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> | |
179a8100 | 95 | #include <linux/workqueue.h> |
3c7b4e6b CM |
96 | |
97 | #include <asm/sections.h> | |
98 | #include <asm/processor.h> | |
99 | #include <asm/atomic.h> | |
100 | ||
8e019366 | 101 | #include <linux/kmemcheck.h> |
3c7b4e6b CM |
102 | #include <linux/kmemleak.h> |
103 | ||
104 | /* | |
105 | * Kmemleak configuration and common defines. | |
106 | */ | |
107 | #define MAX_TRACE 16 /* stack trace length */ | |
3c7b4e6b | 108 | #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */ |
3c7b4e6b CM |
109 | #define SECS_FIRST_SCAN 60 /* delay before the first scan */ |
110 | #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */ | |
2587362e | 111 | #define GRAY_LIST_PASSES 25 /* maximum number of gray list scans */ |
af98603d | 112 | #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */ |
3c7b4e6b CM |
113 | |
114 | #define BYTES_PER_POINTER sizeof(void *) | |
115 | ||
216c04b0 CM |
116 | /* GFP bitmask for kmemleak internal allocations */ |
117 | #define GFP_KMEMLEAK_MASK (GFP_KERNEL | GFP_ATOMIC) | |
118 | ||
3c7b4e6b CM |
119 | /* scanning area inside a memory block */ |
120 | struct kmemleak_scan_area { | |
121 | struct hlist_node node; | |
122 | unsigned long offset; | |
123 | size_t length; | |
124 | }; | |
125 | ||
a1084c87 LR |
126 | #define KMEMLEAK_GREY 0 |
127 | #define KMEMLEAK_BLACK -1 | |
128 | ||
3c7b4e6b CM |
129 | /* |
130 | * Structure holding the metadata for each allocated memory block. | |
131 | * Modifications to such objects should be made while holding the | |
132 | * object->lock. Insertions or deletions from object_list, gray_list or | |
133 | * tree_node are already protected by the corresponding locks or mutex (see | |
134 | * the notes on locking above). These objects are reference-counted | |
135 | * (use_count) and freed using the RCU mechanism. | |
136 | */ | |
137 | struct kmemleak_object { | |
138 | spinlock_t lock; | |
139 | unsigned long flags; /* object status flags */ | |
140 | struct list_head object_list; | |
141 | struct list_head gray_list; | |
142 | struct prio_tree_node tree_node; | |
143 | struct rcu_head rcu; /* object_list lockless traversal */ | |
144 | /* object usage count; object freed when use_count == 0 */ | |
145 | atomic_t use_count; | |
146 | unsigned long pointer; | |
147 | size_t size; | |
148 | /* minimum number of a pointers found before it is considered leak */ | |
149 | int min_count; | |
150 | /* the total number of pointers found pointing to this object */ | |
151 | int count; | |
152 | /* memory ranges to be scanned inside an object (empty for all) */ | |
153 | struct hlist_head area_list; | |
154 | unsigned long trace[MAX_TRACE]; | |
155 | unsigned int trace_len; | |
156 | unsigned long jiffies; /* creation timestamp */ | |
157 | pid_t pid; /* pid of the current task */ | |
158 | char comm[TASK_COMM_LEN]; /* executable name */ | |
159 | }; | |
160 | ||
161 | /* flag representing the memory block allocation status */ | |
162 | #define OBJECT_ALLOCATED (1 << 0) | |
163 | /* flag set after the first reporting of an unreference object */ | |
164 | #define OBJECT_REPORTED (1 << 1) | |
165 | /* flag set to not scan the object */ | |
166 | #define OBJECT_NO_SCAN (1 << 2) | |
2587362e CM |
167 | /* flag set on newly allocated objects */ |
168 | #define OBJECT_NEW (1 << 3) | |
3c7b4e6b | 169 | |
0494e082 SS |
170 | /* number of bytes to print per line; must be 16 or 32 */ |
171 | #define HEX_ROW_SIZE 16 | |
172 | /* number of bytes to print at a time (1, 2, 4, 8) */ | |
173 | #define HEX_GROUP_SIZE 1 | |
174 | /* include ASCII after the hex output */ | |
175 | #define HEX_ASCII 1 | |
176 | /* max number of lines to be printed */ | |
177 | #define HEX_MAX_LINES 2 | |
178 | ||
3c7b4e6b CM |
179 | /* the list of all allocated objects */ |
180 | static LIST_HEAD(object_list); | |
181 | /* the list of gray-colored objects (see color_gray comment below) */ | |
182 | static LIST_HEAD(gray_list); | |
183 | /* prio search tree for object boundaries */ | |
184 | static struct prio_tree_root object_tree_root; | |
185 | /* rw_lock protecting the access to object_list and prio_tree_root */ | |
186 | static DEFINE_RWLOCK(kmemleak_lock); | |
187 | ||
188 | /* allocation caches for kmemleak internal data */ | |
189 | static struct kmem_cache *object_cache; | |
190 | static struct kmem_cache *scan_area_cache; | |
191 | ||
192 | /* set if tracing memory operations is enabled */ | |
193 | static atomic_t kmemleak_enabled = ATOMIC_INIT(0); | |
194 | /* set in the late_initcall if there were no errors */ | |
195 | static atomic_t kmemleak_initialized = ATOMIC_INIT(0); | |
196 | /* enables or disables early logging of the memory operations */ | |
197 | static atomic_t kmemleak_early_log = ATOMIC_INIT(1); | |
198 | /* set if a fata kmemleak error has occurred */ | |
199 | static atomic_t kmemleak_error = ATOMIC_INIT(0); | |
200 | ||
201 | /* minimum and maximum address that may be valid pointers */ | |
202 | static unsigned long min_addr = ULONG_MAX; | |
203 | static unsigned long max_addr; | |
204 | ||
3c7b4e6b | 205 | static struct task_struct *scan_thread; |
acf4968e | 206 | /* used to avoid reporting of recently allocated objects */ |
3c7b4e6b | 207 | static unsigned long jiffies_min_age; |
acf4968e | 208 | static unsigned long jiffies_last_scan; |
3c7b4e6b CM |
209 | /* delay between automatic memory scannings */ |
210 | static signed long jiffies_scan_wait; | |
211 | /* enables or disables the task stacks scanning */ | |
e0a2a160 | 212 | static int kmemleak_stack_scan = 1; |
4698c1f2 | 213 | /* protects the memory scanning, parameters and debug/kmemleak file access */ |
3c7b4e6b | 214 | static DEFINE_MUTEX(scan_mutex); |
3c7b4e6b | 215 | |
3c7b4e6b | 216 | /* |
2030117d | 217 | * Early object allocation/freeing logging. Kmemleak is initialized after the |
3c7b4e6b | 218 | * kernel allocator. However, both the kernel allocator and kmemleak may |
2030117d | 219 | * allocate memory blocks which need to be tracked. Kmemleak defines an |
3c7b4e6b CM |
220 | * arbitrary buffer to hold the allocation/freeing information before it is |
221 | * fully initialized. | |
222 | */ | |
223 | ||
224 | /* kmemleak operation type for early logging */ | |
225 | enum { | |
226 | KMEMLEAK_ALLOC, | |
227 | KMEMLEAK_FREE, | |
53238a60 | 228 | KMEMLEAK_FREE_PART, |
3c7b4e6b CM |
229 | KMEMLEAK_NOT_LEAK, |
230 | KMEMLEAK_IGNORE, | |
231 | KMEMLEAK_SCAN_AREA, | |
232 | KMEMLEAK_NO_SCAN | |
233 | }; | |
234 | ||
235 | /* | |
236 | * Structure holding the information passed to kmemleak callbacks during the | |
237 | * early logging. | |
238 | */ | |
239 | struct early_log { | |
240 | int op_type; /* kmemleak operation type */ | |
241 | const void *ptr; /* allocated/freed memory block */ | |
242 | size_t size; /* memory block size */ | |
243 | int min_count; /* minimum reference count */ | |
244 | unsigned long offset; /* scan area offset */ | |
245 | size_t length; /* scan area length */ | |
fd678967 CM |
246 | unsigned long trace[MAX_TRACE]; /* stack trace */ |
247 | unsigned int trace_len; /* stack trace length */ | |
3c7b4e6b CM |
248 | }; |
249 | ||
250 | /* early logging buffer and current position */ | |
a6186d89 CM |
251 | static struct early_log |
252 | early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata; | |
253 | static int crt_early_log __initdata; | |
3c7b4e6b CM |
254 | |
255 | static void kmemleak_disable(void); | |
256 | ||
257 | /* | |
258 | * Print a warning and dump the stack trace. | |
259 | */ | |
260 | #define kmemleak_warn(x...) do { \ | |
261 | pr_warning(x); \ | |
262 | dump_stack(); \ | |
263 | } while (0) | |
264 | ||
265 | /* | |
266 | * Macro invoked when a serious kmemleak condition occured and cannot be | |
