]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - mm/kmemleak.c
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
[mirror_ubuntu-artful-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 red black 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 * Locks and mutexes are acquired/nested in the following order:
57 *
58 * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING)
59 *
60 * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex
61 * regions.
62 *
63 * The kmemleak_object structures have a use_count incremented or decremented
64 * using the get_object()/put_object() functions. When the use_count becomes
65 * 0, this count can no longer be incremented and put_object() schedules the
66 * kmemleak_object freeing via an RCU callback. All calls to the get_object()
67 * function must be protected by rcu_read_lock() to avoid accessing a freed
68 * structure.
69 */
70
71 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
72
73 #include <linux/init.h>
74 #include <linux/kernel.h>
75 #include <linux/list.h>
76 #include <linux/sched.h>
77 #include <linux/jiffies.h>
78 #include <linux/delay.h>
79 #include <linux/export.h>
80 #include <linux/kthread.h>
81 #include <linux/rbtree.h>
82 #include <linux/fs.h>
83 #include <linux/debugfs.h>
84 #include <linux/seq_file.h>
85 #include <linux/cpumask.h>
86 #include <linux/spinlock.h>
87 #include <linux/mutex.h>
88 #include <linux/rcupdate.h>
89 #include <linux/stacktrace.h>
90 #include <linux/cache.h>
91 #include <linux/percpu.h>
92 #include <linux/hardirq.h>
93 #include <linux/mmzone.h>
94 #include <linux/slab.h>
95 #include <linux/thread_info.h>
96 #include <linux/err.h>
97 #include <linux/uaccess.h>
98 #include <linux/string.h>
99 #include <linux/nodemask.h>
100 #include <linux/mm.h>
101 #include <linux/workqueue.h>
102 #include <linux/crc32.h>
103
104 #include <asm/sections.h>
105 #include <asm/processor.h>
106 #include <linux/atomic.h>
107
108 #include <linux/kasan.h>
109 #include <linux/kmemcheck.h>
110 #include <linux/kmemleak.h>
111 #include <linux/memory_hotplug.h>
112
113 /*
114 * Kmemleak configuration and common defines.
115 */
116 #define MAX_TRACE 16 /* stack trace length */
117 #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
118 #define SECS_FIRST_SCAN 60 /* delay before the first scan */
119 #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
120 #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
121
122 #define BYTES_PER_POINTER sizeof(void *)
123
124 /* GFP bitmask for kmemleak internal allocations */
125 #define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC)) | \
126 __GFP_NORETRY | __GFP_NOMEMALLOC | \
127 __GFP_NOWARN)
128
129 /* scanning area inside a memory block */
130 struct kmemleak_scan_area {
131 struct hlist_node node;
132 unsigned long start;
133 size_t size;
134 };
135
136 #define KMEMLEAK_GREY 0
137 #define KMEMLEAK_BLACK -1
138
139 /*
140 * Structure holding the metadata for each allocated memory block.
141 * Modifications to such objects should be made while holding the
142 * object->lock. Insertions or deletions from object_list, gray_list or
143 * rb_node are already protected by the corresponding locks or mutex (see
144 * the notes on locking above). These objects are reference-counted
145 * (use_count) and freed using the RCU mechanism.
146 */
147 struct kmemleak_object {
148 spinlock_t lock;
149 unsigned long flags; /* object status flags */
150 struct list_head object_list;
151 struct list_head gray_list;
152 struct rb_node rb_node;
153 struct rcu_head rcu; /* object_list lockless traversal */
154 /* object usage count; object freed when use_count == 0 */
155 atomic_t use_count;
156 unsigned long pointer;
157 size_t size;
158 /* minimum number of a pointers found before it is considered leak */
159 int min_count;
160 /* the total number of pointers found pointing to this object */
161 int count;
162 /* checksum for detecting modified objects */
163 u32 checksum;
164 /* memory ranges to be scanned inside an object (empty for all) */
165 struct hlist_head area_list;
166 unsigned long trace[MAX_TRACE];
167 unsigned int trace_len;
168 unsigned long jiffies; /* creation timestamp */
169 pid_t pid; /* pid of the current task */
170 char comm[TASK_COMM_LEN]; /* executable name */
171 };
172
173 /* flag representing the memory block allocation status */
174 #define OBJECT_ALLOCATED (1 << 0)
175 /* flag set after the first reporting of an unreference object */
176 #define OBJECT_REPORTED (1 << 1)
177 /* flag set to not scan the object */
178 #define OBJECT_NO_SCAN (1 << 2)
179
180 /* number of bytes to print per line; must be 16 or 32 */
181 #define HEX_ROW_SIZE 16
182 /* number of bytes to print at a time (1, 2, 4, 8) */
183 #define HEX_GROUP_SIZE 1
184 /* include ASCII after the hex output */
185 #define HEX_ASCII 1
186 /* max number of lines to be printed */
187 #define HEX_MAX_LINES 2
188
189 /* the list of all allocated objects */
190 static LIST_HEAD(object_list);
191 /* the list of gray-colored objects (see color_gray comment below) */
192 static LIST_HEAD(gray_list);
193 /* search tree for object boundaries */
194 static struct rb_root object_tree_root = RB_ROOT;
195 /* rw_lock protecting the access to object_list and object_tree_root */
196 static DEFINE_RWLOCK(kmemleak_lock);
197
198 /* allocation caches for kmemleak internal data */
199 static struct kmem_cache *object_cache;
200 static struct kmem_cache *scan_area_cache;
201
202 /* set if tracing memory operations is enabled */
203 static int kmemleak_enabled;
204 /* same as above but only for the kmemleak_free() callback */
205 static int kmemleak_free_enabled;
206 /* set in the late_initcall if there were no errors */
207 static int kmemleak_initialized;
208 /* enables or disables early logging of the memory operations */
209 static int kmemleak_early_log = 1;
210 /* set if a kmemleak warning was issued */
211 static int kmemleak_warning;
212 /* set if a fatal kmemleak error has occurred */
213 static int kmemleak_error;
214
215 /* minimum and maximum address that may be valid pointers */
216 static unsigned long min_addr = ULONG_MAX;
217 static unsigned long max_addr;
218
219 static struct task_struct *scan_thread;
220 /* used to avoid reporting of recently allocated objects */
221 static unsigned long jiffies_min_age;
222 static unsigned long jiffies_last_scan;
223 /* delay between automatic memory scannings */
224 static signed long jiffies_scan_wait;
225 /* enables or disables the task stacks scanning */
226 static int kmemleak_stack_scan = 1;
227 /* protects the memory scanning, parameters and debug/kmemleak file access */
228 static DEFINE_MUTEX(scan_mutex);
229 /* setting kmemleak=on, will set this var, skipping the disable */
230 static int kmemleak_skip_disable;
231 /* If there are leaks that can be reported */
232 static bool kmemleak_found_leaks;
233
234 /*
235 * Early object allocation/freeing logging. Kmemleak is initialized after the
236 * kernel allocator. However, both the kernel allocator and kmemleak may
237 * allocate memory blocks which need to be tracked. Kmemleak defines an
238 * arbitrary buffer to hold the allocation/freeing information before it is
239 * fully initialized.
240 */
241
242 /* kmemleak operation type for early logging */
243 enum {
244 KMEMLEAK_ALLOC,
245 KMEMLEAK_ALLOC_PERCPU,
246 KMEMLEAK_FREE,
247 KMEMLEAK_FREE_PART,
248 KMEMLEAK_FREE_PERCPU,
249 KMEMLEAK_NOT_LEAK,
250 KMEMLEAK_IGNORE,
251 KMEMLEAK_SCAN_AREA,
252 KMEMLEAK_NO_SCAN
253 };
254
255 /*
256 * Structure holding the information passed to kmemleak callbacks during the
257 * early logging.
