]>
Commit | Line | Data |
---|---|---|
3c7b4e6b CM |
1 | /* |
2 | * mm/kmemleak.c | |
3 | * | |
4 | * Copyright (C) 2008 ARM Limited | |
5 | * Written by Catalin Marinas <catalin.marinas@arm.com> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * along with this program; if not, write to the Free Software | |
18 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
19 | * | |
20 | * | |
21 | * For more information on the algorithm and kmemleak usage, please see | |
22 | * Documentation/kmemleak.txt. | |
23 | * | |
24 | * Notes on locking | |
25 | * ---------------- | |
26 | * | |
27 | * The following locks and mutexes are used by kmemleak: | |
28 | * | |
29 | * - kmemleak_lock (rwlock): protects the object_list modifications and | |
30 | * accesses to the object_tree_root. The object_list is the main list | |
31 | * holding the metadata (struct kmemleak_object) for the allocated memory | |
32 | * blocks. The object_tree_root is a priority search tree used to look-up | |
33 | * metadata based on a pointer to the corresponding memory block. The | |
34 | * kmemleak_object structures are added to the object_list and | |
35 | * object_tree_root in the create_object() function called from the | |
36 | * kmemleak_alloc() callback and removed in delete_object() called from the | |
37 | * kmemleak_free() callback | |
38 | * - kmemleak_object.lock (spinlock): protects a kmemleak_object. Accesses to | |
39 | * the metadata (e.g. count) are protected by this lock. Note that some | |
40 | * members of this structure may be protected by other means (atomic or | |
41 | * kmemleak_lock). This lock is also held when scanning the corresponding | |
42 | * memory block to avoid the kernel freeing it via the kmemleak_free() | |
43 | * callback. This is less heavyweight than holding a global lock like | |
44 | * kmemleak_lock during scanning | |
45 | * - scan_mutex (mutex): ensures that only one thread may scan the memory for | |
46 | * unreferenced objects at a time. The gray_list contains the objects which | |
47 | * are already referenced or marked as false positives and need to be | |
48 | * scanned. This list is only modified during a scanning episode when the | |
49 | * scan_mutex is held. At the end of a scan, the gray_list is always empty. | |
50 | * Note that the kmemleak_object.use_count is incremented when an object is | |
4698c1f2 CM |
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 | |
3c7b4e6b CM |
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 | ||
ae281064 JP |
64 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
65 | ||
3c7b4e6b CM |
66 | #include <linux/init.h> |
67 | #include <linux/kernel.h> | |
68 | #include <linux/list.h> | |
69 | #include <linux/sched.h> | |
70 | #include <linux/jiffies.h> | |
71 | #include <linux/delay.h> | |
72 | #include <linux/module.h> | |
73 | #include <linux/kthread.h> | |
74 | #include <linux/prio_tree.h> | |
75 | #include <linux/gfp.h> | |
76 | #include <linux/fs.h> | |
77 | #include <linux/debugfs.h> | |
78 | #include <linux/seq_file.h> | |
79 | #include <linux/cpumask.h> | |
80 | #include <linux/spinlock.h> | |
81 | #include <linux/mutex.h> | |
82 | #include <linux/rcupdate.h> | |
83 | #include <linux/stacktrace.h> | |
84 | #include <linux/cache.h> | |
85 | #include <linux/percpu.h> | |
86 | #include <linux/hardirq.h> | |
87 | #include <linux/mmzone.h> | |
88 | #include <linux/slab.h> | |
89 | #include <linux/thread_info.h> | |
90 | #include <linux/err.h> | |
91 | #include <linux/uaccess.h> | |
92 | #include <linux/string.h> | |
93 | #include <linux/nodemask.h> | |
94 | #include <linux/mm.h> | |
95 | ||
96 | #include <asm/sections.h> | |
97 | #include <asm/processor.h> | |
98 | #include <asm/atomic.h> | |
99 | ||
100 | #include <linux/kmemleak.h> | |
101 | ||
102 | /* | |
103 | * Kmemleak configuration and common defines. | |
104 | */ | |
105 | #define MAX_TRACE 16 /* stack trace length */ | |
106 | #define REPORTS_NR 50 /* maximum number of reported leaks */ | |
107 | #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */ | |
108 | #define MSECS_SCAN_YIELD 10 /* CPU yielding period */ | |
109 | #define SECS_FIRST_SCAN 60 /* delay before the first scan */ | |
110 | #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */ | |
111 | ||
112 | #define BYTES_PER_POINTER sizeof(void *) | |
113 | ||
216c04b0 CM |
114 | /* GFP bitmask for kmemleak internal allocations */ |
115 | #define GFP_KMEMLEAK_MASK (GFP_KERNEL | GFP_ATOMIC) | |
116 | ||
3c7b4e6b CM |
117 | /* scanning area inside a memory block */ |
118 | struct kmemleak_scan_area { | |
119 | struct hlist_node node; | |
120 | unsigned long offset; | |
121 | size_t length; | |
122 | }; | |
123 | ||
124 | /* | |
125 | * Structure holding the metadata for each allocated memory block. | |
126 | * Modifications to such objects should be made while holding the | |
127 | * object->lock. Insertions or deletions from object_list, gray_list or | |
128 | * tree_node are already protected by the corresponding locks or mutex (see | |
129 | * the notes on locking above). These objects are reference-counted | |
130 | * (use_count) and freed using the RCU mechanism. | |
131 | */ | |
132 | struct kmemleak_object { | |
133 | spinlock_t lock; | |
134 | unsigned long flags; /* object status flags */ | |
135 | struct list_head object_list; | |
136 | struct list_head gray_list; | |
137 | struct prio_tree_node tree_node; | |
138 | struct rcu_head rcu; /* object_list lockless traversal */ | |
139 | /* object usage count; object freed when use_count == 0 */ | |
140 | atomic_t use_count; | |
141 | unsigned long pointer; | |
142 | size_t size; | |
143 | /* minimum number of a pointers found before it is considered leak */ | |
144 | int min_count; | |
145 | /* the total number of pointers found pointing to this object */ | |
146 | int count; | |
147 | /* memory ranges to be scanned inside an object (empty for all) */ | |
148 | struct hlist_head area_list; | |
149 | unsigned long trace[MAX_TRACE]; | |
150 | unsigned int trace_len; | |
151 | unsigned long jiffies; /* creation timestamp */ | |
152 | pid_t pid; /* pid of the current task */ | |
153 | char comm[TASK_COMM_LEN]; /* executable name */ | |
154 | }; | |
155 | ||
156 | /* flag representing the memory block allocation status */ | |
157 | #define OBJECT_ALLOCATED (1 << 0) | |
158 | /* flag set after the first reporting of an unreference object */ | |
159 | #define OBJECT_REPORTED (1 << 1) | |
160 | /* flag set to not scan the object */ | |
161 | #define OBJECT_NO_SCAN (1 << 2) | |
162 | ||
163 | /* the list of all allocated objects */ | |
164 | static LIST_HEAD(object_list); | |
165 | /* the list of gray-colored objects (see color_gray comment below) */ | |
166 | static LIST_HEAD(gray_list); | |
167 | /* prio search tree for object boundaries */ | |
168 | static struct prio_tree_root object_tree_root; | |
