]>
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 | |
85d3a316 | 32 | * blocks. The object_tree_root is a red black tree used to look-up |
3c7b4e6b CM |
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 | 55 | * |
9d5a4c73 CM |
56 | * Locks and mutexes should only be acquired/nested in the following order: |
57 | * | |
58 | * scan_mutex -> object->lock -> other_object->lock (SINGLE_DEPTH_NESTING) | |
59 | * -> kmemleak_lock | |
60 | * | |
3c7b4e6b CM |
61 | * The kmemleak_object structures have a use_count incremented or decremented |
62 | * using the get_object()/put_object() functions. When the use_count becomes | |
63 | * 0, this count can no longer be incremented and put_object() schedules the | |
64 | * kmemleak_object freeing via an RCU callback. All calls to the get_object() | |
65 | * function must be protected by rcu_read_lock() to avoid accessing a freed | |
66 | * structure. | |
67 | */ | |
68 | ||
ae281064 JP |
69 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
70 | ||
3c7b4e6b CM |
71 | #include <linux/init.h> |
72 | #include <linux/kernel.h> | |
73 | #include <linux/list.h> | |
74 | #include <linux/sched.h> | |
75 | #include <linux/jiffies.h> | |
76 | #include <linux/delay.h> | |
b95f1b31 | 77 | #include <linux/export.h> |
3c7b4e6b | 78 | #include <linux/kthread.h> |
85d3a316 | 79 | #include <linux/rbtree.h> |
3c7b4e6b CM |
80 | #include <linux/fs.h> |
81 | #include <linux/debugfs.h> | |
82 | #include <linux/seq_file.h> | |
83 | #include <linux/cpumask.h> | |
84 | #include <linux/spinlock.h> | |
85 | #include <linux/mutex.h> | |
86 | #include <linux/rcupdate.h> | |
87 | #include <linux/stacktrace.h> | |
88 | #include <linux/cache.h> | |
89 | #include <linux/percpu.h> | |
90 | #include <linux/hardirq.h> | |
91 | #include <linux/mmzone.h> | |
92 | #include <linux/slab.h> | |
93 | #include <linux/thread_info.h> | |
94 | #include <linux/err.h> | |
95 | #include <linux/uaccess.h> | |
96 | #include <linux/string.h> | |
97 | #include <linux/nodemask.h> | |
98 | #include <linux/mm.h> | |
179a8100 | 99 | #include <linux/workqueue.h> |
04609ccc | 100 | #include <linux/crc32.h> |
3c7b4e6b CM |
101 | |
102 | #include <asm/sections.h> | |
103 | #include <asm/processor.h> | |
60063497 | 104 | #include <linux/atomic.h> |
3c7b4e6b | 105 | |
e79ed2f1 | 106 | #include <linux/kasan.h> |
8e019366 | 107 | #include <linux/kmemcheck.h> |
3c7b4e6b | 108 | #include <linux/kmemleak.h> |
029aeff5 | 109 | #include <linux/memory_hotplug.h> |
3c7b4e6b CM |
110 | |
111 | /* | |
112 | * Kmemleak configuration and common defines. | |
113 | */ | |
114 | #define MAX_TRACE 16 /* stack trace length */ | |
3c7b4e6b | 115 | #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */ |
3c7b4e6b CM |
116 | #define SECS_FIRST_SCAN 60 /* delay before the first scan */ |
117 | #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */ | |
af98603d | 118 | #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */ |
3c7b4e6b CM |
119 | |
120 | #define BYTES_PER_POINTER sizeof(void *) | |
121 | ||
216c04b0 | 122 | /* GFP bitmask for kmemleak internal allocations */ |
8f4fc071 VD |
123 | #define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC | \ |
124 | __GFP_NOACCOUNT)) | \ | |
6ae4bd1f CM |
125 | __GFP_NORETRY | __GFP_NOMEMALLOC | \ |
126 | __GFP_NOWARN) | |
216c04b0 | 127 | |
3c7b4e6b CM |
128 | /* scanning area inside a memory block */ |
129 | struct kmemleak_scan_area { | |
130 | struct hlist_node node; | |
c017b4be CM |
131 | unsigned long start; |
132 | size_t size; | |
3c7b4e6b CM |
133 | }; |
134 | ||
a1084c87 LR |
135 | #define KMEMLEAK_GREY 0 |
136 | #define KMEMLEAK_BLACK -1 | |
137 | ||
3c7b4e6b CM |
138 | /* |