2030117d | 267 | * recovered from. Kmemleak will be disabled and further allocation/freeing |
3c7b4e6b CM |
268 | * tracing no longer available. |
269 | */ | |
000814f4 | 270 | #define kmemleak_stop(x...) do { \ |
3c7b4e6b CM |
271 | kmemleak_warn(x); \ |
272 | kmemleak_disable(); \ | |
273 | } while (0) | |
274 | ||
0494e082 SS |
275 | /* |
276 | * Printing of the objects hex dump to the seq file. The number of lines to be | |
277 | * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The | |
278 | * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called | |
279 | * with the object->lock held. | |
280 | */ | |
281 | static void hex_dump_object(struct seq_file *seq, | |
282 | struct kmemleak_object *object) | |
283 | { | |
284 | const u8 *ptr = (const u8 *)object->pointer; | |
285 | int i, len, remaining; | |
286 | unsigned char linebuf[HEX_ROW_SIZE * 5]; | |
287 | ||
288 | /* limit the number of lines to HEX_MAX_LINES */ | |
289 | remaining = len = | |
290 | min(object->size, (size_t)(HEX_MAX_LINES * HEX_ROW_SIZE)); | |
291 | ||
292 | seq_printf(seq, " hex dump (first %d bytes):\n", len); | |
293 | for (i = 0; i < len; i += HEX_ROW_SIZE) { | |
294 | int linelen = min(remaining, HEX_ROW_SIZE); | |
295 | ||
296 | remaining -= HEX_ROW_SIZE; | |
297 | hex_dump_to_buffer(ptr + i, linelen, HEX_ROW_SIZE, | |
298 | HEX_GROUP_SIZE, linebuf, sizeof(linebuf), | |
299 | HEX_ASCII); | |
300 | seq_printf(seq, " %s\n", linebuf); | |
301 | } | |
302 | } | |
303 | ||
3c7b4e6b CM |
304 | /* |
305 | * Object colors, encoded with count and min_count: | |
306 | * - white - orphan object, not enough references to it (count < min_count) | |
307 | * - gray - not orphan, not marked as false positive (min_count == 0) or | |
308 | * sufficient references to it (count >= min_count) | |
309 | * - black - ignore, it doesn't contain references (e.g. text section) | |
310 | * (min_count == -1). No function defined for this color. | |
311 | * Newly created objects don't have any color assigned (object->count == -1) | |
312 | * before the next memory scan when they become white. | |
313 | */ | |
4a558dd6 | 314 | static bool color_white(const struct kmemleak_object *object) |
3c7b4e6b | 315 | { |
a1084c87 LR |
316 | return object->count != KMEMLEAK_BLACK && |
317 | object->count < object->min_count; | |
3c7b4e6b CM |
318 | } |
319 | ||
4a558dd6 | 320 | static bool color_gray(const struct kmemleak_object *object) |
3c7b4e6b | 321 | { |
a1084c87 LR |
322 | return object->min_count != KMEMLEAK_BLACK && |
323 | object->count >= object->min_count; | |
3c7b4e6b CM |
324 | } |
325 | ||
4a558dd6 | 326 | static bool color_black(const struct kmemleak_object *object) |
2587362e | 327 | { |
a1084c87 | 328 | return object->min_count == KMEMLEAK_BLACK; |
2587362e CM |
329 | } |
330 | ||
3c7b4e6b CM |
331 | /* |
332 | * Objects are considered unreferenced only if their color is white, they have | |
333 | * not be deleted and have a minimum age to avoid false positives caused by | |
334 | * pointers temporarily stored in CPU registers. | |
335 | */ | |
4a558dd6 | 336 | static bool unreferenced_object(struct kmemleak_object *object) |
3c7b4e6b CM |
337 | { |
338 | return (object->flags & OBJECT_ALLOCATED) && color_white(object) && | |
acf4968e CM |
339 | time_before_eq(object->jiffies + jiffies_min_age, |
340 | jiffies_last_scan); | |
3c7b4e6b CM |
341 | } |
342 | ||
343 | /* | |
bab4a34a CM |
344 | * Printing of the unreferenced objects information to the seq file. The |
345 | * print_unreferenced function must be called with the object->lock held. | |
3c7b4e6b | 346 | */ |
3c7b4e6b CM |
347 | static void print_unreferenced(struct seq_file *seq, |
348 | struct kmemleak_object *object) | |
349 | { | |
350 | int i; | |
351 | ||
bab4a34a CM |
352 | seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n", |
353 | object->pointer, object->size); | |
354 | seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu\n", | |
355 | object->comm, object->pid, object->jiffies); | |
0494e082 | 356 | hex_dump_object(seq, object); |
bab4a34a | 357 | seq_printf(seq, " backtrace:\n"); |
3c7b4e6b CM |
358 | |
359 | for (i = 0; i < object->trace_len; i++) { | |
360 | void *ptr = (void *)object->trace[i]; | |
bab4a34a | 361 | seq_printf(seq, " [<%p>] %pS\n", ptr, ptr); |
3c7b4e6b CM |
362 | } |
363 | } | |
364 | ||
365 | /* | |
366 | * Print the kmemleak_object information. This function is used mainly for | |
367 | * debugging special cases when kmemleak operations. It must be called with | |
368 | * the object->lock held. | |
369 | */ | |
370 | static void dump_object_info(struct kmemleak_object *object) | |
371 | { | |
372 | struct stack_trace trace; | |
373 | ||
374 | trace.nr_entries = object->trace_len; | |
375 | trace.entries = object->trace; | |
376 | ||
ae281064 | 377 | pr_notice("Object 0x%08lx (size %zu):\n", |
3c7b4e6b CM |
378 | object->tree_node.start, object->size); |
379 | pr_notice(" comm \"%s\", pid %d, jiffies %lu\n", | |
380 | object->comm, object->pid, object->jiffies); | |
381 | pr_notice(" min_count = %d\n", object->min_count); | |
382 | pr_notice(" count = %d\n", object->count); | |
189d84ed | 383 | pr_notice(" flags = 0x%lx\n", object->flags); |
3c7b4e6b CM |
384 | pr_notice(" backtrace:\n"); |
385 | print_stack_trace(&trace, 4); | |
386 | } | |
387 | ||
388 | /* | |
389 | * Look-up a memory block metadata (kmemleak_object) in the priority search | |
390 | * tree based on a pointer value. If alias is 0, only values pointing to the | |
391 | * beginning of the memory block are allowed. The kmemleak_lock must be held | |
392 | * when calling this function. | |
393 | */ | |
394 | static struct kmemleak_object *lookup_object(unsigned long ptr, int alias) | |
395 | { | |
396 | struct prio_tree_node *node; | |
397 | struct prio_tree_iter iter; | |
398 | struct kmemleak_object *object; | |
399 | ||
400 | prio_tree_iter_init(&iter, &object_tree_root, ptr, ptr); | |
401 | node = prio_tree_next(&iter); | |
402 | if (node) { | |
403 | object = prio_tree_entry(node, struct kmemleak_object, | |
404 | tree_node); | |
405 | if (!alias && object->pointer != ptr) { | |
ae281064 | 406 | kmemleak_warn("Found object by alias"); |
3c7b4e6b CM |
407 | object = NULL; |
408 | } | |
409 | } else | |
410 | object = NULL; | |
411 | ||
412 | return object; | |
413 | } | |
414 | ||
415 | /* | |
416 | * Increment the object use_count. Return 1 if successful or 0 otherwise. Note | |
417 | * that once an object's use_count reached 0, the RCU freeing was already | |
418 | * registered and the object should no longer be used. This function must be | |
419 | * called under the protection of rcu_read_lock(). | |
420 | */ | |
421 | static int get_object(struct kmemleak_object *object) | |
422 | { | |
423 | return atomic_inc_not_zero(&object->use_count); | |
424 | } | |
425 | ||
426 | /* | |
427 | * RCU callback to free a kmemleak_object. | |
428 | */ | |
429 | static void free_object_rcu(struct rcu_head *rcu) | |
430 | { | |
431 | struct hlist_node *elem, *tmp; | |
432 | struct kmemleak_scan_area *area; | |
433 | struct kmemleak_object *object = | |
434 | container_of(rcu, struct kmemleak_object, rcu); | |
435 | ||
436 | /* | |
437 | * Once use_count is 0 (guaranteed by put_object), there is no other | |
438 | * code accessing this object, hence no need for locking. | |
439 | */ | |
440 | hlist_for_each_entry_safe(area, elem, tmp, &object->area_list, node) { | |
441 | hlist_del(elem); | |
442 | kmem_cache_free(scan_area_cache, area); | |
443 | } | |
444 | kmem_cache_free(object_cache, object); | |
445 | } | |
446 | ||
447 | /* | |
448 | * Decrement the object use_count. Once the count is 0, free the object using | |
449 | * an RCU callback. Since put_object() may be called via the kmemleak_free() -> | |
