]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - mm/kmemleak.c
Merge branch 'acer' into release
[mirror_ubuntu-bionic-kernel.git] / mm / kmemleak.c
1 /*
2 * mm/kmemleak.c
3 *
4 * Copyright (C) 2008 ARM Limited
5 * Written by Catalin Marinas <catalin.marinas@arm.com>
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 *
20 *
21 * For more information on the algorithm and kmemleak usage, please see
22 * Documentation/kmemleak.txt.
23 *
24 * Notes on locking
25 * ----------------
26 *
27 * The following locks and mutexes are used by kmemleak:
28 *
29 * - kmemleak_lock (rwlock): protects the object_list modifications and
30 * accesses to the object_tree_root. The object_list is the main list
31 * holding the metadata (struct kmemleak_object) for the allocated memory
32 * blocks. The object_tree_root is a priority search tree used to look-up
33 * metadata based on a pointer to the corresponding memory block. The
34 * kmemleak_object structures are added to the object_list and
35 * object_tree_root in the create_object() function called from the
36 * kmemleak_alloc() callback and removed in delete_object() called from the
37 * kmemleak_free() callback
38 * - kmemleak_object.lock (spinlock): protects a kmemleak_object. Accesses to
39 * the metadata (e.g. count) are protected by this lock. Note that some
40 * members of this structure may be protected by other means (atomic or
41 * kmemleak_lock). This lock is also held when scanning the corresponding
42 * memory block to avoid the kernel freeing it via the kmemleak_free()
43 * callback. This is less heavyweight than holding a global lock like
44 * kmemleak_lock during scanning
45 * - scan_mutex (mutex): ensures that only one thread may scan the memory for
46 * unreferenced objects at a time. The gray_list contains the objects which
47 * are already referenced or marked as false positives and need to be
48 * scanned. This list is only modified during a scanning episode when the
49 * scan_mutex is held. At the end of a scan, the gray_list is always empty.
50 * Note that the kmemleak_object.use_count is incremented when an object is
51 * added to the gray_list and therefore cannot be freed. This mutex also
52 * prevents multiple users of the "kmemleak" debugfs file together with
53 * modifications to the memory scanning parameters including the scan_thread
54 * pointer
55 *
56 * 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
64 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
65
66 #include <linux/init.h>
67 #include <linux/kernel.h>
68 #include <linux/list.h>
69 #include <linux/sched.h>
70 #include <linux/jiffies.h>
71 #include <linux/delay.h>
72 #include <linux/module.h>
73 #include <linux/kthread.h>
74 #include <linux/prio_tree.h>
75 #include <linux/gfp.h>
76 #include <linux/fs.h>
77 #include <linux/debugfs.h>
78 #include <linux/seq_file.h>
79 #include <linux/cpumask.h>
80 #include <linux/spinlock.h>
81 #include <linux/mutex.h>
82 #include <linux/rcupdate.h>
83 #include <linux/stacktrace.h>
84 #include <linux/cache.h>
85 #include <linux/percpu.h>
86 #include <linux/hardirq.h>
87 #include <linux/mmzone.h>
88 #include <linux/slab.h>
89 #include <linux/thread_info.h>
90 #include <linux/err.h>
91 #include <linux/uaccess.h>
92 #include <linux/string.h>
93 #include <linux/nodemask.h>
94 #include <linux/mm.h>
95 #include <linux/workqueue.h>
96
97 #include <asm/sections.h>
98 #include <asm/processor.h>
99 #include <asm/atomic.h>
100
101 #include <linux/kmemcheck.h>
102 #include <linux/kmemleak.h>
103
104 /*
105 * Kmemleak configuration and common defines.
106 */
107 #define MAX_TRACE 16 /* stack trace length */
108 #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
109 #define SECS_FIRST_SCAN 60 /* delay before the first scan */
110 #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
111 #define GRAY_LIST_PASSES 25 /* maximum number of gray list scans */
112 #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
113
114 #define BYTES_PER_POINTER sizeof(void *)
115
116 /* GFP bitmask for kmemleak internal allocations */
117 #define GFP_KMEMLEAK_MASK (GFP_KERNEL | GFP_ATOMIC)
118
119 /* scanning area inside a memory block */
120 struct kmemleak_scan_area {
121 struct hlist_node node;
122 unsigned long offset;
123 size_t length;
124 };
125
126 #define KMEMLEAK_GREY 0
127 #define KMEMLEAK_BLACK -1
128
129 /*
130 * Structure holding the metadata for each allocated memory block.
131 * Modifications to such objects should be made while holding the
132 * object->lock. Insertions or deletions from object_list, gray_list or
133 * tree_node are already protected by the corresponding locks or mutex (see
134 * the notes on locking above). These objects are reference-counted
135 * (use_count) and freed using the RCU mechanism.
136 */
137 struct kmemleak_object {
138 spinlock_t lock;
139 unsigned long flags; /* object status flags */
140 struct list_head object_list;
141 struct list_head gray_list;
142 struct prio_tree_node tree_node;
143 struct rcu_head rcu; /* object_list lockless traversal */
144 /* object usage count; object freed when use_count == 0 */
145 atomic_t use_count;
146 unsigned long pointer;
147 size_t size;
148 /* minimum number of a pointers found before it is considered leak */
149 int min_count;
150 /* the total number of pointers found pointing to this object */
151 int count;
152 /* memory ranges to be scanned inside an object (empty for all) */
153 struct hlist_head area_list;
154 unsigned long trace[MAX_TRACE];
155 unsigned int trace_len;
156 unsigned long jiffies; /* creation timestamp */
157 pid_t pid; /* pid of the current task */
158 char comm[TASK_COMM_LEN]; /* executable name */
159 };
160
161 /* flag representing the memory block allocation status */
162 #define OBJECT_ALLOCATED (1 << 0)
163 /* flag set after the first reporting of an unreference object */
164 #define OBJECT_REPORTED (1 << 1)
165 /* flag set to not scan the object */
166 #define OBJECT_NO_SCAN (1 << 2)
167 /* flag set on newly allocated objects */
168 #define OBJECT_NEW (1 << 3)
169
170 /* number of bytes to print per line; must be 16 or 32 */
171 #define HEX_ROW_SIZE 16
172 /* number of bytes to print at a time (1, 2, 4, 8) */
173 #define HEX_GROUP_SIZE 1
174 /* include ASCII after the hex output */
175 #define HEX_ASCII 1
176 /* max number of lines to be printed */
177 #define HEX_MAX_LINES 2
178
179 /* the list of all allocated objects */
180 static LIST_HEAD(object_list);
181 /* the list of gray-colored objects (see color_gray comment below) */
182 static LIST_HEAD(gray_list);
183 /* prio search tree for object boundaries */
184 static struct prio_tree_root object_tree_root;
185 /* rw_lock protecting the access to object_list and prio_tree_root */
186 static DEFINE_RWLOCK(kmemleak_lock);
187
188 /* allocation caches for kmemleak internal data */
189 static struct kmem_cache *object_cache;
190 static struct kmem_cache *scan_area_cache;
191
192 /* set if tracing memory operations is enabled */
193 static atomic_t kmemleak_enabled = ATOMIC_INIT(0);
194 /* set in the late_initcall if there were no errors */
195 static atomic_t kmemleak_initialized = ATOMIC_INIT(0);
196 /* enables or disables early logging of the memory operations */
197 static atomic_t kmemleak_early_log = ATOMIC_INIT(1);
198 /* set if a fata kmemleak error has occurred */
199 static atomic_t kmemleak_error = ATOMIC_INIT(0);
200
201 /* minimum and maximum address that may be valid pointers */
202 static unsigned long min_addr = ULONG_MAX;
203 static unsigned long max_addr;
204
205 static struct task_struct *scan_thread;
206 /* used to avoid reporting of recently allocated objects */
207 static unsigned long jiffies_min_age;
208 static unsigned long jiffies_last_scan;
209 /* delay between automatic memory scannings */
210 static signed long jiffies_scan_wait;
211 /* enables or disables the task stacks scanning */
212 static int kmemleak_stack_scan = 1;
213 /* protects the memory scanning, parameters and debug/kmemleak file access */
214 static DEFINE_MUTEX(scan_mutex);
215
216 /*
217 * Early object allocation/freeing logging. Kmemleak is initialized after the
218 * kernel allocator. However, both the kernel allocator and kmemleak may
219 * allocate memory blocks which need to be tracked. Kmemleak defines an
220 * arbitrary buffer to hold the allocation/freeing information before it is
221 * fully initialized.