169 | /* rw_lock protecting the access to object_list and prio_tree_root */ | |
170 | static DEFINE_RWLOCK(kmemleak_lock); | |
171 | ||
172 | /* allocation caches for kmemleak internal data */ | |
173 | static struct kmem_cache *object_cache; | |
174 | static struct kmem_cache *scan_area_cache; | |
175 | ||
176 | /* set if tracing memory operations is enabled */ | |
177 | static atomic_t kmemleak_enabled = ATOMIC_INIT(0); | |
178 | /* set in the late_initcall if there were no errors */ | |
179 | static atomic_t kmemleak_initialized = ATOMIC_INIT(0); | |
180 | /* enables or disables early logging of the memory operations */ | |
181 | static atomic_t kmemleak_early_log = ATOMIC_INIT(1); | |
182 | /* set if a fata kmemleak error has occurred */ | |
183 | static atomic_t kmemleak_error = ATOMIC_INIT(0); | |
184 | ||
185 | /* minimum and maximum address that may be valid pointers */ | |
186 | static unsigned long min_addr = ULONG_MAX; | |
187 | static unsigned long max_addr; | |
188 | ||
189 | /* used for yielding the CPU to other tasks during scanning */ | |
190 | static unsigned long next_scan_yield; | |
191 | static struct task_struct *scan_thread; | |
192 | static unsigned long jiffies_scan_yield; | |
acf4968e | 193 | /* used to avoid reporting of recently allocated objects */ |
3c7b4e6b | 194 | static unsigned long jiffies_min_age; |
acf4968e | 195 | static unsigned long jiffies_last_scan; |
3c7b4e6b CM |
196 | /* delay between automatic memory scannings */ |
197 | static signed long jiffies_scan_wait; | |
198 | /* enables or disables the task stacks scanning */ | |
e0a2a160 | 199 | static int kmemleak_stack_scan = 1; |
4698c1f2 | 200 | /* protects the memory scanning, parameters and debug/kmemleak file access */ |
3c7b4e6b | 201 | static DEFINE_MUTEX(scan_mutex); |
3c7b4e6b CM |
202 | |
203 | /* number of leaks reported (for limitation purposes) */ | |
204 | static int reported_leaks; | |
205 | ||
206 | /* | |
2030117d | 207 | * Early object allocation/freeing logging. Kmemleak is initialized after the |
3c7b4e6b | 208 | * kernel allocator. However, both the kernel allocator and kmemleak may |
2030117d | 209 | * allocate memory blocks which need to be tracked. Kmemleak defines an |
3c7b4e6b CM |
210 | * arbitrary buffer to hold the allocation/freeing information before it is |
211 | * fully initialized. | |
212 | */ | |
213 | ||
214 | /* kmemleak operation type for early logging */ | |
215 | enum { | |
216 | KMEMLEAK_ALLOC, | |
217 | KMEMLEAK_FREE, | |
218 | KMEMLEAK_NOT_LEAK, | |
219 | KMEMLEAK_IGNORE, | |
220 | KMEMLEAK_SCAN_AREA, | |
221 | KMEMLEAK_NO_SCAN | |
222 | }; | |
223 | ||
224 | /* | |
225 | * Structure holding the information passed to kmemleak callbacks during the | |
226 | * early logging. | |
227 | */ | |
228 | struct early_log { | |
229 | int op_type; /* kmemleak operation type */ | |
230 | const void *ptr; /* allocated/freed memory block */ | |
231 | size_t size; /* memory block size */ | |
232 | int min_count; /* minimum reference count */ | |
233 | unsigned long offset; /* scan area offset */ | |
234 | size_t length; /* scan area length */ | |
235 | }; | |
236 | ||
237 | /* early logging buffer and current position */ | |
a9d9058a | 238 | static struct early_log early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE]; |
3c7b4e6b CM |
239 | static int crt_early_log; |
240 | ||
241 | static void kmemleak_disable(void); | |
242 | ||
243 | /* | |
244 | * Print a warning and dump the stack trace. | |
245 | */ | |
246 | #define kmemleak_warn(x...) do { \ | |
247 | pr_warning(x); \ | |
248 | dump_stack(); \ | |
249 | } while (0) | |
250 | ||
251 | /* | |
252 | * Macro invoked when a serious kmemleak condition occured and cannot be | |
2030117d | 253 | * recovered from. Kmemleak will be disabled and further allocation/freeing |
3c7b4e6b CM |
254 | * tracing no longer available. |
255 | */ | |
000814f4 | 256 | #define kmemleak_stop(x...) do { \ |
3c7b4e6b CM |
257 | kmemleak_warn(x); \ |
258 | kmemleak_disable(); \ | |
259 | } while (0) | |
260 | ||
261 | /* | |
262 | * Object colors, encoded with count and min_count: | |
263 | * - white - orphan object, not enough references to it (count < min_count) | |
264 | * - gray - not orphan, not marked as false positive (min_count == 0) or | |
265 | * sufficient references to it (count >= min_count) | |
266 | * - black - ignore, it doesn't contain references (e.g. text section) | |
267 | * (min_count == -1). No function defined for this color. | |
268 | * Newly created objects don't have any color assigned (object->count == -1) | |
269 | * before the next memory scan when they become white. | |
270 | */ | |
271 | static int color_white(const struct kmemleak_object *object) | |
272 | { | |
273 | return object->count != -1 && object->count < object->min_count; | |
274 | } | |
275 | ||
276 | static int color_gray(const struct kmemleak_object *object) | |
277 | { | |
278 | return object->min_count != -1 && object->count >= object->min_count; | |
279 | } | |
280 | ||
3c7b4e6b CM |
281 | /* |
282 | * Objects are considered unreferenced only if their color is white, they have | |
283 | * not be deleted and have a minimum age to avoid false positives caused by | |
284 | * pointers temporarily stored in CPU registers. | |
285 | */ | |
286 | static int unreferenced_object(struct kmemleak_object *object) | |
287 | { | |
288 | return (object->flags & OBJECT_ALLOCATED) && color_white(object) && | |
acf4968e CM |
289 | time_before_eq(object->jiffies + jiffies_min_age, |
290 | jiffies_last_scan); | |
3c7b4e6b CM |
291 | } |
292 | ||
293 | /* | |
bab4a34a CM |
294 | * Printing of the unreferenced objects information to the seq file. The |
295 | * print_unreferenced function must be called with the object->lock held. | |
3c7b4e6b | 296 | */ |
3c7b4e6b CM |
297 | static void print_unreferenced(struct seq_file *seq, |
298 | struct kmemleak_object *object) | |
299 | { | |
300 | int i; | |
301 | ||
bab4a34a CM |
302 | seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n", |
303 | object->pointer, object->size); | |
304 | seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu\n", | |
305 | object->comm, object->pid, object->jiffies); | |
306 | seq_printf(seq, " backtrace:\n"); | |
3c7b4e6b CM |
307 | |
308 | for (i = 0; i < object->trace_len; i++) { | |
309 | void *ptr = (void *)object->trace[i]; | |
bab4a34a | 310 | seq_printf(seq, " [<%p>] %pS\n", ptr, ptr); |
3c7b4e6b CM |
311 | } |
312 | } | |
313 | ||
314 | /* | |
315 | * Print the kmemleak_object information. This function is used mainly for | |
316 | * debugging special cases when kmemleak operations. It must be called with | |
317 | * the object->lock held. | |
318 | */ | |
319 | static void dump_object_info(struct kmemleak_object *object) | |
320 | { | |
321 | struct stack_trace trace; | |
322 | ||