450 | * delete_object() path, the delayed RCU freeing ensures that there is no | |
451 | * recursive call to the kernel allocator. Lock-less RCU object_list traversal | |
452 | * is also possible. | |
453 | */ | |
454 | static void put_object(struct kmemleak_object *object) | |
455 | { | |
456 | if (!atomic_dec_and_test(&object->use_count)) | |
457 | return; | |
458 | ||
459 | /* should only get here after delete_object was called */ | |
460 | WARN_ON(object->flags & OBJECT_ALLOCATED); | |
461 | ||
462 | call_rcu(&object->rcu, free_object_rcu); | |
463 | } | |
464 | ||
465 | /* | |
466 | * Look up an object in the prio search tree and increase its use_count. | |
467 | */ | |
468 | static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias) | |
469 | { | |
470 | unsigned long flags; | |
471 | struct kmemleak_object *object = NULL; | |
472 | ||
473 | rcu_read_lock(); | |
474 | read_lock_irqsave(&kmemleak_lock, flags); | |
475 | if (ptr >= min_addr && ptr < max_addr) | |
476 | object = lookup_object(ptr, alias); | |
477 | read_unlock_irqrestore(&kmemleak_lock, flags); | |
478 | ||
479 | /* check whether the object is still available */ | |
480 | if (object && !get_object(object)) | |
481 | object = NULL; | |
482 | rcu_read_unlock(); | |
483 | ||
484 | return object; | |
485 | } | |
486 | ||
fd678967 CM |
487 | /* |
488 | * Save stack trace to the given array of MAX_TRACE size. | |
489 | */ | |
490 | static int __save_stack_trace(unsigned long *trace) | |
491 | { | |
492 | struct stack_trace stack_trace; | |
493 | ||
494 | stack_trace.max_entries = MAX_TRACE; | |
495 | stack_trace.nr_entries = 0; | |
496 | stack_trace.entries = trace; | |
497 | stack_trace.skip = 2; | |
498 | save_stack_trace(&stack_trace); | |
499 | ||
500 | return stack_trace.nr_entries; | |
501 | } | |
502 | ||
3c7b4e6b CM |
503 | /* |
504 | * Create the metadata (struct kmemleak_object) corresponding to an allocated | |
505 | * memory block and add it to the object_list and object_tree_root. | |
506 | */ | |
fd678967 CM |
507 | static struct kmemleak_object *create_object(unsigned long ptr, size_t size, |
508 | int min_count, gfp_t gfp) | |
3c7b4e6b CM |
509 | { |
510 | unsigned long flags; | |
511 | struct kmemleak_object *object; | |
512 | struct prio_tree_node *node; | |
3c7b4e6b | 513 | |
216c04b0 | 514 | object = kmem_cache_alloc(object_cache, gfp & GFP_KMEMLEAK_MASK); |
3c7b4e6b | 515 | if (!object) { |
ae281064 | 516 | kmemleak_stop("Cannot allocate a kmemleak_object structure\n"); |
fd678967 | 517 | return NULL; |
3c7b4e6b CM |
518 | } |
519 | ||
520 | INIT_LIST_HEAD(&object->object_list); | |
521 | INIT_LIST_HEAD(&object->gray_list); | |
522 | INIT_HLIST_HEAD(&object->area_list); | |
523 | spin_lock_init(&object->lock); | |
524 | atomic_set(&object->use_count, 1); | |
2587362e | 525 | object->flags = OBJECT_ALLOCATED | OBJECT_NEW; |
3c7b4e6b CM |
526 | object->pointer = ptr; |
527 | object->size = size; | |
528 | object->min_count = min_count; | |
529 | object->count = -1; /* no color initially */ | |
530 | object->jiffies = jiffies; | |
531 | ||
532 | /* task information */ | |
533 | if (in_irq()) { | |
534 | object->pid = 0; | |
535 | strncpy(object->comm, "hardirq", sizeof(object->comm)); | |
536 | } else if (in_softirq()) { | |
537 | object->pid = 0; | |
538 | strncpy(object->comm, "softirq", sizeof(object->comm)); | |
539 | } else { | |
540 | object->pid = current->pid; | |
541 | /* | |
542 | * There is a small chance of a race with set_task_comm(), | |
543 | * however using get_task_comm() here may cause locking | |
544 | * dependency issues with current->alloc_lock. In the worst | |
545 | * case, the command line is not correct. | |
546 | */ | |
547 | strncpy(object->comm, current->comm, sizeof(object->comm)); | |
548 | } | |
549 | ||
550 | /* kernel backtrace */ | |
fd678967 | 551 | object->trace_len = __save_stack_trace(object->trace); |
3c7b4e6b CM |
552 | |
553 | INIT_PRIO_TREE_NODE(&object->tree_node); | |
554 | object->tree_node.start = ptr; | |
555 | object->tree_node.last = ptr + size - 1; | |
556 | ||
557 | write_lock_irqsave(&kmemleak_lock, flags); | |
0580a181 | 558 | |
3c7b4e6b CM |
559 | min_addr = min(min_addr, ptr); |
560 | max_addr = max(max_addr, ptr + size); | |
561 | node = prio_tree_insert(&object_tree_root, &object->tree_node); | |
562 | /* | |
563 | * The code calling the kernel does not yet have the pointer to the | |
564 | * memory block to be able to free it. However, we still hold the | |
565 | * kmemleak_lock here in case parts of the kernel started freeing | |
566 | * random memory blocks. | |
567 | */ | |
568 | if (node != &object->tree_node) { | |
ae281064 JP |
569 | kmemleak_stop("Cannot insert 0x%lx into the object search tree " |
570 | "(already existing)\n", ptr); | |
3c7b4e6b | 571 | object = lookup_object(ptr, 1); |
0580a181 | 572 | spin_lock(&object->lock); |
3c7b4e6b | 573 | dump_object_info(object); |
0580a181 | 574 | spin_unlock(&object->lock); |
3c7b4e6b CM |
575 | |
576 | goto out; | |
577 | } | |
578 | list_add_tail_rcu(&object->object_list, &object_list); | |
579 | out: | |
580 | write_unlock_irqrestore(&kmemleak_lock, flags); | |
fd678967 | 581 | return object; |
3c7b4e6b CM |
582 | } |
583 | ||
584 | /* | |
585 | * Remove the metadata (struct kmemleak_object) for a memory block from the | |
586 | * object_list and object_tree_root and decrement its use_count. | |
587 | */ | |
53238a60 | 588 | static void __delete_object(struct kmemleak_object *object) |
3c7b4e6b CM |
589 | { |
590 | unsigned long flags; | |
3c7b4e6b CM |
591 | |
592 | write_lock_irqsave(&kmemleak_lock, flags); | |
3c7b4e6b CM |
593 | prio_tree_remove(&object_tree_root, &object->tree_node); |
594 | list_del_rcu(&object->object_list); | |
595 | write_unlock_irqrestore(&kmemleak_lock, flags); | |
596 | ||
597 | WARN_ON(!(object->flags & OBJECT_ALLOCATED)); | |
53238a60 | 598 | WARN_ON(atomic_read(&object->use_count) < 2); |
3c7b4e6b CM |
599 | |
600 | /* | |
601 | * Locking here also ensures that the corresponding memory block | |
602 | * cannot be freed when it is being scanned. | |
603 | */ | |
604 | spin_lock_irqsave(&object->lock, flags); | |
3c7b4e6b CM |
605 | object->flags &= ~OBJECT_ALLOCATED; |
606 | spin_unlock_irqrestore(&object->lock, flags); | |
607 | put_object(object); | |
608 | } | |
609 | ||
53238a60 CM |
610 | /* |
611 | * Look up the metadata (struct kmemleak_object) corresponding to ptr and | |
612 | * delete it. | |
613 | */ | |
614 | static void delete_object_full(unsigned long ptr) | |
615 | { | |
616 | struct kmemleak_object *object; | |
617 | ||
618 | object = find_and_get_object(ptr, 0); | |
619 | if (!object) { | |
620 | #ifdef DEBUG | |
621 | kmemleak_warn("Freeing unknown object at 0x%08lx\n", | |
622 | ptr); | |
623 | #endif | |
624 | return; | |
625 | } | |
626 | __delete_object(object); | |
627 | put_object(object); | |
628 | } | |
629 | ||
630 | /* | |
631 | * Look up the metadata (struct kmemleak_object) corresponding to ptr and | |
632 | * delete it. If the memory block is partially freed, the function may create | |
633 | * additional metadata for the remaining parts of the block. | |
634 | */ | |
635 | static void delete_object_part(unsigned long ptr, size_t size) | |
636 | { | |
637 | struct kmemleak_object *object; | |
638 | unsigned long start, end; | |
639 | ||
640 | object = find_and_get_object(ptr, 1); | |
641 | if (!object) { | |
642 | #ifdef DEBUG | |
643 | kmemleak_warn("Partially freeing unknown object at 0x%08lx " | |
644 | "(size %zu)\n", ptr, size); | |
645 | #endif | |
646 | return; | |
647 | } | |
648 | __delete_object(object); | |
649 | ||
650 | /* | |
651 | * Create one or two objects that may result from the memory block | |