139 | * Structure holding the metadata for each allocated memory block. | |
140 | * Modifications to such objects should be made while holding the | |
141 | * object->lock. Insertions or deletions from object_list, gray_list or | |
85d3a316 | 142 | * rb_node are already protected by the corresponding locks or mutex (see |
3c7b4e6b CM |
143 | * the notes on locking above). These objects are reference-counted |
144 | * (use_count) and freed using the RCU mechanism. | |
145 | */ | |
146 | struct kmemleak_object { | |
147 | spinlock_t lock; | |
148 | unsigned long flags; /* object status flags */ | |
149 | struct list_head object_list; | |
150 | struct list_head gray_list; | |
85d3a316 | 151 | struct rb_node rb_node; |
3c7b4e6b CM |
152 | struct rcu_head rcu; /* object_list lockless traversal */ |
153 | /* object usage count; object freed when use_count == 0 */ | |
154 | atomic_t use_count; | |
155 | unsigned long pointer; | |
156 | size_t size; | |
157 | /* minimum number of a pointers found before it is considered leak */ | |
158 | int min_count; | |
159 | /* the total number of pointers found pointing to this object */ | |
160 | int count; | |
04609ccc CM |
161 | /* checksum for detecting modified objects */ |
162 | u32 checksum; | |
3c7b4e6b CM |
163 | /* memory ranges to be scanned inside an object (empty for all) */ |
164 | struct hlist_head area_list; | |
165 | unsigned long trace[MAX_TRACE]; | |
166 | unsigned int trace_len; | |
167 | unsigned long jiffies; /* creation timestamp */ | |
168 | pid_t pid; /* pid of the current task */ | |
169 | char comm[TASK_COMM_LEN]; /* executable name */ | |
170 | }; | |
171 | ||
172 | /* flag representing the memory block allocation status */ | |
173 | #define OBJECT_ALLOCATED (1 << 0) | |
174 | /* flag set after the first reporting of an unreference object */ | |
175 | #define OBJECT_REPORTED (1 << 1) | |
176 | /* flag set to not scan the object */ | |
177 | #define OBJECT_NO_SCAN (1 << 2) | |
178 | ||
0494e082 SS |
179 | /* number of bytes to print per line; must be 16 or 32 */ |
180 | #define HEX_ROW_SIZE 16 | |
181 | /* number of bytes to print at a time (1, 2, 4, 8) */ | |
182 | #define HEX_GROUP_SIZE 1 | |
183 | /* include ASCII after the hex output */ | |
184 | #define HEX_ASCII 1 | |
185 | /* max number of lines to be printed */ | |
186 | #define HEX_MAX_LINES 2 | |
187 | ||
3c7b4e6b CM |
188 | /* the list of all allocated objects */ |
189 | static LIST_HEAD(object_list); | |
190 | /* the list of gray-colored objects (see color_gray comment below) */ | |
191 | static LIST_HEAD(gray_list); | |
85d3a316 ML |
192 | /* search tree for object boundaries */ |
193 | static struct rb_root object_tree_root = RB_ROOT; | |
194 | /* rw_lock protecting the access to object_list and object_tree_root */ | |
3c7b4e6b CM |
195 | static DEFINE_RWLOCK(kmemleak_lock); |
196 | ||
197 | /* allocation caches for kmemleak internal data */ | |
198 | static struct kmem_cache *object_cache; | |
199 | static struct kmem_cache *scan_area_cache; | |
200 | ||
201 | /* set if tracing memory operations is enabled */ | |
8910ae89 | 202 | static int kmemleak_enabled; |
c5f3b1a5 CM |
203 | /* same as above but only for the kmemleak_free() callback */ |
204 | static int kmemleak_free_enabled; | |
3c7b4e6b | 205 | /* set in the late_initcall if there were no errors */ |
8910ae89 | 206 | static int kmemleak_initialized; |
3c7b4e6b | 207 | /* enables or disables early logging of the memory operations */ |
8910ae89 | 208 | static int kmemleak_early_log = 1; |
5f79020c | 209 | /* set if a kmemleak warning was issued */ |
8910ae89 | 210 | static int kmemleak_warning; |
5f79020c | 211 | /* set if a fatal kmemleak error has occurred */ |
