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