652 | * split. Note that partial freeing is only done by free_bootmem() and | |
653 | * this happens before kmemleak_init() is called. The path below is | |
654 | * only executed during early log recording in kmemleak_init(), so | |
655 | * GFP_KERNEL is enough. | |
656 | */ | |
657 | start = object->pointer; | |
658 | end = object->pointer + object->size; | |
659 | if (ptr > start) | |
660 | create_object(start, ptr - start, object->min_count, | |
661 | GFP_KERNEL); | |
662 | if (ptr + size < end) | |
663 | create_object(ptr + size, end - ptr - size, object->min_count, | |
664 | GFP_KERNEL); | |
665 | ||
666 | put_object(object); | |
667 | } | |
a1084c87 LR |
668 | |
669 | static void __paint_it(struct kmemleak_object *object, int color) | |
670 | { | |
671 | object->min_count = color; | |
672 | if (color == KMEMLEAK_BLACK) | |
673 | object->flags |= OBJECT_NO_SCAN; | |
674 | } | |
675 | ||
676 | static void paint_it(struct kmemleak_object *object, int color) | |
3c7b4e6b CM |
677 | { |
678 | unsigned long flags; | |
a1084c87 LR |
679 | |
680 | spin_lock_irqsave(&object->lock, flags); | |
681 | __paint_it(object, color); | |
682 | spin_unlock_irqrestore(&object->lock, flags); | |
683 | } | |
684 | ||
685 | static void paint_ptr(unsigned long ptr, int color) | |
686 | { | |
3c7b4e6b CM |
687 | struct kmemleak_object *object; |
688 | ||
689 | object = find_and_get_object(ptr, 0); | |
690 | if (!object) { | |
a1084c87 LR |
691 | kmemleak_warn("Trying to color unknown object " |
692 | "at 0x%08lx as %s\n", ptr, | |
693 | (color == KMEMLEAK_GREY) ? "Grey" : | |
694 | (color == KMEMLEAK_BLACK) ? "Black" : "Unknown"); | |
3c7b4e6b CM |
695 | return; |
696 | } | |
a1084c87 | 697 | paint_it(object, color); |
3c7b4e6b CM |
698 | put_object(object); |
699 | } | |
700 | ||
a1084c87 LR |
701 | /* |
702 | * Make a object permanently as gray-colored so that it can no longer be | |
703 | * reported as a leak. This is used in general to mark a false positive. | |
704 | */ | |
705 | static void make_gray_object(unsigned long ptr) | |
706 | { | |
707 | paint_ptr(ptr, KMEMLEAK_GREY); | |
708 | } | |
709 | ||
3c7b4e6b CM |
710 | /* |
711 | * Mark the object as black-colored so that it is ignored from scans and | |
712 | * reporting. | |
713 | */ | |
714 | static void make_black_object(unsigned long ptr) | |
715 | { | |
a1084c87 | 716 | paint_ptr(ptr, KMEMLEAK_BLACK); |
3c7b4e6b CM |
717 | } |
718 | ||
719 | /* | |
720 | * Add a scanning area to the object. If at least one such area is added, | |
721 | * kmemleak will only scan these ranges rather than the whole memory block. | |
722 | */ | |
723 | static void add_scan_area(unsigned long ptr, unsigned long offset, | |
724 | size_t length, gfp_t gfp) | |
725 | { | |
726 | unsigned long flags; | |
727 | struct kmemleak_object *object; | |
728 | struct kmemleak_scan_area *area; | |
729 | ||
730 | object = find_and_get_object(ptr, 0); | |
731 | if (!object) { | |
ae281064 JP |
732 | kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n", |
733 | ptr); | |
3c7b4e6b CM |
734 | return; |
735 | } | |
736 | ||
216c04b0 | 737 | area = kmem_cache_alloc(scan_area_cache, gfp & GFP_KMEMLEAK_MASK); |
3c7b4e6b | 738 | if (!area) { |
ae281064 | 739 | kmemleak_warn("Cannot allocate a scan area\n"); |
3c7b4e6b CM |
740 | goto out; |
741 | } | |
742 | ||
743 | spin_lock_irqsave(&object->lock, flags); | |
744 | if (offset + length > object->size) { | |
ae281064 | 745 | kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr); |
3c7b4e6b CM |
746 | dump_object_info(object); |
747 | kmem_cache_free(scan_area_cache, area); | |
748 | goto out_unlock; | |
749 | } | |
750 | ||
751 | INIT_HLIST_NODE(&area->node); | |
752 | area->offset = offset; | |
753 | area->length = length; | |
754 | ||
755 | hlist_add_head(&area->node, &object->area_list); | |
756 | out_unlock: | |
757 | spin_unlock_irqrestore(&object->lock, flags); | |
758 | out: | |
759 | put_object(object); | |
760 | } | |
761 | ||
762 | /* | |
763 | * Set the OBJECT_NO_SCAN flag for the object corresponding to the give | |
764 | * pointer. Such object will not be scanned by kmemleak but references to it | |
765 | * are searched. | |
766 | */ | |
767 | static void object_no_scan(unsigned long ptr) | |
768 | { | |
769 | unsigned long flags; | |
770 | struct kmemleak_object *object; | |
771 | ||
772 | object = find_and_get_object(ptr, 0); | |
773 | if (!object) { | |
ae281064 | 774 | kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr); |
3c7b4e6b CM |
775 | return; |
776 | } | |
777 | ||
778 | spin_lock_irqsave(&object->lock, flags); | |
779 | object->flags |= OBJECT_NO_SCAN; | |
780 | spin_unlock_irqrestore(&object->lock, flags); | |
781 | put_object(object); | |
782 | } | |
783 | ||
784 | /* | |
785 | * Log an early kmemleak_* call to the early_log buffer. These calls will be | |
786 | * processed later once kmemleak is fully initialized. | |
787 | */ | |
a6186d89 CM |
788 | static void __init log_early(int op_type, const void *ptr, size_t size, |
789 | int min_count, unsigned long offset, size_t length) | |
3c7b4e6b CM |
790 | { |
791 | unsigned long flags; | |
792 | struct early_log *log; | |
793 | ||
794 | if (crt_early_log >= ARRAY_SIZE(early_log)) { | |
addd72c1 CM |
795 | pr_warning("Early log buffer exceeded, " |
796 | "please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n"); | |
a9d9058a | 797 | kmemleak_disable(); |
3c7b4e6b CM |
798 | return; |
799 | } | |
800 | ||
801 | /* | |
802 | * There is no need for locking since the kernel is still in UP mode | |
803 | * at this stage. Disabling the IRQs is enough. | |
804 | */ | |
805 | local_irq_save(flags); | |
806 | log = &early_log[crt_early_log]; | |
807 | log->op_type = op_type; | |
808 | log->ptr = ptr; | |
809 | log->size = size; | |
810 | log->min_count = min_count; | |
811 | log->offset = offset; | |
812 | log->length = length; | |
fd678967 CM |
813 | if (op_type == KMEMLEAK_ALLOC) |
814 | log->trace_len = __save_stack_trace(log->trace); | |
3c7b4e6b CM |
815 | crt_early_log++; |
816 | local_irq_restore(flags); | |
817 | } | |
818 | ||
fd678967 CM |
819 | /* |
820 | * Log an early allocated block and populate the stack trace. | |
821 | */ | |
822 | static void early_alloc(struct early_log *log) | |
823 | { | |
824 | struct kmemleak_object *object; | |
825 | unsigned long flags; | |
826 | int i; | |
827 | ||
828 | if (!atomic_read(&kmemleak_enabled) || !log->ptr || IS_ERR(log->ptr)) | |
829 | return; | |
830 | ||
831 | /* | |
832 | * RCU locking needed to ensure object is not freed via put_object(). | |
833 | */ | |
834 | rcu_read_lock(); | |
835 | object = create_object((unsigned long)log->ptr, log->size, | |
c1bcd6b3 | 836 | log->min_count, GFP_ATOMIC); |
0d5d1aad CM |
837 | if (!object) |
838 | goto out; | |
fd678967 CM |
839 | spin_lock_irqsave(&object->lock, flags); |
840 | for (i = 0; i < log->trace_len; i++) | |
841 | object->trace[i] = log->trace[i]; | |
842 | object->trace_len = log->trace_len; | |
843 | spin_unlock_irqrestore(&object->lock, flags); | |
0d5d1aad | 844 | out: |
fd678967 CM |
845 | rcu_read_unlock(); |
846 | } | |
847 | ||
3c7b4e6b CM |
848 | /* |
849 | * Memory allocation function callback. This function is called from the | |
850 | * kernel allocators when a new block is allocated (kmem_cache_alloc, kmalloc, | |
851 | * vmalloc etc.). | |
852 | */ | |
a6186d89 CM |
853 | void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count, |
854 | gfp_t gfp) | |
3c7b4e6b CM |
855 | { |
856 | pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count); | |
857 | ||
858 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
859 | create_object((unsigned long)ptr, size, min_count, gfp); | |
860 | else if (atomic_read(&kmemleak_early_log)) | |
