blob: 60065531f60c81d9b055ae8fa2d4286274c76b2e [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 */
106#define REPORTS_NR 50 /* maximum number of reported leaks */
107#define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100108#define SECS_FIRST_SCAN 60 /* delay before the first scan */
109#define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
110
111#define BYTES_PER_POINTER sizeof(void *)
112
Catalin Marinas216c04b2009-06-17 18:29:02 +0100113/* GFP bitmask for kmemleak internal allocations */
114#define GFP_KMEMLEAK_MASK (GFP_KERNEL | GFP_ATOMIC)
115
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100116/* scanning area inside a memory block */
117struct kmemleak_scan_area {
118 struct hlist_node node;
119 unsigned long offset;
120 size_t length;
121};
122
123/*
124 * Structure holding the metadata for each allocated memory block.
125 * Modifications to such objects should be made while holding the
126 * object->lock. Insertions or deletions from object_list, gray_list or
127 * tree_node are already protected by the corresponding locks or mutex (see
128 * the notes on locking above). These objects are reference-counted
129 * (use_count) and freed using the RCU mechanism.
130 */
131struct kmemleak_object {
132 spinlock_t lock;
133 unsigned long flags; /* object status flags */
134 struct list_head object_list;
135 struct list_head gray_list;
136 struct prio_tree_node tree_node;
137 struct rcu_head rcu; /* object_list lockless traversal */
138 /* object usage count; object freed when use_count == 0 */
139 atomic_t use_count;
140 unsigned long pointer;
141 size_t size;
142 /* minimum number of a pointers found before it is considered leak */
143 int min_count;
144 /* the total number of pointers found pointing to this object */
145 int count;
146 /* memory ranges to be scanned inside an object (empty for all) */
147 struct hlist_head area_list;
148 unsigned long trace[MAX_TRACE];
149 unsigned int trace_len;
150 unsigned long jiffies; /* creation timestamp */
151 pid_t pid; /* pid of the current task */
152 char comm[TASK_COMM_LEN]; /* executable name */
153};
154
155/* flag representing the memory block allocation status */
156#define OBJECT_ALLOCATED (1 << 0)
157/* flag set after the first reporting of an unreference object */
158#define OBJECT_REPORTED (1 << 1)
159/* flag set to not scan the object */
160#define OBJECT_NO_SCAN (1 << 2)
161
162/* the list of all allocated objects */
163static LIST_HEAD(object_list);
164/* the list of gray-colored objects (see color_gray comment below) */
165static LIST_HEAD(gray_list);
166/* prio search tree for object boundaries */
167static struct prio_tree_root object_tree_root;
168/* rw_lock protecting the access to object_list and prio_tree_root */
169static DEFINE_RWLOCK(kmemleak_lock);
170
171/* allocation caches for kmemleak internal data */
172static struct kmem_cache *object_cache;
173static struct kmem_cache *scan_area_cache;
174
175/* set if tracing memory operations is enabled */
176static atomic_t kmemleak_enabled = ATOMIC_INIT(0);
177/* set in the late_initcall if there were no errors */
178static atomic_t kmemleak_initialized = ATOMIC_INIT(0);
179/* enables or disables early logging of the memory operations */
180static atomic_t kmemleak_early_log = ATOMIC_INIT(1);
181/* set if a fata kmemleak error has occurred */
182static atomic_t kmemleak_error = ATOMIC_INIT(0);
183
184/* minimum and maximum address that may be valid pointers */
185static unsigned long min_addr = ULONG_MAX;
186static unsigned long max_addr;
187
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100188static struct task_struct *scan_thread;
Catalin Marinasacf49682009-06-26 17:38:29 +0100189/* used to avoid reporting of recently allocated objects */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100190static unsigned long jiffies_min_age;
Catalin Marinasacf49682009-06-26 17:38:29 +0100191static unsigned long jiffies_last_scan;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100192/* delay between automatic memory scannings */
193static signed long jiffies_scan_wait;
194/* enables or disables the task stacks scanning */
Catalin Marinase0a2a162009-06-26 17:38:25 +0100195static int kmemleak_stack_scan = 1;
Catalin Marinas4698c1f2009-06-26 17:38:27 +0100196/* protects the memory scanning, parameters and debug/kmemleak file access */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100197static DEFINE_MUTEX(scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100198
199/* number of leaks reported (for limitation purposes) */
200static int reported_leaks;
201
202/*
Catalin Marinas20301172009-06-17 18:29:04 +0100203 * Early object allocation/freeing logging. Kmemleak is initialized after the
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100204 * kernel allocator. However, both the kernel allocator and kmemleak may
Catalin Marinas20301172009-06-17 18:29:04 +0100205 * allocate memory blocks which need to be tracked. Kmemleak defines an
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100206 * 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,
214 KMEMLEAK_NOT_LEAK,
215 KMEMLEAK_IGNORE,
216 KMEMLEAK_SCAN_AREA,
217 KMEMLEAK_NO_SCAN
218};
219
220/*
221 * Structure holding the information passed to kmemleak callbacks during the
222 * early logging.
223 */
224struct early_log {
225 int op_type; /* kmemleak operation type */
226 const void *ptr; /* allocated/freed memory block */
227 size_t size; /* memory block size */
228 int min_count; /* minimum reference count */
229 unsigned long offset; /* scan area offset */
230 size_t length; /* scan area length */
231};
232
233/* early logging buffer and current position */
Catalin Marinasa9d90582009-06-25 10:16:11 +0100234static struct early_log early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE];
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100235static int crt_early_log;
236
237static void kmemleak_disable(void);
238
239/*
240 * Print a warning and dump the stack trace.