222 */
223
224 /* kmemleak operation type for early logging */
225 enum {
226 KMEMLEAK_ALLOC,
227 KMEMLEAK_FREE,
228 KMEMLEAK_FREE_PART,
229 KMEMLEAK_NOT_LEAK,
230 KMEMLEAK_IGNORE,
231 KMEMLEAK_SCAN_AREA,
232 KMEMLEAK_NO_SCAN
233 };
234
235 /*
236 * Structure holding the information passed to kmemleak callbacks during the
237 * early logging.
238 */
239 struct early_log {
240 int op_type; /* kmemleak operation type */
241 const void *ptr; /* allocated/freed memory block */
242 size_t size; /* memory block size */
243 int min_count; /* minimum reference count */
244 unsigned long offset; /* scan area offset */
245 size_t length; /* scan area length */
246 unsigned long trace[MAX_TRACE]; /* stack trace */
247 unsigned int trace_len; /* stack trace length */
248 };
249
250 /* early logging buffer and current position */
251 static struct early_log
252 early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata;
253 static int crt_early_log __initdata;
254
255 static void kmemleak_disable(void);
256
257 /*
258 * Print a warning and dump the stack trace.
259 */
260 #define kmemleak_warn(x...) do { \
261 pr_warning(x); \
262 dump_stack(); \
263 } while (0)
264
265 /*
266 * Macro invoked when a serious kmemleak condition occured and cannot be
267 * recovered from. Kmemleak will be disabled and further allocation/freeing
268 * tracing no longer available.
269 */
270 #define kmemleak_stop(x...) do { \
271 kmemleak_warn(x); \
272 kmemleak_disable(); \
273 } while (0)
274
275 /*
276 * Printing of the objects hex dump to the seq file. The number of lines to be
277 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
278 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
279 * with the object->lock held.
280 */
281 static void hex_dump_object(struct seq_file *seq,
282 struct kmemleak_object *object)
283 {
284 const u8 *ptr = (const u8 *)object->pointer;
285 int i, len, remaining;
286 unsigned char linebuf[HEX_ROW_SIZE * 5];
287
288 /* limit the number of lines to HEX_MAX_LINES */
289 remaining = len =
290 min(object->size, (size_t)(HEX_MAX_LINES * HEX_ROW_SIZE));
291
292 seq_printf(seq, " hex dump (first %d bytes):\n", len);
293 for (i = 0; i < len; i += HEX_ROW_SIZE) {
294 int linelen = min(remaining, HEX_ROW_SIZE);
295
296 remaining -= HEX_ROW_SIZE;
297 hex_dump_to_buffer(ptr + i, linelen, HEX_ROW_SIZE,
298 HEX_GROUP_SIZE, linebuf, sizeof(linebuf),
299 HEX_ASCII);
300 seq_printf(seq, " %s\n", linebuf);
301 }
302 }
303
304 /*
305 * Object colors, encoded with count and min_count:
306 * - white - orphan object, not enough references to it (count < min_count)
307 * - gray - not orphan, not marked as false positive (min_count == 0) or
308 * sufficient references to it (count >= min_count)
309 * - black - ignore, it doesn't contain references (e.g. text section)
310 * (min_count == -1). No function defined for this color.
311 * Newly created objects don't have any color assigned (object->count == -1)
312 * before the next memory scan when they become white.
313 */
314 static bool color_white(const struct kmemleak_object *object)
315 {
316 return object->count != KMEMLEAK_BLACK &&
317 object->count < object->min_count;
318 }
319
320 static bool color_gray(const struct kmemleak_object *object)
321 {
322 return object->min_count != KMEMLEAK_BLACK &&
323 object->count >= object->min_count;
324 }
325
326 static bool color_black(const struct kmemleak_object *object)
327 {
328 return object->min_count == KMEMLEAK_BLACK;
329 }
330
331 /*
332 * Objects are considered unreferenced only if their color is white, they have
333 * not be deleted and have a minimum age to avoid false positives caused by
334 * pointers temporarily stored in CPU registers.
335 */
336 static bool unreferenced_object(struct kmemleak_object *object)
337 {
338 return (object->flags & OBJECT_ALLOCATED) && color_white(object) &&
339 time_before_eq(object->jiffies + jiffies_min_age,
340 jiffies_last_scan);
341 }
342
343 /*
344 * Printing of the unreferenced objects information to the seq file. The
345 * print_unreferenced function must be called with the object->lock held.
346 */
347 static void print_unreferenced(struct seq_file *seq,
348 struct kmemleak_object *object)
349 {
350 int i;
351
352 seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
353 object->pointer, object->size);
354 seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu\n",
355 object->comm, object->pid, object->jiffies);
356 hex_dump_object(seq, object);
357 seq_printf(seq, " backtrace:\n");
358
359 for (i = 0; i < object->trace_len; i++) {
360 void *ptr = (void *)object->trace[i];
361 seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
362 }
363 }
364
365 /*
366 * Print the kmemleak_object information. This function is used mainly for
367 * debugging special cases when kmemleak operations. It must be called with
368 * the object->lock held.
369 */
370 static void dump_object_info(struct kmemleak_object *object)
371 {
372 struct stack_trace trace;
373
374 trace.nr_entries = object->trace_len;
375 trace.entries = object->trace;
376
377 pr_notice("Object 0x%08lx (size %zu):\n",
378 object->tree_node.start, object->size);
379 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
380 object->comm, object->pid, object->jiffies);
381 pr_notice(" min_count = %d\n", object->min_count);
382 pr_notice(" count = %d\n", object->count);
383 pr_notice(" flags = 0x%lx\n", object->flags);
384 pr_notice(" backtrace:\n");
385 print_stack_trace(&trace, 4);
386 }
387
388 /*
389 * Look-up a memory block metadata (kmemleak_object) in the priority search
390 * tree based on a pointer value. If alias is 0, only values pointing to the
391 * beginning of the memory block are allowed. The kmemleak_lock must be held
392 * when calling this function.