258 */
259 struct early_log {
260 int op_type; /* kmemleak operation type */
261 const void *ptr; /* allocated/freed memory block */
262 size_t size; /* memory block size */
263 int min_count; /* minimum reference count */
264 unsigned long trace[MAX_TRACE]; /* stack trace */
265 unsigned int trace_len; /* stack trace length */
266 };
267
268 /* early logging buffer and current position */
269 static struct early_log
270 early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata;
271 static int crt_early_log __initdata;
272
273 static void kmemleak_disable(void);
274
275 /*
276 * Print a warning and dump the stack trace.
277 */
278 #define kmemleak_warn(x...) do { \
279 pr_warn(x); \
280 dump_stack(); \
281 kmemleak_warning = 1; \
282 } while (0)
283
284 /*
285 * Macro invoked when a serious kmemleak condition occurred and cannot be
286 * recovered from. Kmemleak will be disabled and further allocation/freeing
287 * tracing no longer available.
288 */
289 #define kmemleak_stop(x...) do { \
290 kmemleak_warn(x); \
291 kmemleak_disable(); \
292 } while (0)
293
294 /*
295 * Printing of the objects hex dump to the seq file. The number of lines to be
296 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
297 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
298 * with the object->lock held.
299 */
300 static void hex_dump_object(struct seq_file *seq,
301 struct kmemleak_object *object)
302 {
303 const u8 *ptr = (const u8 *)object->pointer;
304 size_t len;
305
306 /* limit the number of lines to HEX_MAX_LINES */
307 len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
308
309 seq_printf(seq, " hex dump (first %zu bytes):\n", len);
310 seq_hex_dump(seq, " ", DUMP_PREFIX_NONE, HEX_ROW_SIZE,
311 HEX_GROUP_SIZE, ptr, len, HEX_ASCII);
312 }
313
314 /*
315 * Object colors, encoded with count and min_count:
316 * - white - orphan object, not enough references to it (count < min_count)
317 * - gray - not orphan, not marked as false positive (min_count == 0) or
318 * sufficient references to it (count >= min_count)
319 * - black - ignore, it doesn't contain references (e.g. text section)
320 * (min_count == -1). No function defined for this color.
321 * Newly created objects don't have any color assigned (object->count == -1)
322 * before the next memory scan when they become white.
323 */
324 static bool color_white(const struct kmemleak_object *object)
325 {
326 return object->count != KMEMLEAK_BLACK &&
327 object->count < object->min_count;
328 }
329
330 static bool color_gray(const struct kmemleak_object *object)
331 {
332 return object->min_count != KMEMLEAK_BLACK &&
333 object->count >= object->min_count;
334 }
335
336 /*
337 * Objects are considered unreferenced only if their color is white, they have
338 * not be deleted and have a minimum age to avoid false positives caused by
339 * pointers temporarily stored in CPU registers.
340 */
341 static bool unreferenced_object(struct kmemleak_object *object)
342 {
343 return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
344 time_before_eq(object->jiffies + jiffies_min_age,
345 jiffies_last_scan);
346 }
347
348 /*
349 * Printing of the unreferenced objects information to the seq file. The
350 * print_unreferenced function must be called with the object->lock held.
351 */
352 static void print_unreferenced(struct seq_file *seq,
353 struct kmemleak_object *object)
354 {
355 int i;
356 unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
357
358 seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
359 object->pointer, object->size);
360 seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
361 object->comm, object->pid, object->jiffies,
362 msecs_age / 1000, msecs_age % 1000);
363 hex_dump_object(seq, object);
364 seq_printf(seq, " backtrace:\n");
365
366 for (i = 0; i < object->trace_len; i++) {
367 void *ptr = (void *)object->trace[i];
368 seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
369 }
370 }
371
372 /*
373 * Print the kmemleak_object information. This function is used mainly for
374 * debugging special cases when kmemleak operations. It must be called with
375 * the object->lock held.
376 */
377 static void dump_object_info(struct kmemleak_object *object)
378 {
379 struct stack_trace trace;
380
381 trace.nr_entries = object->trace_len;
382 trace.entries = object->trace;
383
384 pr_notice("Object 0x%08lx (size %zu):\n",
385 object->pointer, object->size);
386 pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
387 object->comm, object->pid, object->jiffies);
388 pr_notice(" min_count = %d\n", object->min_count);
389 pr_notice(" count = %d\n", object->count);
390 pr_notice(" flags = 0x%lx\n", object->flags);
391 pr_notice(" checksum = %u\n", object->checksum);
392 pr_notice(" backtrace:\n");
393 print_stack_trace(&trace, 4);
394 }
395
396 /*
397 * Look-up a memory block metadata (kmemleak_object) in the object search
398 * tree based on a pointer value. If alias is 0, only values pointing to the
399 * beginning of the memory block are allowed. The kmemleak_lock must be held
400 * when calling this function.
401 */
402 static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
403 {
404 struct rb_node *rb = object_tree_root.rb_node;
405
406 while (rb) {
407 struct kmemleak_object *object =
408 rb_entry(rb, struct kmemleak_object, rb_node);
409 if (ptr < object->pointer)
410 rb = object->rb_node.rb_left;
411 else if (object->pointer + object->size <= ptr)
412 rb = object->rb_node.rb_right;
413 else if (object->pointer == ptr || alias)
414 return object;
415 else {
416 kmemleak_warn("Found object by alias at 0x%08lx\n",
417 ptr);
418 dump_object_info(object);
419 break;
420 }
421 }
422 return NULL;
423 }
424
425 /*
426 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
427 * that once an object's use_count reached 0, the RCU freeing was already
428 * registered and the object should no longer be used. This function must be
429 * called under the protection of rcu_read_lock().
430 */
431 static int get_object(struct kmemleak_object *object)
432 {
433 return atomic_inc_not_zero(&object->use_count);
434 }
435
436 /*
437 * RCU callback to free a kmemleak_object.
438 */
439 static void free_object_rcu(struct rcu_head *rcu)
440 {
441 struct hlist_node *tmp;
442 struct kmemleak_scan_area *area;
443 struct kmemleak_object *object =
444 container_of(rcu, struct kmemleak_object, rcu);
445
446 /*
447 * Once use_count is 0 (guaranteed by put_object), there is no other
448 * code accessing this object, hence no need for locking.
449 */
450 hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
451 hlist_del(&area->node);
452 kmem_cache_free(scan_area_cache, area);
453 }
454 kmem_cache_free(object_cache, object);
455 }
456
457 /*
458 * Decrement the object use_count. Once the count is 0, free the object using
459 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
460 * delete_object() path, the delayed RCU freeing ensures that there is no
461 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
462 * is also possible.
463 */
464 static void put_object(struct kmemleak_object *object)
465 {
466 if (!atomic_dec_and_test(&object->use_count))
467 return;
468
469 /* should only get here after delete_object was called */
470 WARN_ON(object->flags & OBJECT_ALLOCATED);
471
472 call_rcu(&object->rcu, free_object_rcu);
473 }
474
475 /*
476 * Look up an object in the object search tree and increase its use_count.
477 */
478 static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
479 {
480 unsigned long flags;
481 struct kmemleak_object *object;
482
483 rcu_read_lock();
484 read_lock_irqsave(&kmemleak_lock, flags);
485 object = lookup_object(ptr, alias);
486 read_unlock_irqrestore(&kmemleak_lock, flags);
487
488 /* check whether the object is still available */
489 if (object && !get_object(object))
490 object = NULL;
491 rcu_read_unlock();
492
493 return object;
494 }
495
496 /*
497 * Look up an object in the object search tree and remove it from both
498 * object_tree_root and object_list. The returned object's use_count should be
499 * at least 1, as initially set by create_object().