861 | log_early(KMEMLEAK_ALLOC, ptr, size, min_count, 0, 0); | |
862 | } | |
863 | EXPORT_SYMBOL_GPL(kmemleak_alloc); | |
864 | ||
865 | /* | |
866 | * Memory freeing function callback. This function is called from the kernel | |
867 | * allocators when a block is freed (kmem_cache_free, kfree, vfree etc.). | |
868 | */ | |
a6186d89 | 869 | void __ref kmemleak_free(const void *ptr) |
3c7b4e6b CM |
870 | { |
871 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
872 | ||
873 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
53238a60 | 874 | delete_object_full((unsigned long)ptr); |
3c7b4e6b CM |
875 | else if (atomic_read(&kmemleak_early_log)) |
876 | log_early(KMEMLEAK_FREE, ptr, 0, 0, 0, 0); | |
877 | } | |
878 | EXPORT_SYMBOL_GPL(kmemleak_free); | |
879 | ||
53238a60 CM |
880 | /* |
881 | * Partial memory freeing function callback. This function is usually called | |
882 | * from bootmem allocator when (part of) a memory block is freed. | |
883 | */ | |
a6186d89 | 884 | void __ref kmemleak_free_part(const void *ptr, size_t size) |
53238a60 CM |
885 | { |
886 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
887 | ||
888 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
889 | delete_object_part((unsigned long)ptr, size); | |
890 | else if (atomic_read(&kmemleak_early_log)) | |
891 | log_early(KMEMLEAK_FREE_PART, ptr, size, 0, 0, 0); | |
892 | } | |
893 | EXPORT_SYMBOL_GPL(kmemleak_free_part); | |
894 | ||
3c7b4e6b CM |
895 | /* |
896 | * Mark an already allocated memory block as a false positive. This will cause | |
897 | * the block to no longer be reported as leak and always be scanned. | |
898 | */ | |
a6186d89 | 899 | void __ref kmemleak_not_leak(const void *ptr) |
3c7b4e6b CM |
900 | { |
901 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
902 | ||
903 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
904 | make_gray_object((unsigned long)ptr); | |
905 | else if (atomic_read(&kmemleak_early_log)) | |
906 | log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0, 0, 0); | |
907 | } | |
908 | EXPORT_SYMBOL(kmemleak_not_leak); | |
909 | ||
910 | /* | |
911 | * Ignore a memory block. This is usually done when it is known that the | |
912 | * corresponding block is not a leak and does not contain any references to | |
913 | * other allocated memory blocks. | |
914 | */ | |
a6186d89 | 915 | void __ref kmemleak_ignore(const void *ptr) |
3c7b4e6b CM |
916 | { |
917 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
918 | ||
919 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
920 | make_black_object((unsigned long)ptr); | |
921 | else if (atomic_read(&kmemleak_early_log)) | |
922 | log_early(KMEMLEAK_IGNORE, ptr, 0, 0, 0, 0); | |
923 | } | |
924 | EXPORT_SYMBOL(kmemleak_ignore); | |
925 | ||
926 | /* | |
927 | * Limit the range to be scanned in an allocated memory block. | |
928 | */ | |
a6186d89 CM |
929 | void __ref kmemleak_scan_area(const void *ptr, unsigned long offset, |
930 | size_t length, gfp_t gfp) | |
3c7b4e6b CM |
931 | { |
932 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
933 | ||
934 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
935 | add_scan_area((unsigned long)ptr, offset, length, gfp); | |
936 | else if (atomic_read(&kmemleak_early_log)) | |
937 | log_early(KMEMLEAK_SCAN_AREA, ptr, 0, 0, offset, length); | |
938 | } | |
939 | EXPORT_SYMBOL(kmemleak_scan_area); | |
940 | ||
941 | /* | |
942 | * Inform kmemleak not to scan the given memory block. | |
943 | */ | |
a6186d89 | 944 | void __ref kmemleak_no_scan(const void *ptr) |
3c7b4e6b CM |
945 | { |
946 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
947 | ||
948 | if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr)) | |
949 | object_no_scan((unsigned long)ptr); | |
950 | else if (atomic_read(&kmemleak_early_log)) | |
951 | log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0, 0, 0); | |
952 | } | |
953 | EXPORT_SYMBOL(kmemleak_no_scan); | |
954 | ||
3c7b4e6b CM |
955 | /* |
956 | * Memory scanning is a long process and it needs to be interruptable. This | |
957 | * function checks whether such interrupt condition occured. | |
958 | */ | |
959 | static int scan_should_stop(void) | |
960 | { | |
961 | if (!atomic_read(&kmemleak_enabled)) | |
962 | return 1; | |
963 | ||
964 | /* | |
965 | * This function may be called from either process or kthread context, | |
966 | * hence the need to check for both stop conditions. | |
967 | */ | |
968 | if (current->mm) | |
969 | return signal_pending(current); | |
970 | else | |
971 | return kthread_should_stop(); | |
972 | ||
973 | return 0; | |
974 | } | |
975 | ||
976 | /* | |
977 | * Scan a memory block (exclusive range) for valid pointers and add those | |
978 | * found to the gray list. | |
979 | */ | |
980 | static void scan_block(void *_start, void *_end, | |
4b8a9674 | 981 | struct kmemleak_object *scanned, int allow_resched) |
3c7b4e6b CM |
982 | { |
983 | unsigned long *ptr; | |
984 | unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER); | |
985 | unsigned long *end = _end - (BYTES_PER_POINTER - 1); | |
986 | ||
987 | for (ptr = start; ptr < end; ptr++) { | |
3c7b4e6b | 988 | struct kmemleak_object *object; |
8e019366 PE |
989 | unsigned long flags; |
990 | unsigned long pointer; | |
3c7b4e6b | 991 | |
4b8a9674 CM |
992 | if (allow_resched) |
993 | cond_resched(); | |
3c7b4e6b CM |
994 | if (scan_should_stop()) |
995 | break; | |
996 | ||
8e019366 PE |
997 | /* don't scan uninitialized memory */ |
998 | if (!kmemcheck_is_obj_initialized((unsigned long)ptr, | |
999 | BYTES_PER_POINTER)) | |
1000 | continue; | |
1001 | ||
1002 | pointer = *ptr; | |
1003 | ||
3c7b4e6b CM |
1004 | object = find_and_get_object(pointer, 1); |
1005 | if (!object) | |
1006 | continue; | |
1007 | if (object == scanned) { | |
1008 | /* self referenced, ignore */ | |
1009 | put_object(object); | |
1010 | continue; | |
1011 | } | |
1012 | ||
1013 | /* | |
1014 | * Avoid the lockdep recursive warning on object->lock being | |
1015 | * previously acquired in scan_object(). These locks are | |
1016 | * enclosed by scan_mutex. | |
1017 | */ | |
1018 | spin_lock_irqsave_nested(&object->lock, flags, | |
1019 | SINGLE_DEPTH_NESTING); | |
1020 | if (!color_white(object)) { | |
1021 | /* non-orphan, ignored or new */ | |
1022 | spin_unlock_irqrestore(&object->lock, flags); | |
1023 | put_object(object); | |
1024 | continue; | |
1025 | } | |
1026 | ||
1027 | /* | |
1028 | * Increase the object's reference count (number of pointers | |
1029 | * to the memory block). If this count reaches the required | |
1030 | * minimum, the object's color will become gray and it will be | |
1031 | * added to the gray_list. | |
1032 | */ | |
1033 | object->count++; | |
1034 | if (color_gray(object)) | |
1035 | list_add_tail(&object->gray_list, &gray_list); | |
1036 | else | |
1037 | put_object(object); | |
1038 | spin_unlock_irqrestore(&object->lock, flags); | |
1039 | } | |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * Scan a memory block corresponding to a kmemleak_object. A condition is | |
1044 | * that object->use_count >= 1. | |
1045 | */ | |
1046 | static void scan_object(struct kmemleak_object *object) | |
1047 | { | |
1048 | struct kmemleak_scan_area *area; | |
1049 | struct hlist_node *elem; | |
1050 | unsigned long flags; | |
1051 | ||
1052 | /* | |
1053 | * Once the object->lock is aquired, the corresponding memory block | |
1054 | * cannot be freed (the same lock is aquired in delete_object). | |
1055 | */ | |
1056 | spin_lock_irqsave(&object->lock, flags); | |
1057 | if (object->flags & OBJECT_NO_SCAN) | |
1058 | goto out; | |
1059 | if (!(object->flags & OBJECT_ALLOCATED)) | |
1060 | /* already freed object */ | |
1061 | goto out; | |
af98603d CM |
1062 | if (hlist_empty(&object->area_list)) { |
