blob: 487267310a8447a175ede900f69c35ca55d63dba [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 */
Catalin Marinas25873622009-07-07 10:32:58 +0100109#define GRAY_LIST_PASSES 25 /* maximum number of gray list scans */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100110
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)
Catalin Marinas25873622009-07-07 10:32:58 +0100161/* flag set on newly allocated objects */
162#define OBJECT_NEW (1 << 3)
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100163
164/* the list of all allocated objects */
165static LIST_HEAD(object_list);
166/* the list of gray-colored objects (see color_gray comment below) */
167static LIST_HEAD(gray_list);
168/* prio search tree for object boundaries */
169static struct prio_tree_root object_tree_root;
170/* rw_lock protecting the access to object_list and prio_tree_root */
171static DEFINE_RWLOCK(kmemleak_lock);
172
173/* allocation caches for kmemleak internal data */
174static struct kmem_cache *object_cache;
175static struct kmem_cache *scan_area_cache;
176
177/* set if tracing memory operations is enabled */
178static atomic_t kmemleak_enabled = ATOMIC_INIT(0);
179/* set in the late_initcall if there were no errors */
180static atomic_t kmemleak_initialized = ATOMIC_INIT(0);
181/* enables or disables early logging of the memory operations */
182static atomic_t kmemleak_early_log = ATOMIC_INIT(1);
183/* set if a fata kmemleak error has occurred */
184static atomic_t kmemleak_error = ATOMIC_INIT(0);
185
186/* minimum and maximum address that may be valid pointers */
187static unsigned long min_addr = ULONG_MAX;
188static unsigned long max_addr;
189
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100190static struct task_struct *scan_thread;
Catalin Marinasacf49682009-06-26 17:38:29 +0100191/* used to avoid reporting of recently allocated objects */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100192static unsigned long jiffies_min_age;
Catalin Marinasacf49682009-06-26 17:38:29 +0100193static unsigned long jiffies_last_scan;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100194/* delay between automatic memory scannings */
195static signed long jiffies_scan_wait;
196/* enables or disables the task stacks scanning */
Catalin Marinase0a2a162009-06-26 17:38:25 +0100197static int kmemleak_stack_scan = 1;
Catalin Marinas4698c1f2009-06-26 17:38:27 +0100198/* protects the memory scanning, parameters and debug/kmemleak file access */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100199static DEFINE_MUTEX(scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100200
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100201/*
Catalin Marinas20301172009-06-17 18:29:04 +0100202 * Early object allocation/freeing logging. Kmemleak is initialized after the
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100203 * kernel allocator. However, both the kernel allocator and kmemleak may
Catalin Marinas20301172009-06-17 18:29:04 +0100204 * allocate memory blocks which need to be tracked. Kmemleak defines an
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100205 * arbitrary buffer to hold the allocation/freeing information before it is
206 * fully initialized.
207 */
208
209/* kmemleak operation type for early logging */
210enum {
211 KMEMLEAK_ALLOC,
212 KMEMLEAK_FREE,
Catalin Marinas53238a62009-07-07 10:33:00 +0100213 KMEMLEAK_FREE_PART,
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100214 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
Catalin Marinas25873622009-07-07 10:32:58 +0100277static int color_black(const struct kmemleak_object *object)
278{
279 return object->min_count == -1;
280}
281
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100282/*
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100283 * Objects are considered unreferenced only if their color is white, they have
284 * not be deleted and have a minimum age to avoid false positives caused by
285 * pointers temporarily stored in CPU registers.
286 */
287static int unreferenced_object(struct kmemleak_object *object)
288{
289 return (object->flags & OBJECT_ALLOCATED) && color_white(object) &&
Catalin Marinasacf49682009-06-26 17:38:29 +0100290 time_before_eq(object->jiffies + jiffies_min_age,
291 jiffies_last_scan);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100292}
293
294/*
Catalin Marinasbab4a342009-06-26 17:38:26 +0100295 * Printing of the unreferenced objects information to the seq file. The
296 * print_unreferenced function must be called with the object->lock held.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100297 */
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100298static void print_unreferenced(struct seq_file *seq,
299 struct kmemleak_object *object)
300{
301 int i;
302
Catalin Marinasbab4a342009-06-26 17:38:26 +0100303 seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
304 object->pointer, object->size);
305 seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu\n",
306 object->comm, object->pid, object->jiffies);
307 seq_printf(seq, " backtrace:\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100308
309 for (i = 0; i < object->trace_len; i++) {
310 void *ptr = (void *)object->trace[i];
Catalin Marinasbab4a342009-06-26 17:38:26 +0100311 seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100312 }
313}
314
315/*
316 * Print the kmemleak_object information. This function is used mainly for
317 * debugging special cases when kmemleak operations. It must be called with
318 * the object->lock held.
319 */
320static void dump_object_info(struct kmemleak_object *object)
321{
322 struct stack_trace trace;
323
324 trace.nr_entries = object->trace_len;
325 trace.entries = object->trace;
326
Joe Perchesae281062009-06-23 14:40:26 +0100327 pr_notice("Object 0x%08lx (size %zu):\n",
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100328 object->tree_node.start, object->size);
329 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
330 object->comm, object->pid, object->jiffies);
331 pr_notice(" min_count = %d\n", object->min_count);
332 pr_notice(" count = %d\n", object->count);
333 pr_notice(" backtrace:\n");
334 print_stack_trace(&trace, 4);
335}
336
337/*
338 * Look-up a memory block metadata (kmemleak_object) in the priority search
339 * tree based on a pointer value. If alias is 0, only values pointing to the
340 * beginning of the memory block are allowed. The kmemleak_lock must be held
341 * when calling this function.