8910ae89 | 212 | static int kmemleak_error; |
3c7b4e6b CM |
213 | |
214 | /* minimum and maximum address that may be valid pointers */ | |
215 | static unsigned long min_addr = ULONG_MAX; | |
216 | static unsigned long max_addr; | |
217 | ||
3c7b4e6b | 218 | static struct task_struct *scan_thread; |
acf4968e | 219 | /* used to avoid reporting of recently allocated objects */ |
3c7b4e6b | 220 | static unsigned long jiffies_min_age; |
acf4968e | 221 | static unsigned long jiffies_last_scan; |
3c7b4e6b CM |
222 | /* delay between automatic memory scannings */ |
223 | static signed long jiffies_scan_wait; | |
224 | /* enables or disables the task stacks scanning */ | |
e0a2a160 | 225 | static int kmemleak_stack_scan = 1; |
4698c1f2 | 226 | /* protects the memory scanning, parameters and debug/kmemleak file access */ |
3c7b4e6b | 227 | static DEFINE_MUTEX(scan_mutex); |
ab0155a2 JB |
228 | /* setting kmemleak=on, will set this var, skipping the disable */ |
229 | static int kmemleak_skip_disable; | |
dc9b3f42 LZ |
230 | /* If there are leaks that can be reported */ |
231 | static bool kmemleak_found_leaks; | |
3c7b4e6b | 232 | |
3c7b4e6b | 233 | /* |
2030117d | 234 | * Early object allocation/freeing logging. Kmemleak is initialized after the |
3c7b4e6b | 235 | * kernel allocator. However, both the kernel allocator and kmemleak may |
2030117d | 236 | * allocate memory blocks which need to be tracked. Kmemleak defines an |
3c7b4e6b CM |
237 | * arbitrary buffer to hold the allocation/freeing information before it is |
238 | * fully initialized. | |
239 | */ | |
240 | ||
241 | /* kmemleak operation type for early logging */ | |
242 | enum { | |
243 | KMEMLEAK_ALLOC, | |
f528f0b8 | 244 | KMEMLEAK_ALLOC_PERCPU, |
3c7b4e6b | 245 | KMEMLEAK_FREE, |
53238a60 | 246 | KMEMLEAK_FREE_PART, |
f528f0b8 | 247 | KMEMLEAK_FREE_PERCPU, |
3c7b4e6b CM |
248 | KMEMLEAK_NOT_LEAK, |
249 | KMEMLEAK_IGNORE, | |
250 | KMEMLEAK_SCAN_AREA, | |
251 | KMEMLEAK_NO_SCAN | |
252 | }; | |
253 | ||
254 | /* | |
255 | * Structure holding the information passed to kmemleak callbacks during the | |
256 | * early logging. | |
257 | */ | |
258 | struct early_log { | |
259 | int op_type; /* kmemleak operation type */ | |
260 | const void *ptr; /* allocated/freed memory block */ | |
261 | size_t size; /* memory block size */ | |
262 | int min_count; /* minimum reference count */ | |
fd678967 CM |
263 | unsigned long trace[MAX_TRACE]; /* stack trace */ |
264 | unsigned int trace_len; /* stack trace length */ | |
3c7b4e6b CM |
265 | }; |
266 | ||
267 | /* early logging buffer and current position */ | |
a6186d89 CM |
268 | static struct early_log |
269 | early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata; | |
270 | static int crt_early_log __initdata; | |
3c7b4e6b CM |
271 | |
272 | static void kmemleak_disable(void); | |
273 | ||
274 | /* | |
275 | * Print a warning and dump the stack trace. | |
276 | */ | |
5f79020c CM |
277 | #define kmemleak_warn(x...) do { \ |
278 | pr_warning(x); \ | |
279 | dump_stack(); \ | |
8910ae89 | 280 | kmemleak_warning = 1; \ |
3c7b4e6b CM |
281 | } while (0) |
282 | ||
283 | /* | |
25985edc | 284 | * Macro invoked when a serious kmemleak condition occurred and cannot be |
2030117d | 285 | * recovered from. Kmemleak will be disabled and further allocation/freeing |
3c7b4e6b CM |
286 | * tracing no longer available. |
287 | */ | |
000814f4 | 288 | #define kmemleak_stop(x...) do { \ |
3c7b4e6b CM |
289 | kmemleak_warn(x); \ |
290 | kmemleak_disable(); \ | |
291 | } while (0) | |
292 | ||
0494e082 SS |
293 | /* |
294 | * Printing of the objects hex dump to the seq file. The number of lines to be | |