241 */
242#define kmemleak_warn(x...) do { \
243 pr_warning(x); \
244 dump_stack(); \
245} while (0)
246
247/*
248 * Macro invoked when a serious kmemleak condition occured and cannot be
Catalin Marinas20301172009-06-17 18:29:04 +0100249 * recovered from. Kmemleak will be disabled and further allocation/freeing
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100250 * tracing no longer available.
251 */
Catalin Marinas000814f2009-06-17 18:29:03 +0100252#define kmemleak_stop(x...) do { \
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100253 kmemleak_warn(x); \
254 kmemleak_disable(); \
255} while (0)
256
257/*
258 * Object colors, encoded with count and min_count:
259 * - white - orphan object, not enough references to it (count < min_count)
260 * - gray - not orphan, not marked as false positive (min_count == 0) or
261 * sufficient references to it (count >= min_count)
262 * - black - ignore, it doesn't contain references (e.g. text section)
263 * (min_count == -1). No function defined for this color.
264 * Newly created objects don't have any color assigned (object->count == -1)
265 * before the next memory scan when they become white.
266 */
267static int color_white(const struct kmemleak_object *object)
268{
269 return object->count != -1 && object->count < object->min_count;
270}
271
272static int color_gray(const struct kmemleak_object *object)
273{
274 return object->min_count != -1 && object->count >= object->min_count;
275}
276
277/*
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100278 * Objects are considered unreferenced only if their color is white, they have
279 * not be deleted and have a minimum age to avoid false positives caused by
280 * pointers temporarily stored in CPU registers.
281 */
282static int unreferenced_object(struct kmemleak_object *object)
283{
284 return (object->flags & OBJECT_ALLOCATED) && color_white(object) &&
Catalin Marinasacf49682009-06-26 17:38:29 +0100285 time_before_eq(object->jiffies + jiffies_min_age,
286 jiffies_last_scan);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100287}
288
289/*
Catalin Marinasbab4a342009-06-26 17:38:26 +0100290 * Printing of the unreferenced objects information to the seq file. The
291 * print_unreferenced function must be called with the object->lock held.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100292 */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100293static void print_unreferenced(struct seq_file *seq,
294 struct kmemleak_object *object)
295{
296 int i;
297
Catalin Marinasbab4a342009-06-26 17:38:26 +0100298 seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
299 object->pointer, object->size);
300 seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu\n",
301 object->comm, object->pid, object->jiffies);
302 seq_printf(seq, " backtrace:\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100303
304 for (i = 0; i < object->trace_len; i++) {
305 void *ptr = (void *)object->trace[i];
Catalin Marinasbab4a342009-06-26 17:38:26 +0100306 seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100307 }
308}
309
310/*
311 * Print the kmemleak_object information. This function is used mainly for
312 * debugging special cases when kmemleak operations. It must be called with
313 * the object->lock held.
314 */
315static void dump_object_info(struct kmemleak_object *object)
316{
317 struct stack_trace trace;
318
319 trace.nr_entries = object->trace_len;
320 trace.entries = object->trace;
321
Joe Perchesae281062009-06-23 14:40:26 +0100322 pr_notice("Object 0x%08lx (size %zu):\n",
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100323 object->tree_node.start, object->size);
324 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
325 object->comm, object->pid, object->jiffies);
326 pr_notice(" min_count = %d\n", object->min_count);
327 pr_notice(" count = %d\n", object->count);
328 pr_notice(" backtrace:\n");
329 print_stack_trace(&trace, 4);
330}
331
332/*
333 * Look-up a memory block metadata (kmemleak_object) in the priority search
334 * tree based on a pointer value. If alias is 0, only values pointing to the
335 * beginning of the memory block are allowed. The kmemleak_lock must be held
336 * when calling this function.
337 */
338static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
339{
340 struct prio_tree_node *node;
341 struct prio_tree_iter iter;
342 struct kmemleak_object *object;
343
344 prio_tree_iter_init(&iter, &object_tree_root, ptr, ptr);
345 node = prio_tree_next(&iter);
346 if (node) {
347 object = prio_tree_entry(node, struct kmemleak_object,
348 tree_node);
349 if (!alias && object->pointer != ptr) {
Joe Perchesae281062009-06-23 14:40:26 +0100350 kmemleak_warn("Found object by alias");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100351 object = NULL;
352 }
353 } else
354 object = NULL;
355
356 return object;
357}
358
359/*
360 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
361 * that once an object's use_count reached 0, the RCU freeing was already
362 * registered and the object should no longer be used. This function must be
363 * called under the protection of rcu_read_lock().
364 */
365static int get_object(struct kmemleak_object *object)
366{
367 return atomic_inc_not_zero(&object->use_count);
368}
369
370/*
371 * RCU callback to free a kmemleak_object.
372 */
373static void free_object_rcu(struct rcu_head *rcu)
374{
375 struct hlist_node *elem, *tmp;
376 struct kmemleak_scan_area *area;
377 struct kmemleak_object *object =
378 container_of(rcu, struct kmemleak_object, rcu);
379
380 /*
381 * Once use_count is 0 (guaranteed by put_object), there is no other
382 * code accessing this object, hence no need for locking.
383 */
384 hlist_for_each_entry_safe(area, elem, tmp, &object->area_list, node) {
385 hlist_del(elem);
386 kmem_cache_free(scan_area_cache, area);
387 }
388 kmem_cache_free(object_cache, object);
389}
390
391/*
392 * Decrement the object use_count. Once the count is 0, free the object using
393 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
394 * delete_object() path, the delayed RCU freeing ensures that there is no
395 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
396 * is also possible.
397 */
398static void put_object(struct kmemleak_object *object)
399{
400 if (!atomic_dec_and_test(&object->use_count))
401 return;
402
403 /* should only get here after delete_object was called */
404 WARN_ON(object->flags & OBJECT_ALLOCATED);
405
406 call_rcu(&object->rcu, free_object_rcu);
407}
408
409/*
410 * Look up an object in the prio search tree and increase its use_count.