342 */
343static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
344{
345 struct prio_tree_node *node;
346 struct prio_tree_iter iter;
347 struct kmemleak_object *object;
348
349 prio_tree_iter_init(&iter, &object_tree_root, ptr, ptr);
350 node = prio_tree_next(&iter);
351 if (node) {
352 object = prio_tree_entry(node, struct kmemleak_object,
353 tree_node);
354 if (!alias && object->pointer != ptr) {
Joe Perchesae281062009-06-23 14:40:26 +0100355 kmemleak_warn("Found object by alias");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100356 object = NULL;
357 }
358 } else
359 object = NULL;
360
361 return object;
362}
363
364/*
365 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
366 * that once an object's use_count reached 0, the RCU freeing was already
367 * registered and the object should no longer be used. This function must be
368 * called under the protection of rcu_read_lock().
369 */
370static int get_object(struct kmemleak_object *object)
371{
372 return atomic_inc_not_zero(&object->use_count);
373}
374
375/*
376 * RCU callback to free a kmemleak_object.
377 */
378static void free_object_rcu(struct rcu_head *rcu)
379{
380 struct hlist_node *elem, *tmp;
381 struct kmemleak_scan_area *area;
382 struct kmemleak_object *object =
383 container_of(rcu, struct kmemleak_object, rcu);
384
385 /*
386 * Once use_count is 0 (guaranteed by put_object), there is no other
387 * code accessing this object, hence no need for locking.
388 */
389 hlist_for_each_entry_safe(area, elem, tmp, &object->area_list, node) {
390 hlist_del(elem);
391 kmem_cache_free(scan_area_cache, area);
392 }
393 kmem_cache_free(object_cache, object);
394}
395
396/*
397 * Decrement the object use_count. Once the count is 0, free the object using
398 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
399 * delete_object() path, the delayed RCU freeing ensures that there is no
400 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
401 * is also possible.
402 */
403static void put_object(struct kmemleak_object *object)
404{
405 if (!atomic_dec_and_test(&object->use_count))
406 return;
407
408 /* should only get here after delete_object was called */
409 WARN_ON(object->flags & OBJECT_ALLOCATED);
410
411 call_rcu(&object->rcu, free_object_rcu);
412}
413
414/*
415 * Look up an object in the prio search tree and increase its use_count.
416 */
417static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
418{
419 unsigned long flags;
420 struct kmemleak_object *object = NULL;
421
422 rcu_read_lock();
423 read_lock_irqsave(&kmemleak_lock, flags);
424 if (ptr >= min_addr && ptr < max_addr)
425 object = lookup_object(ptr, alias);
426 read_unlock_irqrestore(&kmemleak_lock, flags);
427
428 /* check whether the object is still available */
429 if (object && !get_object(object))
430 object = NULL;
431 rcu_read_unlock();
432
433 return object;
434}
435
436/*
437 * Create the metadata (struct kmemleak_object) corresponding to an allocated
438 * memory block and add it to the object_list and object_tree_root.
439 */
440static void create_object(unsigned long ptr, size_t size, int min_count,
441 gfp_t gfp)
442{
443 unsigned long flags;
444 struct kmemleak_object *object;
445 struct prio_tree_node *node;
446 struct stack_trace trace;
447
Catalin Marinas216c04b2009-06-17 18:29:02 +0100448 object = kmem_cache_alloc(object_cache, gfp & GFP_KMEMLEAK_MASK);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100449 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100450 kmemleak_stop("Cannot allocate a kmemleak_object structure\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100451 return;
452 }
453
454 INIT_LIST_HEAD(&object->object_list);
455 INIT_LIST_HEAD(&object->gray_list);
456 INIT_HLIST_HEAD(&object->area_list);
457 spin_lock_init(&object->lock);
458 atomic_set(&object->use_count, 1);
Catalin Marinas25873622009-07-07 10:32:58 +0100459 object->flags = OBJECT_ALLOCATED | OBJECT_NEW;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100460 object->pointer = ptr;
461 object->size = size;
462 object->min_count = min_count;
463 object->count = -1; /* no color initially */
464 object->jiffies = jiffies;
465
466 /* task information */
467 if (in_irq()) {
468 object->pid = 0;
469 strncpy(object->comm, "hardirq", sizeof(object->comm));
470 } else if (in_softirq()) {
471 object->pid = 0;
472 strncpy(object->comm, "softirq", sizeof(object->comm));
473 } else {
474 object->pid = current->pid;
475 /*
476 * There is a small chance of a race with set_task_comm(),
477 * however using get_task_comm() here may cause locking
478 * dependency issues with current->alloc_lock. In the worst
479 * case, the command line is not correct.
480 */
481 strncpy(object->comm, current->comm, sizeof(object->comm));
482 }
483
484 /* kernel backtrace */
485 trace.max_entries = MAX_TRACE;
486 trace.nr_entries = 0;
487 trace.entries = object->trace;
488 trace.skip = 1;
489 save_stack_trace(&trace);
490 object->trace_len = trace.nr_entries;
491
492 INIT_PRIO_TREE_NODE(&object->tree_node);
493 object->tree_node.start = ptr;
494 object->tree_node.last = ptr + size - 1;
495
496 write_lock_irqsave(&kmemleak_lock, flags);
497 min_addr = min(min_addr, ptr);
498 max_addr = max(max_addr, ptr + size);
499 node = prio_tree_insert(&object_tree_root, &object->tree_node);
500 /*
501 * The code calling the kernel does not yet have the pointer to the
502 * memory block to be able to free it. However, we still hold the
503 * kmemleak_lock here in case parts of the kernel started freeing
504 * random memory blocks.
505 */
506 if (node != &object->tree_node) {
507 unsigned long flags;
508
Joe Perchesae281062009-06-23 14:40:26 +0100509 kmemleak_stop("Cannot insert 0x%lx into the object search tree "
510 "(already existing)\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100511 object = lookup_object(ptr, 1);
512 spin_lock_irqsave(&object->lock, flags);
513 dump_object_info(object);
514 spin_unlock_irqrestore(&object->lock, flags);
515
516 goto out;
517 }
518 list_add_tail_rcu(&object->object_list, &object_list);
519out:
520 write_unlock_irqrestore(&kmemleak_lock, flags);
521}
522
523/*
524 * Remove the metadata (struct kmemleak_object) for a memory block from the
525 * object_list and object_tree_root and decrement its use_count.