500 */
501 static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias)
502 {
503 unsigned long flags;
504 struct kmemleak_object *object;
505
506 write_lock_irqsave(&kmemleak_lock, flags);
507 object = lookup_object(ptr, alias);
508 if (object) {
509 rb_erase(&object->rb_node, &object_tree_root);
510 list_del_rcu(&object->object_list);
511 }
512 write_unlock_irqrestore(&kmemleak_lock, flags);
513
514 return object;
515 }
516
517 /*
518 * Save stack trace to the given array of MAX_TRACE size.
519 */
520 static int __save_stack_trace(unsigned long *trace)
521 {
522 struct stack_trace stack_trace;
523
524 stack_trace.max_entries = MAX_TRACE;
525 stack_trace.nr_entries = 0;
526 stack_trace.entries = trace;
527 stack_trace.skip = 2;
528 save_stack_trace(&stack_trace);
529
530 return stack_trace.nr_entries;
531 }
532
533 /*
534 * Create the metadata (struct kmemleak_object) corresponding to an allocated
535 * memory block and add it to the object_list and object_tree_root.
536 */
537 static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
538 int min_count, gfp_t gfp)
539 {
540 unsigned long flags;
541 struct kmemleak_object *object, *parent;
542 struct rb_node **link, *rb_parent;
543
544 object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
545 if (!object) {
546 pr_warn("Cannot allocate a kmemleak_object structure\n");
547 kmemleak_disable();
548 return NULL;
549 }
550
551 INIT_LIST_HEAD(&object->object_list);
552 INIT_LIST_HEAD(&object->gray_list);
553 INIT_HLIST_HEAD(&object->area_list);
554 spin_lock_init(&object->lock);
555 atomic_set(&object->use_count, 1);
556 object->flags = OBJECT_ALLOCATED;
557 object->pointer = ptr;
558 object->size = size;
559 object->min_count = min_count;
560 object->count = 0; /* white color initially */
561 object->jiffies = jiffies;
562 object->checksum = 0;
563
564 /* task information */
565 if (in_irq()) {
566 object->pid = 0;
567 strncpy(object->comm, "hardirq", sizeof(object->comm));
568 } else if (in_softirq()) {
569 object->pid = 0;
570 strncpy(object->comm, "softirq", sizeof(object->comm));
571 } else {
572 object->pid = current->pid;
573 /*
574 * There is a small chance of a race with set_task_comm(),
575 * however using get_task_comm() here may cause locking
576 * dependency issues with current->alloc_lock. In the worst
577 * case, the command line is not correct.
578 */
579 strncpy(object->comm, current->comm, sizeof(object->comm));
580 }
581
582 /* kernel backtrace */
583 object->trace_len = __save_stack_trace(object->trace);
584
585 write_lock_irqsave(&kmemleak_lock, flags);
586
587 min_addr = min(min_addr, ptr);
588 max_addr = max(max_addr, ptr + size);
589 link = &object_tree_root.rb_node;
590 rb_parent = NULL;
591 while (*link) {
592 rb_parent = *link;
593 parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
594 if (ptr + size <= parent->pointer)
595 link = &parent->rb_node.rb_left;
596 else if (parent->pointer + parent->size <= ptr)
597 link = &parent->rb_node.rb_right;
598 else {
599 kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
600 ptr);
601 /*
602 * No need for parent->lock here since "parent" cannot
603 * be freed while the kmemleak_lock is held.
604 */
605 dump_object_info(parent);
606 kmem_cache_free(object_cache, object);
607 object = NULL;
608 goto out;
609 }
610 }
611 rb_link_node(&object->rb_node, rb_parent, link);
612 rb_insert_color(&object->rb_node, &object_tree_root);
613
614 list_add_tail_rcu(&object->object_list, &object_list);
615 out:
616 write_unlock_irqrestore(&kmemleak_lock, flags);
617 return object;
618 }
619
620 /*
621 * Mark the object as not allocated and schedule RCU freeing via put_object().
622 */
623 static void __delete_object(struct kmemleak_object *object)
624 {
625 unsigned long flags;
626
627 WARN_ON(!(object->flags & OBJECT_ALLOCATED));
628 WARN_ON(atomic_read(&object->use_count) < 1);
629
630 /*
631 * Locking here also ensures that the corresponding memory block
632 * cannot be freed when it is being scanned.
633 */
634 spin_lock_irqsave(&object->lock, flags);
635 object->flags &= ~OBJECT_ALLOCATED;
636 spin_unlock_irqrestore(&object->lock, flags);
637 put_object(object);
638 }
639
640 /*
641 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
642 * delete it.
643 */
644 static void delete_object_full(unsigned long ptr)
645 {
646 struct kmemleak_object *object;
647
648 object = find_and_remove_object(ptr, 0);
649 if (!object) {
650 #ifdef DEBUG
651 kmemleak_warn("Freeing unknown object at 0x%08lx\n",
652 ptr);
653 #endif
654 return;
655 }
656 __delete_object(object);
657 }
658
659 /*
660 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
661 * delete it. If the memory block is partially freed, the function may create
662 * additional metadata for the remaining parts of the block.
663 */
664 static void delete_object_part(unsigned long ptr, size_t size)
665 {
666 struct kmemleak_object *object;
667 unsigned long start, end;
668
669 object = find_and_remove_object(ptr, 1);
670 if (!object) {
671 #ifdef DEBUG
672 kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
673 ptr, size);
674 #endif
675 return;
676 }
677
678 /*
679 * Create one or two objects that may result from the memory block
680 * split. Note that partial freeing is only done by free_bootmem() and
681 * this happens before kmemleak_init() is called. The path below is
682 * only executed during early log recording in kmemleak_init(), so
683 * GFP_KERNEL is enough.
684 */
685 start = object->pointer;
686 end = object->pointer + object->size;
687 if (ptr > start)
688 create_object(start, ptr - start, object->min_count,
689 GFP_KERNEL);
690 if (ptr + size < end)
691 create_object(ptr + size, end - ptr - size, object->min_count,
692 GFP_KERNEL);
693
694 __delete_object(object);
695 }
696
697 static void __paint_it(struct kmemleak_object *object, int color)
698 {
699 object->min_count = color;
700 if (color == KMEMLEAK_BLACK)
701 object->flags |= OBJECT_NO_SCAN;
702 }
703
704 static void paint_it(struct kmemleak_object *object, int color)
705 {
706 unsigned long flags;
707
708 spin_lock_irqsave(&object->lock, flags);
709 __paint_it(object, color);
710 spin_unlock_irqrestore(&object->lock, flags);
711 }
712
713 static void paint_ptr(unsigned long ptr, int color)
714 {
715 struct kmemleak_object *object;
716
717 object = find_and_get_object(ptr, 0);
718 if (!object) {
719 kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n",
720 ptr,
721 (color == KMEMLEAK_GREY) ? "Grey" :
722 (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
723 return;
724 }
725 paint_it(object, color);
726 put_object(object);
727 }
728
729 /*
730 * Mark an object permanently as gray-colored so that it can no longer be
731 * reported as a leak. This is used in general to mark a false positive.
732 */
733 static void make_gray_object(unsigned long ptr)
734 {
735 paint_ptr(ptr, KMEMLEAK_GREY);
736 }
737
738 /*
739 * Mark the object as black-colored so that it is ignored from scans and
740 * reporting.
741 */
742 static void make_black_object(unsigned long ptr)
743 {
744 paint_ptr(ptr, KMEMLEAK_BLACK);
745 }
746
747 /*
748 * Add a scanning area to the object. If at least one such area is added,
749 * kmemleak will only scan these ranges rather than the whole memory block.
