longer be reported as leak and always be scanned.
723 */
724void kmemleak_not_leak(const void *ptr)
725{
726 pr_debug("%s(0x%p)\n", __func__, ptr);
727
728 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
729 make_gray_object((unsigned long)ptr);
730 else if (atomic_read(&kmemleak_early_log))
731 log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0, 0, 0);
732}
733EXPORT_SYMBOL(kmemleak_not_leak);
734
735/*
736 * Ignore a memory block. This is usually done when it is known that the
737 * corresponding block is not a leak and does not contain any references to
738 * other allocated memory blocks.
739 */
740void kmemleak_ignore(const void *ptr)
741{
742 pr_debug("%s(0x%p)\n", __func__, ptr);
743
744 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
745 make_black_object((unsigned long)ptr);
746 else if (atomic_read(&kmemleak_early_log))
747 log_early(KMEMLEAK_IGNORE, ptr, 0, 0, 0, 0);
748}
749EXPORT_SYMBOL(kmemleak_ignore);
750
751/*
752 * Limit the range to be scanned in an allocated memory block.
753 */
754void kmemleak_scan_area(const void *ptr, unsigned long offset, size_t length,
755 gfp_t gfp)
756{
757 pr_debug("%s(0x%p)\n", __func__, ptr);
758
759 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
760 add_scan_area((unsigned long)ptr, offset, length, gfp);
761 else if (atomic_read(&kmemleak_early_log))
762 log_early(KMEMLEAK_SCAN_AREA, ptr, 0, 0, offset, length);
763}
764EXPORT_SYMBOL(kmemleak_scan_area);
765
766/*
767 * Inform kmemleak not to scan the given memory block.
768 */
769void kmemleak_no_scan(const void *ptr)
770{
771 pr_debug("%s(0x%p)\n", __func__, ptr);
772
773 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
774 object_no_scan((unsigned long)ptr);
775 else if (atomic_read(&kmemleak_early_log))
776 log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0, 0, 0);
777}
778EXPORT_SYMBOL(kmemleak_no_scan);
779
780/*
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100781 * Memory scanning is a long process and it needs to be interruptable. This
782 * function checks whether such interrupt condition occured.
783 */
784static int scan_should_stop(void)
785{
786 if (!atomic_read(&kmemleak_enabled))
787 return 1;
788
789 /*
790 * This function may be called from either process or kthread context,
791 * hence the need to check for both stop conditions.
792 */
793 if (current->mm)
794 return signal_pending(current);
795 else
796 return kthread_should_stop();
797
798 return 0;
799}
800
801/*
802 * Scan a memory block (exclusive range) for valid pointers and add those
803 * found to the gray list.
804 */
805static void scan_block(void *_start, void *_end,
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100806 struct kmemleak_object *scanned, int allow_resched)
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100807{
808 unsigned long *ptr;
809 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
810 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
811
812 for (ptr = start; ptr < end; ptr++) {
813 unsigned long flags;
814 unsigned long pointer = *ptr;
815 struct kmemleak_object *object;
816
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100817 if (allow_resched)
818 cond_resched();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100819 if (scan_should_stop())
820 break;
821
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100822 object = find_and_get_object(pointer, 1);
823 if (!object)
824 continue;
825 if (object == scanned) {
826 /* self referenced, ignore */
827 put_object(object);
828 continue;
829 }
830
831 /*
832 * Avoid the lockdep recursive warning on object->lock being
833 * previously acquired in scan_object(). These locks are
834 * enclosed by scan_mutex.
835 */
836 spin_lock_irqsave_nested(&object->lock, flags,
837 SINGLE_DEPTH_NESTING);
838 if (!color_white(object)) {
839 /* non-orphan, ignored or new */
840 spin_unlock_irqrestore(&object->lock, flags);
841 put_object(object);
842 continue;
843 }
844
845 /*
846 * Increase the object's reference count (number of pointers
847 * to the memory block). If this count reaches the required
848 * minimum, the object's color will become gray and it will be
849 * added to the gray_list.