526 */
Catalin Marinas53238a62009-07-07 10:33:00 +0100527static void __delete_object(struct kmemleak_object *object)
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100528{
529 unsigned long flags;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100530
531 write_lock_irqsave(&kmemleak_lock, flags);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100532 prio_tree_remove(&object_tree_root, &object->tree_node);
533 list_del_rcu(&object->object_list);
534 write_unlock_irqrestore(&kmemleak_lock, flags);
535
536 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
Catalin Marinas53238a62009-07-07 10:33:00 +0100537 WARN_ON(atomic_read(&object->use_count) < 2);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100538
539 /*
540 * Locking here also ensures that the corresponding memory block
541 * cannot be freed when it is being scanned.
542 */
543 spin_lock_irqsave(&object->lock, flags);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100544 object->flags &= ~OBJECT_ALLOCATED;
545 spin_unlock_irqrestore(&object->lock, flags);
546 put_object(object);
547}
548
549/*
Catalin Marinas53238a62009-07-07 10:33:00 +0100550 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
551 * delete it.
552 */
553static void delete_object_full(unsigned long ptr)
554{
555 struct kmemleak_object *object;
556
557 object = find_and_get_object(ptr, 0);
558 if (!object) {
559#ifdef DEBUG
560 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
561 ptr);
562#endif
563 return;
564 }
565 __delete_object(object);
566 put_object(object);
567}
568
569/*
570 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
571 * delete it. If the memory block is partially freed, the function may create
572 * additional metadata for the remaining parts of the block.
573 */
574static void delete_object_part(unsigned long ptr, size_t size)
575{
576 struct kmemleak_object *object;
577 unsigned long start, end;
578
579 object = find_and_get_object(ptr, 1);
580 if (!object) {
581#ifdef DEBUG
582 kmemleak_warn("Partially freeing unknown object at 0x%08lx "
583 "(size %zu)\n", ptr, size);
584#endif
585 return;
586 }
587 __delete_object(object);
588
589 /*
590 * Create one or two objects that may result from the memory block
591 * split. Note that partial freeing is only done by free_bootmem() and
592 * this happens before kmemleak_init() is called. The path below is
593 * only executed during early log recording in kmemleak_init(), so
594 * GFP_KERNEL is enough.
595 */
596 start = object->pointer;
597 end = object->pointer + object->size;
598 if (ptr > start)
599 create_object(start, ptr - start, object->min_count,
600 GFP_KERNEL);
601 if (ptr + size < end)
602 create_object(ptr + size, end - ptr - size, object->min_count,
603 GFP_KERNEL);
604
605 put_object(object);
606}
607/*
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100608 * Make a object permanently as gray-colored so that it can no longer be
609 * reported as a leak. This is used in general to mark a false positive.
610 */
611static void make_gray_object(unsigned long ptr)
612{
613 unsigned long flags;
614 struct kmemleak_object *object;
615
616 object = find_and_get_object(ptr, 0);
617 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100618 kmemleak_warn("Graying unknown object at 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100619 return;
620 }
621
622 spin_lock_irqsave(&object->lock, flags);
623 object->min_count = 0;
624 spin_unlock_irqrestore(&object->lock, flags);
625 put_object(object);
626}
627
628/*
629 * Mark the object as black-colored so that it is ignored from scans and
630 * reporting.
631 */
632static void make_black_object(unsigned long ptr)
633{
634 unsigned long flags;
635 struct kmemleak_object *object;
636
637 object = find_and_get_object(ptr, 0);
638 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100639 kmemleak_warn("Blacking unknown object at 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100640 return;
641 }
642
643 spin_lock_irqsave(&object->lock, flags);
644 object->min_count = -1;
645 spin_unlock_irqrestore(&object->lock, flags);
646 put_object(object);
647}
648
649/*
650 * Add a scanning area to the object. If at least one such area is added,
651 * kmemleak will only scan these ranges rather than the whole memory block.
652 */
653static void add_scan_area(unsigned long ptr, unsigned long offset,
654 size_t length, gfp_t gfp)
655{
656 unsigned long flags;
657 struct kmemleak_object *object;
658 struct kmemleak_scan_area *area;
659
660 object = find_and_get_object(ptr, 0);
661 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100662 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
663 ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100664 return;
665 }
666
Catalin Marinas216c04b2009-06-17 18:29:02 +0100667 area = kmem_cache_alloc(scan_area_cache, gfp & GFP_KMEMLEAK_MASK);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100668 if (!area) {
Joe Perchesae281062009-06-23 14:40:26 +0100669 kmemleak_warn("Cannot allocate a scan area\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100670 goto out;
671 }
672
673 spin_lock_irqsave(&object->lock, flags);
674 if (offset + length > object->size) {
Joe Perchesae281062009-06-23 14:40:26 +0100675 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100676 dump_object_info(object);
677 kmem_cache_free(scan_area_cache, area);
678 goto out_unlock;
679 }
680
681 INIT_HLIST_NODE(&area->node);
682 area->offset = offset;
683 area->length = length;
684
685 hlist_add_head(&area->node, &object->area_list);
686out_unlock:
687 spin_unlock_irqrestore(&object->lock, flags);
688out:
689 put_object(object);
690}
691
692/*
693 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
694 * pointer. Such object will not be scanned by kmemleak but references to it
695 * are searched.
696 */
697static void object_no_scan(unsigned long ptr)
698{
699 unsigned long flags;
700 struct kmemleak_object *object;
701
702 object = find_and_get_object(ptr, 0);
703 if (!object) {
Joe Perchesae281062009-06-23 14:40:26 +0100704 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100705 return;
706 }
707
708 spin_lock_irqsave(&object->lock, flags);
709 object->flags |= OBJECT_NO_SCAN;
710 spin_unlock_irqrestore(&object->lock, flags);
711 put_object(object);
712}
713
714/*
715 * Log an early kmemleak_* call to the early_log buffer. These calls will be
716 * processed later once kmemleak is fully initialized.