295 | * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The | |
296 | * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called | |
297 | * with the object->lock held. | |
298 | */ | |
299 | static void hex_dump_object(struct seq_file *seq, | |
300 | struct kmemleak_object *object) | |
301 | { | |
302 | const u8 *ptr = (const u8 *)object->pointer; | |
303 | int i, len, remaining; | |
304 | unsigned char linebuf[HEX_ROW_SIZE * 5]; | |
305 | ||
306 | /* limit the number of lines to HEX_MAX_LINES */ | |
307 | remaining = len = | |
308 | min(object->size, (size_t)(HEX_MAX_LINES * HEX_ROW_SIZE)); | |
309 | ||
310 | seq_printf(seq, " hex dump (first %d bytes):\n", len); | |
311 | for (i = 0; i < len; i += HEX_ROW_SIZE) { | |
312 | int linelen = min(remaining, HEX_ROW_SIZE); | |
313 | ||
314 | remaining -= HEX_ROW_SIZE; | |
315 | hex_dump_to_buffer(ptr + i, linelen, HEX_ROW_SIZE, | |
316 | HEX_GROUP_SIZE, linebuf, sizeof(linebuf), | |
317 | HEX_ASCII); | |
318 | seq_printf(seq, " %s\n", linebuf); | |
319 | } | |
320 | } | |
321 | ||
3c7b4e6b CM |
322 | /* |
323 | * Object colors, encoded with count and min_count: | |
324 | * - white - orphan object, not enough references to it (count < min_count) | |
325 | * - gray - not orphan, not marked as false positive (min_count == 0) or | |
326 | * sufficient references to it (count >= min_count) | |
327 | * - black - ignore, it doesn't contain references (e.g. text section) | |
328 | * (min_count == -1). No function defined for this color. | |
329 | * Newly created objects don't have any color assigned (object->count == -1) | |
330 | * before the next memory scan when they become white. | |
331 | */ | |
4a558dd6 | 332 | static bool color_white(const struct kmemleak_object *object) |
3c7b4e6b | 333 | { |
a1084c87 LR |
334 | return object->count != KMEMLEAK_BLACK && |
335 | object->count < object->min_count; | |
3c7b4e6b CM |
336 | } |
337 | ||
4a558dd6 | 338 | static bool color_gray(const struct kmemleak_object *object) |
3c7b4e6b | 339 | { |
a1084c87 LR |
340 | return object->min_count != KMEMLEAK_BLACK && |
341 | object->count >= object->min_count; | |
3c7b4e6b CM |
342 | } |
343 | ||
3c7b4e6b CM |
344 | /* |
345 | * Objects are considered unreferenced only if their color is white, they have | |
346 | * not be deleted and have a minimum age to avoid false positives caused by | |
347 | * pointers temporarily stored in CPU registers. | |
348 | */ | |
4a558dd6 | 349 | static bool unreferenced_object(struct kmemleak_object *object) |
3c7b4e6b | 350 | { |
04609ccc | 351 | return (color_white(object) && object->flags & OBJECT_ALLOCATED) && |
acf4968e CM |
352 | time_before_eq(object->jiffies + jiffies_min_age, |
353 | jiffies_last_scan); | |
3c7b4e6b CM |
354 | } |
355 | ||
356 | /* | |
bab4a34a CM |
357 | * Printing of the unreferenced objects information to the seq file. The |
358 | * print_unreferenced function must be called with the object->lock held. | |
3c7b4e6b | 359 | */ |
3c7b4e6b CM |
360 | static void print_unreferenced(struct seq_file *seq, |
361 | struct kmemleak_object *object) | |
362 | { | |
363 | int i; | |
fefdd336 | 364 | unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies); |
3c7b4e6b | 365 | |
bab4a34a CM |
366 | seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n", |
367 | object->pointer, object->size); | |
fefdd336 CM |
368 | seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n", |
369 | object->comm, object->pid, object->jiffies, | |
370 | msecs_age / 1000, msecs_age % 1000); | |
0494e082 | 371 | hex_dump_object(seq, object); |
bab4a34a | 372 | seq_printf(seq, " backtrace:\n"); |
3c7b4e6b CM |
373 | |
374 | for (i = 0; i < object->trace_len; i++) { | |