850 */
851 object->count++;
852 if (color_gray(object))
853 list_add_tail(&object->gray_list, &gray_list);
854 else
855 put_object(object);
856 spin_unlock_irqrestore(&object->lock, flags);
857 }
858}
859
860/*
861 * Scan a memory block corresponding to a kmemleak_object. A condition is
862 * that object->use_count >= 1.
863 */
864static void scan_object(struct kmemleak_object *object)
865{
866 struct kmemleak_scan_area *area;
867 struct hlist_node *elem;
868 unsigned long flags;
869
870 /*
871 * Once the object->lock is aquired, the corresponding memory block
872 * cannot be freed (the same lock is aquired in delete_object).
873 */
874 spin_lock_irqsave(&object->lock, flags);
875 if (object->flags & OBJECT_NO_SCAN)
876 goto out;
877 if (!(object->flags & OBJECT_ALLOCATED))
878 /* already freed object */
879 goto out;
880 if (hlist_empty(&object->area_list))
881 scan_block((void *)object->pointer,
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100882 (void *)(object->pointer + object->size), object, 0);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100883 else
884 hlist_for_each_entry(area, elem, &object->area_list, node)
885 scan_block((void *)(object->pointer + area->offset),
886 (void *)(object->pointer + area->offset
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100887 + area->length), object, 0);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100888out:
889 spin_unlock_irqrestore(&object->lock, flags);
890}
891
892/*
893 * Scan data sections and all the referenced memory blocks allocated via the
894 * kernel's standard allocators. This function must be called with the
895 * scan_mutex held.
896 */
897static void kmemleak_scan(void)
898{
899 unsigned long flags;
900 struct kmemleak_object *object, *tmp;
901 struct task_struct *task;
902 int i;
Catalin Marinas4698c1f2009-06-26 17:38:27 +0100903 int new_leaks = 0;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100904
Catalin Marinasacf49682009-06-26 17:38:29 +0100905 jiffies_last_scan = jiffies;
906
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100907 /* prepare the kmemleak_object's */
908 rcu_read_lock();
909 list_for_each_entry_rcu(object, &object_list, object_list) {
910 spin_lock_irqsave(&object->lock, flags);
911#ifdef DEBUG
912 /*
913 * With a few exceptions there should be a maximum of
914 * 1 reference to any object at this point.
915 */
916 if (atomic_read(&object->use_count) > 1) {
Joe Perchesae281062009-06-23 14:40:26 +0100917 pr_debug("object->use_count = %d\n",
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100918 atomic_read(&object->use_count));
919 dump_object_info(object);
920 }
921#endif
922 /* reset the reference count (whiten the object) */
923 object->count = 0;
924 if (color_gray(object) && get_object(object))
925 list_add_tail(&object->gray_list, &gray_list);
926
927 spin_unlock_irqrestore(&object->lock, flags);
928 }
929 rcu_read_unlock();
930
931 /* data/bss scanning */
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100932 scan_block(_sdata, _edata, NULL, 1);
933 scan_block(__bss_start, __bss_stop, NULL, 1);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100934
935#ifdef CONFIG_SMP
936 /* per-cpu sections scanning */
937 for_each_possible_cpu(i)
938 scan_block(__per_cpu_start + per_cpu_offset(i),
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100939 __per_cpu_end + per_cpu_offset(i), NULL, 1);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100940#endif
941
942 /*
943 * Struct page scanning for each node. The code below is not yet safe
944 * with MEMORY_HOTPLUG.
945 */
946 for_each_online_node(i) {
947 pg_data_t *pgdat = NODE_DATA(i);
948 unsigned long start_pfn = pgdat->node_start_pfn;
949 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
950 unsigned long pfn;
951
952 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
953 struct page *page;
954
955 if (!pfn_valid(pfn))
956 continue;
957 page = pfn_to_page(pfn);
958 /* only scan if page is in use */
959 if (page_count(page) == 0)
960 continue;
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100961 scan_block(page, page + 1, NULL, 1);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100962 }
963 }
964
965 /*
966 * Scanning the task stacks may introduce false negatives and it is
967 * not enabled by default.