717 */
718static void log_early(int op_type, const void *ptr, size_t size,
719 int min_count, unsigned long offset, size_t length)
720{
721 unsigned long flags;
722 struct early_log *log;
723
724 if (crt_early_log >= ARRAY_SIZE(early_log)) {
Catalin Marinasa9d90582009-06-25 10:16:11 +0100725 pr_warning("Early log buffer exceeded\n");
726 kmemleak_disable();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100727 return;
728 }
729
730 /*
731 * There is no need for locking since the kernel is still in UP mode
732 * at this stage. Disabling the IRQs is enough.
733 */
734 local_irq_save(flags);
735 log = &early_log[crt_early_log];
736 log->op_type = op_type;
737 log->ptr = ptr;
738 log->size = size;
739 log->min_count = min_count;
740 log->offset = offset;
741 log->length = length;
742 crt_early_log++;
743 local_irq_restore(flags);
744}
745
746/*
747 * Memory allocation function callback. This function is called from the
748 * kernel allocators when a new block is allocated (kmem_cache_alloc, kmalloc,
749 * vmalloc etc.).
750 */
751void kmemleak_alloc(const void *ptr, size_t size, int min_count, gfp_t gfp)
752{
753 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
754
755 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
756 create_object((unsigned long)ptr, size, min_count, gfp);
757 else if (atomic_read(&kmemleak_early_log))
758 log_early(KMEMLEAK_ALLOC, ptr, size, min_count, 0, 0);
759}
760EXPORT_SYMBOL_GPL(kmemleak_alloc);
761
762/*
763 * Memory freeing function callback. This function is called from the kernel
764 * allocators when a block is freed (kmem_cache_free, kfree, vfree etc.).
765 */
766void kmemleak_free(const void *ptr)
767{
768 pr_debug("%s(0x%p)\n", __func__, ptr);
769
770 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
Catalin Marinas53238a62009-07-07 10:33:00 +0100771 delete_object_full((unsigned long)ptr);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100772 else if (atomic_read(&kmemleak_early_log))
773 log_early(KMEMLEAK_FREE, ptr, 0, 0, 0, 0);
774}
775EXPORT_SYMBOL_GPL(kmemleak_free);
776
777/*
Catalin Marinas53238a62009-07-07 10:33:00 +0100778 * Partial memory freeing function callback. This function is usually called
779 * from bootmem allocator when (part of) a memory block is freed.
780 */
781void kmemleak_free_part(const void *ptr, size_t size)
782{
783 pr_debug("%s(0x%p)\n", __func__, ptr);
784
785 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
786 delete_object_part((unsigned long)ptr, size);
787 else if (atomic_read(&kmemleak_early_log))
788 log_early(KMEMLEAK_FREE_PART, ptr, size, 0, 0, 0);
789}
790EXPORT_SYMBOL_GPL(kmemleak_free_part);
791
792/*
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100793 * Mark an already allocated memory block as a false positive. This will cause
794 * the block to no longer be reported as leak and always be scanned.
795 */
796void kmemleak_not_leak(const void *ptr)
797{
798 pr_debug("%s(0x%p)\n", __func__, ptr);
799
800 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
801 make_gray_object((unsigned long)ptr);
802 else if (atomic_read(&kmemleak_early_log))
803 log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0, 0, 0);
804}
805EXPORT_SYMBOL(kmemleak_not_leak);
806
807/*
808 * Ignore a memory block. This is usually done when it is known that the
809 * corresponding block is not a leak and does not contain any references to
810 * other allocated memory blocks.
811 */
812void kmemleak_ignore(const void *ptr)
813{
814 pr_debug("%s(0x%p)\n", __func__, ptr);
815
816 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
817 make_black_object((unsigned long)ptr);
818 else if (atomic_read(&kmemleak_early_log))
819 log_early(KMEMLEAK_IGNORE, ptr, 0, 0, 0, 0);
820}
821EXPORT_SYMBOL(kmemleak_ignore);
822
823/*
824 * Limit the range to be scanned in an allocated memory block.
825 */
826void kmemleak_scan_area(const void *ptr, unsigned long offset, size_t length,
827 gfp_t gfp)
828{
829 pr_debug("%s(0x%p)\n", __func__, ptr);
830
831 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
832 add_scan_area((unsigned long)ptr, offset, length, gfp);
833 else if (atomic_read(&kmemleak_early_log))
834 log_early(KMEMLEAK_SCAN_AREA, ptr, 0, 0, offset, length);
835}
836EXPORT_SYMBOL(kmemleak_scan_area);
837
838/*
839 * Inform kmemleak not to scan the given memory block.
840 */
841void kmemleak_no_scan(const void *ptr)
842{
843 pr_debug("%s(0x%p)\n", __func__, ptr);
844
845 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
846 object_no_scan((unsigned long)ptr);
847 else if (atomic_read(&kmemleak_early_log))
848 log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0, 0, 0);
849}
850EXPORT_SYMBOL(kmemleak_no_scan);
851
852/*
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100853 * Memory scanning is a long process and it needs to be interruptable. This
854 * function checks whether such interrupt condition occured.
855 */
856static int scan_should_stop(void)
857{
858 if (!atomic_read(&kmemleak_enabled))
859 return 1;
860
861 /*
862 * This function may be called from either process or kthread context,
863 * hence the need to check for both stop conditions.
864 */
865 if (current->mm)
866 return signal_pending(current);
867 else
868 return kthread_should_stop();
869
870 return 0;
871}
872
873/*
874 * Scan a memory block (exclusive range) for valid pointers and add those
875 * found to the gray list.