411 */
412static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
413{
414 unsigned long flags;
415 struct kmemleak_object *object = NULL;
416
417 rcu_read_lock();
418 read_lock_irqsave(&kmemleak_lock, flags);
419 if (ptr >= min_addr && ptr < max_addr)
420 object = lookup_object(ptr, alias);
421 read_unlock_irqrestore(&kmemleak_lock, flags);
422
423 /* check whether the object is still available */
424 if (object && !get_object(object))
425 object = NULL;
426 rcu_read_unlock();
427
428 return object;
429}
430
431/*
432 * Create the metadata (struct kmemleak_object) corresponding to an allocated
433 * memory block and add it to the object_list and object_tree_root.
434 */
435static void create_object(unsigned long ptr, size_t size, int min_count,
436 gfp_t gfp)
437{
438 unsigned long flags;
439 struct kmemleak_object *object;
440 struct prio_tree_node *node;
441 struct stack_trace trace;
442
Catalin Marinas216c04b2009-06-17 18:29:02 +0100443 object = kmem_cache_alloc(object_cache, gfp & GFP_KMEMLEAK_MASK);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100444 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100445 kmemleak_stop("Cannot allocate a kmemleak_object structure\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100446 return;
447 }
448
449 INIT_LIST_HEAD(&object->object_list);
450 INIT_LIST_HEAD(&object->gray_list);
451 INIT_HLIST_HEAD(&object->area_list);
452 spin_lock_init(&object->lock);
453 atomic_set(&object->use_count, 1);
454 object->flags = OBJECT_ALLOCATED;
455 object->pointer = ptr;
456 object->size = size;
457 object->min_count = min_count;
458 object->count = -1; /* no color initially */
459 object->jiffies = jiffies;
460
461 /* task information */
462 if (in_irq()) {
463 object->pid = 0;
464 strncpy(object->comm, "hardirq", sizeof(object->comm));
465 } else if (in_softirq()) {
466 object->pid = 0;
467 strncpy(object->comm, "softirq", sizeof(object->comm));
468 } else {
469 object->pid = current->pid;
470 /*
471 * There is a small chance of a race with set_task_comm(),
472 * however using get_task_comm() here may cause locking
473 * dependency issues with current->alloc_lock. In the worst
474 * case, the command line is not correct.
475 */
476 strncpy(object->comm, current->comm, sizeof(object->comm));
477 }
478
479 /* kernel backtrace */
480 trace.max_entries = MAX_TRACE;
481 trace.nr_entries = 0;
482 trace.entries = object->trace;
483 trace.skip = 1;
484 save_stack_trace(&trace);
485 object->trace_len = trace.nr_entries;
486
487 INIT_PRIO_TREE_NODE(&object->tree_node);
488 object->tree_node.start = ptr;
489 object->tree_node.last = ptr + size - 1;
490
491 write_lock_irqsave(&kmemleak_lock, flags);
492 min_addr = min(min_addr, ptr);
493 max_addr = max(max_addr, ptr + size);
494 node = prio_tree_insert(&object_tree_root, &object->tree_node);
495 /*
496 * The code calling the kernel does not yet have the pointer to the
497 * memory block to be able to free it. However, we still hold the
498 * kmemleak_lock here in case parts of the kernel started freeing
499 * random memory blocks.
500 */
501 if (node != &object->tree_node) {
502 unsigned long flags;
503
Joe Perchesae281062009-06-23 14:40:26 +0100504 kmemleak_stop("Cannot insert 0x%lx into the object search tree "
505 "(already existing)\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100506 object = lookup_object(ptr, 1);
507 spin_lock_irqsave(&object->lock, flags);
508 dump_object_info(object);
509 spin_unlock_irqrestore(&object->lock, flags);
510
511 goto out;
512 }
513 list_add_tail_rcu(&object->object_list, &object_list);
514out:
515 write_unlock_irqrestore(&kmemleak_lock, flags);
516}
517
518/*
519 * Remove the metadata (struct kmemleak_object) for a memory block from the
520 * object_list and object_tree_root and decrement its use_count.
521 */
522static void delete_object(unsigned long ptr)
523{
524 unsigned long flags;
525 struct kmemleak_object *object;
526
527 write_lock_irqsave(&kmemleak_lock, flags);
528 object = lookup_object(ptr, 0);
529 if (!object) {
Catalin Marinasb6e68722009-06-29 17:13:57 +0100530#ifdef DEBUG
Joe Perchesae281062009-06-23 14:40:26 +0100531 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100532 ptr);
Catalin Marinasb6e68722009-06-29 17:13:57 +0100533#endif
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100534 write_unlock_irqrestore(&kmemleak_lock, flags);
535 return;
536 }
537 prio_tree_remove(&object_tree_root, &object->tree_node);
538 list_del_rcu(&object->object_list);
539 write_unlock_irqrestore(&kmemleak_lock, flags);
540
541 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
542 WARN_ON(atomic_read(&object->use_count) < 1);
543
544 /*
545 * Locking here also ensures that the corresponding memory block
546 * cannot be freed when it is being scanned.
547 */
548 spin_lock_irqsave(&object->lock, flags);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100549 object->flags &= ~OBJECT_ALLOCATED;
550 spin_unlock_irqrestore(&object->lock, flags);
551 put_object(object);
552}
553
554/*
555 * Make a object permanently as gray-colored so that it can no longer be
556 * reported as a leak. This is used in general to mark a false positive.