968 */
969 if (kmemleak_stack_scan) {
970 read_lock(&tasklist_lock);
971 for_each_process(task)
972 scan_block(task_stack_page(task),
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100973 task_stack_page(task) + THREAD_SIZE,
974 NULL, 0);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100975 read_unlock(&tasklist_lock);
976 }
977
978 /*
979 * Scan the objects already referenced from the sections scanned
980 * above. More objects will be referenced and, if there are no memory
981 * leaks, all the objects will be scanned. The list traversal is safe
982 * for both tail additions and removals from inside the loop. The
983 * kmemleak objects cannot be freed from outside the loop because their
984 * use_count was increased.
985 */
986 object = list_entry(gray_list.next, typeof(*object), gray_list);
987 while (&object->gray_list != &gray_list) {
Ingo Molnar57d81f62009-07-01 09:43:53 +0200988 cond_resched();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100989
990 /* may add new objects to the list */
991 if (!scan_should_stop())
992 scan_object(object);
993
994 tmp = list_entry(object->gray_list.next, typeof(*object),
995 gray_list);
996
997 /* remove the object from the list and release it */
998 list_del(&object->gray_list);
999 put_object(object);
1000
1001 object = tmp;
1002 }
1003 WARN_ON(!list_empty(&gray_list));
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001004
1005 /*
Catalin Marinas17bb9e02009-06-29 17:13:56 +01001006 * If scanning was stopped do not report any new unreferenced objects.
1007 */
1008 if (scan_should_stop())
1009 return;
1010
1011 /*
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001012 * Scanning result reporting.
1013 */
1014 rcu_read_lock();
1015 list_for_each_entry_rcu(object, &object_list, object_list) {
1016 spin_lock_irqsave(&object->lock, flags);
1017 if (unreferenced_object(object) &&
1018 !(object->flags & OBJECT_REPORTED)) {
1019 object->flags |= OBJECT_REPORTED;
1020 new_leaks++;
1021 }
1022 spin_unlock_irqrestore(&object->lock, flags);
1023 }
1024 rcu_read_unlock();
1025
1026 if (new_leaks)
1027 pr_info("%d new suspected memory leaks (see "
1028 "/sys/kernel/debug/kmemleak)\n", new_leaks);
1029
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001030}
1031
1032/*
1033 * Thread function performing automatic memory scanning. Unreferenced objects
1034 * at the end of a memory scan are reported but only the first time.
1035 */
1036static int kmemleak_scan_thread(void *arg)
1037{
1038 static int first_run = 1;
1039
Joe Perchesae281062009-06-23 14:40:26 +01001040 pr_info("Automatic memory scanning thread started\n");
Catalin Marinasbf2a76b2009-07-07 10:32:55 +01001041 set_user_nice(current, 10);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001042
1043 /*
1044 * Wait before the first scan to allow the system to fully initialize.
1045 */
1046 if (first_run) {
1047 first_run = 0;
1048 ssleep(SECS_FIRST_SCAN);
1049 }
1050
1051 while (!kthread_should_stop()) {
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001052 signed long timeout = jiffies_scan_wait;
1053
1054 mutex_lock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001055 kmemleak_scan();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001056 mutex_unlock(&scan_mutex);
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001057
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001058 /* wait before the next scan */
1059 while (timeout && !kthread_should_stop())
1060 timeout = schedule_timeout_interruptible(timeout);
1061 }
1062
Joe Perchesae281062009-06-23 14:40:26 +01001063 pr_info("Automatic memory scanning thread ended\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001064
1065 return 0;
1066}
1067
1068/*
1069 * Start the automatic memory scanning thread. This function must be called
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001070 * with the scan_mutex held.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001071 */
1072void start_scan_thread(void)
1073{
1074 if (scan_thread)
1075 return;
1076 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1077 if (IS_ERR(scan_thread)) {
Joe Perchesae281062009-06-23 14:40:26 +01001078 pr_warning("Failed to create the scan thread\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001079 scan_thread = NULL;
1080 }
1081}
1082
1083/*
1084 * Stop the automatic memory scanning thread. This function must be called
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001085 * with the scan_mutex held.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001086 */
1087void stop_scan_thread(void)
1088{
1089 if (scan_thread) {
1090 kthread_stop(scan_thread);
1091 scan_thread = NULL;
1092 }
1093}
1094
1095/*
1096 * Iterate over the object_list and return the first valid object at or after
1097 * the required position with its use_count incremented. The function triggers
1098 * a memory scanning when the pos argument points to the first position.