393 */
394 static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
395 {
396 struct prio_tree_node *node;
397 struct prio_tree_iter iter;
398 struct kmemleak_object *object;
399
400 prio_tree_iter_init(&iter, &object_tree_root, ptr, ptr);
401 node = prio_tree_next(&iter);
402 if (node) {
403 object = prio_tree_entry(node, struct kmemleak_object,
404 tree_node);
405 if (!alias && object->pointer != ptr) {
406 kmemleak_warn("Found object by alias");
407 object = NULL;
408 }
409 } else
410 object = NULL;
411
412 return object;
413 }
414
415 /*
416 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
417 * that once an object's use_count reached 0, the RCU freeing was already
418 * registered and the object should no longer be used. This function must be
419 * called under the protection of rcu_read_lock().
420 */
421 static int get_object(struct kmemleak_object *object)
422 {
423 return atomic_inc_not_zero(&object->use_count);
424 }
425
426 /*
427 * RCU callback to free a kmemleak_object.
428 */
429 static void free_object_rcu(struct rcu_head *rcu)
430 {
431 struct hlist_node *elem, *tmp;
432 struct kmemleak_scan_area *area;
433 struct kmemleak_object *object =
434 container_of(rcu, struct kmemleak_object, rcu);
435
436 /*
437 * Once use_count is 0 (guaranteed by put_object), there is no other
438 * code accessing this object, hence no need for locking.
439 */
440 hlist_for_each_entry_safe(area, elem, tmp, &object->area_list, node) {
441 hlist_del(elem);
442 kmem_cache_free(scan_area_cache, area);
443 }
444 kmem_cache_free(object_cache, object);
445 }
446
447 /*
448 * Decrement the object use_count. Once the count is 0, free the object using
449 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
450 * delete_object() path, the delayed RCU freeing ensures that there is no
451 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
452 * is also possible.
453 */
454 static void put_object(struct kmemleak_object *object)
455 {
456 if (!atomic_dec_and_test(&object->use_count))
457 return;
458
459 /* should only get here after delete_object was called */
460 WARN_ON(object->flags & OBJECT_ALLOCATED);
461
462 call_rcu(&object->rcu, free_object_rcu);
463 }
464
465 /*
466 * Look up an object in the prio search tree and increase its use_count.
467 */
468 static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
469 {
470 unsigned long flags;
471 struct kmemleak_object *object = NULL;
472
473 rcu_read_lock();
474 read_lock_irqsave(&kmemleak_lock, flags);
475 if (ptr >= min_addr && ptr < max_addr)
476 object = lookup_object(ptr, alias);
477 read_unlock_irqrestore(&kmemleak_lock, flags);
478
479 /* check whether the object is still available */
480 if (object && !get_object(object))
481 object = NULL;
482 rcu_read_unlock();
483
484 return object;
485 }
486
487 /*
488 * Save stack trace to the given array of MAX_TRACE size.
489 */
490 static int __save_stack_trace(unsigned long *trace)
491 {
492 struct stack_trace stack_trace;
493
494 stack_trace.max_entries = MAX_TRACE;
495 stack_trace.nr_entries = 0;
496 stack_trace.entries = trace;
497 stack_trace.skip = 2;
498 save_stack_trace(&stack_trace);
499
500 return stack_trace.nr_entries;
501 }
502
503 /*
504 * Create the metadata (struct kmemleak_object) corresponding to an allocated
505 * memory block and add it to the object_list and object_tree_root.
506 */
507 static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
508 int min_count, gfp_t gfp)
509 {
510 unsigned long flags;
511 struct kmemleak_object *object;
512 struct prio_tree_node *node;
513
514 object = kmem_cache_alloc(object_cache, gfp & GFP_KMEMLEAK_MASK);
515 if (!object) {
516 kmemleak_stop("Cannot allocate a kmemleak_object structure\n");
517 return NULL;
518 }
519
520 INIT_LIST_HEAD(&object->object_list);
521 INIT_LIST_HEAD(&object->gray_list);
522 INIT_HLIST_HEAD(&object->area_list);
523 spin_lock_init(&object->lock);
524 atomic_set(&object->use_count, 1);
525 object->flags = OBJECT_ALLOCATED | OBJECT_NEW;
526 object->pointer = ptr;
527 object->size = size;
528 object->min_count = min_count;
529 object->count = -1; /* no color initially */
530 object->jiffies = jiffies;
531
532 /* task information */
533 if (in_irq()) {
534 object->pid = 0;
535 strncpy(object->comm, "hardirq", sizeof(object->comm));
536 } else if (in_softirq()) {
537 object->pid = 0;
538 strncpy(object->comm, "softirq", sizeof(object->comm));
539 } else {
540 object->pid = current->pid;
541 /*
542 * There is a small chance of a race with set_task_comm(),
543 * however using get_task_comm() here may cause locking
544 * dependency issues with current->alloc_lock. In the worst
545 * case, the command line is not correct.
546 */
547 strncpy(object->comm, current->comm, sizeof(object->comm));
548 }
549
550 /* kernel backtrace */
551 object->trace_len = __save_stack_trace(object->trace);
552
553 INIT_PRIO_TREE_NODE(&object->tree_node);
554 object->tree_node.start = ptr;
555 object->tree_node.last = ptr + size - 1;
556
557 write_lock_irqsave(&kmemleak_lock, flags);
558
559 min_addr = min(min_addr, ptr);
560 max_addr = max(max_addr, ptr + size);
561 node = prio_tree_insert(&object_tree_root, &object->tree_node);
562 /*
563 * The code calling the kernel does not yet have the pointer to the
564 * memory block to be able to free it. However, we still hold the
565 * kmemleak_lock here in case parts of the kernel started freeing
566 * random memory blocks.
567 */
568 if (node != &object->tree_node) {
569 kmemleak_stop("Cannot insert 0x%lx into the object search tree "
570 "(already existing)\n", ptr);
571 object = lookup_object(ptr, 1);
572 spin_lock(&object->lock);
573 dump_object_info(object);
574 spin_unlock(&object->lock);
575
576 goto out;
577 }
578 list_add_tail_rcu(&object->object_list, &object_list);
579 out:
580 write_unlock_irqrestore(&kmemleak_lock, flags);
581 return object;
582 }
583
584 /*
585 * Remove the metadata (struct kmemleak_object) for a memory block from the
586 * object_list and object_tree_root and decrement its use_count.
587 */
588 static void __delete_object(struct kmemleak_object *object)
589 {
590 unsigned long flags;
591
592 write_lock_irqsave(&kmemleak_lock, flags);
593 prio_tree_remove(&object_tree_root, &object->tree_node);
594 list_del_rcu(&object->object_list);
595 write_unlock_irqrestore(&kmemleak_lock, flags);
596
597 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
598 WARN_ON(atomic_read(&object->use_count) < 2);
599
600 /*
601 * Locking here also ensures that the corresponding memory block
602 * cannot be freed when it is being scanned.