375 | void *ptr = (void *)object->trace[i]; | |
bab4a34a | 376 | seq_printf(seq, " [<%p>] %pS\n", ptr, ptr); |
3c7b4e6b CM |
377 | } |
378 | } | |
379 | ||
380 | /* | |
381 | * Print the kmemleak_object information. This function is used mainly for | |
382 | * debugging special cases when kmemleak operations. It must be called with | |
383 | * the object->lock held. | |
384 | */ | |
385 | static void dump_object_info(struct kmemleak_object *object) | |
386 | { | |
387 | struct stack_trace trace; | |
388 | ||
389 | trace.nr_entries = object->trace_len; | |
390 | trace.entries = object->trace; | |
391 | ||
ae281064 | 392 | pr_notice("Object 0x%08lx (size %zu):\n", |
85d3a316 | 393 | object->pointer, object->size); |
3c7b4e6b CM |
394 | pr_notice(" comm \"%s\", pid %d, jiffies %lu\n", |
395 | object->comm, object->pid, object->jiffies); | |
396 | pr_notice(" min_count = %d\n", object->min_count); | |
397 | pr_notice(" count = %d\n", object->count); | |
189d84ed | 398 | pr_notice(" flags = 0x%lx\n", object->flags); |
aae0ad7a | 399 | pr_notice(" checksum = %u\n", object->checksum); |
3c7b4e6b CM |
400 | pr_notice(" backtrace:\n"); |
401 | print_stack_trace(&trace, 4); | |
402 | } | |
403 | ||
404 | /* | |
85d3a316 | 405 | * Look-up a memory block metadata (kmemleak_object) in the object search |
3c7b4e6b CM |
406 | * tree based on a pointer value. If alias is 0, only values pointing to the |
407 | * beginning of the memory block are allowed. The kmemleak_lock must be held | |
408 | * when calling this function. | |
409 | */ | |
410 | static struct kmemleak_object *lookup_object(unsigned long ptr, int alias) | |
411 | { | |
85d3a316 ML |
412 | struct rb_node *rb = object_tree_root.rb_node; |
413 | ||
414 | while (rb) { | |
415 | struct kmemleak_object *object = | |
416 | rb_entry(rb, struct kmemleak_object, rb_node); | |
417 | if (ptr < object->pointer) | |
418 | rb = object->rb_node.rb_left; | |
419 | else if (object->pointer + object->size <= ptr) | |
420 | rb = object->rb_node.rb_right; | |
421 | else if (object->pointer == ptr || alias) | |
422 | return object; | |
423 | else { | |
5f79020c CM |
424 | kmemleak_warn("Found object by alias at 0x%08lx\n", |
425 | ptr); | |
a7686a45 | 426 | dump_object_info(object); |
85d3a316 | 427 | break; |
3c7b4e6b | 428 | } |
85d3a316 ML |
429 | } |
430 | return NULL; | |
3c7b4e6b CM |
431 | } |
432 | ||
433 | /* | |
434 | * Increment the object use_count. Return 1 if successful or 0 otherwise. Note | |
435 | * that once an object's use_count reached 0, the RCU freeing was already | |
436 | * registered and the object should no longer be used. This function must be | |
437 | * called under the protection of rcu_read_lock(). | |
438 | */ | |
439 | static int get_object(struct kmemleak_object *object) | |
440 | { | |
441 | return atomic_inc_not_zero(&object->use_count); | |
442 | } | |
443 | ||
444 | /* | |
445 | * RCU callback to free a kmemleak_object. | |
446 | */ | |
447 | static void free_object_rcu(struct rcu_head *rcu) | |
448 | { | |
b67bfe0d | 449 | struct hlist_node *tmp; |
3c7b4e6b CM |
450 | struct kmemleak_scan_area *area; |
451 | struct kmemleak_object *object = | |
452 | container_of(rcu, struct kmemleak_object, rcu); | |
453 | ||
454 | /* | |
455 | * Once use_count is 0 (guaranteed by put_object), there is no other | |
456 | * code accessing this object, hence no need for locking. | |
457 | */ | |
b67bfe0d SL |
458 | hlist_for_each_entry_safe(area, tmp, &object->area_list, node) { |
459 | hlist_del(&area->node); | |
3c7b4e6b CM |
460 | kmem_cache_free(scan_area_cache, area); |
461 | } | |
462 | kmem_cache_free(object_cache, object); | |
463 | } | |
464 | ||
465 | /* | |