1099 */
1100static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1101{
1102 struct kmemleak_object *object;
1103 loff_t n = *pos;
1104
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001105 rcu_read_lock();
1106 list_for_each_entry_rcu(object, &object_list, object_list) {
1107 if (n-- > 0)
1108 continue;
1109 if (get_object(object))
1110 goto out;
1111 }
1112 object = NULL;
1113out:
1114 rcu_read_unlock();
1115 return object;
1116}
1117
1118/*
1119 * Return the next object in the object_list. The function decrements the
1120 * use_count of the previous object and increases that of the next one.
1121 */
1122static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1123{
1124 struct kmemleak_object *prev_obj = v;
1125 struct kmemleak_object *next_obj = NULL;
1126 struct list_head *n = &prev_obj->object_list;
1127
1128 ++(*pos);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001129
1130 rcu_read_lock();
1131 list_for_each_continue_rcu(n, &object_list) {
1132 next_obj = list_entry(n, struct kmemleak_object, object_list);
1133 if (get_object(next_obj))
1134 break;
1135 }
1136 rcu_read_unlock();
Catalin Marinas288c8572009-07-07 10:32:57 +01001137
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001138 put_object(prev_obj);
1139 return next_obj;
1140}
1141
1142/*
1143 * Decrement the use_count of the last object required, if any.
1144 */
1145static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1146{
1147 if (v)
1148 put_object(v);
1149}
1150
1151/*
1152 * Print the information for an unreferenced object to the seq file.
1153 */
1154static int kmemleak_seq_show(struct seq_file *seq, void *v)
1155{
1156 struct kmemleak_object *object = v;
1157 unsigned long flags;
1158
1159 spin_lock_irqsave(&object->lock, flags);
Catalin Marinas288c8572009-07-07 10:32:57 +01001160 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
Catalin Marinas17bb9e02009-06-29 17:13:56 +01001161 print_unreferenced(seq, object);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001162 spin_unlock_irqrestore(&object->lock, flags);
1163 return 0;
1164}
1165
1166static const struct seq_operations kmemleak_seq_ops = {
1167 .start = kmemleak_seq_start,
1168 .next = kmemleak_seq_next,
1169 .stop = kmemleak_seq_stop,
1170 .show = kmemleak_seq_show,
1171};
1172
1173static int kmemleak_open(struct inode *inode, struct file *file)
1174{
1175 int ret = 0;
1176
1177 if (!atomic_read(&kmemleak_enabled))
1178 return -EBUSY;
1179
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001180 ret = mutex_lock_interruptible(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001181 if (ret < 0)
1182 goto out;
1183 if (file->f_mode & FMODE_READ) {
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001184 ret = seq_open(file, &kmemleak_seq_ops);
1185 if (ret < 0)
1186 goto scan_unlock;
1187 }
1188 return ret;
1189
1190scan_unlock:
1191 mutex_unlock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001192out:
1193 return ret;
1194}
1195
1196static int kmemleak_release(struct inode *inode, struct file *file)
1197{
1198 int ret = 0;
1199
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001200 if (file->f_mode & FMODE_READ)
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001201 seq_release(inode, file);
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001202 mutex_unlock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001203
1204 return ret;
1205}
1206
1207/*
1208 * File write operation to configure kmemleak at run-time. The following