876 */
877static void scan_block(void *_start, void *_end,
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100878 struct kmemleak_object *scanned, int allow_resched)
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100879{
880 unsigned long *ptr;
881 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
882 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
883
884 for (ptr = start; ptr < end; ptr++) {
885 unsigned long flags;
886 unsigned long pointer = *ptr;
887 struct kmemleak_object *object;
888
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100889 if (allow_resched)
890 cond_resched();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100891 if (scan_should_stop())
892 break;
893
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100894 object = find_and_get_object(pointer, 1);
895 if (!object)
896 continue;
897 if (object == scanned) {
898 /* self referenced, ignore */
899 put_object(object);
900 continue;
901 }
902
903 /*
904 * Avoid the lockdep recursive warning on object->lock being
905 * previously acquired in scan_object(). These locks are
906 * enclosed by scan_mutex.
907 */
908 spin_lock_irqsave_nested(&object->lock, flags,
909 SINGLE_DEPTH_NESTING);
910 if (!color_white(object)) {
911 /* non-orphan, ignored or new */
912 spin_unlock_irqrestore(&object->lock, flags);
913 put_object(object);
914 continue;
915 }
916
917 /*
918 * Increase the object's reference count (number of pointers
919 * to the memory block). If this count reaches the required
920 * minimum, the object's color will become gray and it will be
921 * added to the gray_list.
922 */
923 object->count++;
924 if (color_gray(object))
925 list_add_tail(&object->gray_list, &gray_list);
926 else
927 put_object(object);
928 spin_unlock_irqrestore(&object->lock, flags);
929 }
930}
931
932/*
933 * Scan a memory block corresponding to a kmemleak_object. A condition is
934 * that object->use_count >= 1.
935 */
936static void scan_object(struct kmemleak_object *object)
937{
938 struct kmemleak_scan_area *area;
939 struct hlist_node *elem;
940 unsigned long flags;
941
942 /*
943 * Once the object->lock is aquired, the corresponding memory block
944 * cannot be freed (the same lock is aquired in delete_object).
945 */
946 spin_lock_irqsave(&object->lock, flags);
947 if (object->flags & OBJECT_NO_SCAN)
948 goto out;
949 if (!(object->flags & OBJECT_ALLOCATED))
950 /* already freed object */
951 goto out;
952 if (hlist_empty(&object->area_list))
953 scan_block((void *)object->pointer,
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100954 (void *)(object->pointer + object->size), object, 0);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100955 else
956 hlist_for_each_entry(area, elem, &object->area_list, node)
957 scan_block((void *)(object->pointer + area->offset),
958 (void *)(object->pointer + area->offset
Catalin Marinas4b8a9672009-07-07 10:32:56 +0100959 + area->length), object, 0);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100960out:
961 spin_unlock_irqrestore(&object->lock, flags);
962}
963
964/*
965 * Scan data sections and all the referenced memory blocks allocated via the
966 * kernel's standard allocators. This function must be called with the
967 * scan_mutex held.
968 */
969static void kmemleak_scan(void)
970{
971 unsigned long flags;
972 struct kmemleak_object *object, *tmp;
973 struct task_struct *task;
974 int i;
Catalin Marinas4698c1f2009-06-26 17:38:27 +0100975 int new_leaks = 0;
Catalin Marinas25873622009-07-07 10:32:58 +0100976 int gray_list_pass = 0;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100977
Catalin Marinasacf49682009-06-26 17:38:29 +0100978 jiffies_last_scan = jiffies;
979
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100980 /* prepare the kmemleak_object's */
981 rcu_read_lock();
982 list_for_each_entry_rcu(object, &object_list, object_list) {
983 spin_lock_irqsave(&object->lock, flags);
984#ifdef DEBUG
985 /*
986 * With a few exceptions there should be a maximum of
987 * 1 reference to any object at this point.
988 */
989 if (atomic_read(&object->use_count) > 1) {
Joe Perchesae281062009-06-23 14:40:26 +0100990 pr_debug("object->use_count = %d\n",
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100991 atomic_read(&object->use_count));
992 dump_object_info(object);
993 }
994#endif
995 /* reset the reference count (whiten the object) */
996 object->count = 0;
Catalin Marinas25873622009-07-07 10:32:58 +0100997 object->flags &= ~OBJECT_NEW;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100998 if (color_gray(object) && get_object(object))
999 list_add_tail(&object->gray_list, &gray_list);
1000
1001 spin_unlock_irqrestore(&object->lock, flags);
1002 }
1003 rcu_read_unlock();
1004
1005 /* data/bss scanning */
Catalin Marinas4b8a9672009-07-07 10:32:56 +01001006 scan_block(_sdata, _edata, NULL, 1);
1007 scan_block(__bss_start, __bss_stop, NULL, 1);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001008
1009#ifdef CONFIG_SMP
1010 /* per-cpu sections scanning */
1011 for_each_possible_cpu(i)
1012 scan_block(__per_cpu_start + per_cpu_offset(i),
Catalin Marinas4b8a9672009-07-07 10:32:56 +01001013 __per_cpu_end + per_cpu_offset(i), NULL, 1);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001014#endif
1015
1016 /*
1017 * Struct page scanning for each node. The code below is not yet safe
1018 * with MEMORY_HOTPLUG.
1019 */
1020 for_each_online_node(i) {
1021 pg_data_t *pgdat = NODE_DATA(i);
1022 unsigned long start_pfn = pgdat->node_start_pfn;
1023 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1024 unsigned long pfn;
1025
1026 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1027 struct page *page;
1028
1029 if (!pfn_valid(pfn))
1030 continue;
1031 page = pfn_to_page(pfn);
1032 /* only scan if page is in use */
1033 if (page_count(page) == 0)
1034 continue;
Catalin Marinas4b8a9672009-07-07 10:32:56 +01001035 scan_block(page, page + 1, NULL, 1);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001036 }
1037 }
1038
1039 /*
1040 * Scanning the task stacks may introduce false negatives and it is
1041 * not enabled by default.