750 */
751 static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
752 {
753 unsigned long flags;
754 struct kmemleak_object *object;
755 struct kmemleak_scan_area *area;
756
757 object = find_and_get_object(ptr, 1);
758 if (!object) {
759 kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
760 ptr);
761 return;
762 }
763
764 area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
765 if (!area) {
766 pr_warn("Cannot allocate a scan area\n");
767 goto out;
768 }
769
770 spin_lock_irqsave(&object->lock, flags);
771 if (size == SIZE_MAX) {
772 size = object->pointer + object->size - ptr;
773 } else if (ptr + size > object->pointer + object->size) {
774 kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
775 dump_object_info(object);
776 kmem_cache_free(scan_area_cache, area);
777 goto out_unlock;
778 }
779
780 INIT_HLIST_NODE(&area->node);
781 area->start = ptr;
782 area->size = size;
783
784 hlist_add_head(&area->node, &object->area_list);
785 out_unlock:
786 spin_unlock_irqrestore(&object->lock, flags);
787 out:
788 put_object(object);
789 }
790
791 /*
792 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
793 * pointer. Such object will not be scanned by kmemleak but references to it
794 * are searched.
795 */
796 static void object_no_scan(unsigned long ptr)
797 {
798 unsigned long flags;
799 struct kmemleak_object *object;
800
801 object = find_and_get_object(ptr, 0);
802 if (!object) {
803 kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
804 return;
805 }
806
807 spin_lock_irqsave(&object->lock, flags);
808 object->flags |= OBJECT_NO_SCAN;
809 spin_unlock_irqrestore(&object->lock, flags);
810 put_object(object);
811 }
812
813 /*
814 * Log an early kmemleak_* call to the early_log buffer. These calls will be
815 * processed later once kmemleak is fully initialized.
816 */
817 static void __init log_early(int op_type, const void *ptr, size_t size,
818 int min_count)
819 {
820 unsigned long flags;
821 struct early_log *log;
822
823 if (kmemleak_error) {
824 /* kmemleak stopped recording, just count the requests */
825 crt_early_log++;
826 return;
827 }
828
829 if (crt_early_log >= ARRAY_SIZE(early_log)) {
830 crt_early_log++;
831 kmemleak_disable();
832 return;
833 }
834
835 /*
836 * There is no need for locking since the kernel is still in UP mode
837 * at this stage. Disabling the IRQs is enough.
838 */
839 local_irq_save(flags);
840 log = &early_log[crt_early_log];
841 log->op_type = op_type;
842 log->ptr = ptr;
843 log->size = size;
844 log->min_count = min_count;
845 log->trace_len = __save_stack_trace(log->trace);
846 crt_early_log++;
847 local_irq_restore(flags);
848 }
849
850 /*
851 * Log an early allocated block and populate the stack trace.
852 */
853 static void early_alloc(struct early_log *log)
854 {
855 struct kmemleak_object *object;
856 unsigned long flags;
857 int i;
858
859 if (!kmemleak_enabled || !log->ptr || IS_ERR(log->ptr))
860 return;
861
862 /*
863 * RCU locking needed to ensure object is not freed via put_object().
864 */
865 rcu_read_lock();
866 object = create_object((unsigned long)log->ptr, log->size,
867 log->min_count, GFP_ATOMIC);
868 if (!object)
869 goto out;
870 spin_lock_irqsave(&object->lock, flags);
871 for (i = 0; i < log->trace_len; i++)
872 object->trace[i] = log->trace[i];
873 object->trace_len = log->trace_len;
874 spin_unlock_irqrestore(&object->lock, flags);
875 out:
876 rcu_read_unlock();
877 }
878
879 /*
880 * Log an early allocated block and populate the stack trace.
881 */
882 static void early_alloc_percpu(struct early_log *log)
883 {
884 unsigned int cpu;
885 const void __percpu *ptr = log->ptr;
886
887 for_each_possible_cpu(cpu) {
888 log->ptr = per_cpu_ptr(ptr, cpu);
889 early_alloc(log);
890 }
891 }
892
893 /**
894 * kmemleak_alloc - register a newly allocated object
895 * @ptr: pointer to beginning of the object
896 * @size: size of the object
897 * @min_count: minimum number of references to this object. If during memory
898 * scanning a number of references less than @min_count is found,
899 * the object is reported as a memory leak. If @min_count is 0,
900 * the object is never reported as a leak. If @min_count is -1,
901 * the object is ignored (not scanned and not reported as a leak)
902 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
903 *
904 * This function is called from the kernel allocators when a new object
905 * (memory block) is allocated (kmem_cache_alloc, kmalloc, vmalloc etc.).
906 */
907 void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
908 gfp_t gfp)
909 {
910 pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
911
912 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
913 create_object((unsigned long)ptr, size, min_count, gfp);
914 else if (kmemleak_early_log)
915 log_early(KMEMLEAK_ALLOC, ptr, size, min_count);
916 }
917 EXPORT_SYMBOL_GPL(kmemleak_alloc);
918
919 /**
920 * kmemleak_alloc_percpu - register a newly allocated __percpu object
921 * @ptr: __percpu pointer to beginning of the object
922 * @size: size of the object
923 * @gfp: flags used for kmemleak internal memory allocations
924 *
925 * This function is called from the kernel percpu allocator when a new object
926 * (memory block) is allocated (alloc_percpu).
927 */
928 void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
929 gfp_t gfp)
930 {
931 unsigned int cpu;
932
933 pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size);
934
935 /*
936 * Percpu allocations are only scanned and not reported as leaks
937 * (min_count is set to 0).
938 */
939 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
940 for_each_possible_cpu(cpu)
941 create_object((unsigned long)per_cpu_ptr(ptr, cpu),
942 size, 0, gfp);
943 else if (kmemleak_early_log)
944 log_early(KMEMLEAK_ALLOC_PERCPU, ptr, size, 0);
945 }
946 EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
947
948 /**
949 * kmemleak_free - unregister a previously registered object
950 * @ptr: pointer to beginning of the object
951 *
952 * This function is called from the kernel allocators when an object (memory
953 * block) is freed (kmem_cache_free, kfree, vfree etc.).
954 */
955 void __ref kmemleak_free(const void *ptr)
956 {
957 pr_debug("%s(0x%p)\n", __func__, ptr);
958
959 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
960 delete_object_full((unsigned long)ptr);
961 else if (kmemleak_early_log)
962 log_early(KMEMLEAK_FREE, ptr, 0, 0);
963 }
964 EXPORT_SYMBOL_GPL(kmemleak_free);
965
966 /**
967 * kmemleak_free_part - partially unregister a previously registered object
968 * @ptr: pointer to the beginning or inside the object. This also
969 * represents the start of the range to be freed
970 * @size: size to be unregistered
971 *
972 * This function is called when only a part of a memory block is freed
973 * (usually from the bootmem allocator).
974 */
975 void __ref kmemleak_free_part(const void *ptr, size_t size)
976 {
977 pr_debug("%s(0x%p)\n", __func__, ptr);
978
979 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
980 delete_object_part((unsigned long)ptr, size);
981 else if (kmemleak_early_log)
982 log_early(KMEMLEAK_FREE_PART, ptr, size, 0);
983 }
984 EXPORT_SYMBOL_GPL(kmemleak_free_part);
985
986 /**
987 * kmemleak_free_percpu - unregister a previously registered __percpu object
988 * @ptr: __percpu pointer to beginning of the object
989 *
990 * This function is called from the kernel percpu allocator when an object
991 * (memory block) is freed (free_percpu).
992 */
993 void __ref kmemleak_free_percpu(const void __percpu *ptr)
994 {
995 unsigned int cpu;
996
997 pr_debug("%s(0x%p)\n", __func__, ptr);
998
999 if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
1000 for_each_possible_cpu(cpu)
1001 delete_object_full((unsigned long)per_cpu_ptr(ptr,
1002 cpu));
1003 else if (kmemleak_early_log)
1004 log_early(KMEMLEAK_FREE_PERCPU, ptr, 0, 0);
1005 }
1006 EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
1007
1008 /**
1009 * kmemleak_update_trace - update object allocation stack trace
1010 * @ptr: pointer to beginning of the object
1011 *
1012 * Override the object allocation stack trace for cases where the actual
1013 * allocation place is not always useful.