1209 * commands can be written to the /sys/kernel/debug/kmemleak file:
1210 * off - disable kmemleak (irreversible)
1211 * stack=on - enable the task stacks scanning
1212 * stack=off - disable the tasks stacks scanning
1213 * scan=on - start the automatic memory scanning thread
1214 * scan=off - stop the automatic memory scanning thread
1215 * scan=... - set the automatic memory scanning period in seconds (0 to
1216 * disable it)
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001217 * scan - trigger a memory scan
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001218 */
1219static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1220 size_t size, loff_t *ppos)
1221{
1222 char buf[64];
1223 int buf_size;
1224
1225 if (!atomic_read(&kmemleak_enabled))
1226 return -EBUSY;
1227
1228 buf_size = min(size, (sizeof(buf) - 1));
1229 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1230 return -EFAULT;
1231 buf[buf_size] = 0;
1232
1233 if (strncmp(buf, "off", 3) == 0)
1234 kmemleak_disable();
1235 else if (strncmp(buf, "stack=on", 8) == 0)
1236 kmemleak_stack_scan = 1;
1237 else if (strncmp(buf, "stack=off", 9) == 0)
1238 kmemleak_stack_scan = 0;
1239 else if (strncmp(buf, "scan=on", 7) == 0)
1240 start_scan_thread();
1241 else if (strncmp(buf, "scan=off", 8) == 0)
1242 stop_scan_thread();
1243 else if (strncmp(buf, "scan=", 5) == 0) {
1244 unsigned long secs;
1245 int err;
1246
1247 err = strict_strtoul(buf + 5, 0, &secs);
1248 if (err < 0)
1249 return err;
1250 stop_scan_thread();
1251 if (secs) {
1252 jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1253 start_scan_thread();
1254 }
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001255 } else if (strncmp(buf, "scan", 4) == 0)
1256 kmemleak_scan();
1257 else
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001258 return -EINVAL;
1259
1260 /* ignore the rest of the buffer, only one command at a time */
1261 *ppos += size;
1262 return size;
1263}
1264
1265static const struct file_operations kmemleak_fops = {
1266 .owner = THIS_MODULE,
1267 .open = kmemleak_open,
1268 .read = seq_read,
1269 .write = kmemleak_write,
1270 .llseek = seq_lseek,
1271 .release = kmemleak_release,
1272};
1273
1274/*
1275 * Perform the freeing of the kmemleak internal objects after waiting for any
1276 * current memory scan to complete.
1277 */
1278static int kmemleak_cleanup_thread(void *arg)
1279{
1280 struct kmemleak_object *object;
1281
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001282 mutex_lock(&scan_mutex);
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001283 stop_scan_thread();
1284
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001285 rcu_read_lock();
1286 list_for_each_entry_rcu(object, &object_list, object_list)
1287 delete_object(object->pointer);
1288 rcu_read_unlock();
1289 mutex_unlock(&scan_mutex);
1290
1291 return 0;
1292}
1293
1294/*
1295 * Start the clean-up thread.
1296 */
1297static void kmemleak_cleanup(void)
1298{
1299 struct task_struct *cleanup_thread;
1300
1301 cleanup_thread = kthread_run(kmemleak_cleanup_thread, NULL,
1302 "kmemleak-clean");
1303 if (IS_ERR(cleanup_thread))
Joe Perchesae281062009-06-23 14:40:26 +01001304 pr_warning("Failed to create the clean-up thread\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001305}
1306
1307/*
1308 * Disable kmemleak. No memory allocation/freeing will be traced once this
1309 * function is called. Disabling kmemleak is an irreversible operation.