323 | trace.nr_entries = object->trace_len; | |
324 | trace.entries = object->trace; | |
325 | ||
ae281064 | 326 | pr_notice("Object 0x%08lx (size %zu):\n", |
3c7b4e6b CM |
327 | object->tree_node.start, object->size); |
328 | pr_notice(" comm \"%s\", pid %d, jiffies %lu\n", | |
329 | object->comm, object->pid, object->jiffies); | |
330 | pr_notice(" min_count = %d\n", object->min_count); | |
331 | pr_notice(" count = %d\n", object->count); | |
332 | pr_notice(" backtrace:\n"); | |
333 | print_stack_trace(&trace, 4); | |
334 | } | |
335 | ||
336 | /* | |
337 | * Look-up a memory block metadata (kmemleak_object) in the priority search | |
338 | * tree based on a pointer value. If alias is 0, only values pointing to the | |
339 | * beginning of the memory block are allowed. The kmemleak_lock must be held | |
340 | * when calling this function. | |
341 | */ | |
342 | static struct kmemleak_object *lookup_object(unsigned long ptr, int alias) | |
343 | { | |
344 | struct prio_tree_node *node; | |
345 | struct prio_tree_iter iter; | |
346 | struct kmemleak_object *object; | |
347 | ||
348 | prio_tree_iter_init(&iter, &object_tree_root, ptr, ptr); | |
349 | node = prio_tree_next(&iter); | |
350 | if (node) { | |
351 | object = prio_tree_entry(node, struct kmemleak_object, | |
352 | tree_node); | |
353 | if (!alias && object->pointer != ptr) { | |
ae281064 | 354 | kmemleak_warn("Found object by alias"); |
3c7b4e6b CM |
355 | object = NULL; |
356 | } | |
357 | } else | |
358 | object = NULL; | |
359 | ||
360 | return object; | |
361 | } | |
362 | ||
363 | /* | |
364 | * Increment the object use_count. Return 1 if successful or 0 otherwise. Note | |
365 | * that once an object's use_count reached 0, the RCU freeing was already | |
366 | * registered and the object should no longer be used. This function must be | |
367 | * called under the protection of rcu_read_lock(). | |
368 | */ | |
369 | static int get_object(struct kmemleak_object *object) | |
370 | { | |
371 | return atomic_inc_not_zero(&object->use_count); | |
372 | } | |
373 | ||
374 | /* | |
375 | * RCU callback to free a kmemleak_object. | |
376 | */ | |
377 | static void free_object_rcu(struct rcu_head *rcu) | |
378 | { | |
379 | struct hlist_node *elem, *tmp; | |
380 | struct kmemleak_scan_area *area; | |
381 | struct kmemleak_object *object = | |
382 | container_of(rcu, struct kmemleak_object, rcu); | |
383 | ||
384 | /* | |
385 | * Once use_count is 0 (guaranteed by put_object), there is no other | |
386 | * code accessing this object, hence no need for locking. | |
387 | */ | |
388 | hlist_for_each_entry_safe(area, elem, tmp, &object->area_list, node) { | |
389 | hlist_del(elem); | |
390 | kmem_cache_free(scan_area_cache, area); | |
391 | } | |
392 | kmem_cache_free(object_cache, object); | |
393 | } | |
394 | ||
395 | /* | |
396 | * Decrement the object use_count. Once the count is 0, free the object using | |
397 | * an RCU callback. Since put_object() may be called via the kmemleak_free() -> | |
398 | * delete_object() path, the delayed RCU freeing ensures that there is no | |
399 | * recursive call to the kernel allocator. Lock-less RCU object_list traversal | |
400 | * is also possible. | |
401 | */ | |
402 | static void put_object(struct kmemleak_object *object) | |
403 | { | |
404 | if (!atomic_dec_and_test(&object->use_count)) | |
405 | return; | |
406 | ||
407 | /* should only get here after delete_object was called */ | |
408 | WARN_ON(object->flags & OBJECT_ALLOCATED); | |
409 | ||
410 | call_rcu(&object->rcu, free_object_rcu); | |
411 | } | |
412 | ||
413 | /* | |
414 | * Look up an object in the prio search tree and increase its use_count. | |
415 | */ | |
416 | static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias) | |
417 | { | |
418 | unsigned long flags; | |
419 | struct kmemleak_object *object = NULL; | |
420 | ||
421 | rcu_read_lock(); | |
422 | read_lock_irqsave(&kmemleak_lock, flags); | |
423 | if (ptr >= min_addr && ptr < max_addr) | |
424 | object = lookup_object(ptr, alias); | |
425 | read_unlock_irqrestore(&kmemleak_lock, flags); | |
426 | ||
427 | /* check whether the object is still available */ | |
428 | if (object && !get_object(object)) | |
429 | object = NULL; | |
430 | rcu_read_unlock(); | |
431 | ||
432 | return object; | |
433 | } | |
434 | ||
435 | /* | |
436 | * Create the metadata (struct kmemleak_object) corresponding to an allocated | |
437 | * memory block and add it to the object_list and object_tree_root. | |
438 | */ | |
439 | static void create_object(unsigned long ptr, size_t size, int min_count, | |
440 | gfp_t gfp) | |
441 | { | |
442 | unsigned long flags; | |
443 | struct kmemleak_object *object; | |
444 | struct prio_tree_node *node; | |
445 | struct stack_trace trace; | |
446 | ||
216c04b0 | 447 | object = kmem_cache_alloc(object_cache, gfp & GFP_KMEMLEAK_MASK); |
3c7b4e6b | 448 | if (!object) { |
ae281064 | 449 | kmemleak_stop("Cannot allocate a kmemleak_object structure\n"); |
3c7b4e6b CM |
450 | return; |
451 | } | |
452 | ||
453 | INIT_LIST_HEAD(&object->object_list); | |
454 | INIT_LIST_HEAD(&object->gray_list); | |
455 | INIT_HLIST_HEAD(&object->area_list); | |
456 | spin_lock_init(&object->lock); | |
457 | atomic_set(&object->use_count, 1); | |
458 | object->flags = OBJECT_ALLOCATED; | |
459 | object->pointer = ptr; | |
460 | object->size = size; | |
461 | object->min_count = min_count; | |
462 | object->count = -1; /* no color initially */ | |
463 | object->jiffies = jiffies; | |
464 | ||
465 | /* task information */ | |
466 | if (in_irq()) { | |
467 | object->pid = 0; | |
468 | strncpy(object->comm, "hardirq", sizeof(object->comm)); | |
469 | } else if (in_softirq()) { | |
470 | object->pid = 0; | |
471 | strncpy(object->comm, "softirq", sizeof(object->comm)); | |
472 | } else { | |
473 | object->pid = current->pid; | |
474 | /* | |
475 | * There is a small chance of a race with set_task_comm(), | |
476 | * however using get_task_comm() here may cause locking | |
477 | * dependency issues with current->alloc_lock. In the worst | |
478 | * case, the command line is not correct. | |
479 | */ | |
480 | strncpy(object->comm, current->comm, sizeof(object->comm)); | |
481 | } | |
482 | ||
483 | /* kernel backtrace */ | |
484 | trace.max_entries = MAX_TRACE; | |
485 | trace.nr_entries = 0; | |
486 | trace.entries = object->trace; | |
487 | trace.skip = 1; | |
488 | save_stack_trace(&trace); | |
489 | object->trace_len = trace.nr_entries; | |
490 | ||
491 | INIT_PRIO_TREE_NODE(&object->tree_node); | |
492 | object->tree_node.start = ptr; | |
493 | object->tree_node.last = ptr + size - 1; | |
494 | ||
495 | write_lock_irqsave(&kmemleak_lock, flags); | |
496 | min_addr = min(min_addr, ptr); | |
497 | max_addr = max(max_addr, ptr + size); | |
498 | node = prio_tree_insert(&object_tree_root, &object->tree_node); | |