1014 */
1015 void __ref kmemleak_update_trace(const void *ptr)
1016 {
1017 struct kmemleak_object *object;
1018 unsigned long flags;
1019
1020 pr_debug("%s(0x%p)\n", __func__, ptr);
1021
1022 if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
1023 return;
1024
1025 object = find_and_get_object((unsigned long)ptr, 1);
1026 if (!object) {
1027 #ifdef DEBUG
1028 kmemleak_warn("Updating stack trace for unknown object at %p\n",
1029 ptr);
1030 #endif
1031 return;
1032 }
1033
1034 spin_lock_irqsave(&object->lock, flags);
1035 object->trace_len = __save_stack_trace(object->trace);
1036 spin_unlock_irqrestore(&object->lock, flags);
1037
1038 put_object(object);
1039 }
1040 EXPORT_SYMBOL(kmemleak_update_trace);
1041
1042 /**
1043 * kmemleak_not_leak - mark an allocated object as false positive
1044 * @ptr: pointer to beginning of the object
1045 *
1046 * Calling this function on an object will cause the memory block to no longer
1047 * be reported as leak and always be scanned.
1048 */
1049 void __ref kmemleak_not_leak(const void *ptr)
1050 {
1051 pr_debug("%s(0x%p)\n", __func__, ptr);
1052
1053 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1054 make_gray_object((unsigned long)ptr);
1055 else if (kmemleak_early_log)
1056 log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0);
1057 }
1058 EXPORT_SYMBOL(kmemleak_not_leak);
1059
1060 /**
1061 * kmemleak_ignore - ignore an allocated object
1062 * @ptr: pointer to beginning of the object
1063 *
1064 * Calling this function on an object will cause the memory block to be
1065 * ignored (not scanned and not reported as a leak). This is usually done when
1066 * it is known that the corresponding block is not a leak and does not contain
1067 * any references to other allocated memory blocks.
1068 */
1069 void __ref kmemleak_ignore(const void *ptr)
1070 {
1071 pr_debug("%s(0x%p)\n", __func__, ptr);
1072
1073 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1074 make_black_object((unsigned long)ptr);
1075 else if (kmemleak_early_log)
1076 log_early(KMEMLEAK_IGNORE, ptr, 0, 0);
1077 }
1078 EXPORT_SYMBOL(kmemleak_ignore);
1079
1080 /**
1081 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1082 * @ptr: pointer to beginning or inside the object. This also
1083 * represents the start of the scan area
1084 * @size: size of the scan area
1085 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1086 *
1087 * This function is used when it is known that only certain parts of an object
1088 * contain references to other objects. Kmemleak will only scan these areas
1089 * reducing the number false negatives.
1090 */
1091 void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1092 {
1093 pr_debug("%s(0x%p)\n", __func__, ptr);
1094
1095 if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
1096 add_scan_area((unsigned long)ptr, size, gfp);
1097 else if (kmemleak_early_log)
1098 log_early(KMEMLEAK_SCAN_AREA, ptr, size, 0);
1099 }
1100 EXPORT_SYMBOL(kmemleak_scan_area);
1101
1102 /**
1103 * kmemleak_no_scan - do not scan an allocated object
1104 * @ptr: pointer to beginning of the object
1105 *
1106 * This function notifies kmemleak not to scan the given memory block. Useful
1107 * in situations where it is known that the given object does not contain any
1108 * references to other objects. Kmemleak will not scan such objects reducing
1109 * the number of false negatives.
1110 */
1111 void __ref kmemleak_no_scan(const void *ptr)
1112 {
1113 pr_debug("%s(0x%p)\n", __func__, ptr);
1114
1115 if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1116 object_no_scan((unsigned long)ptr);
1117 else if (kmemleak_early_log)
1118 log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0);
1119 }
1120 EXPORT_SYMBOL(kmemleak_no_scan);
1121
1122 /*
1123 * Update an object's checksum and return true if it was modified.
1124 */
1125 static bool update_checksum(struct kmemleak_object *object)
1126 {
1127 u32 old_csum = object->checksum;
1128
1129 if (!kmemcheck_is_obj_initialized(object->pointer, object->size))
1130 return false;
1131
1132 kasan_disable_current();
1133 object->checksum = crc32(0, (void *)object->pointer, object->size);
1134 kasan_enable_current();
1135
1136 return object->checksum != old_csum;
1137 }
1138
1139 /*
1140 * Memory scanning is a long process and it needs to be interruptable. This
1141 * function checks whether such interrupt condition occurred.
1142 */
1143 static int scan_should_stop(void)
1144 {
1145 if (!kmemleak_enabled)
1146 return 1;
1147
1148 /*
1149 * This function may be called from either process or kthread context,
1150 * hence the need to check for both stop conditions.
1151 */
1152 if (current->mm)
1153 return signal_pending(current);
1154 else
1155 return kthread_should_stop();
1156
1157 return 0;
1158 }
1159
1160 /*
1161 * Scan a memory block (exclusive range) for valid pointers and add those
1162 * found to the gray list.
1163 */
1164 static void scan_block(void *_start, void *_end,
1165 struct kmemleak_object *scanned)
1166 {
1167 unsigned long *ptr;
1168 unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1169 unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1170 unsigned long flags;
1171
1172 read_lock_irqsave(&kmemleak_lock, flags);
1173 for (ptr = start; ptr < end; ptr++) {
1174 struct kmemleak_object *object;
1175 unsigned long pointer;
1176
1177 if (scan_should_stop())
1178 break;
1179
1180 /* don't scan uninitialized memory */
1181 if (!kmemcheck_is_obj_initialized((unsigned long)ptr,
1182 BYTES_PER_POINTER))
1183 continue;
1184
1185 kasan_disable_current();
1186 pointer = *ptr;
1187 kasan_enable_current();
1188
1189 if (pointer < min_addr || pointer >= max_addr)
1190 continue;
1191
1192 /*
1193 * No need for get_object() here since we hold kmemleak_lock.
1194 * object->use_count cannot be dropped to 0 while the object
1195 * is still present in object_tree_root and object_list
1196 * (with updates protected by kmemleak_lock).
1197 */
1198 object = lookup_object(pointer, 1);
1199 if (!object)
1200 continue;
1201 if (object == scanned)
1202 /* self referenced, ignore */
1203 continue;
1204
1205 /*
1206 * Avoid the lockdep recursive warning on object->lock being
1207 * previously acquired in scan_object(). These locks are
1208 * enclosed by scan_mutex.
1209 */
1210 spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
1211 if (!color_white(object)) {
1212 /* non-orphan, ignored or new */
1213 spin_unlock(&object->lock);
1214 continue;
1215 }
1216
1217 /*
1218 * Increase the object's reference count (number of pointers
1219 * to the memory block). If this count reaches the required
1220 * minimum, the object's color will become gray and it will be
1221 * added to the gray_list.
1222 */
1223 object->count++;
1224 if (color_gray(object)) {
1225 /* put_object() called when removing from gray_list */
1226 WARN_ON(!get_object(object));
1227 list_add_tail(&object->gray_list, &gray_list);
1228 }
1229 spin_unlock(&object->lock);
1230 }
1231 read_unlock_irqrestore(&kmemleak_lock, flags);
1232 }
1233
1234 /*
1235 * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
1236 */
1237 static void scan_large_block(void *start, void *end)
1238 {
1239 void *next;
1240
1241 while (start < end) {
1242 next = min(start + MAX_SCAN_SIZE, end);
1243 scan_block(start, next, NULL);
1244 start = next;
1245 cond_resched();
1246 }
1247 }
1248
1249 /*
1250 * Scan a memory block corresponding to a kmemleak_object. A condition is
1251 * that object->use_count >= 1.