1310 */
1311static void kmemleak_disable(void)
1312{
1313 /* atomically check whether it was already invoked */
1314 if (atomic_cmpxchg(&kmemleak_error, 0, 1))
1315 return;
1316
1317 /* stop any memory operation tracing */
1318 atomic_set(&kmemleak_early_log, 0);
1319 atomic_set(&kmemleak_enabled, 0);
1320
1321 /* check whether it is too early for a kernel thread */
1322 if (atomic_read(&kmemleak_initialized))
1323 kmemleak_cleanup();
1324
1325 pr_info("Kernel memory leak detector disabled\n");
1326}
1327
1328/*
1329 * Allow boot-time kmemleak disabling (enabled by default).
1330 */
1331static int kmemleak_boot_config(char *str)
1332{
1333 if (!str)
1334 return -EINVAL;
1335 if (strcmp(str, "off") == 0)
1336 kmemleak_disable();
1337 else if (strcmp(str, "on") != 0)
1338 return -EINVAL;
1339 return 0;
1340}
1341early_param("kmemleak", kmemleak_boot_config);
1342
1343/*
Catalin Marinas20301172009-06-17 18:29:04 +01001344 * Kmemleak initialization.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001345 */
1346void __init kmemleak_init(void)
1347{
1348 int i;
1349 unsigned long flags;
1350
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001351 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1352 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1353
1354 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1355 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1356 INIT_PRIO_TREE_ROOT(&object_tree_root);
1357
1358 /* the kernel is still in UP mode, so disabling the IRQs is enough */
1359 local_irq_save(flags);
1360 if (!atomic_read(&kmemleak_error)) {
1361 atomic_set(&kmemleak_enabled, 1);
1362 atomic_set(&kmemleak_early_log, 0);
1363 }
1364 local_irq_restore(flags);
1365
1366 /*
1367 * This is the point where tracking allocations is safe. Automatic
1368 * scanning is started during the late initcall. Add the early logged
1369 * callbacks to the kmemleak infrastructure.
1370 */
1371 for (i = 0; i < crt_early_log; i++) {
1372 struct early_log *log = &early_log[i];
1373
1374 switch (log->op_type) {
1375 case KMEMLEAK_ALLOC:
1376 kmemleak_alloc(log->ptr, log->size, log->min_count,
1377 GFP_KERNEL);
1378 break;
1379 case KMEMLEAK_FREE:
1380 kmemleak_free(log->ptr);
1381 break;
1382 case KMEMLEAK_NOT_LEAK:
1383 kmemleak_not_leak(log->ptr);
1384 break;
1385 case KMEMLEAK_IGNORE:
1386 kmemleak_ignore(log->ptr);
1387 break;
1388 case KMEMLEAK_SCAN_AREA:
1389 kmemleak_scan_area(log->ptr, log->offset, log->length,
1390 GFP_KERNEL);
1391 break;
1392 case KMEMLEAK_NO_SCAN:
1393 kmemleak_no_scan(log->ptr);
1394 break;
1395 default:
1396 WARN_ON(1);
1397 }
1398 }
1399}
1400
1401/*
1402 * Late initialization function.
1403 */
1404static int __init kmemleak_late_init(void)
1405{
1406 struct dentry *dentry;
1407
1408 atomic_set(&kmemleak_initialized, 1);
1409
1410 if (atomic_read(&kmemleak_error)) {
1411 /*
1412 * Some error occured and kmemleak was disabled. There is a
1413 * small chance that kmemleak_disable() was called immediately
1414 * after setting kmemleak_initialized and we may end up with
1415 * two clean-up threads but serialized by scan_mutex.
1416 */
1417 kmemleak_cleanup();
1418 return -ENOMEM;
1419 }
1420
1421 dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL,
1422 &kmemleak_fops);
1423 if (!dentry)
Joe Perchesae281062009-06-23 14:40:26 +01001424 pr_warning("Failed to create the debugfs kmemleak file\n");
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001425 mutex_lock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001426 start_scan_thread();
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001427 mutex_unlock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001428
1429 pr_info("Kernel memory leak detector initialized\n");
1430
1431 return 0;
1432}
1433late_initcall(kmemleak_late_init);