1042 */
1043 if (kmemleak_stack_scan) {
1044 read_lock(&tasklist_lock);
1045 for_each_process(task)
1046 scan_block(task_stack_page(task),
Catalin Marinas4b8a9672009-07-07 10:32:56 +01001047 task_stack_page(task) + THREAD_SIZE,
1048 NULL, 0);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001049 read_unlock(&tasklist_lock);
1050 }
1051
1052 /*
1053 * Scan the objects already referenced from the sections scanned
1054 * above. More objects will be referenced and, if there are no memory
1055 * leaks, all the objects will be scanned. The list traversal is safe
1056 * for both tail additions and removals from inside the loop. The
1057 * kmemleak objects cannot be freed from outside the loop because their
1058 * use_count was increased.
1059 */
Catalin Marinas25873622009-07-07 10:32:58 +01001060repeat:
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001061 object = list_entry(gray_list.next, typeof(*object), gray_list);
1062 while (&object->gray_list != &gray_list) {
Ingo Molnar57d81f62009-07-01 09:43:53 +02001063 cond_resched();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001064
1065 /* may add new objects to the list */
1066 if (!scan_should_stop())
1067 scan_object(object);
1068
1069 tmp = list_entry(object->gray_list.next, typeof(*object),
1070 gray_list);
1071
1072 /* remove the object from the list and release it */
1073 list_del(&object->gray_list);
1074 put_object(object);
1075
1076 object = tmp;
1077 }
Catalin Marinas25873622009-07-07 10:32:58 +01001078
1079 if (scan_should_stop() || ++gray_list_pass >= GRAY_LIST_PASSES)
1080 goto scan_end;
1081
1082 /*
1083 * Check for new objects allocated during this scanning and add them
1084 * to the gray list.
1085 */
1086 rcu_read_lock();
1087 list_for_each_entry_rcu(object, &object_list, object_list) {
1088 spin_lock_irqsave(&object->lock, flags);
1089 if ((object->flags & OBJECT_NEW) && !color_black(object) &&
1090 get_object(object)) {
1091 object->flags &= ~OBJECT_NEW;
1092 list_add_tail(&object->gray_list, &gray_list);
1093 }
1094 spin_unlock_irqrestore(&object->lock, flags);
1095 }
1096 rcu_read_unlock();
1097
1098 if (!list_empty(&gray_list))
1099 goto repeat;
1100
1101scan_end:
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001102 WARN_ON(!list_empty(&gray_list));
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001103
1104 /*
Catalin Marinas25873622009-07-07 10:32:58 +01001105 * If scanning was stopped or new objects were being allocated at a
1106 * higher rate than gray list scanning, do not report any new
1107 * unreferenced objects.
Catalin Marinas17bb9e02009-06-29 17:13:56 +01001108 */
Catalin Marinas25873622009-07-07 10:32:58 +01001109 if (scan_should_stop() || gray_list_pass >= GRAY_LIST_PASSES)
Catalin Marinas17bb9e02009-06-29 17:13:56 +01001110 return;
1111
1112 /*
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001113 * Scanning result reporting.
1114 */
1115 rcu_read_lock();
1116 list_for_each_entry_rcu(object, &object_list, object_list) {
1117 spin_lock_irqsave(&object->lock, flags);
1118 if (unreferenced_object(object) &&
1119 !(object->flags & OBJECT_REPORTED)) {
1120 object->flags |= OBJECT_REPORTED;
1121 new_leaks++;
1122 }
1123 spin_unlock_irqrestore(&object->lock, flags);
1124 }
1125 rcu_read_unlock();
1126
1127 if (new_leaks)
1128 pr_info("%d new suspected memory leaks (see "
1129 "/sys/kernel/debug/kmemleak)\n", new_leaks);
1130
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001131}
1132
1133/*
1134 * Thread function performing automatic memory scanning. Unreferenced objects
1135 * at the end of a memory scan are reported but only the first time.
1136 */
1137static int kmemleak_scan_thread(void *arg)
1138{
1139 static int first_run = 1;
1140
Joe Perchesae281062009-06-23 14:40:26 +01001141 pr_info("Automatic memory scanning thread started\n");
Catalin Marinasbf2a76b2009-07-07 10:32:55 +01001142 set_user_nice(current, 10);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001143
1144 /*
1145 * Wait before the first scan to allow the system to fully initialize.
1146 */
1147 if (first_run) {
1148 first_run = 0;
1149 ssleep(SECS_FIRST_SCAN);
1150 }
1151
1152 while (!kthread_should_stop()) {
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001153 signed long timeout = jiffies_scan_wait;
1154
1155 mutex_lock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001156 kmemleak_scan();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001157 mutex_unlock(&scan_mutex);
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001158
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001159 /* wait before the next scan */
1160 while (timeout && !kthread_should_stop())
1161 timeout = schedule_timeout_interruptible(timeout);
1162 }
1163
Joe Perchesae281062009-06-23 14:40:26 +01001164 pr_info("Automatic memory scanning thread ended\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001165
1166 return 0;
1167}
1168
1169/*
1170 * Start the automatic memory scanning thread. This function must be called
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001171 * with the scan_mutex held.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001172 */
1173void start_scan_thread(void)
1174{
1175 if (scan_thread)
1176 return;
1177 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1178 if (IS_ERR(scan_thread)) {
Joe Perchesae281062009-06-23 14:40:26 +01001179 pr_warning("Failed to create the scan thread\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001180 scan_thread = NULL;
1181 }
1182}
1183
1184/*
1185 * Stop the automatic memory scanning thread. This function must be called
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001186 * with the scan_mutex held.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001187 */
1188void stop_scan_thread(void)
1189{
1190 if (scan_thread) {
1191 kthread_stop(scan_thread);
1192 scan_thread = NULL;
1193 }
1194}
1195
1196/*
1197 * Iterate over the object_list and return the first valid object at or after
1198 * the required position with its use_count incremented. The function triggers
1199 * a memory scanning when the pos argument points to the first position.