557 */
558static void make_gray_object(unsigned long ptr)
559{
560 unsigned long flags;
561 struct kmemleak_object *object;
562
563 object = find_and_get_object(ptr, 0);
564 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100565 kmemleak_warn("Graying unknown object at 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100566 return;
567 }
568
569 spin_lock_irqsave(&object->lock, flags);
570 object->min_count = 0;
571 spin_unlock_irqrestore(&object->lock, flags);
572 put_object(object);
573}
574
575/*
576 * Mark the object as black-colored so that it is ignored from scans and
577 * reporting.
578 */
579static void make_black_object(unsigned long ptr)
580{
581 unsigned long flags;
582 struct kmemleak_object *object;
583
584 object = find_and_get_object(ptr, 0);
585 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100586 kmemleak_warn("Blacking unknown object at 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100587 return;
588 }
589
590 spin_lock_irqsave(&object->lock, flags);
591 object->min_count = -1;
592 spin_unlock_irqrestore(&object->lock, flags);
593 put_object(object);
594}
595
596/*
597 * Add a scanning area to the object. If at least one such area is added,
598 * kmemleak will only scan these ranges rather than the whole memory block.
599 */
600static void add_scan_area(unsigned long ptr, unsigned long offset,
601 size_t length, gfp_t gfp)
602{
603 unsigned long flags;
604 struct kmemleak_object *object;
605 struct kmemleak_scan_area *area;
606
607 object = find_and_get_object(ptr, 0);
608 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100609 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
610 ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100611 return;
612 }
613
Catalin Marinas216c04b2009-06-17 18:29:02 +0100614 area = kmem_cache_alloc(scan_area_cache, gfp & GFP_KMEMLEAK_MASK);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100615 if (!area) {
Joe Perchesae281062009-06-23 14:40:26 +0100616 kmemleak_warn("Cannot allocate a scan area\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100617 goto out;
618 }
619
620 spin_lock_irqsave(&object->lock, flags);
621 if (offset + length > object->size) {
Joe Perchesae281062009-06-23 14:40:26 +0100622 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100623 dump_object_info(object);
624 kmem_cache_free(scan_area_cache, area);
625 goto out_unlock;
626 }
627
628 INIT_HLIST_NODE(&area->node);
629 area->offset = offset;
630 area->length = length;
631
632 hlist_add_head(&area->node, &object->area_list);
633out_unlock:
634 spin_unlock_irqrestore(&object->lock, flags);
635out:
636 put_object(object);
637}
638
639/*
640 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
641 * pointer. Such object will not be scanned by kmemleak but references to it
642 * are searched.
643 */
644static void object_no_scan(unsigned long ptr)
645{
646 unsigned long flags;
647 struct kmemleak_object *object;
648
649 object = find_and_get_object(ptr, 0);
650 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100651 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100652 return;
653 }
654
655 spin_lock_irqsave(&object->lock, flags);
656 object->flags |= OBJECT_NO_SCAN;
657 spin_unlock_irqrestore(&object->lock, flags);
658 put_object(object);
659}
660
661/*
662 * Log an early kmemleak_* call to the early_log buffer. These calls will be
663 * processed later once kmemleak is fully initialized.
664 */
665static void log_early(int op_type, const void *ptr, size_t size,
666 int min_count, unsigned long offset, size_t length)
667{
668 unsigned long flags;
669 struct early_log *log;
670
671 if (crt_early_log >= ARRAY_SIZE(early_log)) {
Catalin Marinasa9d90582009-06-25 10:16:11 +0100672 pr_warning("Early log buffer exceeded\n");
673 kmemleak_disable();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100674 return;
675 }
676
677 /*
678 * There is no need for locking since the kernel is still in UP mode
679 * at this stage. Disabling the IRQs is enough.
680 */
681 local_irq_save(flags);
682 log = &early_log[crt_early_log];
683 log->op_type = op_type;
684 log->ptr = ptr;
685 log->size = size;
686 log->min_count = min_count;
687 log->offset = offset;
688 log->length = length;
689 crt_early_log++;
690 local_irq_restore(flags);
691}
692
693/*
694 * Memory allocation function callback. This function is called from the
695 * kernel allocators when a new block is allocated (kmem_cache_alloc, kmalloc,
696 * vmalloc etc.).
697 */
698void kmemleak_alloc(const void *ptr, size_t size, int min_count, gfp_t gfp)
699{
700 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
701
702 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
703 create_object((unsigned long)ptr, size, min_count, gfp);
704 else if (atomic_read(&kmemleak_early_log))
705 log_early(KMEMLEAK_ALLOC, ptr, size, min_count, 0, 0);
706}
707EXPORT_SYMBOL_GPL(kmemleak_alloc);
708
709/*
710 * Memory freeing function callback. This function is called from the kernel
711 * allocators when a block is freed (kmem_cache_free, kfree, vfree etc.).
712 */
713void kmemleak_free(const void *ptr)
714{
715 pr_debug("%s(0x%p)\n", __func__, ptr);
716
717 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
718 delete_object((unsigned long)ptr);
719 else if (atomic_read(&kmemleak_early_log))
720 log_early(KMEMLEAK_FREE, ptr, 0, 0, 0, 0);
721}
722EXPORT_SYMBOL_GPL(kmemleak_free);
723
724/*
725 * Mark an already allocated memory block as a false positive. This will cause
726 * the block to no longer be reported as leak and always be scanned.
727 */
728void kmemleak_not_leak(const void *ptr)
729{
730 pr_debug("%s(0x%p)\n", __func__, ptr);
731
732 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
733 make_gray_object((unsigned long)ptr);
734 else if (atomic_read(&kmemleak_early_log))
735 log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0, 0, 0);
736}
737EXPORT_SYMBOL(kmemleak_not_leak);
738
739/*
740 * Ignore a memory block. This is usually done when it is known that the
741 * corresponding block is not a leak and does not contain any references to
742 * other allocated memory blocks.