603 */
604 spin_lock_irqsave(&object->lock, flags);
605 object->flags &= ~OBJECT_ALLOCATED;
606 spin_unlock_irqrestore(&object->lock, flags);
607 put_object(object);
608 }
609
610 /*
611 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
612 * delete it.
613 */
614 static void delete_object_full(unsigned long ptr)
615 {
616 struct kmemleak_object *object;
617
618 object = find_and_get_object(ptr, 0);
619 if (!object) {
620 #ifdef DEBUG
621 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
622 ptr);
623 #endif
624 return;
625 }
626 __delete_object(object);
627 put_object(object);
628 }
629
630 /*
631 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
632 * delete it. If the memory block is partially freed, the function may create
633 * additional metadata for the remaining parts of the block.
634 */
635 static void delete_object_part(unsigned long ptr, size_t size)
636 {
637 struct kmemleak_object *object;
638 unsigned long start, end;
639
640 object = find_and_get_object(ptr, 1);
641 if (!object) {
642 #ifdef DEBUG
643 kmemleak_warn("Partially freeing unknown object at 0x%08lx "
644 "(size %zu)\n", ptr, size);
645 #endif
646 return;
647 }
648 __delete_object(object);
649
650 /*
651 * Create one or two objects that may result from the memory block
652 * split. Note that partial freeing is only done by free_bootmem() and
653 * this happens before kmemleak_init() is called. The path below is
654 * only executed during early log recording in kmemleak_init(), so
655 * GFP_KERNEL is enough.
656 */
657 start = object->pointer;
658 end = object->pointer + object->size;
659 if (ptr > start)
660 create_object(start, ptr - start, object->min_count,
661 GFP_KERNEL);
662 if (ptr + size < end)
663 create_object(ptr + size, end - ptr - size, object->min_count,
664 GFP_KERNEL);
665
666 put_object(object);
667 }
668
669 static void __paint_it(struct kmemleak_object *object, int color)
670 {
671 object->min_count = color;
672 if (color == KMEMLEAK_BLACK)
673 object->flags |= OBJECT_NO_SCAN;
674 }
675
676 static void paint_it(struct kmemleak_object *object, int color)
677 {
678 unsigned long flags;
679
680 spin_lock_irqsave(&object->lock, flags);
681 __paint_it(object, color);
682 spin_unlock_irqrestore(&object->lock, flags);
683 }
684
685 static void paint_ptr(unsigned long ptr, int color)
686 {
687 struct kmemleak_object *object;
688
689 object = find_and_get_object(ptr, 0);
690 if (!object) {
691 kmemleak_warn("Trying to color unknown object "
692 "at 0x%08lx as %s\n", ptr,
693 (color == KMEMLEAK_GREY) ? "Grey" :
694 (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
695 return;
696 }
697 paint_it(object, color);
698 put_object(object);
699 }
700
701 /*
702 * Make a object permanently as gray-colored so that it can no longer be
703 * reported as a leak. This is used in general to mark a false positive.
704 */
705 static void make_gray_object(unsigned long ptr)
706 {
707 paint_ptr(ptr, KMEMLEAK_GREY);
708 }
709
710 /*
711 * Mark the object as black-colored so that it is ignored from scans and
712 * reporting.
713 */
714 static void make_black_object(unsigned long ptr)
715 {
716 paint_ptr(ptr, KMEMLEAK_BLACK);
717 }
718
719 /*
720 * Add a scanning area to the object. If at least one such area is added,
721 * kmemleak will only scan these ranges rather than the whole memory block.
722 */
723 static void add_scan_area(unsigned long ptr, unsigned long offset,
724 size_t length, gfp_t gfp)
725 {
726 unsigned long flags;
727 struct kmemleak_object *object;
728 struct kmemleak_scan_area *area;
729
730 object = find_and_get_object(ptr, 0);
731 if (!object) {
732 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
733 ptr);
734 return;
735 }
736
737 area = kmem_cache_alloc(scan_area_cache, gfp & GFP_KMEMLEAK_MASK);
738 if (!area) {
739 kmemleak_warn("Cannot allocate a scan area\n");
740 goto out;
741 }
742
743 spin_lock_irqsave(&object->lock, flags);
744 if (offset + length > object->size) {
745 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
746 dump_object_info(object);
747 kmem_cache_free(scan_area_cache, area);
748 goto out_unlock;
749 }
750
751 INIT_HLIST_NODE(&area->node);
752 area->offset = offset;
753 area->length = length;
754
755 hlist_add_head(&area->node, &object->area_list);
756 out_unlock:
757 spin_unlock_irqrestore(&object->lock, flags);
758 out:
759 put_object(object);
760 }
761
762 /*
763 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
764 * pointer. Such object will not be scanned by kmemleak but references to it
765 * are searched.
766 */
767 static void object_no_scan(unsigned long ptr)
768 {
769 unsigned long flags;
770 struct kmemleak_object *object;
771
772 object = find_and_get_object(ptr, 0);
773 if (!object) {
774 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
775 return;
776 }
777
778 spin_lock_irqsave(&object->lock, flags);
779 object->flags |= OBJECT_NO_SCAN;
780 spin_unlock_irqrestore(&object->lock, flags);
781 put_object(object);
782 }
783
784 /*
785 * Log an early kmemleak_* call to the early_log buffer. These calls will be
786 * processed later once kmemleak is fully initialized.
787 */
788 static void __init log_early(int op_type, const void *ptr, size_t size,
789 int min_count, unsigned long offset, size_t length)
790 {
791 unsigned long flags;
792 struct early_log *log;
793
794 if (crt_early_log >= ARRAY_SIZE(early_log)) {
795 pr_warning("Early log buffer exceeded, "
796 "please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n");
797 kmemleak_disable();
798 return;
799 }
800
801 /*
802 * There is no need for locking since the kernel is still in UP mode
803 * at this stage. Disabling the IRQs is enough.
804 */
805 local_irq_save(flags);
806 log = &early_log[crt_early_log];
807 log->op_type = op_type;
808 log->ptr = ptr;
809 log->size = size;
810 log->min_count = min_count;
811 log->offset = offset;
812 log->length = length;
813 if (op_type == KMEMLEAK_ALLOC)
814 log->trace_len = __save_stack_trace(log->trace);
815 crt_early_log++;
816 local_irq_restore(flags);
817 }
818
819 /*
820 * Log an early allocated block and populate the stack trace.
821 */
822 static void early_alloc(struct early_log *log)
823 {
824 struct kmemleak_object *object;
825 unsigned long flags;
826 int i;
827
828 if (!atomic_read(&kmemleak_enabled) || !log->ptr || IS_ERR(log->ptr))
829 return;
830
831 /*
832 * RCU locking needed to ensure object is not freed via put_object().