499 | /* | |
500 | * The code calling the kernel does not yet have the pointer to the | |
501 | * memory block to be able to free it. However, we still hold the | |
502 | * kmemleak_lock here in case parts of the kernel started freeing | |
503 | * random memory blocks. | |
504 | */ | |
505 | if (node != &object->tree_node) { | |
506 | unsigned long flags; | |
507 | ||
ae281064 JP |
508 | kmemleak_stop("Cannot insert 0x%lx into the object search tree " |
509 | "(already existing)\n", ptr); | |
3c7b4e6b CM |
510 | object = lookup_object(ptr, 1); |
511 | spin_lock_irqsave(&object->lock, flags); | |
512 | dump_object_info(object); | |
513 | spin_unlock_irqrestore(&object->lock, flags); | |
514 | ||
515 | goto out; | |
516 | } | |
517 | list_add_tail_rcu(&object->object_list, &object_list); | |
518 | out: | |
519 | write_unlock_irqrestore(&kmemleak_lock, flags); | |
520 | } | |
521 | ||
522 | /* | |
523 | * Remove the metadata (struct kmemleak_object) for a memory block from the | |
524 | * object_list and object_tree_root and decrement its use_count. | |
525 | */ | |
526 | static void delete_object(unsigned long ptr) | |
527 | { | |
528 | unsigned long flags; | |
529 | struct kmemleak_object *object; | |
530 | ||
531 | write_lock_irqsave(&kmemleak_lock, flags); | |
532 | object = lookup_object(ptr, 0); | |
533 | if (!object) { | |
ae281064 | 534 | kmemleak_warn("Freeing unknown object at 0x%08lx\n", |
3c7b4e6b CM |
535 | ptr); |
536 | write_unlock_irqrestore(&kmemleak_lock, flags); | |
537 | return; | |
538 | } | |
539 | prio_tree_remove(&object_tree_root, &object->tree_node); | |
540 | list_del_rcu(&object->object_list); | |
541 | write_unlock_irqrestore(&kmemleak_lock, flags); | |
542 | ||
543 | WARN_ON(!(object->flags & OBJECT_ALLOCATED)); | |
544 | WARN_ON(atomic_read(&object->use_count) < 1); | |
545 | ||
546 | /* | |
547 | * Locking here also ensures that the corresponding memory block | |
548 | * cannot be freed when it is being scanned. | |
549 | */ | |
550 | spin_lock_irqsave(&object->lock, flags); | |
3c7b4e6b CM |
551 | object->flags &= ~OBJECT_ALLOCATED; |
552 | spin_unlock_irqrestore(&object->lock, flags); | |
553 | put_object(object); | |
554 | } | |
555 | ||
556 | /* | |
557 | * Make a object permanently as gray-colored so that it can no longer be | |
558 | * reported as a leak. This is used in general to mark a false positive. | |
559 | */ | |
560 | static void make_gray_object(unsigned long ptr) | |
561 | { | |
562 | unsigned long flags; | |
563 | struct kmemleak_object *object; | |
564 | ||
565 | object = find_and_get_object(ptr, 0); | |
566 | if (!object) { | |
ae281064 | 567 | kmemleak_warn("Graying unknown object at 0x%08lx\n", ptr); |
3c7b4e6b CM |
568 | return; |
569 | } | |
570 | ||
571 | spin_lock_irqsave(&object->lock, flags); | |
572 | object->min_count = 0; | |
573 | spin_unlock_irqrestore(&object->lock, flags); | |
574 | put_object(object); | |
575 | } | |
576 | ||
577 | /* | |
578 | * Mark the object as black-colored so that it is ignored from scans and | |
579 | * reporting. | |
580 | */ | |
581 | static void make_black_object(unsigned long ptr) | |
582 | { | |
583 | unsigned long flags; | |
584 | struct kmemleak_object *object; | |
585 | ||
586 | object = find_and_get_object(ptr, 0); | |
587 | if (!object) { | |
ae281064 | 588 | kmemleak_warn("Blacking unknown object at 0x%08lx\n", ptr); |
3c7b4e6b CM |
589 | return; |
590 | } | |
591 | ||
592 | spin_lock_irqsave(&object->lock, flags); | |
593 | object->min_count = -1; | |
594 | spin_unlock_irqrestore(&object->lock, flags); | |
595 | put_object(object); | |
596 | } | |
597 | ||
598 | /* | |
599 | * Add a scanning area to the object. If at least one such area is added, | |
600 | * kmemleak will only scan these ranges rather than the whole memory block. | |
601 | */ | |
602 | static void add_scan_area(unsigned long ptr, unsigned long offset, | |
603 | size_t length, gfp_t gfp) | |
604 | { | |
605 | unsigned long flags; | |
606 | struct kmemleak_object *object; | |
607 | struct kmemleak_scan_area *area; | |
608 | ||
609 | object = find_and_get_object(ptr, 0); | |
610 | if (!object) { | |
ae281064 JP |
611 | kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n", |
612 | ptr); | |
3c7b4e6b CM |
613 | return; |
614 | } | |
615 | ||
216c04b0 | 616 | area = kmem_cache_alloc(scan_area_cache, gfp & GFP_KMEMLEAK_MASK); |
3c7b4e6b | 617 | if (!area) { |
ae281064 | 618 | kmemleak_warn("Cannot allocate a scan area\n"); |
3c7b4e6b CM |
619 | goto out; |
620 | } | |
621 | ||
622 | spin_lock_irqsave(&object->lock, flags); | |
623 | if (offset + length > object->size) { | |
ae281064 | 624 | kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr); |
3c7b4e6b CM |
625 | dump_object_info(object); |
626 | kmem_cache_free(scan_area_cache, area); | |
627 | goto out_unlock; | |
628 | } | |
629 | ||
630 | INIT_HLIST_NODE(&area->node); | |
631 | area->offset = offset; | |
632 | area->length = length; | |
633 | ||
634 | hlist_add_head(&area->node, &object->area_list); | |
635 | out_unlock: | |
636 | spin_unlock_irqrestore(&object->lock, flags); | |
637 | out: | |
638 | put_object(object); | |
639 | } | |
640 | ||
641 | /* | |
642 | * Set the OBJECT_NO_SCAN flag for the object corresponding to the give | |
643 | * pointer. Such object will not be scanned by kmemleak but references to it | |
644 | * are searched. | |
645 | */ | |
646 | static void object_no_scan(unsigned long ptr) | |
647 | { | |
648 | unsigned long flags; | |
649 | struct kmemleak_object *object; | |
650 | ||
651 | object = find_and_get_object(ptr, 0); | |
652 | if (!object) { | |
ae281064 | 653 | kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr); |
3c7b4e6b CM |
654 | return; |
655 | } | |
656 | ||
657 | spin_lock_irqsave(&object->lock, flags); | |
658 | object->flags |= OBJECT_NO_SCAN; | |
659 | spin_unlock_irqrestore(&object->lock, flags); | |
660 | put_object(object); | |
661 | } | |
662 | ||
663 | /* | |
664 | * Log an early kmemleak_* call to the early_log buffer. These calls will be | |
665 | * processed later once kmemleak is fully initialized. | |
666 | */ | |
667 | static void log_early(int op_type, const void *ptr, size_t size, | |
668 | int min_count, unsigned long offset, size_t length) | |
669 | { | |
670 | unsigned long flags; | |
671 | struct early_log *log; | |
672 | ||
673 | if (crt_early_log >= ARRAY_SIZE(early_log)) { | |
a9d9058a CM |
674 | pr_warning("Early log buffer exceeded\n"); |
675 | kmemleak_disable(); | |
3c7b4e6b CM |
676 | return; |
677 | } | |
678 | ||
679 | /* | |
680 | * There is no need for locking since the kernel is still in UP mode | |
681 | * at this stage. Disabling the IRQs is enough. | |
682 | */ | |
683 | local_irq_save(flags); | |
684 | log = &early_log[crt_early_log]; | |
685 | log->op_type = op_type; | |
686 | log->ptr = ptr; | |