1063 | void *start = (void *)object->pointer; | |
1064 | void *end = (void *)(object->pointer + object->size); | |
1065 | ||
1066 | while (start < end && (object->flags & OBJECT_ALLOCATED) && | |
1067 | !(object->flags & OBJECT_NO_SCAN)) { | |
1068 | scan_block(start, min(start + MAX_SCAN_SIZE, end), | |
1069 | object, 0); | |
1070 | start += MAX_SCAN_SIZE; | |
1071 | ||
1072 | spin_unlock_irqrestore(&object->lock, flags); | |
1073 | cond_resched(); | |
1074 | spin_lock_irqsave(&object->lock, flags); | |
1075 | } | |
1076 | } else | |
3c7b4e6b CM |
1077 | hlist_for_each_entry(area, elem, &object->area_list, node) |
1078 | scan_block((void *)(object->pointer + area->offset), | |
1079 | (void *)(object->pointer + area->offset | |
4b8a9674 | 1080 | + area->length), object, 0); |
3c7b4e6b CM |
1081 | out: |
1082 | spin_unlock_irqrestore(&object->lock, flags); | |
1083 | } | |
1084 | ||
1085 | /* | |
1086 | * Scan data sections and all the referenced memory blocks allocated via the | |
1087 | * kernel's standard allocators. This function must be called with the | |
1088 | * scan_mutex held. | |
1089 | */ | |
1090 | static void kmemleak_scan(void) | |
1091 | { | |
1092 | unsigned long flags; | |
1093 | struct kmemleak_object *object, *tmp; | |
3c7b4e6b | 1094 | int i; |
4698c1f2 | 1095 | int new_leaks = 0; |
2587362e | 1096 | int gray_list_pass = 0; |
3c7b4e6b | 1097 | |
acf4968e CM |
1098 | jiffies_last_scan = jiffies; |
1099 | ||
3c7b4e6b CM |
1100 | /* prepare the kmemleak_object's */ |
1101 | rcu_read_lock(); | |
1102 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1103 | spin_lock_irqsave(&object->lock, flags); | |
1104 | #ifdef DEBUG | |
1105 | /* | |
1106 | * With a few exceptions there should be a maximum of | |
1107 | * 1 reference to any object at this point. | |
1108 | */ | |
1109 | if (atomic_read(&object->use_count) > 1) { | |
ae281064 | 1110 | pr_debug("object->use_count = %d\n", |
3c7b4e6b CM |
1111 | atomic_read(&object->use_count)); |
1112 | dump_object_info(object); | |
1113 | } | |
1114 | #endif | |
1115 | /* reset the reference count (whiten the object) */ | |
1116 | object->count = 0; | |
2587362e | 1117 | object->flags &= ~OBJECT_NEW; |
3c7b4e6b CM |
1118 | if (color_gray(object) && get_object(object)) |
1119 | list_add_tail(&object->gray_list, &gray_list); | |
1120 | ||
1121 | spin_unlock_irqrestore(&object->lock, flags); | |
1122 | } | |
1123 | rcu_read_unlock(); | |
1124 | ||
1125 | /* data/bss scanning */ | |
4b8a9674 CM |
1126 | scan_block(_sdata, _edata, NULL, 1); |
1127 | scan_block(__bss_start, __bss_stop, NULL, 1); | |
3c7b4e6b CM |
1128 | |
1129 | #ifdef CONFIG_SMP | |
1130 | /* per-cpu sections scanning */ | |
1131 | for_each_possible_cpu(i) | |
1132 | scan_block(__per_cpu_start + per_cpu_offset(i), | |
4b8a9674 | 1133 | __per_cpu_end + per_cpu_offset(i), NULL, 1); |
3c7b4e6b CM |
1134 | #endif |
1135 | ||
1136 | /* | |
1137 | * Struct page scanning for each node. The code below is not yet safe | |
1138 | * with MEMORY_HOTPLUG. | |
1139 | */ | |
1140 | for_each_online_node(i) { | |
1141 | pg_data_t *pgdat = NODE_DATA(i); | |
1142 | unsigned long start_pfn = pgdat->node_start_pfn; | |
1143 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; | |
1144 | unsigned long pfn; | |
1145 | ||
1146 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
1147 | struct page *page; | |
1148 | ||
1149 | if (!pfn_valid(pfn)) | |
1150 | continue; | |
1151 | page = pfn_to_page(pfn); | |
1152 | /* only scan if page is in use */ | |
1153 | if (page_count(page) == 0) | |
1154 | continue; | |
4b8a9674 | 1155 | scan_block(page, page + 1, NULL, 1); |
3c7b4e6b CM |
1156 | } |
1157 | } | |
1158 | ||
1159 | /* | |
43ed5d6e | 1160 | * Scanning the task stacks (may introduce false negatives). |
3c7b4e6b CM |
1161 | */ |
1162 | if (kmemleak_stack_scan) { | |
43ed5d6e CM |
1163 | struct task_struct *p, *g; |
1164 | ||
3c7b4e6b | 1165 | read_lock(&tasklist_lock); |
43ed5d6e CM |
1166 | do_each_thread(g, p) { |
1167 | scan_block(task_stack_page(p), task_stack_page(p) + | |
1168 | THREAD_SIZE, NULL, 0); | |
1169 | } while_each_thread(g, p); | |
3c7b4e6b CM |
1170 | read_unlock(&tasklist_lock); |
1171 | } | |
1172 | ||
1173 | /* | |
1174 | * Scan the objects already referenced from the sections scanned | |
1175 | * above. More objects will be referenced and, if there are no memory | |
1176 | * leaks, all the objects will be scanned. The list traversal is safe | |
1177 | * for both tail additions and removals from inside the loop. The | |
1178 | * kmemleak objects cannot be freed from outside the loop because their | |
1179 | * use_count was increased. | |
1180 | */ | |
2587362e | 1181 | repeat: |
3c7b4e6b CM |
1182 | object = list_entry(gray_list.next, typeof(*object), gray_list); |
1183 | while (&object->gray_list != &gray_list) { | |
57d81f6f | 1184 | cond_resched(); |
3c7b4e6b CM |
1185 | |
1186 | /* may add new objects to the list */ | |
1187 | if (!scan_should_stop()) | |
1188 | scan_object(object); | |
1189 | ||
1190 | tmp = list_entry(object->gray_list.next, typeof(*object), | |
1191 | gray_list); | |
1192 | ||
1193 | /* remove the object from the list and release it */ | |
1194 | list_del(&object->gray_list); | |
1195 | put_object(object); | |
1196 | ||
1197 | object = tmp; | |
1198 | } | |
2587362e CM |
1199 | |
1200 | if (scan_should_stop() || ++gray_list_pass >= GRAY_LIST_PASSES) | |
1201 | goto scan_end; | |
1202 | ||
1203 | /* | |
1204 | * Check for new objects allocated during this scanning and add them | |
1205 | * to the gray list. | |
1206 | */ | |
1207 | rcu_read_lock(); | |
1208 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1209 | spin_lock_irqsave(&object->lock, flags); | |
1210 | if ((object->flags & OBJECT_NEW) && !color_black(object) && | |
1211 | get_object(object)) { | |
1212 | object->flags &= ~OBJECT_NEW; | |
1213 | list_add_tail(&object->gray_list, &gray_list); | |
1214 | } | |
1215 | spin_unlock_irqrestore(&object->lock, flags); | |
1216 | } | |
1217 | rcu_read_unlock(); | |
1218 | ||
1219 | if (!list_empty(&gray_list)) | |
1220 | goto repeat; | |
1221 | ||
1222 | scan_end: | |
3c7b4e6b | 1223 | WARN_ON(!list_empty(&gray_list)); |
4698c1f2 | 1224 | |
17bb9e0d | 1225 | /* |
2587362e CM |
1226 | * If scanning was stopped or new objects were being allocated at a |
1227 | * higher rate than gray list scanning, do not report any new | |
1228 | * unreferenced objects. | |
17bb9e0d | 1229 | */ |
2587362e | 1230 | if (scan_should_stop() || gray_list_pass >= GRAY_LIST_PASSES) |
17bb9e0d CM |
1231 | return; |
1232 | ||
4698c1f2 CM |
1233 | /* |
1234 | * Scanning result reporting. | |
1235 | */ | |
1236 | rcu_read_lock(); | |
1237 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1238 | spin_lock_irqsave(&object->lock, flags); | |
1239 | if (unreferenced_object(object) && | |
1240 | !(object->flags & OBJECT_REPORTED)) { | |
1241 | object->flags |= OBJECT_REPORTED; | |
1242 | new_leaks++; | |
1243 | } | |
1244 | spin_unlock_irqrestore(&object->lock, flags); | |
1245 | } | |
1246 | rcu_read_unlock(); | |
1247 | ||
1248 | if (new_leaks) | |
1249 | pr_info("%d new suspected memory leaks (see " | |
1250 | "/sys/kernel/debug/kmemleak)\n", new_leaks); | |
1251 | ||
3c7b4e6b CM |
1252 | } |
1253 | ||
1254 | /* | |
1255 | * Thread function performing automatic memory scanning. Unreferenced objects | |
1256 | * at the end of a memory scan are reported but only the first time. | |
1257 | */ | |
1258 | static int kmemleak_scan_thread(void *arg) | |
1259 | { | |
1260 | static int first_run = 1; | |
1261 | ||
ae281064 | 1262 | pr_info("Automatic memory scanning thread started\n"); |
bf2a76b3 | 1263 | set_user_nice(current, 10); |
3c7b4e6b CM |
1264 | |
1265 | /* | |