833 */
834 rcu_read_lock();
835 object = create_object((unsigned long)log->ptr, log->size,
836 log->min_count, GFP_KERNEL);
837 spin_lock_irqsave(&object->lock, flags);
838 for (i = 0; i < log->trace_len; i++)
839 object->trace[i] = log->trace[i];
840 object->trace_len = log->trace_len;
841 spin_unlock_irqrestore(&object->lock, flags);
842 rcu_read_unlock();
843 }
844
845 /*
846 * Memory allocation function callback. This function is called from the
847 * kernel allocators when a new block is allocated (kmem_cache_alloc, kmalloc,
848 * vmalloc etc.).
849 */
850 void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
851 gfp_t gfp)
852 {
853 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
854
855 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
856 create_object((unsigned long)ptr, size, min_count, gfp);
857 else if (atomic_read(&kmemleak_early_log))
858 log_early(KMEMLEAK_ALLOC, ptr, size, min_count, 0, 0);
859 }
860 EXPORT_SYMBOL_GPL(kmemleak_alloc);
861
862 /*
863 * Memory freeing function callback. This function is called from the kernel
864 * allocators when a block is freed (kmem_cache_free, kfree, vfree etc.).
865 */
866 void __ref kmemleak_free(const void *ptr)
867 {
868 pr_debug("%s(0x%p)\n", __func__, ptr);
869
870 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
871 delete_object_full((unsigned long)ptr);
872 else if (atomic_read(&kmemleak_early_log))
873 log_early(KMEMLEAK_FREE, ptr, 0, 0, 0, 0);
874 }
875 EXPORT_SYMBOL_GPL(kmemleak_free);
876
877 /*
878 * Partial memory freeing function callback. This function is usually called
879 * from bootmem allocator when (part of) a memory block is freed.
880 */
881 void __ref kmemleak_free_part(const void *ptr, size_t size)
882 {
883 pr_debug("%s(0x%p)\n", __func__, ptr);
884
885 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
886 delete_object_part((unsigned long)ptr, size);
887 else if (atomic_read(&kmemleak_early_log))
888 log_early(KMEMLEAK_FREE_PART, ptr, size, 0, 0, 0);
889 }
890 EXPORT_SYMBOL_GPL(kmemleak_free_part);
891
892 /*
893 * Mark an already allocated memory block as a false positive. This will cause
894 * the block to no longer be reported as leak and always be scanned.
895 */
896 void __ref kmemleak_not_leak(const void *ptr)
897 {
898 pr_debug("%s(0x%p)\n", __func__, ptr);
899
900 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
901 make_gray_object((unsigned long)ptr);
902 else if (atomic_read(&kmemleak_early_log))
903 log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0, 0, 0);
904 }
905 EXPORT_SYMBOL(kmemleak_not_leak);
906
907 /*
908 * Ignore a memory block. This is usually done when it is known that the
909 * corresponding block is not a leak and does not contain any references to
910 * other allocated memory blocks.
911 */
912 void __ref kmemleak_ignore(const void *ptr)
913 {
914 pr_debug("%s(0x%p)\n", __func__, ptr);
915
916 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
917 make_black_object((unsigned long)ptr);
918 else if (atomic_read(&kmemleak_early_log))
919 log_early(KMEMLEAK_IGNORE, ptr, 0, 0, 0, 0);
920 }
921 EXPORT_SYMBOL(kmemleak_ignore);
922
923 /*
924 * Limit the range to be scanned in an allocated memory block.
925 */
926 void __ref kmemleak_scan_area(const void *ptr, unsigned long offset,
927 size_t length, gfp_t gfp)
928 {
929 pr_debug("%s(0x%p)\n", __func__, ptr);
930
931 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
932 add_scan_area((unsigned long)ptr, offset, length, gfp);
933 else if (atomic_read(&kmemleak_early_log))
934 log_early(KMEMLEAK_SCAN_AREA, ptr, 0, 0, offset, length);
935 }
936 EXPORT_SYMBOL(kmemleak_scan_area);
937
938 /*
939 * Inform kmemleak not to scan the given memory block.
940 */
941 void __ref kmemleak_no_scan(const void *ptr)
942 {
943 pr_debug("%s(0x%p)\n", __func__, ptr);
944
945 if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
946 object_no_scan((unsigned long)ptr);
947 else if (atomic_read(&kmemleak_early_log))
948 log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0, 0, 0);
949 }
950 EXPORT_SYMBOL(kmemleak_no_scan);
951
952 /*
953 * Memory scanning is a long process and it needs to be interruptable. This
954 * function checks whether such interrupt condition occured.
955 */
956 static int scan_should_stop(void)
957 {
958 if (!atomic_read(&kmemleak_enabled))
959 return 1;
960
961 /*
962 * This function may be called from either process or kthread context,
963 * hence the need to check for both stop conditions.
964 */
965 if (current->mm)
966 return signal_pending(current);
967 else
968 return kthread_should_stop();
969
970 return 0;
971 }
972
973 /*
974 * Scan a memory block (exclusive range) for valid pointers and add those
975 * found to the gray list.
976 */
977 static void scan_block(void *_start, void *_end,
978 struct kmemleak_object *scanned, int allow_resched)
979 {
980 unsigned long *ptr;
981 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
982 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
983
984 for (ptr = start; ptr < end; ptr++) {
985 struct kmemleak_object *object;
986 unsigned long flags;
987 unsigned long pointer;
988
989 if (allow_resched)
990 cond_resched();
991 if (scan_should_stop())
992 break;
993
994 /* don't scan uninitialized memory */
995 if (!kmemcheck_is_obj_initialized((unsigned long)ptr,
996 BYTES_PER_POINTER))
997 continue;
998
999 pointer = *ptr;
1000
1001 object = find_and_get_object(pointer, 1);
1002 if (!object)
1003 continue;
1004 if (object == scanned) {
1005 /* self referenced, ignore */
1006 put_object(object);
1007 continue;
1008 }
1009
1010 /*
1011 * Avoid the lockdep recursive warning on object->lock being
1012 * previously acquired in scan_object(). These locks are
1013 * enclosed by scan_mutex.
1014 */
1015 spin_lock_irqsave_nested(&object->lock, flags,
1016 SINGLE_DEPTH_NESTING);
1017 if (!color_white(object)) {
1018 /* non-orphan, ignored or new */
1019 spin_unlock_irqrestore(&object->lock, flags);
1020 put_object(object);
1021 continue;
1022 }
1023
1024 /*
1025 * Increase the object's reference count (number of pointers
1026 * to the memory block). If this count reaches the required
1027 * minimum, the object's color will become gray and it will be
1028 * added to the gray_list.
1029 */
1030 object->count++;
1031 if (color_gray(object))
1032 list_add_tail(&object->gray_list, &gray_list);
1033 else
1034 put_object(object);
1035 spin_unlock_irqrestore(&object->lock, flags);
1036 }
1037 }
1038
1039 /*
1040 * Scan a memory block corresponding to a kmemleak_object. A condition is
1041 * that object->use_count >= 1.
1042 */
1043 static void scan_object(struct kmemleak_object *object)
1044 {
1045 struct kmemleak_scan_area *area;
1046 struct hlist_node *elem;
1047 unsigned long flags;
1048
1049 /*
1050 * Once the object->lock is aquired, the corresponding memory block
1051 * cannot be freed (the same lock is aquired in delete_object).