1252 */
1253 static void scan_object(struct kmemleak_object *object)
1254 {
1255 struct kmemleak_scan_area *area;
1256 unsigned long flags;
1257
1258 /*
1259 * Once the object->lock is acquired, the corresponding memory block
1260 * cannot be freed (the same lock is acquired in delete_object).
1261 */
1262 spin_lock_irqsave(&object->lock, flags);
1263 if (object->flags & OBJECT_NO_SCAN)
1264 goto out;
1265 if (!(object->flags & OBJECT_ALLOCATED))
1266 /* already freed object */
1267 goto out;
1268 if (hlist_empty(&object->area_list)) {
1269 void *start = (void *)object->pointer;
1270 void *end = (void *)(object->pointer + object->size);
1271 void *next;
1272
1273 do {
1274 next = min(start + MAX_SCAN_SIZE, end);
1275 scan_block(start, next, object);
1276
1277 start = next;
1278 if (start >= end)
1279 break;
1280
1281 spin_unlock_irqrestore(&object->lock, flags);
1282 cond_resched();
1283 spin_lock_irqsave(&object->lock, flags);
1284 } while (object->flags & OBJECT_ALLOCATED);
1285 } else
1286 hlist_for_each_entry(area, &object->area_list, node)
1287 scan_block((void *)area->start,
1288 (void *)(area->start + area->size),
1289 object);
1290 out:
1291 spin_unlock_irqrestore(&object->lock, flags);
1292 }
1293
1294 /*
1295 * Scan the objects already referenced (gray objects). More objects will be
1296 * referenced and, if there are no memory leaks, all the objects are scanned.
1297 */
1298 static void scan_gray_list(void)
1299 {
1300 struct kmemleak_object *object, *tmp;
1301
1302 /*
1303 * The list traversal is safe for both tail additions and removals
1304 * from inside the loop. The kmemleak objects cannot be freed from
1305 * outside the loop because their use_count was incremented.
1306 */
1307 object = list_entry(gray_list.next, typeof(*object), gray_list);
1308 while (&object->gray_list != &gray_list) {
1309 cond_resched();
1310
1311 /* may add new objects to the list */
1312 if (!scan_should_stop())
1313 scan_object(object);
1314
1315 tmp = list_entry(object->gray_list.next, typeof(*object),
1316 gray_list);
1317
1318 /* remove the object from the list and release it */
1319 list_del(&object->gray_list);
1320 put_object(object);
1321
1322 object = tmp;
1323 }
1324 WARN_ON(!list_empty(&gray_list));
1325 }
1326
1327 /*
1328 * Scan data sections and all the referenced memory blocks allocated via the
1329 * kernel's standard allocators. This function must be called with the
1330 * scan_mutex held.
1331 */
1332 static void kmemleak_scan(void)
1333 {
1334 unsigned long flags;
1335 struct kmemleak_object *object;
1336 int i;
1337 int new_leaks = 0;
1338
1339 jiffies_last_scan = jiffies;
1340
1341 /* prepare the kmemleak_object's */
1342 rcu_read_lock();
1343 list_for_each_entry_rcu(object, &object_list, object_list) {
1344 spin_lock_irqsave(&object->lock, flags);
1345 #ifdef DEBUG
1346 /*
1347 * With a few exceptions there should be a maximum of
1348 * 1 reference to any object at this point.
1349 */
1350 if (atomic_read(&object->use_count) > 1) {
1351 pr_debug("object->use_count = %d\n",
1352 atomic_read(&object->use_count));
1353 dump_object_info(object);
1354 }
1355 #endif
1356 /* reset the reference count (whiten the object) */
1357 object->count = 0;
1358 if (color_gray(object) && get_object(object))
1359 list_add_tail(&object->gray_list, &gray_list);
1360
1361 spin_unlock_irqrestore(&object->lock, flags);
1362 }
1363 rcu_read_unlock();
1364
1365 /* data/bss scanning */
1366 scan_large_block(_sdata, _edata);
1367 scan_large_block(__bss_start, __bss_stop);
1368
1369 #ifdef CONFIG_SMP
1370 /* per-cpu sections scanning */
1371 for_each_possible_cpu(i)
1372 scan_large_block(__per_cpu_start + per_cpu_offset(i),
1373 __per_cpu_end + per_cpu_offset(i));
1374 #endif
1375
1376 /*
1377 * Struct page scanning for each node.
1378 */
1379 get_online_mems();
1380 for_each_online_node(i) {
1381 unsigned long start_pfn = node_start_pfn(i);
1382 unsigned long end_pfn = node_end_pfn(i);
1383 unsigned long pfn;
1384
1385 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1386 struct page *page;
1387
1388 if (!pfn_valid(pfn))
1389 continue;
1390 page = pfn_to_page(pfn);
1391 /* only scan if page is in use */
1392 if (page_count(page) == 0)
1393 continue;
1394 scan_block(page, page + 1, NULL);
1395 }
1396 }
1397 put_online_mems();
1398
1399 /*
1400 * Scanning the task stacks (may introduce false negatives).
1401 */
1402 if (kmemleak_stack_scan) {
1403 struct task_struct *p, *g;
1404
1405 read_lock(&tasklist_lock);
1406 do_each_thread(g, p) {
1407 scan_block(task_stack_page(p), task_stack_page(p) +
1408 THREAD_SIZE, NULL);
1409 } while_each_thread(g, p);
1410 read_unlock(&tasklist_lock);
1411 }
1412
1413 /*
1414 * Scan the objects already referenced from the sections scanned
1415 * above.
1416 */
1417 scan_gray_list();
1418
1419 /*
1420 * Check for new or unreferenced objects modified since the previous
1421 * scan and color them gray until the next scan.
1422 */
1423 rcu_read_lock();
1424 list_for_each_entry_rcu(object, &object_list, object_list) {
1425 spin_lock_irqsave(&object->lock, flags);
1426 if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1427 && update_checksum(object) && get_object(object)) {
1428 /* color it gray temporarily */
1429 object->count = object->min_count;
1430 list_add_tail(&object->gray_list, &gray_list);
1431 }
1432 spin_unlock_irqrestore(&object->lock, flags);
1433 }
1434 rcu_read_unlock();
1435
1436 /*
1437 * Re-scan the gray list for modified unreferenced objects.
1438 */
1439 scan_gray_list();
1440
1441 /*
1442 * If scanning was stopped do not report any new unreferenced objects.
1443 */
1444 if (scan_should_stop())
1445 return;
1446
1447 /*
1448 * Scanning result reporting.
1449 */
1450 rcu_read_lock();
1451 list_for_each_entry_rcu(object, &object_list, object_list) {
1452 spin_lock_irqsave(&object->lock, flags);
1453 if (unreferenced_object(object) &&
1454 !(object->flags & OBJECT_REPORTED)) {
1455 object->flags |= OBJECT_REPORTED;
1456 new_leaks++;
1457 }
1458 spin_unlock_irqrestore(&object->lock, flags);
1459 }
1460 rcu_read_unlock();
1461
1462 if (new_leaks) {
1463 kmemleak_found_leaks = true;
1464
1465 pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
1466 new_leaks);
1467 }
1468
1469 }
1470
1471 /*
1472 * Thread function performing automatic memory scanning. Unreferenced objects
1473 * at the end of a memory scan are reported but only the first time.
1474 */
1475 static int kmemleak_scan_thread(void *arg)
1476 {
1477 static int first_run = 1;
1478
1479 pr_info("Automatic memory scanning thread started\n");
1480 set_user_nice(current, 10);
1481
1482 /*
1483 * Wait before the first scan to allow the system to fully initialize.