1266 | * Wait before the first scan to allow the system to fully initialize. | |
1267 | */ | |
1268 | if (first_run) { | |
1269 | first_run = 0; | |
1270 | ssleep(SECS_FIRST_SCAN); | |
1271 | } | |
1272 | ||
1273 | while (!kthread_should_stop()) { | |
3c7b4e6b CM |
1274 | signed long timeout = jiffies_scan_wait; |
1275 | ||
1276 | mutex_lock(&scan_mutex); | |
3c7b4e6b | 1277 | kmemleak_scan(); |
3c7b4e6b | 1278 | mutex_unlock(&scan_mutex); |
4698c1f2 | 1279 | |
3c7b4e6b CM |
1280 | /* wait before the next scan */ |
1281 | while (timeout && !kthread_should_stop()) | |
1282 | timeout = schedule_timeout_interruptible(timeout); | |
1283 | } | |
1284 | ||
ae281064 | 1285 | pr_info("Automatic memory scanning thread ended\n"); |
3c7b4e6b CM |
1286 | |
1287 | return 0; | |
1288 | } | |
1289 | ||
1290 | /* | |
1291 | * Start the automatic memory scanning thread. This function must be called | |
4698c1f2 | 1292 | * with the scan_mutex held. |
3c7b4e6b | 1293 | */ |
7eb0d5e5 | 1294 | static void start_scan_thread(void) |
3c7b4e6b CM |
1295 | { |
1296 | if (scan_thread) | |
1297 | return; | |
1298 | scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak"); | |
1299 | if (IS_ERR(scan_thread)) { | |
ae281064 | 1300 | pr_warning("Failed to create the scan thread\n"); |
3c7b4e6b CM |
1301 | scan_thread = NULL; |
1302 | } | |
1303 | } | |
1304 | ||
1305 | /* | |
1306 | * Stop the automatic memory scanning thread. This function must be called | |
4698c1f2 | 1307 | * with the scan_mutex held. |
3c7b4e6b | 1308 | */ |
7eb0d5e5 | 1309 | static void stop_scan_thread(void) |
3c7b4e6b CM |
1310 | { |
1311 | if (scan_thread) { | |
1312 | kthread_stop(scan_thread); | |
1313 | scan_thread = NULL; | |
1314 | } | |
1315 | } | |
1316 | ||
1317 | /* | |
1318 | * Iterate over the object_list and return the first valid object at or after | |
1319 | * the required position with its use_count incremented. The function triggers | |
1320 | * a memory scanning when the pos argument points to the first position. | |
1321 | */ | |
1322 | static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos) | |
1323 | { | |
1324 | struct kmemleak_object *object; | |
1325 | loff_t n = *pos; | |
b87324d0 CM |
1326 | int err; |
1327 | ||
1328 | err = mutex_lock_interruptible(&scan_mutex); | |
1329 | if (err < 0) | |
1330 | return ERR_PTR(err); | |
3c7b4e6b | 1331 | |
3c7b4e6b CM |
1332 | rcu_read_lock(); |
1333 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1334 | if (n-- > 0) | |
1335 | continue; | |
1336 | if (get_object(object)) | |
1337 | goto out; | |
1338 | } | |
1339 | object = NULL; | |
1340 | out: | |
3c7b4e6b CM |
1341 | return object; |
1342 | } | |
1343 | ||
1344 | /* | |
1345 | * Return the next object in the object_list. The function decrements the | |
1346 | * use_count of the previous object and increases that of the next one. | |
1347 | */ | |
1348 | static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
1349 | { | |
1350 | struct kmemleak_object *prev_obj = v; | |
1351 | struct kmemleak_object *next_obj = NULL; | |
1352 | struct list_head *n = &prev_obj->object_list; | |
1353 | ||
1354 | ++(*pos); | |
3c7b4e6b | 1355 | |
3c7b4e6b CM |
1356 | list_for_each_continue_rcu(n, &object_list) { |
1357 | next_obj = list_entry(n, struct kmemleak_object, object_list); | |
1358 | if (get_object(next_obj)) | |
1359 | break; | |
1360 | } | |
288c857d | 1361 | |
3c7b4e6b CM |
1362 | put_object(prev_obj); |
1363 | return next_obj; | |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * Decrement the use_count of the last object required, if any. | |
1368 | */ | |
1369 | static void kmemleak_seq_stop(struct seq_file *seq, void *v) | |
1370 | { | |
b87324d0 CM |
1371 | if (!IS_ERR(v)) { |
1372 | /* | |
1373 | * kmemleak_seq_start may return ERR_PTR if the scan_mutex | |
1374 | * waiting was interrupted, so only release it if !IS_ERR. | |
1375 | */ | |
f5886c7f | 1376 | rcu_read_unlock(); |
b87324d0 CM |
1377 | mutex_unlock(&scan_mutex); |
1378 | if (v) | |
1379 | put_object(v); | |
1380 | } | |
3c7b4e6b CM |
1381 | } |
1382 | ||
1383 | /* | |
1384 | * Print the information for an unreferenced object to the seq file. | |
1385 | */ | |
1386 | static int kmemleak_seq_show(struct seq_file *seq, void *v) | |
1387 | { | |
1388 | struct kmemleak_object *object = v; | |
1389 | unsigned long flags; | |
1390 | ||
1391 | spin_lock_irqsave(&object->lock, flags); | |
288c857d | 1392 | if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object)) |
17bb9e0d | 1393 | print_unreferenced(seq, object); |
3c7b4e6b CM |
1394 | spin_unlock_irqrestore(&object->lock, flags); |
1395 | return 0; | |
1396 | } | |
1397 | ||
1398 | static const struct seq_operations kmemleak_seq_ops = { | |
1399 | .start = kmemleak_seq_start, | |
1400 | .next = kmemleak_seq_next, | |
1401 | .stop = kmemleak_seq_stop, | |
1402 | .show = kmemleak_seq_show, | |
1403 | }; | |
1404 | ||
1405 | static int kmemleak_open(struct inode *inode, struct file *file) | |
1406 | { | |
3c7b4e6b CM |
1407 | if (!atomic_read(&kmemleak_enabled)) |
1408 | return -EBUSY; | |
1409 | ||
b87324d0 | 1410 | return seq_open(file, &kmemleak_seq_ops); |
3c7b4e6b CM |
1411 | } |
1412 | ||
1413 | static int kmemleak_release(struct inode *inode, struct file *file) | |
1414 | { | |
b87324d0 | 1415 | return seq_release(inode, file); |
3c7b4e6b CM |
1416 | } |
1417 | ||
189d84ed CM |
1418 | static int dump_str_object_info(const char *str) |
1419 | { | |
1420 | unsigned long flags; | |
1421 | struct kmemleak_object *object; | |
1422 | unsigned long addr; | |
1423 | ||
1424 | addr= simple_strtoul(str, NULL, 0); | |
1425 | object = find_and_get_object(addr, 0); | |
1426 | if (!object) { | |
1427 | pr_info("Unknown object at 0x%08lx\n", addr); | |
1428 | return -EINVAL; | |
1429 | } | |
1430 | ||
1431 | spin_lock_irqsave(&object->lock, flags); | |
1432 | dump_object_info(object); | |
1433 | spin_unlock_irqrestore(&object->lock, flags); | |
1434 | ||
1435 | put_object(object); | |
1436 | return 0; | |
1437 | } | |
1438 | ||
30b37101 LR |
1439 | /* |
1440 | * We use grey instead of black to ensure we can do future scans on the same | |
1441 | * objects. If we did not do future scans these black objects could | |
1442 | * potentially contain references to newly allocated objects in the future and | |
1443 | * we'd end up with false positives. | |
1444 | */ | |
1445 | static void kmemleak_clear(void) | |
1446 | { | |
1447 | struct kmemleak_object *object; | |
1448 | unsigned long flags; | |
1449 | ||
1450 | rcu_read_lock(); | |
1451 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1452 | spin_lock_irqsave(&object->lock, flags); | |
1453 | if ((object->flags & OBJECT_REPORTED) && | |
1454 | unreferenced_object(object)) | |
a1084c87 | 1455 | __paint_it(object, KMEMLEAK_GREY); |
30b37101 LR |
1456 | spin_unlock_irqrestore(&object->lock, flags); |
1457 | } | |
1458 | rcu_read_unlock(); | |
1459 | } | |
1460 | ||
3c7b4e6b CM |
1461 | /* |
1462 | * File write operation to configure kmemleak at run-time. The following | |
1463 | * commands can be written to the /sys/kernel/debug/kmemleak file: | |
1464 | * off - disable kmemleak (irreversible) | |
1465 | * stack=on - enable the task stacks scanning | |
1466 | * stack=off - disable the tasks stacks scanning | |
1467 | * scan=on - start the automatic memory scanning thread | |
1468 | * scan=off - stop the automatic memory scanning thread | |
1469 | * scan=... - set the automatic memory scanning period in seconds (0 to | |
1470 | * disable it) | |
4698c1f2 | 1471 | * scan - trigger a memory scan |
30b37101 LR |
1472 | * clear - mark all current reported unreferenced kmemleak objects as |
1473 | * grey to ignore printing them | |
189d84ed | 1474 | * dump=... - dump information about the object found at the given address |