743 */
744void kmemleak_ignore(const void *ptr)
745{
746 pr_debug("%s(0x%p)\n", __func__, ptr);
747
748 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
749 make_black_object((unsigned long)ptr);
750 else if (atomic_read(&kmemleak_early_log))
751 log_early(KMEMLEAK_IGNORE, ptr, 0, 0, 0, 0);
752}
753EXPORT_SYMBOL(kmemleak_ignore);
754
755/*
756 * Limit the range to be scanned in an allocated memory block.
757 */
758void kmemleak_scan_area(const void *ptr, unsigned long offset, size_t length,
759 gfp_t gfp)
760{
761 pr_debug("%s(0x%p)\n", __func__, ptr);
762
763 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
764 add_scan_area((unsigned long)ptr, offset, length, gfp);
765 else if (atomic_read(&kmemleak_early_log))
766 log_early(KMEMLEAK_SCAN_AREA, ptr, 0, 0, offset, length);
767}
768EXPORT_SYMBOL(kmemleak_scan_area);
769
770/*
771 * Inform kmemleak not to scan the given memory block.
772 */
773void kmemleak_no_scan(const void *ptr)
774{
775 pr_debug("%s(0x%p)\n", __func__, ptr);
776
777 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
778 object_no_scan((unsigned long)ptr);
779 else if (atomic_read(&kmemleak_early_log))
780 log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0, 0, 0);
781}
782EXPORT_SYMBOL(kmemleak_no_scan);
783
784/*
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100785 * Memory scanning is a long process and it needs to be interruptable. This
786 * function checks whether such interrupt condition occured.
787 */
788static int scan_should_stop(void)
789{
790 if (!atomic_read(&kmemleak_enabled))
791 return 1;
792
793 /*
794 * This function may be called from either process or kthread context,
795 * hence the need to check for both stop conditions.
796 */
797 if (current->mm)
798 return signal_pending(current);
799 else
800 return kthread_should_stop();
801
802 return 0;
803}
804
805/*
806 * Scan a memory block (exclusive range) for valid pointers and add those
807 * found to the gray list.
808 */
809static void scan_block(void *_start, void *_end,
810 struct kmemleak_object *scanned)
811{
812 unsigned long *ptr;
813 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
814 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
815
816 for (ptr = start; ptr < end; ptr++) {
817 unsigned long flags;
818 unsigned long pointer = *ptr;
819 struct kmemleak_object *object;
820
821 if (scan_should_stop())
822 break;
823
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100824 object = find_and_get_object(pointer, 1);
825 if (!object)
826 continue;
827 if (object == scanned) {
828 /* self referenced, ignore */
829 put_object(object);
830 continue;
831 }
832
833 /*
834 * Avoid the lockdep recursive warning on object->lock being
835 * previously acquired in scan_object(). These locks are
836 * enclosed by scan_mutex.
837 */
838 spin_lock_irqsave_nested(&object->lock, flags,
839 SINGLE_DEPTH_NESTING);
840 if (!color_white(object)) {
841 /* non-orphan, ignored or new */
842 spin_unlock_irqrestore(&object->lock, flags);
843 put_object(object);
844 continue;
845 }
846
847 /*
848 * Increase the object's reference count (number of pointers
849 * to the memory block). If this count reaches the required
850 * minimum, the object's color will become gray and it will be
851 * added to the gray_list.
852 */
853 object->count++;
854 if (color_gray(object))
855 list_add_tail(&object->gray_list, &gray_list);
856 else
857 put_object(object);
858 spin_unlock_irqrestore(&object->lock, flags);
859 }
860}
861
862/*
863 * Scan a memory block corresponding to a kmemleak_object. A condition is
864 * that object->use_count >= 1.
865 */
866static void scan_object(struct kmemleak_object *object)
867{
868 struct kmemleak_scan_area *area;
869 struct hlist_node *elem;
870 unsigned long flags;
871
872 /*
873 * Once the object->lock is aquired, the corresponding memory block
874 * cannot be freed (the same lock is aquired in delete_object).
875 */
876 spin_lock_irqsave(&object->lock, flags);
877 if (object->flags & OBJECT_NO_SCAN)
878 goto out;
879 if (!(object->flags & OBJECT_ALLOCATED))
880 /* already freed object */
881 goto out;
882 if (hlist_empty(&object->area_list))
883 scan_block((void *)object->pointer,
884 (void *)(object->pointer + object->size), object);
885 else
886 hlist_for_each_entry(area, elem, &object->area_list, node)
887 scan_block((void *)(object->pointer + area->offset),
888 (void *)(object->pointer + area->offset
889 + area->length), object);
890out:
891 spin_unlock_irqrestore(&object->lock, flags);
892}
893
894/*
895 * Scan data sections and all the referenced memory blocks allocated via the
896 * kernel's standard allocators. This function must be called with the
897 * scan_mutex held.
898 */
899static void kmemleak_scan(void)
900{
901 unsigned long flags;
902 struct kmemleak_object *object, *tmp;
903 struct task_struct *task;
904 int i;
Catalin Marinas4698c1f2009-06-26 17:38:27 +0100905 int new_leaks = 0;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100906
Catalin Marinasacf49682009-06-26 17:38:29 +0100907 jiffies_last_scan = jiffies;
908
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100909 /* prepare the kmemleak_object's */
910 rcu_read_lock();
911 list_for_each_entry_rcu(object, &object_list, object_list) {
912 spin_lock_irqsave(&object->lock, flags);
913#ifdef DEBUG
914 /*
915 * With a few exceptions there should be a maximum of
916 * 1 reference to any object at this point.