466 | * Decrement the object use_count. Once the count is 0, free the object using | |
467 | * an RCU callback. Since put_object() may be called via the kmemleak_free() -> | |
468 | * delete_object() path, the delayed RCU freeing ensures that there is no | |
469 | * recursive call to the kernel allocator. Lock-less RCU object_list traversal | |
470 | * is also possible. | |
471 | */ | |
472 | static void put_object(struct kmemleak_object *object) | |
473 | { | |
474 | if (!atomic_dec_and_test(&object->use_count)) | |
475 | return; | |
476 | ||
477 | /* should only get here after delete_object was called */ | |
478 | WARN_ON(object->flags & OBJECT_ALLOCATED); | |
479 | ||
480 | call_rcu(&object->rcu, free_object_rcu); | |
481 | } | |
482 | ||
483 | /* | |
85d3a316 | 484 | * Look up an object in the object search tree and increase its use_count. |
3c7b4e6b CM |
485 | */ |
486 | static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias) | |
487 | { | |
488 | unsigned long flags; | |
489 | struct kmemleak_object *object = NULL; | |
490 | ||
491 | rcu_read_lock(); | |
492 | read_lock_irqsave(&kmemleak_lock, flags); | |
493 | if (ptr >= min_addr && ptr < max_addr) | |
494 | object = lookup_object(ptr, alias); | |
495 | read_unlock_irqrestore(&kmemleak_lock, flags); | |
496 | ||
497 | /* check whether the object is still available */ | |
498 | if (object && !get_object(object)) | |
499 | object = NULL; | |
500 | rcu_read_unlock(); | |
501 | ||
502 | return object; | |
503 | } | |
504 | ||
e781a9ab CM |
505 | /* |
506 | * Look up an object in the object search tree and remove it from both | |
507 | * object_tree_root and object_list. The returned object's use_count should be | |
508 | * at least 1, as initially set by create_object(). | |
509 | */ | |
510 | static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias) | |
511 | { | |
512 | unsigned long flags; | |
513 | struct kmemleak_object *object; | |
514 | ||
515 | write_lock_irqsave(&kmemleak_lock, flags); | |
516 | object = lookup_object(ptr, alias); | |
517 | if (object) { | |
518 | rb_erase(&object->rb_node, &object_tree_root); | |
519 | list_del_rcu(&object->object_list); | |
520 | } | |
521 | write_unlock_irqrestore(&kmemleak_lock, flags); | |
522 | ||
523 | return object; | |
524 | } | |
525 | ||
fd678967 CM |
526 | /* |
527 | * Save stack trace to the given array of MAX_TRACE size. | |
528 | */ | |
529 | static int __save_stack_trace(unsigned long *trace) | |
530 | { | |
531 | struct stack_trace stack_trace; | |
532 | ||
533 | stack_trace.max_entries = MAX_TRACE; | |
534 | stack_trace.nr_entries = 0; | |
535 | stack_trace.entries = trace; | |
536 | stack_trace.skip = 2; | |
537 | save_stack_trace(&stack_trace); | |
538 | ||
539 | return stack_trace.nr_entries; | |
540 | } | |
541 | ||
3c7b4e6b CM |
542 | /* |
543 | * Create the metadata (struct kmemleak_object) corresponding to an allocated | |
544 | * memory block and add it to the object_list and object_tree_root. | |
545 | */ | |
fd678967 CM |
546 | static struct kmemleak_object *create_object(unsigned long ptr, size_t size, |
547 | int min_count, gfp_t gfp) | |
3c7b4e6b CM |
548 | { |
549 | unsigned long flags; | |
85d3a316 ML |
550 | struct kmemleak_object *object, *parent; |
551 | struct rb_node **link, *rb_parent; | |
3c7b4e6b | 552 | |
6ae4bd1f | 553 | object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp)); |
3c7b4e6b | 554 | if (!object) { |
6ae4bd1f CM |
555 | pr_warning("Cannot allocate a kmemleak_object structure\n"); |
556 | kmemleak_disable(); | |
fd678967 | 557 | return NULL; |
3c7b4e6b CM |
558 | } |
559 | ||
560 | INIT_LIST_HEAD(&object->object_list); | |
561 | INIT_LIST_HEAD(&object->gray_list); | |