687 | log->size = size; | |
688 | log->min_count = min_count; | |
689 | log->offset = offset; | |
690 | log->length = length; | |
691 | crt_early_log++; | |
692 | local_irq_restore(flags); | |
693 | } | |
694 | ||
695 | /* | |
696 | * Memory allocation function callback. This function is called from the | |
697 | * kernel allocators when a new block is allocated (kmem_cache_alloc, kmalloc, | |
698 | * vmalloc etc.). | |
699 | */ | |
700 | void kmemleak_alloc(const void *ptr, size_t size, int min_count, gfp_t gfp) | |
701 | { | |
702 | pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count); | |
703 | ||
704 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
705 | create_object((unsigned long)ptr, size, min_count, gfp); | |
706 | else if (atomic_read(&kmemleak_early_log)) | |
707 | log_early(KMEMLEAK_ALLOC, ptr, size, min_count, 0, 0); | |
708 | } | |
709 | EXPORT_SYMBOL_GPL(kmemleak_alloc); | |
710 | ||
711 | /* | |
712 | * Memory freeing function callback. This function is called from the kernel | |
713 | * allocators when a block is freed (kmem_cache_free, kfree, vfree etc.). | |
714 | */ | |
715 | void kmemleak_free(const void *ptr) | |
716 | { | |
717 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
718 | ||
719 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
720 | delete_object((unsigned long)ptr); | |
721 | else if (atomic_read(&kmemleak_early_log)) | |
722 | log_early(KMEMLEAK_FREE, ptr, 0, 0, 0, 0); | |
723 | } | |
724 | EXPORT_SYMBOL_GPL(kmemleak_free); | |
725 | ||
726 | /* | |
727 | * Mark an already allocated memory block as a false positive. This will cause | |
728 | * the block to no longer be reported as leak and always be scanned. | |
729 | */ | |
730 | void kmemleak_not_leak(const void *ptr) | |
731 | { | |
732 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
733 | ||
734 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
735 | make_gray_object((unsigned long)ptr); | |
736 | else if (atomic_read(&kmemleak_early_log)) | |
737 | log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0, 0, 0); | |
738 | } | |
739 | EXPORT_SYMBOL(kmemleak_not_leak); | |
740 | ||
741 | /* | |
742 | * Ignore a memory block. This is usually done when it is known that the | |
743 | * corresponding block is not a leak and does not contain any references to | |
744 | * other allocated memory blocks. | |
745 | */ | |
746 | void kmemleak_ignore(const void *ptr) | |
747 | { | |
748 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
749 | ||
750 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
751 | make_black_object((unsigned long)ptr); | |
752 | else if (atomic_read(&kmemleak_early_log)) | |
753 | log_early(KMEMLEAK_IGNORE, ptr, 0, 0, 0, 0); | |
754 | } | |
755 | EXPORT_SYMBOL(kmemleak_ignore); | |
756 | ||
757 | /* | |
758 | * Limit the range to be scanned in an allocated memory block. | |
759 | */ | |
760 | void kmemleak_scan_area(const void *ptr, unsigned long offset, size_t length, | |
761 | gfp_t gfp) | |
762 | { | |
763 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
764 | ||
765 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
766 | add_scan_area((unsigned long)ptr, offset, length, gfp); | |
767 | else if (atomic_read(&kmemleak_early_log)) | |
768 | log_early(KMEMLEAK_SCAN_AREA, ptr, 0, 0, offset, length); | |
769 | } | |
770 | EXPORT_SYMBOL(kmemleak_scan_area); | |
771 | ||
772 | /* | |
773 | * Inform kmemleak not to scan the given memory block. | |
774 | */ | |
775 | void kmemleak_no_scan(const void *ptr) | |
776 | { | |
777 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
778 | ||
779 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
780 | object_no_scan((unsigned long)ptr); | |
781 | else if (atomic_read(&kmemleak_early_log)) | |
782 | log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0, 0, 0); | |
783 | } | |
784 | EXPORT_SYMBOL(kmemleak_no_scan); | |
785 | ||
786 | /* | |
787 | * Yield the CPU so that other tasks get a chance to run. The yielding is | |
788 | * rate-limited to avoid excessive number of calls to the schedule() function | |
789 | * during memory scanning. | |
790 | */ | |
791 | static void scan_yield(void) | |
792 | { | |
793 | might_sleep(); | |
794 | ||
795 | if (time_is_before_eq_jiffies(next_scan_yield)) { | |
796 | schedule(); | |
797 | next_scan_yield = jiffies + jiffies_scan_yield; | |
798 | } | |
799 | } | |
800 | ||
801 | /* | |
802 | * Memory scanning is a long process and it needs to be interruptable. This | |
803 | * function checks whether such interrupt condition occured. | |
804 | */ | |
805 | static int scan_should_stop(void) | |
806 | { | |
807 | if (!atomic_read(&kmemleak_enabled)) | |
808 | return 1; | |
809 | ||
810 | /* | |
811 | * This function may be called from either process or kthread context, | |
812 | * hence the need to check for both stop conditions. | |
813 | */ | |
814 | if (current->mm) | |
815 | return signal_pending(current); | |
816 | else | |
817 | return kthread_should_stop(); | |
818 | ||
819 | return 0; | |
820 | } | |
821 | ||
822 | /* | |
823 | * Scan a memory block (exclusive range) for valid pointers and add those | |
824 | * found to the gray list. | |
825 | */ | |
826 | static void scan_block(void *_start, void *_end, | |
827 | struct kmemleak_object *scanned) | |
828 | { | |
829 | unsigned long *ptr; | |
830 | unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER); | |
831 | unsigned long *end = _end - (BYTES_PER_POINTER - 1); | |
832 | ||
833 | for (ptr = start; ptr < end; ptr++) { | |
834 | unsigned long flags; | |
835 | unsigned long pointer = *ptr; | |
836 | struct kmemleak_object *object; | |
837 | ||
838 | if (scan_should_stop()) | |
839 | break; | |
840 | ||
841 | /* | |
842 | * When scanning a memory block with a corresponding | |
843 | * kmemleak_object, the CPU yielding is handled in the calling | |
844 | * code since it holds the object->lock to avoid the block | |
845 | * freeing. | |
846 | */ | |
847 | if (!scanned) | |
848 | scan_yield(); | |
849 | ||
850 | object = find_and_get_object(pointer, 1); | |
851 | if (!object) | |
852 | continue; | |
853 | if (object == scanned) { | |
854 | /* self referenced, ignore */ | |
855 | put_object(object); | |
856 | continue; | |
857 | } | |
858 | ||
859 | /* | |
860 | * Avoid the lockdep recursive warning on object->lock being | |
861 | * previously acquired in scan_object(). These locks are | |
862 | * enclosed by scan_mutex. | |
863 | */ | |
864 | spin_lock_irqsave_nested(&object->lock, flags, | |
865 | SINGLE_DEPTH_NESTING); | |
866 | if (!color_white(object)) { | |
867 | /* non-orphan, ignored or new */ | |
868 | spin_unlock_irqrestore(&object->lock, flags); | |
869 | put_object(object); | |
870 | continue; | |
871 | } | |
872 | ||
873 | /* | |
874 | * Increase the object's reference count (number of pointers | |