917 */
918 if (atomic_read(&object->use_count) > 1) {
Joe Perchesae281062009-06-23 14:40:26 +0100919 pr_debug("object->use_count = %d\n",
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100920 atomic_read(&object->use_count));
921 dump_object_info(object);
922 }
923#endif
924 /* reset the reference count (whiten the object) */
925 object->count = 0;
926 if (color_gray(object) && get_object(object))
927 list_add_tail(&object->gray_list, &gray_list);
928
929 spin_unlock_irqrestore(&object->lock, flags);
930 }
931 rcu_read_unlock();
932
933 /* data/bss scanning */
934 scan_block(_sdata, _edata, NULL);
935 scan_block(__bss_start, __bss_stop, NULL);
936
937#ifdef CONFIG_SMP
938 /* per-cpu sections scanning */
939 for_each_possible_cpu(i)
940 scan_block(__per_cpu_start + per_cpu_offset(i),
941 __per_cpu_end + per_cpu_offset(i), NULL);
942#endif
943
944 /*
945 * Struct page scanning for each node. The code below is not yet safe
946 * with MEMORY_HOTPLUG.
947 */
948 for_each_online_node(i) {
949 pg_data_t *pgdat = NODE_DATA(i);
950 unsigned long start_pfn = pgdat->node_start_pfn;
951 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
952 unsigned long pfn;
953
954 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
955 struct page *page;
956
957 if (!pfn_valid(pfn))
958 continue;
959 page = pfn_to_page(pfn);
960 /* only scan if page is in use */
961 if (page_count(page) == 0)
962 continue;
963 scan_block(page, page + 1, NULL);
964 }
965 }
966
967 /*
968 * Scanning the task stacks may introduce false negatives and it is
969 * not enabled by default.
970 */
971 if (kmemleak_stack_scan) {
972 read_lock(&tasklist_lock);
973 for_each_process(task)
974 scan_block(task_stack_page(task),
975 task_stack_page(task) + THREAD_SIZE, NULL);
976 read_unlock(&tasklist_lock);
977 }
978
979 /*
980 * Scan the objects already referenced from the sections scanned
981 * above. More objects will be referenced and, if there are no memory
982 * leaks, all the objects will be scanned. The list traversal is safe
983 * for both tail additions and removals from inside the loop. The
984 * kmemleak objects cannot be freed from outside the loop because their
985 * use_count was increased.
986 */
987 object = list_entry(gray_list.next, typeof(*object), gray_list);
988 while (&object->gray_list != &gray_list) {
Ingo Molnar57d81f62009-07-01 09:43:53 +0200989 cond_resched();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100990
991 /* may add new objects to the list */
992 if (!scan_should_stop())
993 scan_object(object);
994
995 tmp = list_entry(object->gray_list.next, typeof(*object),
996 gray_list);
997
998 /* remove the object from the list and release it */
999 list_del(&object->gray_list);
1000 put_object(object);
1001
1002 object = tmp;
1003 }
1004 WARN_ON(!list_empty(&gray_list));
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001005
1006 /*
Catalin Marinas17bb9e02009-06-29 17:13:56 +01001007 * If scanning was stopped do not report any new unreferenced objects.
1008 */
1009 if (scan_should_stop())
1010 return;
1011
1012 /*
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001013 * Scanning result reporting.
1014 */
1015 rcu_read_lock();
1016 list_for_each_entry_rcu(object, &object_list, object_list) {
1017 spin_lock_irqsave(&object->lock, flags);
1018 if (unreferenced_object(object) &&
1019 !(object->flags & OBJECT_REPORTED)) {
1020 object->flags |= OBJECT_REPORTED;
1021 new_leaks++;
1022 }
1023 spin_unlock_irqrestore(&object->lock, flags);
1024 }
1025 rcu_read_unlock();
1026
1027 if (new_leaks)
1028 pr_info("%d new suspected memory leaks (see "
1029 "/sys/kernel/debug/kmemleak)\n", new_leaks);
1030
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001031}
1032
1033/*
1034 * Thread function performing automatic memory scanning. Unreferenced objects
1035 * at the end of a memory scan are reported but only the first time.
1036 */
1037static int kmemleak_scan_thread(void *arg)
1038{
1039 static int first_run = 1;
1040
Joe Perchesae281062009-06-23 14:40:26 +01001041 pr_info("Automatic memory scanning thread started\n");
Catalin Marinasbf2a76b2009-07-07 10:32:55 +01001042 set_user_nice(current, 10);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001043
1044 /*
1045 * Wait before the first scan to allow the system to fully initialize.
1046 */
1047 if (first_run) {
1048 first_run = 0;
1049 ssleep(SECS_FIRST_SCAN);
1050 }
1051
1052 while (!kthread_should_stop()) {
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001053 signed long timeout = jiffies_scan_wait;
1054
1055 mutex_lock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001056 kmemleak_scan();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001057 mutex_unlock(&scan_mutex);
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001058
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001059 /* wait before the next scan */
1060 while (timeout && !kthread_should_stop())
1061 timeout = schedule_timeout_interruptible(timeout);
1062 }
1063
Joe Perchesae281062009-06-23 14:40:26 +01001064 pr_info("Automatic memory scanning thread ended\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001065
1066 return 0;
1067}
1068
1069/*
1070 * Start the automatic memory scanning thread. This function must be called
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001071 * with the scan_mutex held.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001072 */
1073void start_scan_thread(void)
1074{
1075 if (scan_thread)
1076 return;
1077 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1078 if (IS_ERR(scan_thread)) {
Joe Perchesae281062009-06-23 14:40:26 +01001079 pr_warning("Failed to create the scan thread\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001080 scan_thread = NULL;
1081 }
1082}
1083
1084/*
1085 * Stop the automatic memory scanning thread. This function must be called
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001086 * with the scan_mutex held.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001087 */
1088void stop_scan_thread(void)
1089{
1090 if (scan_thread) {
1091 kthread_stop(scan_thread);
1092 scan_thread = NULL;
1093 }
1094}
1095
1096/*
1097 * Iterate over the object_list and return the first valid object at or after
1098 * the required position with its use_count incremented. The function triggers
1099 * a memory scanning when the pos argument points to the first position.
