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