875 | * to the memory block). If this count reaches the required | |
876 | * minimum, the object's color will become gray and it will be | |
877 | * added to the gray_list. | |
878 | */ | |
879 | object->count++; | |
880 | if (color_gray(object)) | |
881 | list_add_tail(&object->gray_list, &gray_list); | |
882 | else | |
883 | put_object(object); | |
884 | spin_unlock_irqrestore(&object->lock, flags); | |
885 | } | |
886 | } | |
887 | ||
888 | /* | |
889 | * Scan a memory block corresponding to a kmemleak_object. A condition is | |
890 | * that object->use_count >= 1. | |
891 | */ | |
892 | static void scan_object(struct kmemleak_object *object) | |
893 | { | |
894 | struct kmemleak_scan_area *area; | |
895 | struct hlist_node *elem; | |
896 | unsigned long flags; | |
897 | ||
898 | /* | |
899 | * Once the object->lock is aquired, the corresponding memory block | |
900 | * cannot be freed (the same lock is aquired in delete_object). | |
901 | */ | |
902 | spin_lock_irqsave(&object->lock, flags); | |
903 | if (object->flags & OBJECT_NO_SCAN) | |
904 | goto out; | |
905 | if (!(object->flags & OBJECT_ALLOCATED)) | |
906 | /* already freed object */ | |
907 | goto out; | |
908 | if (hlist_empty(&object->area_list)) | |
909 | scan_block((void *)object->pointer, | |
910 | (void *)(object->pointer + object->size), object); | |
911 | else | |
912 | hlist_for_each_entry(area, elem, &object->area_list, node) | |
913 | scan_block((void *)(object->pointer + area->offset), | |
914 | (void *)(object->pointer + area->offset | |
915 | + area->length), object); | |
916 | out: | |
917 | spin_unlock_irqrestore(&object->lock, flags); | |
918 | } | |
919 | ||
920 | /* | |
921 | * Scan data sections and all the referenced memory blocks allocated via the | |
922 | * kernel's standard allocators. This function must be called with the | |
923 | * scan_mutex held. | |
924 | */ | |
925 | static void kmemleak_scan(void) | |
926 | { | |
927 | unsigned long flags; | |
928 | struct kmemleak_object *object, *tmp; | |
929 | struct task_struct *task; | |
930 | int i; | |
4698c1f2 | 931 | int new_leaks = 0; |
3c7b4e6b | 932 | |
acf4968e CM |
933 | jiffies_last_scan = jiffies; |
934 | ||
3c7b4e6b CM |
935 | /* prepare the kmemleak_object's */ |
936 | rcu_read_lock(); | |
937 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
938 | spin_lock_irqsave(&object->lock, flags); | |
939 | #ifdef DEBUG | |
940 | /* | |
941 | * With a few exceptions there should be a maximum of | |
942 | * 1 reference to any object at this point. | |
943 | */ | |
944 | if (atomic_read(&object->use_count) > 1) { | |
ae281064 | 945 | pr_debug("object->use_count = %d\n", |
3c7b4e6b CM |
946 | atomic_read(&object->use_count)); |
947 | dump_object_info(object); | |
948 | } | |
949 | #endif | |
950 | /* reset the reference count (whiten the object) */ | |
951 | object->count = 0; | |
952 | if (color_gray(object) && get_object(object)) | |
953 | list_add_tail(&object->gray_list, &gray_list); | |
954 | ||
955 | spin_unlock_irqrestore(&object->lock, flags); | |
956 | } | |
957 | rcu_read_unlock(); | |
958 | ||
959 | /* data/bss scanning */ | |
960 | scan_block(_sdata, _edata, NULL); | |
961 | scan_block(__bss_start, __bss_stop, NULL); | |
962 | ||
963 | #ifdef CONFIG_SMP | |
964 | /* per-cpu sections scanning */ | |
965 | for_each_possible_cpu(i) | |
966 | scan_block(__per_cpu_start + per_cpu_offset(i), | |
967 | __per_cpu_end + per_cpu_offset(i), NULL); | |
968 | #endif | |
969 | ||
970 | /* | |
971 | * Struct page scanning for each node. The code below is not yet safe | |
972 | * with MEMORY_HOTPLUG. | |
973 | */ | |
974 | for_each_online_node(i) { | |
975 | pg_data_t *pgdat = NODE_DATA(i); | |
976 | unsigned long start_pfn = pgdat->node_start_pfn; | |
977 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
978 | unsigned long pfn; | |
979 | ||
980 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
981 | struct page *page; | |
982 | ||
983 | if (!pfn_valid(pfn)) | |
984 | continue; | |
985 | page = pfn_to_page(pfn); | |
986 | /* only scan if page is in use */ | |
987 | if (page_count(page) == 0) | |
988 | continue; | |
989 | scan_block(page, page + 1, NULL); | |
990 | } | |
991 | } | |
992 | ||
993 | /* | |
994 | * Scanning the task stacks may introduce false negatives and it is | |
995 | * not enabled by default. | |
996 | */ | |
997 | if (kmemleak_stack_scan) { | |
998 | read_lock(&tasklist_lock); | |
999 | for_each_process(task) | |
1000 | scan_block(task_stack_page(task), | |
1001 | task_stack_page(task) + THREAD_SIZE, NULL); | |
1002 | read_unlock(&tasklist_lock); | |
1003 | } | |
1004 | ||
1005 | /* | |
1006 | * Scan the objects already referenced from the sections scanned | |
1007 | * above. More objects will be referenced and, if there are no memory | |
1008 | * leaks, all the objects will be scanned. The list traversal is safe | |
1009 | * for both tail additions and removals from inside the loop. The | |
1010 | * kmemleak objects cannot be freed from outside the loop because their | |
1011 | * use_count was increased. | |
1012 | */ | |
1013 | object = list_entry(gray_list.next, typeof(*object), gray_list); | |
1014 | while (&object->gray_list != &gray_list) { | |
1015 | scan_yield(); | |
1016 | ||
1017 | /* may add new objects to the list */ | |
1018 | if (!scan_should_stop()) | |
1019 | scan_object(object); | |
1020 | ||
1021 | tmp = list_entry(object->gray_list.next, typeof(*object), | |
1022 | gray_list); | |
1023 | ||
1024 | /* remove the object from the list and release it */ | |
1025 | list_del(&object->gray_list); | |
1026 | put_object(object); | |
1027 | ||
1028 | object = tmp; | |
1029 | } | |
1030 | WARN_ON(!list_empty(&gray_list)); | |
4698c1f2 CM |
1031 | |
1032 | /* | |
1033 | * Scanning result reporting. | |
1034 | */ | |
1035 | rcu_read_lock(); | |
1036 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1037 | spin_lock_irqsave(&object->lock, flags); | |
1038 | if (unreferenced_object(object) && | |
1039 | !(object->flags & OBJECT_REPORTED)) { | |
1040 | object->flags |= OBJECT_REPORTED; | |
1041 | new_leaks++; | |
1042 | } | |
1043 | spin_unlock_irqrestore(&object->lock, flags); | |
1044 | } | |
1045 | rcu_read_unlock(); | |
1046 | ||
1047 | if (new_leaks) | |
1048 | pr_info("%d new suspected memory leaks (see " | |
1049 | "/sys/kernel/debug/kmemleak)\n", new_leaks); | |
1050 | ||
3c7b4e6b CM |
1051 | } |
1052 | ||
1053 | /* | |
1054 | * Thread function performing automatic memory scanning. Unreferenced objects | |
1055 | * at the end of a memory scan are reported but only the first time. | |
1056 | */ | |
1057 | static int kmemleak_scan_thread(void *arg) | |
1058 | { | |
1059 | static int first_run = 1; | |
1060 | ||
ae281064 | 1061 | pr_info("Automatic memory scanning thread started\n"); |
3c7b4e6b CM |
1062 | |
1063 | /* | |
1064 | * Wait before the first scan to allow the system to fully initialize. | |