1100 */
1101static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1102{
1103 struct kmemleak_object *object;
1104 loff_t n = *pos;
1105
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001106 if (!n)
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001107 reported_leaks = 0;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001108 if (reported_leaks >= REPORTS_NR)
1109 return NULL;
1110
1111 rcu_read_lock();
1112 list_for_each_entry_rcu(object, &object_list, object_list) {
1113 if (n-- > 0)
1114 continue;
1115 if (get_object(object))
1116 goto out;
1117 }
1118 object = NULL;
1119out:
1120 rcu_read_unlock();
1121 return object;
1122}
1123
1124/*
1125 * Return the next object in the object_list. The function decrements the
1126 * use_count of the previous object and increases that of the next one.
1127 */
1128static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1129{
1130 struct kmemleak_object *prev_obj = v;
1131 struct kmemleak_object *next_obj = NULL;
1132 struct list_head *n = &prev_obj->object_list;
1133
1134 ++(*pos);
1135 if (reported_leaks >= REPORTS_NR)
1136 goto out;
1137
1138 rcu_read_lock();
1139 list_for_each_continue_rcu(n, &object_list) {
1140 next_obj = list_entry(n, struct kmemleak_object, object_list);
1141 if (get_object(next_obj))
1142 break;
1143 }
1144 rcu_read_unlock();
1145out:
1146 put_object(prev_obj);
1147 return next_obj;
1148}
1149
1150/*
1151 * Decrement the use_count of the last object required, if any.
1152 */
1153static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1154{
1155 if (v)
1156 put_object(v);
1157}
1158
1159/*
1160 * Print the information for an unreferenced object to the seq file.
1161 */
1162static int kmemleak_seq_show(struct seq_file *seq, void *v)
1163{
1164 struct kmemleak_object *object = v;
1165 unsigned long flags;
1166
1167 spin_lock_irqsave(&object->lock, flags);
Catalin Marinas17bb9e02009-06-29 17:13:56 +01001168 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object)) {
1169 print_unreferenced(seq, object);
1170 reported_leaks++;
1171 }
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001172 spin_unlock_irqrestore(&object->lock, flags);
1173 return 0;
1174}
1175
1176static const struct seq_operations kmemleak_seq_ops = {
1177 .start = kmemleak_seq_start,
1178 .next = kmemleak_seq_next,
1179 .stop = kmemleak_seq_stop,
1180 .show = kmemleak_seq_show,
1181};
1182
1183static int kmemleak_open(struct inode *inode, struct file *file)
1184{
1185 int ret = 0;
1186
1187 if (!atomic_read(&kmemleak_enabled))
1188 return -EBUSY;
1189
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001190 ret = mutex_lock_interruptible(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001191 if (ret < 0)
1192 goto out;
1193 if (file->f_mode & FMODE_READ) {
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001194 ret = seq_open(file, &kmemleak_seq_ops);
1195 if (ret < 0)
1196 goto scan_unlock;
1197 }
1198 return ret;
1199
1200scan_unlock:
1201 mutex_unlock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001202out:
1203 return ret;
1204}
1205
1206static int kmemleak_release(struct inode *inode, struct file *file)
1207{
1208 int ret = 0;
1209
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001210 if (file->f_mode & FMODE_READ)
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001211 seq_release(inode, file);
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001212 mutex_unlock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001213
1214 return ret;
1215}
1216
1217/*
1218 * File write operation to configure kmemleak at run-time. The following
1219 * commands can be written to the /sys/kernel/debug/kmemleak file:
1220 * off - disable kmemleak (irreversible)
1221 * stack=on - enable the task stacks scanning
1222 * stack=off - disable the tasks stacks scanning
1223 * scan=on - start the automatic memory scanning thread
1224 * scan=off - stop the automatic memory scanning thread
1225 * scan=... - set the automatic memory scanning period in seconds (0 to
1226 * disable it)
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001227 * scan - trigger a memory scan
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001228 */
1229static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1230 size_t size, loff_t *ppos)
1231{
1232 char buf[64];
1233 int buf_size;
1234
1235 if (!atomic_read(&kmemleak_enabled))
1236 return -EBUSY;
1237
1238 buf_size = min(size, (sizeof(buf) - 1));
1239 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1240 return -EFAULT;
1241 buf[buf_size] = 0;
1242
1243 if (strncmp(buf, "off", 3) == 0)
1244 kmemleak_disable();
1245 else if (strncmp(buf, "stack=on", 8) == 0)
1246 kmemleak_stack_scan = 1;
1247 else if (strncmp(buf, "stack=off", 9) == 0)
1248 kmemleak_stack_scan = 0;
1249 else if (strncmp(buf, "scan=on", 7) == 0)
1250 start_scan_thread();
1251 else if (strncmp(buf, "scan=off", 8) == 0)
1252 stop_scan_thread();
1253 else if (strncmp(buf, "scan=", 5) == 0) {
1254 unsigned long secs;
1255 int err;
1256
1257 err = strict_strtoul(buf + 5, 0, &secs);
1258 if (err < 0)
1259 return err;
1260 stop_scan_thread();
1261 if (secs) {
1262 jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1263 start_scan_thread();
1264 }
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001265 } else if (strncmp(buf, "scan", 4) == 0)
1266 kmemleak_scan();
1267 else
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001268 return -EINVAL;
1269
1270 /* ignore the rest of the buffer, only one command at a time */
1271 *ppos += size;
1272 return size;
1273}
1274
1275static const struct file_operations kmemleak_fops = {
1276 .owner = THIS_MODULE,
1277 .open = kmemleak_open,
1278 .read = seq_read,
1279 .write = kmemleak_write,
1280 .llseek = seq_lseek,
1281 .release = kmemleak_release,
1282};
1283
1284/*
1285 * Perform the freeing of the kmemleak internal objects after waiting for any
1286 * current memory scan to complete.