1200 */
1201static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1202{
1203 struct kmemleak_object *object;
1204 loff_t n = *pos;
Catalin Marinasb87324d2009-07-07 10:32:58 +01001205 int err;
1206
1207 err = mutex_lock_interruptible(&scan_mutex);
1208 if (err < 0)
1209 return ERR_PTR(err);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001210
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001211 rcu_read_lock();
1212 list_for_each_entry_rcu(object, &object_list, object_list) {
1213 if (n-- > 0)
1214 continue;
1215 if (get_object(object))
1216 goto out;
1217 }
1218 object = NULL;
1219out:
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001220 return object;
1221}
1222
1223/*
1224 * Return the next object in the object_list. The function decrements the
1225 * use_count of the previous object and increases that of the next one.
1226 */
1227static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1228{
1229 struct kmemleak_object *prev_obj = v;
1230 struct kmemleak_object *next_obj = NULL;
1231 struct list_head *n = &prev_obj->object_list;
1232
1233 ++(*pos);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001234
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001235 list_for_each_continue_rcu(n, &object_list) {
1236 next_obj = list_entry(n, struct kmemleak_object, object_list);
1237 if (get_object(next_obj))
1238 break;
1239 }
Catalin Marinas288c8572009-07-07 10:32:57 +01001240
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001241 put_object(prev_obj);
1242 return next_obj;
1243}
1244
1245/*
1246 * Decrement the use_count of the last object required, if any.
1247 */
1248static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1249{
Catalin Marinasb87324d2009-07-07 10:32:58 +01001250 if (!IS_ERR(v)) {
1251 /*
1252 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1253 * waiting was interrupted, so only release it if !IS_ERR.
1254 */
Catalin Marinasf5886c72009-07-29 16:26:57 +01001255 rcu_read_unlock();
Catalin Marinasb87324d2009-07-07 10:32:58 +01001256 mutex_unlock(&scan_mutex);
1257 if (v)
1258 put_object(v);
1259 }
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001260}
1261
1262/*
1263 * Print the information for an unreferenced object to the seq file.
1264 */
1265static int kmemleak_seq_show(struct seq_file *seq, void *v)
1266{
1267 struct kmemleak_object *object = v;
1268 unsigned long flags;
1269
1270 spin_lock_irqsave(&object->lock, flags);
Catalin Marinas288c8572009-07-07 10:32:57 +01001271 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
Catalin Marinas17bb9e02009-06-29 17:13:56 +01001272 print_unreferenced(seq, object);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001273 spin_unlock_irqrestore(&object->lock, flags);
1274 return 0;
1275}
1276
1277static const struct seq_operations kmemleak_seq_ops = {
1278 .start = kmemleak_seq_start,
1279 .next = kmemleak_seq_next,
1280 .stop = kmemleak_seq_stop,
1281 .show = kmemleak_seq_show,
1282};
1283
1284static int kmemleak_open(struct inode *inode, struct file *file)
1285{
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001286 if (!atomic_read(&kmemleak_enabled))
1287 return -EBUSY;
1288
Catalin Marinasb87324d2009-07-07 10:32:58 +01001289 return seq_open(file, &kmemleak_seq_ops);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001290}
1291
1292static int kmemleak_release(struct inode *inode, struct file *file)
1293{
Catalin Marinasb87324d2009-07-07 10:32:58 +01001294 return seq_release(inode, file);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001295}
1296
1297/*
1298 * File write operation to configure kmemleak at run-time. The following
1299 * commands can be written to the /sys/kernel/debug/kmemleak file:
1300 * off - disable kmemleak (irreversible)
1301 * stack=on - enable the task stacks scanning
1302 * stack=off - disable the tasks stacks scanning
1303 * scan=on - start the automatic memory scanning thread
1304 * scan=off - stop the automatic memory scanning thread
1305 * scan=... - set the automatic memory scanning period in seconds (0 to
1306 * disable it)
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001307 * scan - trigger a memory scan
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001308 */
1309static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1310 size_t size, loff_t *ppos)
1311{
1312 char buf[64];
1313 int buf_size;
Catalin Marinasb87324d2009-07-07 10:32:58 +01001314 int ret;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001315
1316 buf_size = min(size, (sizeof(buf) - 1));
1317 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1318 return -EFAULT;
1319 buf[buf_size] = 0;
1320
Catalin Marinasb87324d2009-07-07 10:32:58 +01001321 ret = mutex_lock_interruptible(&scan_mutex);
1322 if (ret < 0)
1323 return ret;
1324
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001325 if (strncmp(buf, "off", 3) == 0)
1326 kmemleak_disable();
1327 else if (strncmp(buf, "stack=on", 8) == 0)
1328 kmemleak_stack_scan = 1;
1329 else if (strncmp(buf, "stack=off", 9) == 0)
1330 kmemleak_stack_scan = 0;
1331 else if (strncmp(buf, "scan=on", 7) == 0)
1332 start_scan_thread();
1333 else if (strncmp(buf, "scan=off", 8) == 0)
1334 stop_scan_thread();
1335 else if (strncmp(buf, "scan=", 5) == 0) {
1336 unsigned long secs;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001337
Catalin Marinasb87324d2009-07-07 10:32:58 +01001338 ret = strict_strtoul(buf + 5, 0, &secs);
1339 if (ret < 0)
1340 goto out;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001341 stop_scan_thread();
1342 if (secs) {
1343 jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1344 start_scan_thread();
1345 }
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001346 } else if (strncmp(buf, "scan", 4) == 0)
1347 kmemleak_scan();
1348 else
Catalin Marinasb87324d2009-07-07 10:32:58 +01001349 ret = -EINVAL;
1350
1351out:
1352 mutex_unlock(&scan_mutex);
1353 if (ret < 0)
1354 return ret;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001355
1356 /* ignore the rest of the buffer, only one command at a time */
1357 *ppos += size;
1358 return size;
1359}
1360
1361static const struct file_operations kmemleak_fops = {
1362 .owner = THIS_MODULE,
1363 .open = kmemleak_open,
1364 .read = seq_read,
1365 .write = kmemleak_write,
1366 .llseek = seq_lseek,
1367 .release = kmemleak_release,
1368};
1369
1370/*
1371 * Perform the freeing of the kmemleak internal objects after waiting for any
1372 * current memory scan to complete.