1065 | */ | |
1066 | if (first_run) { | |
1067 | first_run = 0; | |
1068 | ssleep(SECS_FIRST_SCAN); | |
1069 | } | |
1070 | ||
1071 | while (!kthread_should_stop()) { | |
3c7b4e6b CM |
1072 | signed long timeout = jiffies_scan_wait; |
1073 | ||
1074 | mutex_lock(&scan_mutex); | |
3c7b4e6b | 1075 | kmemleak_scan(); |
3c7b4e6b | 1076 | mutex_unlock(&scan_mutex); |
4698c1f2 | 1077 | |
3c7b4e6b CM |
1078 | /* wait before the next scan */ |
1079 | while (timeout && !kthread_should_stop()) | |
1080 | timeout = schedule_timeout_interruptible(timeout); | |
1081 | } | |
1082 | ||
ae281064 | 1083 | pr_info("Automatic memory scanning thread ended\n"); |
3c7b4e6b CM |
1084 | |
1085 | return 0; | |
1086 | } | |
1087 | ||
1088 | /* | |
1089 | * Start the automatic memory scanning thread. This function must be called | |
4698c1f2 | 1090 | * with the scan_mutex held. |
3c7b4e6b CM |
1091 | */ |
1092 | void start_scan_thread(void) | |
1093 | { | |
1094 | if (scan_thread) | |
1095 | return; | |
1096 | scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak"); | |
1097 | if (IS_ERR(scan_thread)) { | |
ae281064 | 1098 | pr_warning("Failed to create the scan thread\n"); |
3c7b4e6b CM |
1099 | scan_thread = NULL; |
1100 | } | |
1101 | } | |
1102 | ||
1103 | /* | |
1104 | * Stop the automatic memory scanning thread. This function must be called | |
4698c1f2 | 1105 | * with the scan_mutex held. |
3c7b4e6b CM |
1106 | */ |
1107 | void stop_scan_thread(void) | |
1108 | { | |
1109 | if (scan_thread) { | |
1110 | kthread_stop(scan_thread); | |
1111 | scan_thread = NULL; | |
1112 | } | |
1113 | } | |
1114 | ||
1115 | /* | |
1116 | * Iterate over the object_list and return the first valid object at or after | |
1117 | * the required position with its use_count incremented. The function triggers | |
1118 | * a memory scanning when the pos argument points to the first position. | |
1119 | */ | |
1120 | static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos) | |
1121 | { | |
1122 | struct kmemleak_object *object; | |
1123 | loff_t n = *pos; | |
1124 | ||
4698c1f2 | 1125 | if (!n) |
3c7b4e6b | 1126 | reported_leaks = 0; |
3c7b4e6b CM |
1127 | if (reported_leaks >= REPORTS_NR) |
1128 | return NULL; | |
1129 | ||
1130 | rcu_read_lock(); | |
1131 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1132 | if (n-- > 0) | |
1133 | continue; | |
1134 | if (get_object(object)) | |
1135 | goto out; | |
1136 | } | |
1137 | object = NULL; | |
1138 | out: | |
1139 | rcu_read_unlock(); | |
1140 | return object; | |
1141 | } | |
1142 | ||
1143 | /* | |
1144 | * Return the next object in the object_list. The function decrements the | |
1145 | * use_count of the previous object and increases that of the next one. | |
1146 | */ | |
1147 | static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
1148 | { | |
1149 | struct kmemleak_object *prev_obj = v; | |
1150 | struct kmemleak_object *next_obj = NULL; | |
1151 | struct list_head *n = &prev_obj->object_list; | |
1152 | ||
1153 | ++(*pos); | |
1154 | if (reported_leaks >= REPORTS_NR) | |
1155 | goto out; | |
1156 | ||
1157 | rcu_read_lock(); | |
1158 | list_for_each_continue_rcu(n, &object_list) { | |
1159 | next_obj = list_entry(n, struct kmemleak_object, object_list); | |
1160 | if (get_object(next_obj)) | |
1161 | break; | |
1162 | } | |
1163 | rcu_read_unlock(); | |
1164 | out: | |
1165 | put_object(prev_obj); | |
1166 | return next_obj; | |
1167 | } | |
1168 | ||
1169 | /* | |
1170 | * Decrement the use_count of the last object required, if any. | |
1171 | */ | |
1172 | static void kmemleak_seq_stop(struct seq_file *seq, void *v) | |
1173 | { | |
1174 | if (v) | |
1175 | put_object(v); | |
1176 | } | |
1177 | ||
1178 | /* | |
1179 | * Print the information for an unreferenced object to the seq file. | |
1180 | */ | |
1181 | static int kmemleak_seq_show(struct seq_file *seq, void *v) | |
1182 | { | |
1183 | struct kmemleak_object *object = v; | |
1184 | unsigned long flags; | |
1185 | ||
1186 | spin_lock_irqsave(&object->lock, flags); | |
1187 | if (!unreferenced_object(object)) | |
1188 | goto out; | |
1189 | print_unreferenced(seq, object); | |
1190 | reported_leaks++; | |
1191 | out: | |
1192 | spin_unlock_irqrestore(&object->lock, flags); | |
1193 | return 0; | |
1194 | } | |
1195 | ||
1196 | static const struct seq_operations kmemleak_seq_ops = { | |
1197 | .start = kmemleak_seq_start, | |
1198 | .next = kmemleak_seq_next, | |
1199 | .stop = kmemleak_seq_stop, | |
1200 | .show = kmemleak_seq_show, | |
1201 | }; | |
1202 | ||
1203 | static int kmemleak_open(struct inode *inode, struct file *file) | |
1204 | { | |
1205 | int ret = 0; | |
1206 | ||
1207 | if (!atomic_read(&kmemleak_enabled)) | |
1208 | return -EBUSY; | |
1209 | ||
4698c1f2 | 1210 | ret = mutex_lock_interruptible(&scan_mutex); |
3c7b4e6b CM |
1211 | if (ret < 0) |
1212 | goto out; | |
1213 | if (file->f_mode & FMODE_READ) { | |
3c7b4e6b CM |
1214 | ret = seq_open(file, &kmemleak_seq_ops); |
1215 | if (ret < 0) | |
1216 | goto scan_unlock; | |
1217 | } | |
1218 | return ret; | |
1219 | ||
1220 | scan_unlock: | |
1221 | mutex_unlock(&scan_mutex); | |
3c7b4e6b CM |
1222 | out: |
1223 | return ret; | |
1224 | } | |
1225 | ||
1226 | static int kmemleak_release(struct inode *inode, struct file *file) | |
1227 | { | |
1228 | int ret = 0; | |
1229 | ||
4698c1f2 | 1230 | if (file->f_mode & FMODE_READ) |
3c7b4e6b | 1231 | seq_release(inode, file); |
4698c1f2 | 1232 | mutex_unlock(&scan_mutex); |
3c7b4e6b CM |
1233 | |
1234 | return ret; | |
1235 | } | |
1236 | ||
1237 | /* | |
1238 | * File write operation to configure kmemleak at run-time. The following | |
1239 | * commands can be written to the /sys/kernel/debug/kmemleak file: | |
1240 | * off - disable kmemleak (irreversible) | |
1241 | * stack=on - enable the task stacks scanning | |
1242 | * stack=off - disable the tasks stacks scanning | |
1243 | * scan=on - start the automatic memory scanning thread | |
1244 | * scan=off - stop the automatic memory scanning thread | |
1245 | * scan=... - set the automatic memory scanning period in seconds (0 to | |
1246 | * disable it) | |
4698c1f2 | 1247 | * scan - trigger a memory scan |
3c7b4e6b CM |
1248 | */ |
1249 | static ssize_t kmemleak_write(struct file *file, const char __user *user_buf, | |
1250 | size_t size, loff_t *ppos) | |
1251 | { | |
1252 | char buf[64]; | |
1253 | int buf_size; | |
1254 | ||
1255 | if (!atomic_read(&kmemleak_enabled)) | |
1256 | return -EBUSY; | |
1257 | ||
1258 | buf_size = min(size, (sizeof(buf) - 1)); | |
1259 | if (strncpy_from_user(buf, user_buf, buf_size) < 0) | |
1260 | return -EFAULT; | |
1261 | buf[buf_size] = 0; | |
1262 | ||
1263 | if (strncmp(buf, "off", 3) == 0) | |
1264 | kmemleak_disable(); | |
1265 | else if (strncmp(buf, "stack=on", 8) == 0) | |