1052 */
1053 spin_lock_irqsave(&object->lock, flags);
1054 if (object->flags & OBJECT_NO_SCAN)
1055 goto out;
1056 if (!(object->flags & OBJECT_ALLOCATED))
1057 /* already freed object */
1058 goto out;
1059 if (hlist_empty(&object->area_list)) {
1060 void *start = (void *)object->pointer;
1061 void *end = (void *)(object->pointer + object->size);
1062
1063 while (start < end && (object->flags & OBJECT_ALLOCATED) &&
1064 !(object->flags & OBJECT_NO_SCAN)) {
1065 scan_block(start, min(start + MAX_SCAN_SIZE, end),
1066 object, 0);
1067 start += MAX_SCAN_SIZE;
1068
1069 spin_unlock_irqrestore(&object->lock, flags);
1070 cond_resched();
1071 spin_lock_irqsave(&object->lock, flags);
1072 }
1073 } else
1074 hlist_for_each_entry(area, elem, &object->area_list, node)
1075 scan_block((void *)(object->pointer + area->offset),
1076 (void *)(object->pointer + area->offset
1077 + area->length), object, 0);
1078 out:
1079 spin_unlock_irqrestore(&object->lock, flags);
1080 }
1081
1082 /*
1083 * Scan data sections and all the referenced memory blocks allocated via the
1084 * kernel's standard allocators. This function must be called with the
1085 * scan_mutex held.
1086 */
1087 static void kmemleak_scan(void)
1088 {
1089 unsigned long flags;
1090 struct kmemleak_object *object, *tmp;
1091 int i;
1092 int new_leaks = 0;
1093 int gray_list_pass = 0;
1094
1095 jiffies_last_scan = jiffies;
1096
1097 /* prepare the kmemleak_object's */
1098 rcu_read_lock();
1099 list_for_each_entry_rcu(object, &object_list, object_list) {
1100 spin_lock_irqsave(&object->lock, flags);
1101 #ifdef DEBUG
1102 /*
1103 * With a few exceptions there should be a maximum of
1104 * 1 reference to any object at this point.
1105 */
1106 if (atomic_read(&object->use_count) > 1) {
1107 pr_debug("object->use_count = %d\n",
1108 atomic_read(&object->use_count));
1109 dump_object_info(object);
1110 }
1111 #endif
1112 /* reset the reference count (whiten the object) */
1113 object->count = 0;
1114 object->flags &= ~OBJECT_NEW;
1115 if (color_gray(object) && get_object(object))
1116 list_add_tail(&object->gray_list, &gray_list);
1117
1118 spin_unlock_irqrestore(&object->lock, flags);
1119 }
1120 rcu_read_unlock();
1121
1122 /* data/bss scanning */
1123 scan_block(_sdata, _edata, NULL, 1);
1124 scan_block(__bss_start, __bss_stop, NULL, 1);
1125
1126 #ifdef CONFIG_SMP
1127 /* per-cpu sections scanning */
1128 for_each_possible_cpu(i)
1129 scan_block(__per_cpu_start + per_cpu_offset(i),
1130 __per_cpu_end + per_cpu_offset(i), NULL, 1);
1131 #endif
1132
1133 /*
1134 * Struct page scanning for each node. The code below is not yet safe
1135 * with MEMORY_HOTPLUG.
1136 */
1137 for_each_online_node(i) {
1138 pg_data_t *pgdat = NODE_DATA(i);
1139 unsigned long start_pfn = pgdat->node_start_pfn;
1140 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1141 unsigned long pfn;
1142
1143 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1144 struct page *page;
1145
1146 if (!pfn_valid(pfn))
1147 continue;
1148 page = pfn_to_page(pfn);
1149 /* only scan if page is in use */
1150 if (page_count(page) == 0)
1151 continue;
1152 scan_block(page, page + 1, NULL, 1);
1153 }
1154 }
1155
1156 /*
1157 * Scanning the task stacks (may introduce false negatives).
1158 */
1159 if (kmemleak_stack_scan) {
1160 struct task_struct *p, *g;
1161
1162 read_lock(&tasklist_lock);
1163 do_each_thread(g, p) {
1164 scan_block(task_stack_page(p), task_stack_page(p) +
1165 THREAD_SIZE, NULL, 0);
1166 } while_each_thread(g, p);
1167 read_unlock(&tasklist_lock);
1168 }
1169
1170 /*
1171 * Scan the objects already referenced from the sections scanned
1172 * above. More objects will be referenced and, if there are no memory
1173 * leaks, all the objects will be scanned. The list traversal is safe
1174 * for both tail additions and removals from inside the loop. The
1175 * kmemleak objects cannot be freed from outside the loop because their
1176 * use_count was increased.
1177 */
1178 repeat:
1179 object = list_entry(gray_list.next, typeof(*object), gray_list);
1180 while (&object->gray_list != &gray_list) {
1181 cond_resched();
1182
1183 /* may add new objects to the list */
1184 if (!scan_should_stop())
1185 scan_object(object);
1186
1187 tmp = list_entry(object->gray_list.next, typeof(*object),
1188 gray_list);
1189
1190 /* remove the object from the list and release it */
1191 list_del(&object->gray_list);
1192 put_object(object);
1193
1194 object = tmp;
1195 }
1196
1197 if (scan_should_stop() || ++gray_list_pass >= GRAY_LIST_PASSES)
1198 goto scan_end;
1199
1200 /*
1201 * Check for new objects allocated during this scanning and add them
1202 * to the gray list.
1203 */
1204 rcu_read_lock();
1205 list_for_each_entry_rcu(object, &object_list, object_list) {
1206 spin_lock_irqsave(&object->lock, flags);
1207 if ((object->flags & OBJECT_NEW) && !color_black(object) &&
1208 get_object(object)) {
1209 object->flags &= ~OBJECT_NEW;
1210 list_add_tail(&object->gray_list, &gray_list);
1211 }
1212 spin_unlock_irqrestore(&object->lock, flags);
1213 }
1214 rcu_read_unlock();
1215
1216 if (!list_empty(&gray_list))
1217 goto repeat;
1218
1219 scan_end:
1220 WARN_ON(!list_empty(&gray_list));
1221
1222 /*
1223 * If scanning was stopped or new objects were being allocated at a
1224 * higher rate than gray list scanning, do not report any new
1225 * unreferenced objects.
1226 */
1227 if (scan_should_stop() || gray_list_pass >= GRAY_LIST_PASSES)
1228 return;
1229
1230 /*
1231 * Scanning result reporting.
1232 */
1233 rcu_read_lock();
1234 list_for_each_entry_rcu(object, &object_list, object_list) {
1235 spin_lock_irqsave(&object->lock, flags);
1236 if (unreferenced_object(object) &&
1237 !(object->flags & OBJECT_REPORTED)) {
1238 object->flags |= OBJECT_REPORTED;
1239 new_leaks++;
1240 }
1241 spin_unlock_irqrestore(&object->lock, flags);
1242 }
1243 rcu_read_unlock();
1244
1245 if (new_leaks)
1246 pr_info("%d new suspected memory leaks (see "
1247 "/sys/kernel/debug/kmemleak)\n", new_leaks);
1248
1249 }
1250
1251 /*
1252 * Thread function performing automatic memory scanning. Unreferenced objects
1253 * at the end of a memory scan are reported but only the first time.