1484 */
1485 if (first_run) {
1486 first_run = 0;
1487 ssleep(SECS_FIRST_SCAN);
1488 }
1489
1490 while (!kthread_should_stop()) {
1491 signed long timeout = jiffies_scan_wait;
1492
1493 mutex_lock(&scan_mutex);
1494 kmemleak_scan();
1495 mutex_unlock(&scan_mutex);
1496
1497 /* wait before the next scan */
1498 while (timeout && !kthread_should_stop())
1499 timeout = schedule_timeout_interruptible(timeout);
1500 }
1501
1502 pr_info("Automatic memory scanning thread ended\n");
1503
1504 return 0;
1505 }
1506
1507 /*
1508 * Start the automatic memory scanning thread. This function must be called
1509 * with the scan_mutex held.
1510 */
1511 static void start_scan_thread(void)
1512 {
1513 if (scan_thread)
1514 return;
1515 scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1516 if (IS_ERR(scan_thread)) {
1517 pr_warn("Failed to create the scan thread\n");
1518 scan_thread = NULL;
1519 }
1520 }
1521
1522 /*
1523 * Stop the automatic memory scanning thread. This function must be called
1524 * with the scan_mutex held.
1525 */
1526 static void stop_scan_thread(void)
1527 {
1528 if (scan_thread) {
1529 kthread_stop(scan_thread);
1530 scan_thread = NULL;
1531 }
1532 }
1533
1534 /*
1535 * Iterate over the object_list and return the first valid object at or after
1536 * the required position with its use_count incremented. The function triggers
1537 * a memory scanning when the pos argument points to the first position.
1538 */
1539 static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1540 {
1541 struct kmemleak_object *object;
1542 loff_t n = *pos;
1543 int err;
1544
1545 err = mutex_lock_interruptible(&scan_mutex);
1546 if (err < 0)
1547 return ERR_PTR(err);
1548
1549 rcu_read_lock();
1550 list_for_each_entry_rcu(object, &object_list, object_list) {
1551 if (n-- > 0)
1552 continue;
1553 if (get_object(object))
1554 goto out;
1555 }
1556 object = NULL;
1557 out:
1558 return object;
1559 }
1560
1561 /*
1562 * Return the next object in the object_list. The function decrements the
1563 * use_count of the previous object and increases that of the next one.
1564 */
1565 static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1566 {
1567 struct kmemleak_object *prev_obj = v;
1568 struct kmemleak_object *next_obj = NULL;
1569 struct kmemleak_object *obj = prev_obj;
1570
1571 ++(*pos);
1572
1573 list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
1574 if (get_object(obj)) {
1575 next_obj = obj;
1576 break;
1577 }
1578 }
1579
1580 put_object(prev_obj);
1581 return next_obj;
1582 }
1583
1584 /*
1585 * Decrement the use_count of the last object required, if any.
1586 */
1587 static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1588 {
1589 if (!IS_ERR(v)) {
1590 /*
1591 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1592 * waiting was interrupted, so only release it if !IS_ERR.
1593 */
1594 rcu_read_unlock();
1595 mutex_unlock(&scan_mutex);
1596 if (v)
1597 put_object(v);
1598 }
1599 }
1600
1601 /*
1602 * Print the information for an unreferenced object to the seq file.
1603 */
1604 static int kmemleak_seq_show(struct seq_file *seq, void *v)
1605 {
1606 struct kmemleak_object *object = v;
1607 unsigned long flags;
1608
1609 spin_lock_irqsave(&object->lock, flags);
1610 if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1611 print_unreferenced(seq, object);
1612 spin_unlock_irqrestore(&object->lock, flags);
1613 return 0;
1614 }
1615
1616 static const struct seq_operations kmemleak_seq_ops = {
1617 .start = kmemleak_seq_start,
1618 .next = kmemleak_seq_next,
1619 .stop = kmemleak_seq_stop,
1620 .show = kmemleak_seq_show,
1621 };
1622
1623 static int kmemleak_open(struct inode *inode, struct file *file)
1624 {
1625 return seq_open(file, &kmemleak_seq_ops);
1626 }
1627
1628 static int dump_str_object_info(const char *str)
1629 {
1630 unsigned long flags;
1631 struct kmemleak_object *object;
1632 unsigned long addr;
1633
1634 if (kstrtoul(str, 0, &addr))
1635 return -EINVAL;
1636 object = find_and_get_object(addr, 0);
1637 if (!object) {
1638 pr_info("Unknown object at 0x%08lx\n", addr);
1639 return -EINVAL;
1640 }
1641
1642 spin_lock_irqsave(&object->lock, flags);
1643 dump_object_info(object);
1644 spin_unlock_irqrestore(&object->lock, flags);
1645
1646 put_object(object);
1647 return 0;
1648 }
1649
1650 /*
1651 * We use grey instead of black to ensure we can do future scans on the same
1652 * objects. If we did not do future scans these black objects could
1653 * potentially contain references to newly allocated objects in the future and
1654 * we'd end up with false positives.
1655 */
1656 static void kmemleak_clear(void)
1657 {
1658 struct kmemleak_object *object;
1659 unsigned long flags;
1660
1661 rcu_read_lock();
1662 list_for_each_entry_rcu(object, &object_list, object_list) {
1663 spin_lock_irqsave(&object->lock, flags);
1664 if ((object->flags & OBJECT_REPORTED) &&
1665 unreferenced_object(object))
1666 __paint_it(object, KMEMLEAK_GREY);
1667 spin_unlock_irqrestore(&object->lock, flags);
1668 }
1669 rcu_read_unlock();
1670
1671 kmemleak_found_leaks = false;
1672 }
1673
1674 static void __kmemleak_do_cleanup(void);
1675
1676 /*
1677 * File write operation to configure kmemleak at run-time. The following
1678 * commands can be written to the /sys/kernel/debug/kmemleak file:
1679 * off - disable kmemleak (irreversible)
1680 * stack=on - enable the task stacks scanning
1681 * stack=off - disable the tasks stacks scanning
1682 * scan=on - start the automatic memory scanning thread
1683 * scan=off - stop the automatic memory scanning thread
1684 * scan=... - set the automatic memory scanning period in seconds (0 to
1685 * disable it)
1686 * scan - trigger a memory scan
1687 * clear - mark all current reported unreferenced kmemleak objects as
1688 * grey to ignore printing them, or free all kmemleak objects
1689 * if kmemleak has been disabled.