1266 | kmemleak_stack_scan = 1; | |
1267 | else if (strncmp(buf, "stack=off", 9) == 0) | |
1268 | kmemleak_stack_scan = 0; | |
1269 | else if (strncmp(buf, "scan=on", 7) == 0) | |
1270 | start_scan_thread(); | |
1271 | else if (strncmp(buf, "scan=off", 8) == 0) | |
1272 | stop_scan_thread(); | |
1273 | else if (strncmp(buf, "scan=", 5) == 0) { | |
1274 | unsigned long secs; | |
1275 | int err; | |
1276 | ||
1277 | err = strict_strtoul(buf + 5, 0, &secs); | |
1278 | if (err < 0) | |
1279 | return err; | |
1280 | stop_scan_thread(); | |
1281 | if (secs) { | |
1282 | jiffies_scan_wait = msecs_to_jiffies(secs * 1000); | |
1283 | start_scan_thread(); | |
1284 | } | |
4698c1f2 CM |
1285 | } else if (strncmp(buf, "scan", 4) == 0) |
1286 | kmemleak_scan(); | |
1287 | else | |
3c7b4e6b CM |
1288 | return -EINVAL; |
1289 | ||
1290 | /* ignore the rest of the buffer, only one command at a time */ | |
1291 | *ppos += size; | |
1292 | return size; | |
1293 | } | |
1294 | ||
1295 | static const struct file_operations kmemleak_fops = { | |
1296 | .owner = THIS_MODULE, | |
1297 | .open = kmemleak_open, | |
1298 | .read = seq_read, | |
1299 | .write = kmemleak_write, | |
1300 | .llseek = seq_lseek, | |
1301 | .release = kmemleak_release, | |
1302 | }; | |
1303 | ||
1304 | /* | |
1305 | * Perform the freeing of the kmemleak internal objects after waiting for any | |
1306 | * current memory scan to complete. | |
1307 | */ | |
1308 | static int kmemleak_cleanup_thread(void *arg) | |
1309 | { | |
1310 | struct kmemleak_object *object; | |
1311 | ||
4698c1f2 | 1312 | mutex_lock(&scan_mutex); |
3c7b4e6b | 1313 | stop_scan_thread(); |
3c7b4e6b | 1314 | |
3c7b4e6b CM |
1315 | rcu_read_lock(); |
1316 | list_for_each_entry_rcu(object, &object_list, object_list) | |
1317 | delete_object(object->pointer); | |
1318 | rcu_read_unlock(); | |
1319 | mutex_unlock(&scan_mutex); | |
1320 | ||
1321 | return 0; | |
1322 | } | |
1323 | ||
1324 | /* | |
1325 | * Start the clean-up thread. | |
1326 | */ | |
1327 | static void kmemleak_cleanup(void) | |
1328 | { | |
1329 | struct task_struct *cleanup_thread; | |
1330 | ||
1331 | cleanup_thread = kthread_run(kmemleak_cleanup_thread, NULL, | |
1332 | "kmemleak-clean"); | |
1333 | if (IS_ERR(cleanup_thread)) | |
ae281064 | 1334 | pr_warning("Failed to create the clean-up thread\n"); |
3c7b4e6b CM |
1335 | } |
1336 | ||
1337 | /* | |
1338 | * Disable kmemleak. No memory allocation/freeing will be traced once this | |
1339 | * function is called. Disabling kmemleak is an irreversible operation. | |
1340 | */ | |
1341 | static void kmemleak_disable(void) | |
1342 | { | |
1343 | /* atomically check whether it was already invoked */ | |
1344 | if (atomic_cmpxchg(&kmemleak_error, 0, 1)) | |
1345 | return; | |
1346 | ||
1347 | /* stop any memory operation tracing */ | |
1348 | atomic_set(&kmemleak_early_log, 0); | |
1349 | atomic_set(&kmemleak_enabled, 0); | |
1350 | ||
1351 | /* check whether it is too early for a kernel thread */ | |
1352 | if (atomic_read(&kmemleak_initialized)) | |
1353 | kmemleak_cleanup(); | |
1354 | ||
1355 | pr_info("Kernel memory leak detector disabled\n"); | |
1356 | } | |
1357 | ||
1358 | /* | |
1359 | * Allow boot-time kmemleak disabling (enabled by default). | |
1360 | */ | |
1361 | static int kmemleak_boot_config(char *str) | |
1362 | { | |
1363 | if (!str) | |
1364 | return -EINVAL; | |
1365 | if (strcmp(str, "off") == 0) | |
1366 | kmemleak_disable(); | |
1367 | else if (strcmp(str, "on") != 0) | |
1368 | return -EINVAL; | |
1369 | return 0; | |
1370 | } | |
1371 | early_param("kmemleak", kmemleak_boot_config); | |
1372 | ||
1373 | /* | |
2030117d | 1374 | * Kmemleak initialization. |
3c7b4e6b CM |
1375 | */ |
1376 | void __init kmemleak_init(void) | |
1377 | { | |
1378 | int i; | |
1379 | unsigned long flags; | |
1380 | ||
1381 | jiffies_scan_yield = msecs_to_jiffies(MSECS_SCAN_YIELD); | |
1382 | jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE); | |
1383 | jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000); | |
1384 | ||
1385 | object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE); | |
1386 | scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE); | |
1387 | INIT_PRIO_TREE_ROOT(&object_tree_root); | |
1388 | ||
1389 | /* the kernel is still in UP mode, so disabling the IRQs is enough */ | |
1390 | local_irq_save(flags); | |
1391 | if (!atomic_read(&kmemleak_error)) { | |
1392 | atomic_set(&kmemleak_enabled, 1); | |
1393 | atomic_set(&kmemleak_early_log, 0); | |
1394 | } | |
1395 | local_irq_restore(flags); | |
1396 | ||
1397 | /* | |
1398 | * This is the point where tracking allocations is safe. Automatic | |
1399 | * scanning is started during the late initcall. Add the early logged | |
1400 | * callbacks to the kmemleak infrastructure. | |
1401 | */ | |
1402 | for (i = 0; i < crt_early_log; i++) { | |
1403 | struct early_log *log = &early_log[i]; | |
1404 | ||
1405 | switch (log->op_type) { | |
1406 | case KMEMLEAK_ALLOC: | |
1407 | kmemleak_alloc(log->ptr, log->size, log->min_count, | |
1408 | GFP_KERNEL); | |
1409 | break; | |
1410 | case KMEMLEAK_FREE: | |
1411 | kmemleak_free(log->ptr); | |
1412 | break; | |
1413 | case KMEMLEAK_NOT_LEAK: | |
1414 | kmemleak_not_leak(log->ptr); | |
1415 | break; | |
1416 | case KMEMLEAK_IGNORE: | |
1417 | kmemleak_ignore(log->ptr); | |
1418 | break; | |
1419 | case KMEMLEAK_SCAN_AREA: | |
1420 | kmemleak_scan_area(log->ptr, log->offset, log->length, | |
1421 | GFP_KERNEL); | |
1422 | break; | |
1423 | case KMEMLEAK_NO_SCAN: | |
1424 | kmemleak_no_scan(log->ptr); | |
1425 | break; | |
1426 | default: | |
1427 | WARN_ON(1); | |
1428 | } | |
1429 | } | |
1430 | } | |
1431 | ||
1432 | /* | |
1433 | * Late initialization function. | |
1434 | */ | |
1435 | static int __init kmemleak_late_init(void) | |
1436 | { | |
1437 | struct dentry *dentry; | |
1438 | ||
1439 | atomic_set(&kmemleak_initialized, 1); | |
1440 | ||
1441 | if (atomic_read(&kmemleak_error)) { | |
1442 | /* | |
1443 | * Some error occured and kmemleak was disabled. There is a | |
1444 | * small chance that kmemleak_disable() was called immediately | |
1445 | * after setting kmemleak_initialized and we may end up with | |
1446 | * two clean-up threads but serialized by scan_mutex. | |
1447 | */ | |
1448 | kmemleak_cleanup(); | |
1449 | return -ENOMEM; | |
1450 | } | |
1451 | ||
1452 | dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL, | |
1453 | &kmemleak_fops); | |
1454 | if (!dentry) | |
ae281064 | 1455 | pr_warning("Failed to create the debugfs kmemleak file\n"); |
4698c1f2 | 1456 | mutex_lock(&scan_mutex); |
3c7b4e6b | 1457 | start_scan_thread(); |
4698c1f2 | 1458 | mutex_unlock(&scan_mutex); |
3c7b4e6b CM |
1459 | |
1460 | pr_info("Kernel memory leak detector initialized\n"); | |
1461 | ||
1462 | return 0; | |
1463 | } | |
1464 | late_initcall(kmemleak_late_init); |