1373 */
1374static int kmemleak_cleanup_thread(void *arg)
1375{
1376 struct kmemleak_object *object;
1377
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001378 mutex_lock(&scan_mutex);
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001379 stop_scan_thread();
1380
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001381 rcu_read_lock();
1382 list_for_each_entry_rcu(object, &object_list, object_list)
Catalin Marinas53238a62009-07-07 10:33:00 +01001383 delete_object_full(object->pointer);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001384 rcu_read_unlock();
1385 mutex_unlock(&scan_mutex);
1386
1387 return 0;
1388}
1389
1390/*
1391 * Start the clean-up thread.
1392 */
1393static void kmemleak_cleanup(void)
1394{
1395 struct task_struct *cleanup_thread;
1396
1397 cleanup_thread = kthread_run(kmemleak_cleanup_thread, NULL,
1398 "kmemleak-clean");
1399 if (IS_ERR(cleanup_thread))
Joe Perchesae281062009-06-23 14:40:26 +01001400 pr_warning("Failed to create the clean-up thread\n");
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001401}
1402
1403/*
1404 * Disable kmemleak. No memory allocation/freeing will be traced once this
1405 * function is called. Disabling kmemleak is an irreversible operation.
1406 */
1407static void kmemleak_disable(void)
1408{
1409 /* atomically check whether it was already invoked */
1410 if (atomic_cmpxchg(&kmemleak_error, 0, 1))
1411 return;
1412
1413 /* stop any memory operation tracing */
1414 atomic_set(&kmemleak_early_log, 0);
1415 atomic_set(&kmemleak_enabled, 0);
1416
1417 /* check whether it is too early for a kernel thread */
1418 if (atomic_read(&kmemleak_initialized))
1419 kmemleak_cleanup();
1420
1421 pr_info("Kernel memory leak detector disabled\n");
1422}
1423
1424/*
1425 * Allow boot-time kmemleak disabling (enabled by default).
1426 */
1427static int kmemleak_boot_config(char *str)
1428{
1429 if (!str)
1430 return -EINVAL;
1431 if (strcmp(str, "off") == 0)
1432 kmemleak_disable();
1433 else if (strcmp(str, "on") != 0)
1434 return -EINVAL;
1435 return 0;
1436}
1437early_param("kmemleak", kmemleak_boot_config);
1438
1439/*
Catalin Marinas20301172009-06-17 18:29:04 +01001440 * Kmemleak initialization.
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001441 */
1442void __init kmemleak_init(void)
1443{
1444 int i;
1445 unsigned long flags;
1446
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001447 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1448 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1449
1450 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1451 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1452 INIT_PRIO_TREE_ROOT(&object_tree_root);
1453
1454 /* the kernel is still in UP mode, so disabling the IRQs is enough */
1455 local_irq_save(flags);
1456 if (!atomic_read(&kmemleak_error)) {
1457 atomic_set(&kmemleak_enabled, 1);
1458 atomic_set(&kmemleak_early_log, 0);
1459 }
1460 local_irq_restore(flags);
1461
1462 /*
1463 * This is the point where tracking allocations is safe. Automatic
1464 * scanning is started during the late initcall. Add the early logged
1465 * callbacks to the kmemleak infrastructure.
1466 */
1467 for (i = 0; i < crt_early_log; i++) {
1468 struct early_log *log = &early_log[i];
1469
1470 switch (log->op_type) {
1471 case KMEMLEAK_ALLOC:
1472 kmemleak_alloc(log->ptr, log->size, log->min_count,
1473 GFP_KERNEL);
1474 break;
1475 case KMEMLEAK_FREE:
1476 kmemleak_free(log->ptr);
1477 break;
Catalin Marinas53238a62009-07-07 10:33:00 +01001478 case KMEMLEAK_FREE_PART:
1479 kmemleak_free_part(log->ptr, log->size);
1480 break;
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001481 case KMEMLEAK_NOT_LEAK:
1482 kmemleak_not_leak(log->ptr);
1483 break;
1484 case KMEMLEAK_IGNORE:
1485 kmemleak_ignore(log->ptr);
1486 break;
1487 case KMEMLEAK_SCAN_AREA:
1488 kmemleak_scan_area(log->ptr, log->offset, log->length,
1489 GFP_KERNEL);
1490 break;
1491 case KMEMLEAK_NO_SCAN:
1492 kmemleak_no_scan(log->ptr);
1493 break;
1494 default:
1495 WARN_ON(1);
1496 }
1497 }
1498}
1499
1500/*
1501 * Late initialization function.
1502 */
1503static int __init kmemleak_late_init(void)
1504{
1505 struct dentry *dentry;
1506
1507 atomic_set(&kmemleak_initialized, 1);
1508
1509 if (atomic_read(&kmemleak_error)) {
1510 /*
1511 * Some error occured and kmemleak was disabled. There is a
1512 * small chance that kmemleak_disable() was called immediately
1513 * after setting kmemleak_initialized and we may end up with
1514 * two clean-up threads but serialized by scan_mutex.
1515 */
1516 kmemleak_cleanup();
1517 return -ENOMEM;
1518 }
1519
1520 dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL,
1521 &kmemleak_fops);
1522 if (!dentry)
Joe Perchesae281062009-06-23 14:40:26 +01001523 pr_warning("Failed to create the debugfs kmemleak file\n");
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001524 mutex_lock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001525 start_scan_thread();
Catalin Marinas4698c1f2009-06-26 17:38:27 +01001526 mutex_unlock(&scan_mutex);
Catalin Marinas3c7b4e62009-06-11 13:22:39 +01001527
1528 pr_info("Kernel memory leak detector initialized\n");
1529
1530 return 0;
1531}
1532late_initcall(kmemleak_late_init);