3c7b4e6b CM |
1475 | */ |
1476 | static ssize_t kmemleak_write(struct file *file, const char __user *user_buf, | |
1477 | size_t size, loff_t *ppos) | |
1478 | { | |
1479 | char buf[64]; | |
1480 | int buf_size; | |
b87324d0 | 1481 | int ret; |
3c7b4e6b CM |
1482 | |
1483 | buf_size = min(size, (sizeof(buf) - 1)); | |
1484 | if (strncpy_from_user(buf, user_buf, buf_size) < 0) | |
1485 | return -EFAULT; | |
1486 | buf[buf_size] = 0; | |
1487 | ||
b87324d0 CM |
1488 | ret = mutex_lock_interruptible(&scan_mutex); |
1489 | if (ret < 0) | |
1490 | return ret; | |
1491 | ||
3c7b4e6b CM |
1492 | if (strncmp(buf, "off", 3) == 0) |
1493 | kmemleak_disable(); | |
1494 | else if (strncmp(buf, "stack=on", 8) == 0) | |
1495 | kmemleak_stack_scan = 1; | |
1496 | else if (strncmp(buf, "stack=off", 9) == 0) | |
1497 | kmemleak_stack_scan = 0; | |
1498 | else if (strncmp(buf, "scan=on", 7) == 0) | |
1499 | start_scan_thread(); | |
1500 | else if (strncmp(buf, "scan=off", 8) == 0) | |
1501 | stop_scan_thread(); | |
1502 | else if (strncmp(buf, "scan=", 5) == 0) { | |
1503 | unsigned long secs; | |
3c7b4e6b | 1504 | |
b87324d0 CM |
1505 | ret = strict_strtoul(buf + 5, 0, &secs); |
1506 | if (ret < 0) | |
1507 | goto out; | |
3c7b4e6b CM |
1508 | stop_scan_thread(); |
1509 | if (secs) { | |
1510 | jiffies_scan_wait = msecs_to_jiffies(secs * 1000); | |
1511 | start_scan_thread(); | |
1512 | } | |
4698c1f2 CM |
1513 | } else if (strncmp(buf, "scan", 4) == 0) |
1514 | kmemleak_scan(); | |
30b37101 LR |
1515 | else if (strncmp(buf, "clear", 5) == 0) |
1516 | kmemleak_clear(); | |
189d84ed CM |
1517 | else if (strncmp(buf, "dump=", 5) == 0) |
1518 | ret = dump_str_object_info(buf + 5); | |
4698c1f2 | 1519 | else |
b87324d0 CM |
1520 | ret = -EINVAL; |
1521 | ||
1522 | out: | |
1523 | mutex_unlock(&scan_mutex); | |
1524 | if (ret < 0) | |
1525 | return ret; | |
3c7b4e6b CM |
1526 | |
1527 | /* ignore the rest of the buffer, only one command at a time */ | |
1528 | *ppos += size; | |
1529 | return size; | |
1530 | } | |
1531 | ||
1532 | static const struct file_operations kmemleak_fops = { | |
1533 | .owner = THIS_MODULE, | |
1534 | .open = kmemleak_open, | |
1535 | .read = seq_read, | |
1536 | .write = kmemleak_write, | |
1537 | .llseek = seq_lseek, | |
1538 | .release = kmemleak_release, | |
1539 | }; | |
1540 | ||
1541 | /* | |
1542 | * Perform the freeing of the kmemleak internal objects after waiting for any | |
1543 | * current memory scan to complete. | |
1544 | */ | |
179a8100 | 1545 | static void kmemleak_do_cleanup(struct work_struct *work) |
3c7b4e6b CM |
1546 | { |
1547 | struct kmemleak_object *object; | |
1548 | ||
4698c1f2 | 1549 | mutex_lock(&scan_mutex); |
3c7b4e6b | 1550 | stop_scan_thread(); |
3c7b4e6b | 1551 | |
3c7b4e6b CM |
1552 | rcu_read_lock(); |
1553 | list_for_each_entry_rcu(object, &object_list, object_list) | |
53238a60 | 1554 | delete_object_full(object->pointer); |
3c7b4e6b CM |
1555 | rcu_read_unlock(); |
1556 | mutex_unlock(&scan_mutex); | |
3c7b4e6b CM |
1557 | } |
1558 | ||
179a8100 | 1559 | static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup); |
3c7b4e6b CM |
1560 | |
1561 | /* | |
1562 | * Disable kmemleak. No memory allocation/freeing will be traced once this | |
1563 | * function is called. Disabling kmemleak is an irreversible operation. | |
1564 | */ | |
1565 | static void kmemleak_disable(void) | |
1566 | { | |
1567 | /* atomically check whether it was already invoked */ | |
1568 | if (atomic_cmpxchg(&kmemleak_error, 0, 1)) | |
1569 | return; | |
1570 | ||
1571 | /* stop any memory operation tracing */ | |
1572 | atomic_set(&kmemleak_early_log, 0); | |
1573 | atomic_set(&kmemleak_enabled, 0); | |
1574 | ||
1575 | /* check whether it is too early for a kernel thread */ | |
1576 | if (atomic_read(&kmemleak_initialized)) | |
179a8100 | 1577 | schedule_work(&cleanup_work); |
3c7b4e6b CM |
1578 | |
1579 | pr_info("Kernel memory leak detector disabled\n"); | |
1580 | } | |
1581 | ||
1582 | /* | |
1583 | * Allow boot-time kmemleak disabling (enabled by default). | |
1584 | */ | |
1585 | static int kmemleak_boot_config(char *str) | |
1586 | { | |
1587 | if (!str) | |
1588 | return -EINVAL; | |
1589 | if (strcmp(str, "off") == 0) | |
1590 | kmemleak_disable(); | |
1591 | else if (strcmp(str, "on") != 0) | |
1592 | return -EINVAL; | |
1593 | return 0; | |
1594 | } | |
1595 | early_param("kmemleak", kmemleak_boot_config); | |
1596 | ||
1597 | /* | |
2030117d | 1598 | * Kmemleak initialization. |
3c7b4e6b CM |
1599 | */ |
1600 | void __init kmemleak_init(void) | |
1601 | { | |
1602 | int i; | |
1603 | unsigned long flags; | |
1604 | ||
3c7b4e6b CM |
1605 | jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE); |
1606 | jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000); | |
1607 | ||
1608 | object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE); | |
1609 | scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE); | |
1610 | INIT_PRIO_TREE_ROOT(&object_tree_root); | |
1611 | ||
1612 | /* the kernel is still in UP mode, so disabling the IRQs is enough */ | |
1613 | local_irq_save(flags); | |
1614 | if (!atomic_read(&kmemleak_error)) { | |
1615 | atomic_set(&kmemleak_enabled, 1); | |
1616 | atomic_set(&kmemleak_early_log, 0); | |
1617 | } | |
1618 | local_irq_restore(flags); | |
1619 | ||
1620 | /* | |
1621 | * This is the point where tracking allocations is safe. Automatic | |
1622 | * scanning is started during the late initcall. Add the early logged | |
1623 | * callbacks to the kmemleak infrastructure. | |
1624 | */ | |
1625 | for (i = 0; i < crt_early_log; i++) { | |
1626 | struct early_log *log = &early_log[i]; | |
1627 | ||
1628 | switch (log->op_type) { | |
1629 | case KMEMLEAK_ALLOC: | |
fd678967 | 1630 | early_alloc(log); |
3c7b4e6b CM |
1631 | break; |
1632 | case KMEMLEAK_FREE: | |
1633 | kmemleak_free(log->ptr); | |
1634 | break; | |
53238a60 CM |
1635 | case KMEMLEAK_FREE_PART: |
1636 | kmemleak_free_part(log->ptr, log->size); | |
1637 | break; | |
3c7b4e6b CM |
1638 | case KMEMLEAK_NOT_LEAK: |
1639 | kmemleak_not_leak(log->ptr); | |
1640 | break; | |
1641 | case KMEMLEAK_IGNORE: | |
1642 | kmemleak_ignore(log->ptr); | |
1643 | break; | |
1644 | case KMEMLEAK_SCAN_AREA: | |
1645 | kmemleak_scan_area(log->ptr, log->offset, log->length, | |
1646 | GFP_KERNEL); | |
1647 | break; | |
1648 | case KMEMLEAK_NO_SCAN: | |
1649 | kmemleak_no_scan(log->ptr); | |
1650 | break; | |
1651 | default: | |
1652 | WARN_ON(1); | |
1653 | } | |
1654 | } | |
1655 | } | |
1656 | ||
1657 | /* | |
1658 | * Late initialization function. | |
1659 | */ | |
1660 | static int __init kmemleak_late_init(void) | |
1661 | { | |
1662 | struct dentry *dentry; | |
1663 | ||
1664 | atomic_set(&kmemleak_initialized, 1); | |
1665 | ||
1666 | if (atomic_read(&kmemleak_error)) { | |
1667 | /* | |
1668 | * Some error occured and kmemleak was disabled. There is a | |
1669 | * small chance that kmemleak_disable() was called immediately | |
1670 | * after setting kmemleak_initialized and we may end up with | |
1671 | * two clean-up threads but serialized by scan_mutex. | |
1672 | */ | |
179a8100 | 1673 | schedule_work(&cleanup_work); |
3c7b4e6b CM |
1674 | return -ENOMEM; |
1675 | } | |
1676 | ||
1677 | dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL, | |
1678 | &kmemleak_fops); | |
1679 | if (!dentry) | |
ae281064 | 1680 | pr_warning("Failed to create the debugfs kmemleak file\n"); |
4698c1f2 | 1681 | mutex_lock(&scan_mutex); |
3c7b4e6b | 1682 | start_scan_thread(); |
4698c1f2 | 1683 | mutex_unlock(&scan_mutex); |
3c7b4e6b CM |
1684 | |
1685 | pr_info("Kernel memory leak detector initialized\n"); | |
1686 | ||
1687 | return 0; | |
1688 | } | |
1689 | late_initcall(kmemleak_late_init); |