1690 * dump=... - dump information about the object found at the given address
1691 */
1692 static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1693 size_t size, loff_t *ppos)
1694 {
1695 char buf[64];
1696 int buf_size;
1697 int ret;
1698
1699 buf_size = min(size, (sizeof(buf) - 1));
1700 if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1701 return -EFAULT;
1702 buf[buf_size] = 0;
1703
1704 ret = mutex_lock_interruptible(&scan_mutex);
1705 if (ret < 0)
1706 return ret;
1707
1708 if (strncmp(buf, "clear", 5) == 0) {
1709 if (kmemleak_enabled)
1710 kmemleak_clear();
1711 else
1712 __kmemleak_do_cleanup();
1713 goto out;
1714 }
1715
1716 if (!kmemleak_enabled) {
1717 ret = -EBUSY;
1718 goto out;
1719 }
1720
1721 if (strncmp(buf, "off", 3) == 0)
1722 kmemleak_disable();
1723 else if (strncmp(buf, "stack=on", 8) == 0)
1724 kmemleak_stack_scan = 1;
1725 else if (strncmp(buf, "stack=off", 9) == 0)
1726 kmemleak_stack_scan = 0;
1727 else if (strncmp(buf, "scan=on", 7) == 0)
1728 start_scan_thread();
1729 else if (strncmp(buf, "scan=off", 8) == 0)
1730 stop_scan_thread();
1731 else if (strncmp(buf, "scan=", 5) == 0) {
1732 unsigned long secs;
1733
1734 ret = kstrtoul(buf + 5, 0, &secs);
1735 if (ret < 0)
1736 goto out;
1737 stop_scan_thread();
1738 if (secs) {
1739 jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1740 start_scan_thread();
1741 }
1742 } else if (strncmp(buf, "scan", 4) == 0)
1743 kmemleak_scan();
1744 else if (strncmp(buf, "dump=", 5) == 0)
1745 ret = dump_str_object_info(buf + 5);
1746 else
1747 ret = -EINVAL;
1748
1749 out:
1750 mutex_unlock(&scan_mutex);
1751 if (ret < 0)
1752 return ret;
1753
1754 /* ignore the rest of the buffer, only one command at a time */
1755 *ppos += size;
1756 return size;
1757 }
1758
1759 static const struct file_operations kmemleak_fops = {
1760 .owner = THIS_MODULE,
1761 .open = kmemleak_open,
1762 .read = seq_read,
1763 .write = kmemleak_write,
1764 .llseek = seq_lseek,
1765 .release = seq_release,
1766 };
1767
1768 static void __kmemleak_do_cleanup(void)
1769 {
1770 struct kmemleak_object *object;
1771
1772 rcu_read_lock();
1773 list_for_each_entry_rcu(object, &object_list, object_list)
1774 delete_object_full(object->pointer);
1775 rcu_read_unlock();
1776 }
1777
1778 /*
1779 * Stop the memory scanning thread and free the kmemleak internal objects if
1780 * no previous scan thread (otherwise, kmemleak may still have some useful
1781 * information on memory leaks).
1782 */
1783 static void kmemleak_do_cleanup(struct work_struct *work)
1784 {
1785 stop_scan_thread();
1786
1787 /*
1788 * Once the scan thread has stopped, it is safe to no longer track
1789 * object freeing. Ordering of the scan thread stopping and the memory
1790 * accesses below is guaranteed by the kthread_stop() function.
1791 */
1792 kmemleak_free_enabled = 0;
1793
1794 if (!kmemleak_found_leaks)
1795 __kmemleak_do_cleanup();
1796 else
1797 pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
1798 }
1799
1800 static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
1801
1802 /*
1803 * Disable kmemleak. No memory allocation/freeing will be traced once this
1804 * function is called. Disabling kmemleak is an irreversible operation.
1805 */
1806 static void kmemleak_disable(void)
1807 {
1808 /* atomically check whether it was already invoked */
1809 if (cmpxchg(&kmemleak_error, 0, 1))
1810 return;
1811
1812 /* stop any memory operation tracing */
1813 kmemleak_enabled = 0;
1814
1815 /* check whether it is too early for a kernel thread */
1816 if (kmemleak_initialized)
1817 schedule_work(&cleanup_work);
1818 else
1819 kmemleak_free_enabled = 0;
1820
1821 pr_info("Kernel memory leak detector disabled\n");
1822 }
1823
1824 /*
1825 * Allow boot-time kmemleak disabling (enabled by default).
1826 */
1827 static int kmemleak_boot_config(char *str)
1828 {
1829 if (!str)
1830 return -EINVAL;
1831 if (strcmp(str, "off") == 0)
1832 kmemleak_disable();
1833 else if (strcmp(str, "on") == 0)
1834 kmemleak_skip_disable = 1;
1835 else
1836 return -EINVAL;
1837 return 0;
1838 }
1839 early_param("kmemleak", kmemleak_boot_config);
1840
1841 static void __init print_log_trace(struct early_log *log)
1842 {
1843 struct stack_trace trace;
1844
1845 trace.nr_entries = log->trace_len;
1846 trace.entries = log->trace;
1847
1848 pr_notice("Early log backtrace:\n");
1849 print_stack_trace(&trace, 2);
1850 }
1851
1852 /*
1853 * Kmemleak initialization.
1854 */
1855 void __init kmemleak_init(void)
1856 {
1857 int i;
1858 unsigned long flags;
1859
1860 #ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
1861 if (!kmemleak_skip_disable) {
1862 kmemleak_early_log = 0;
1863 kmemleak_disable();
1864 return;
1865 }
1866 #endif
1867
1868 jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1869 jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1870
1871 object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1872 scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1873
1874 if (crt_early_log > ARRAY_SIZE(early_log))
1875 pr_warn("Early log buffer exceeded (%d), please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n",
1876 crt_early_log);
1877
1878 /* the kernel is still in UP mode, so disabling the IRQs is enough */
1879 local_irq_save(flags);
1880 kmemleak_early_log = 0;
1881 if (kmemleak_error) {
1882 local_irq_restore(flags);
1883 return;
1884 } else {
1885 kmemleak_enabled = 1;
1886 kmemleak_free_enabled = 1;
1887 }
1888 local_irq_restore(flags);
1889
1890 /*
1891 * This is the point where tracking allocations is safe. Automatic
1892 * scanning is started during the late initcall. Add the early logged
1893 * callbacks to the kmemleak infrastructure.
1894 */
1895 for (i = 0; i < crt_early_log; i++) {
1896 struct early_log *log = &early_log[i];
1897
1898 switch (log->op_type) {
1899 case KMEMLEAK_ALLOC:
1900 early_alloc(log);
1901 break;
1902 case KMEMLEAK_ALLOC_PERCPU:
1903 early_alloc_percpu(log);
1904 break;
1905 case KMEMLEAK_FREE:
1906 kmemleak_free(log->ptr);
1907 break;
1908 case KMEMLEAK_FREE_PART:
1909 kmemleak_free_part(log->ptr, log->size);
1910 break;
1911 case KMEMLEAK_FREE_PERCPU:
1912 kmemleak_free_percpu(log->ptr);
1913 break;
1914 case KMEMLEAK_NOT_LEAK:
1915 kmemleak_not_leak(log->ptr);
1916 break;
1917 case KMEMLEAK_IGNORE:
1918 kmemleak_ignore(log->ptr);
1919 break;
1920 case KMEMLEAK_SCAN_AREA:
1921 kmemleak_scan_area(log->ptr, log->size, GFP_KERNEL);
1922 break;
1923 case KMEMLEAK_NO_SCAN:
1924 kmemleak_no_scan(log->ptr);
1925 break;
1926 default:
1927 kmemleak_warn("Unknown early log operation: %d\n",
1928 log->op_type);
1929 }
1930
1931 if (kmemleak_warning) {
1932 print_log_trace(log);
1933 kmemleak_warning = 0;
1934 }
1935 }
1936 }
1937
1938 /*
1939 * Late initialization function.
1940 */
1941 static int __init kmemleak_late_init(void)
1942 {
1943 struct dentry *dentry;
1944
1945 kmemleak_initialized = 1;
1946
1947 if (kmemleak_error) {
1948 /*
1949 * Some error occurred and kmemleak was disabled. There is a
1950 * small chance that kmemleak_disable() was called immediately
1951 * after setting kmemleak_initialized and we may end up with
1952 * two clean-up threads but serialized by scan_mutex.
1953 */
1954 schedule_work(&cleanup_work);
1955 return -ENOMEM;
1956 }
1957
1958 dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL,
1959 &kmemleak_fops);
1960 if (!dentry)
1961 pr_warn("Failed to create the debugfs kmemleak file\n");
1962 mutex_lock(&scan_mutex);
1963 start_scan_thread();
1964 mutex_unlock(&scan_mutex);
1965
1966 pr_info("Kernel memory leak detector initialized\n");
1967
1968 return 0;
1969 }
1970 late_initcall(kmemleak_late_init);