1287 */
1288static int kmemleak_cleanup_thread(void *arg)
1289{
1290 struct kmemleak_object *object;
1291
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001292 mutex_lock(&scan_mutex);
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001293 stop_scan_thread();
1294
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001295 rcu_read_lock();
1296 list_for_each_entry_rcu(object, &object_list, object_list)
1297 delete_object(object->pointer);
1298 rcu_read_unlock();
1299 mutex_unlock(&scan_mutex);
1300
1301 return 0;
1302}
1303
1304/*
1305 * Start the clean-up thread.
1306 */
1307static void kmemleak_cleanup(void)
1308{
1309 struct task_struct *cleanup_thread;
1310
1311 cleanup_thread = kthread_run(kmemleak_cleanup_thread, NULL,
1312 "kmemleak-clean");
1313 if (IS_ERR(cleanup_thread))
Joe Perchesae281062009-06-23 14:40:26 +01001314 pr_warning("Failed to create the clean-up thread\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001315}
1316
1317/*
1318 * Disable kmemleak. No memory allocation/freeing will be traced once this
1319 * function is called. Disabling kmemleak is an irreversible operation.
1320 */
1321static void kmemleak_disable(void)
1322{
1323 /* atomically check whether it was already invoked */
1324 if (atomic_cmpxchg(&kmemleak_error, 0, 1))
1325 return;
1326
1327 /* stop any memory operation tracing */
1328 atomic_set(&kmemleak_early_log, 0);
1329 atomic_set(&kmemleak_enabled, 0);
1330
1331 /* check whether it is too early for a kernel thread */
1332 if (atomic_read(&kmemleak_initialized))
1333 kmemleak_cleanup();
1334
1335 pr_info("Kernel memory leak detector disabled\n");
1336}
1337
1338/*
1339 * Allow boot-time kmemleak disabling (enabled by default).
1340 */
1341static int kmemleak_boot_config(char *str)
1342{
1343 if (!str)
1344 return -EINVAL;
1345 if (strcmp(str, "off") == 0)
1346 kmemleak_disable();
1347 else if (strcmp(str, "on") != 0)
1348 return -EINVAL;
1349 return 0;
1350}
1351early_param("kmemleak", kmemleak_boot_config);
1352
1353/*
Catalin Marinas20301172009-06-17 18:29:04 +01001354 * Kmemleak initialization.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001355 */
1356void __init kmemleak_init(void)
1357{
1358 int i;
1359 unsigned long flags;
1360
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001361 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1362 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1363
1364 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1365 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1366 INIT_PRIO_TREE_ROOT(&object_tree_root);
1367
1368 /* the kernel is still in UP mode, so disabling the IRQs is enough */
1369 local_irq_save(flags);
1370 if (!atomic_read(&kmemleak_error)) {
1371 atomic_set(&kmemleak_enabled, 1);
1372 atomic_set(&kmemleak_early_log, 0);
1373 }
1374 local_irq_restore(flags);
1375
1376 /*
1377 * This is the point where tracking allocations is safe. Automatic
1378 * scanning is started during the late initcall. Add the early logged
1379 * callbacks to the kmemleak infrastructure.
1380 */
1381 for (i = 0; i < crt_early_log; i++) {
1382 struct early_log *log = &early_log[i];
1383
1384 switch (log->op_type) {
1385 case KMEMLEAK_ALLOC:
1386 kmemleak_alloc(log->ptr, log->size, log->min_count,
1387 GFP_KERNEL);
1388 break;
1389 case KMEMLEAK_FREE:
1390 kmemleak_free(log->ptr);
1391 break;
1392 case KMEMLEAK_NOT_LEAK:
1393 kmemleak_not_leak(log->ptr);
1394 break;
1395 case KMEMLEAK_IGNORE:
1396 kmemleak_ignore(log->ptr);
1397 break;
1398 case KMEMLEAK_SCAN_AREA:
1399 kmemleak_scan_area(log->ptr, log->offset, log->length,
1400 GFP_KERNEL);
1401 break;
1402 case KMEMLEAK_NO_SCAN:
1403 kmemleak_no_scan(log->ptr);
1404 break;
1405 default:
1406 WARN_ON(1);
1407 }
1408 }
1409}
1410
1411/*
1412 * Late initialization function.
1413 */
1414static int __init kmemleak_late_init(void)
1415{
1416 struct dentry *dentry;
1417
1418 atomic_set(&kmemleak_initialized, 1);
1419
1420 if (atomic_read(&kmemleak_error)) {
1421 /*
1422 * Some error occured and kmemleak was disabled. There is a
1423 * small chance that kmemleak_disable() was called immediately
1424 * after setting kmemleak_initialized and we may end up with
1425 * two clean-up threads but serialized by scan_mutex.
1426 */
1427 kmemleak_cleanup();
1428 return -ENOMEM;
1429 }
1430
1431 dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL,
1432 &kmemleak_fops);
1433 if (!dentry)
Joe Perchesae281062009-06-23 14:40:26 +01001434 pr_warning("Failed to create the debugfs kmemleak file\n");
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001435 mutex_lock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001436 start_scan_thread();
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001437 mutex_unlock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001438
1439 pr_info("Kernel memory leak detector initialized\n");
1440
1441 return 0;
1442}
1443late_initcall(kmemleak_late_init);