1254 */
1255 static int kmemleak_scan_thread(void *arg)
1256 {
1257 static int first_run = 1;
1258
1259 pr_info("Automatic memory scanning thread started\n");
1260 set_user_nice(current, 10);
1261
1262 /*
1263 * Wait before the first scan to allow the system to fully initialize.
1264 */
1265 if (first_run) {
1266 first_run = 0;
1267 ssleep(SECS_FIRST_SCAN);
1268 }
1269
1270 while (!kthread_should_stop()) {
1271 signed long timeout = jiffies_scan_wait;
1272
1273 mutex_lock(&scan_mutex);
1274 kmemleak_scan();
1275 mutex_unlock(&scan_mutex);
1276
1277 /* wait before the next scan */
1278 while (timeout && !kthread_should_stop())
1279 timeout = schedule_timeout_interruptible(timeout);
1280 }
1281
1282 pr_info("Automatic memory scanning thread ended\n");
1283
1284 return 0;
1285 }
1286
1287 /*
1288 * Start the automatic memory scanning thread. This function must be called
1289 * with the scan_mutex held.
1290 */
1291 static void start_scan_thread(void)
1292 {
1293 if (scan_thread)
1294 return;
1295 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1296 if (IS_ERR(scan_thread)) {
1297 pr_warning("Failed to create the scan thread\n");
1298 scan_thread = NULL;
1299 }
1300 }
1301
1302 /*
1303 * Stop the automatic memory scanning thread. This function must be called
1304 * with the scan_mutex held.
1305 */
1306 static void stop_scan_thread(void)
1307 {
1308 if (scan_thread) {
1309 kthread_stop(scan_thread);
1310 scan_thread = NULL;
1311 }
1312 }
1313
1314 /*
1315 * Iterate over the object_list and return the first valid object at or after
1316 * the required position with its use_count incremented. The function triggers
1317 * a memory scanning when the pos argument points to the first position.
1318 */
1319 static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1320 {
1321 struct kmemleak_object *object;
1322 loff_t n = *pos;
1323 int err;
1324
1325 err = mutex_lock_interruptible(&scan_mutex);
1326 if (err < 0)
1327 return ERR_PTR(err);
1328
1329 rcu_read_lock();
1330 list_for_each_entry_rcu(object, &object_list, object_list) {
1331 if (n-- > 0)
1332 continue;
1333 if (get_object(object))
1334 goto out;
1335 }
1336 object = NULL;
1337 out:
1338 return object;
1339 }
1340
1341 /*
1342 * Return the next object in the object_list. The function decrements the
1343 * use_count of the previous object and increases that of the next one.
1344 */
1345 static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1346 {
1347 struct kmemleak_object *prev_obj = v;
1348 struct kmemleak_object *next_obj = NULL;
1349 struct list_head *n = &prev_obj->object_list;
1350
1351 ++(*pos);
1352
1353 list_for_each_continue_rcu(n, &object_list) {
1354 next_obj = list_entry(n, struct kmemleak_object, object_list);
1355 if (get_object(next_obj))
1356 break;
1357 }
1358
1359 put_object(prev_obj);
1360 return next_obj;
1361 }
1362
1363 /*
1364 * Decrement the use_count of the last object required, if any.
1365 */
1366 static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1367 {
1368 if (!IS_ERR(v)) {
1369 /*
1370 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1371 * waiting was interrupted, so only release it if !IS_ERR.
1372 */
1373 rcu_read_unlock();
1374 mutex_unlock(&scan_mutex);
1375 if (v)
1376 put_object(v);
1377 }
1378 }
1379
1380 /*
1381 * Print the information for an unreferenced object to the seq file.
1382 */
1383 static int kmemleak_seq_show(struct seq_file *seq, void *v)
1384 {
1385 struct kmemleak_object *object = v;
1386 unsigned long flags;
1387
1388 spin_lock_irqsave(&object->lock, flags);
1389 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1390 print_unreferenced(seq, object);
1391 spin_unlock_irqrestore(&object->lock, flags);
1392 return 0;
1393 }
1394
1395 static const struct seq_operations kmemleak_seq_ops = {
1396 .start = kmemleak_seq_start,
1397 .next = kmemleak_seq_next,
1398 .stop = kmemleak_seq_stop,
1399 .show = kmemleak_seq_show,
1400 };
1401
1402 static int kmemleak_open(struct inode *inode, struct file *file)
1403 {
1404 if (!atomic_read(&kmemleak_enabled))
1405 return -EBUSY;
1406
1407 return seq_open(file, &kmemleak_seq_ops);
1408 }
1409
1410 static int kmemleak_release(struct inode *inode, struct file *file)
1411 {
1412 return seq_release(inode, file);
1413 }
1414
1415 static int dump_str_object_info(const char *str)
1416 {
1417 unsigned long flags;
1418 struct kmemleak_object *object;
1419 unsigned long addr;
1420
1421 addr= simple_strtoul(str, NULL, 0);
1422 object = find_and_get_object(addr, 0);
1423 if (!object) {
1424 pr_info("Unknown object at 0x%08lx\n", addr);
1425 return -EINVAL;
1426 }
1427
1428 spin_lock_irqsave(&object->lock, flags);
1429 dump_object_info(object);
1430 spin_unlock_irqrestore(&object->lock, flags);
1431
1432 put_object(object);
1433 return 0;
1434 }
1435
1436 /*
1437 * We use grey instead of black to ensure we can do future scans on the same
1438 * objects. If we did not do future scans these black objects could
1439 * potentially contain references to newly allocated objects in the future and
1440 * we'd end up with false positives.
1441 */
1442 static void kmemleak_clear(void)
1443 {
1444 struct kmemleak_object *object;
1445 unsigned long flags;
1446
1447 rcu_read_lock();
1448 list_for_each_entry_rcu(object, &object_list, object_list) {
1449 spin_lock_irqsave(&object->lock, flags);
1450 if ((object->flags & OBJECT_REPORTED) &&
1451 unreferenced_object(object))
1452 __paint_it(object, KMEMLEAK_GREY);
1453 spin_unlock_irqrestore(&object->lock, flags);
1454 }
1455 rcu_read_unlock();
1456 }
1457
1458 /*
1459 * File write operation to configure kmemleak at run-time. The following
1460 * commands can be written to the /sys/kernel/debug/kmemleak file:
1461 * off - disable kmemleak (irreversible)
1462 * stack=on - enable the task stacks scanning
1463 * stack=off - disable the tasks stacks scanning
1464 * scan=on - start the automatic memory scanning thread
1465 * scan=off - stop the automatic memory scanning thread
1466 * scan=... - set the automatic memory scanning period in seconds (0 to
1467 * disable it)
1468 * scan - trigger a memory scan
1469 * clear - mark all current reported unreferenced kmemleak objects as
1470 * grey to ignore printing them
1471 * dump=... - dump information about the object found at the given address
1472 */
1473 static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1474 size_t size, loff_t *ppos)
1475 {
1476 char buf[64];
1477 int buf_size;
1478 int ret;
1479
1480 buf_size = min(size, (sizeof(buf) - 1));
1481 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1482 return -EFAULT;
1483 buf[buf_size] = 0;
1484
1485 ret = mutex_lock_interruptible(&scan_mutex);
1486 if (ret < 0)
1487 return ret;
1488
1489 if (strncmp(buf, "off", 3) == 0)
1490 kmemleak_disable();
1491 else if (strncmp(buf, "stack=on", 8) == 0)
1492 kmemleak_stack_scan = 1;
1493 else if (strncmp(buf, "stack=off", 9) == 0)
1494 kmemleak_stack_scan = 0;
1495 else if (strncmp(buf, "scan=on", 7) == 0)
1496 start_scan_thread();
1497 else if (strncmp(buf, "scan=off", 8) == 0)
1498 stop_scan_thread();
1499 else if (strncmp(buf, "scan=", 5) == 0) {
1500 unsigned long secs;
1501
1502 ret = strict_strtoul(buf + 5, 0, &secs);
1503 if (ret < 0)
1504 goto out;
1505 stop_scan_thread();
1506 if (secs) {
1507 jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1508 start_scan_thread();
1509 }
1510 } else if (strncmp(buf, "scan", 4) == 0)
1511 kmemleak_scan();
1512 else if (strncmp(buf, "clear", 5) == 0)
1513 kmemleak_clear();
1514 else if (strncmp(buf, "dump=", 5) == 0)
1515 ret = dump_str_object_info(buf + 5);
1516 else
1517 ret = -EINVAL;
1518
1519 out:
1520 mutex_unlock(&scan_mutex);
1521 if (ret < 0)
1522 return ret;
1523
1524 /* ignore the rest of the buffer, only one command at a time */
1525 *ppos += size;
1526 return size;
1527 }
1528
1529 static const struct file_operations kmemleak_fops = {
1530 .owner = THIS_MODULE,
1531 .open = kmemleak_open,
1532 .read = seq_read,
1533 .write = kmemleak_write,
1534 .llseek = seq_lseek,
1535 .release = kmemleak_release,
1536 };
1537
1538 /*
1539 * Perform the freeing of the kmemleak internal objects after waiting for any
1540 * current memory scan to complete.
1541 */
1542 static void kmemleak_do_cleanup(struct work_struct *work)
1543 {
1544 struct kmemleak_object *object;
1545
1546 mutex_lock(&scan_mutex);
1547 stop_scan_thread();
1548
1549 rcu_read_lock();
1550 list_for_each_entry_rcu(object, &object_list, object_list)
1551 delete_object_full(object->pointer);
1552 rcu_read_unlock();
1553 mutex_unlock(&scan_mutex);
1554 }
1555
1556 static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
1557
1558 /*
1559 * Disable kmemleak. No memory allocation/freeing will be traced once this
1560 * function is called. Disabling kmemleak is an irreversible operation.
1561 */
1562 static void kmemleak_disable(void)
1563 {
1564 /* atomically check whether it was already invoked */
1565 if (atomic_cmpxchg(&kmemleak_error, 0, 1))
1566 return;
1567
1568 /* stop any memory operation tracing */
1569 atomic_set(&kmemleak_early_log, 0);
1570 atomic_set(&kmemleak_enabled, 0);
1571
1572 /* check whether it is too early for a kernel thread */
1573 if (atomic_read(&kmemleak_initialized))
1574 schedule_work(&cleanup_work);
1575
1576 pr_info("Kernel memory leak detector disabled\n");
1577 }
1578
1579 /*
1580 * Allow boot-time kmemleak disabling (enabled by default).
1581 */
1582 static int kmemleak_boot_config(char *str)
1583 {
1584 if (!str)
1585 return -EINVAL;
1586 if (strcmp(str, "off") == 0)
1587 kmemleak_disable();
1588 else if (strcmp(str, "on") != 0)
1589 return -EINVAL;
1590 return 0;
1591 }
1592 early_param("kmemleak", kmemleak_boot_config);
1593
1594 /*
1595 * Kmemleak initialization.
1596 */
1597 void __init kmemleak_init(void)
1598 {
1599 int i;
1600 unsigned long flags;
1601
1602 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1603 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1604
1605 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1606 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1607 INIT_PRIO_TREE_ROOT(&object_tree_root);
1608
1609 /* the kernel is still in UP mode, so disabling the IRQs is enough */
1610 local_irq_save(flags);
1611 if (!atomic_read(&kmemleak_error)) {
1612 atomic_set(&kmemleak_enabled, 1);
1613 atomic_set(&kmemleak_early_log, 0);
1614 }
1615 local_irq_restore(flags);
1616
1617 /*
1618 * This is the point where tracking allocations is safe. Automatic
1619 * scanning is started during the late initcall. Add the early logged
1620 * callbacks to the kmemleak infrastructure.
1621 */
1622 for (i = 0; i < crt_early_log; i++) {
1623 struct early_log *log = &early_log[i];
1624
1625 switch (log->op_type) {
1626 case KMEMLEAK_ALLOC:
1627 early_alloc(log);
1628 break;
1629 case KMEMLEAK_FREE:
1630 kmemleak_free(log->ptr);
1631 break;
1632 case KMEMLEAK_FREE_PART:
1633 kmemleak_free_part(log->ptr, log->size);
1634 break;
1635 case KMEMLEAK_NOT_LEAK:
1636 kmemleak_not_leak(log->ptr);
1637 break;
1638 case KMEMLEAK_IGNORE:
1639 kmemleak_ignore(log->ptr);
1640 break;
1641 case KMEMLEAK_SCAN_AREA:
1642 kmemleak_scan_area(log->ptr, log->offset, log->length,
1643 GFP_KERNEL);
1644 break;
1645 case KMEMLEAK_NO_SCAN:
1646 kmemleak_no_scan(log->ptr);
1647 break;
1648 default:
1649 WARN_ON(1);
1650 }
1651 }
1652 }
1653
1654 /*
1655 * Late initialization function.
1656 */
1657 static int __init kmemleak_late_init(void)
1658 {
1659 struct dentry *dentry;
1660
1661 atomic_set(&kmemleak_initialized, 1);
1662
1663 if (atomic_read(&kmemleak_error)) {
1664 /*
1665 * Some error occured and kmemleak was disabled. There is a
1666 * small chance that kmemleak_disable() was called immediately
1667 * after setting kmemleak_initialized and we may end up with
1668 * two clean-up threads but serialized by scan_mutex.
1669 */
1670 schedule_work(&cleanup_work);
1671 return -ENOMEM;
1672 }
1673
1674 dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL,
1675 &kmemleak_fops);
1676 if (!dentry)
1677 pr_warning("Failed to create the debugfs kmemleak file\n");
1678 mutex_lock(&scan_mutex);
1679 start_scan_thread();
1680 mutex_unlock(&scan_mutex);
1681
1682 pr_info("Kernel memory leak detector initialized\n");
1683
1684 return 0;
1685 }
1686 late_initcall(kmemleak_late_init);