562 | INIT_HLIST_HEAD(&object->area_list); | |
563 | spin_lock_init(&object->lock); | |
564 | atomic_set(&object->use_count, 1); | |
04609ccc | 565 | object->flags = OBJECT_ALLOCATED; |
3c7b4e6b CM |
566 | object->pointer = ptr; |
567 | object->size = size; | |
568 | object->min_count = min_count; | |
04609ccc | 569 | object->count = 0; /* white color initially */ |
3c7b4e6b | 570 | object->jiffies = jiffies; |
04609ccc | 571 | object->checksum = 0; |
3c7b4e6b CM |
572 | |
573 | /* task information */ | |
574 | if (in_irq()) { | |
575 | object->pid = 0; | |
576 | strncpy(object->comm, "hardirq", sizeof(object->comm)); | |
577 | } else if (in_softirq()) { | |
578 | object->pid = 0; | |
579 | strncpy(object->comm, "softirq", sizeof(object->comm)); | |
580 | } else { | |
581 | object->pid = current->pid; | |
582 | /* | |
583 | * There is a small chance of a race with set_task_comm(), | |
584 | * however using get_task_comm() here may cause locking | |
585 | * dependency issues with current->alloc_lock. In the worst | |
586 | * case, the command line is not correct. | |
587 | */ | |
588 | strncpy(object->comm, current->comm, sizeof(object->comm)); | |
589 | } | |
590 | ||
591 | /* kernel backtrace */ | |
fd678967 | 592 | object->trace_len = __save_stack_trace(object->trace); |
3c7b4e6b | 593 | |
3c7b4e6b | 594 | write_lock_irqsave(&kmemleak_lock, flags); |
0580a181 | 595 | |
3c7b4e6b CM |
596 | min_addr = min(min_addr, ptr); |
597 | max_addr = max(max_addr, ptr + size); | |
85d3a316 ML |
598 | link = &object_tree_root.rb_node; |
599 | rb_parent = NULL; | |
600 | while (*link) { | |
601 | rb_parent = *link; | |
602 | parent = rb_entry(rb_parent, struct kmemleak_object, rb_node); | |
603 | if (ptr + size <= parent->pointer) | |
604 | link = &parent->rb_node.rb_left; | |
605 | else if (parent->pointer + parent->size <= ptr) | |
606 | link = &parent->rb_node.rb_right; | |
607 | else { | |
608 | kmemleak_stop("Cannot insert 0x%lx into the object " | |
609 | "search tree (overlaps existing)\n", | |
610 | ptr); | |
9d5a4c73 CM |
611 | /* |
612 | * No need for parent->lock here since "parent" cannot | |
613 | * be freed while the kmemleak_lock is held. | |
614 | */ | |
615 | dump_object_info(parent); | |
85d3a316 | 616 | kmem_cache_free(object_cache, object); |
9d5a4c73 | 617 | object = NULL; |
85d3a316 ML |
618 | goto out; |
619 | } | |
3c7b4e6b | 620 | } |
85d3a316 ML |
621 | rb_link_node(&object->rb_node, rb_parent, link); |
622 | rb_insert_color(&object->rb_node, &object_tree_root); | |
623 | ||
3c7b4e6b CM |
624 | list_add_tail_rcu(&object->object_list, &object_list); |
625 | out: | |
626 | write_unlock_irqrestore(&kmemleak_lock, flags); | |
fd678967 | 627 | return object; |
3c7b4e6b CM |
628 | } |
629 | ||
630 | /* | |
e781a9ab | 631 | * Mark the object as not allocated and schedule RCU freeing via put_object(). |
3c7b4e6b | 632 | */ |
53238a60 | 633 | static void __delete_object(struct kmemleak_object *object) |
3c7b4e6b CM |
634 | { |
635 | unsigned long flags; | |
3c7b4e6b | 636 | |
3c7b4e6b | 637 | WARN_ON(!(object->flags & OBJECT_ALLOCATED)); |
e781a9ab | 638 | WARN_ON(atomic_read(&object->use_count) < 1); |
3c7b4e6b CM |
639 | |
640 | /* | |
641 | * Locking here also ensures that the corresponding memory block | |
642 | * cannot be freed when it is being scanned. | |
643 | */ | |
644 | spin_lock_irqsave(&object->lock, flags); | |
3c7b4e6b CM |
645 | object->flags &= ~OBJECT_ALLOCATED; |
646 | spin_unlock_irqrestore(&object->lock, flags); | |
647 | put_object(object); | |
648 | } | |
649 | ||
53238a60 CM |
650 | /* |
651 | * Look up the metadata (struct kmemleak_object) corresponding to ptr and | |
652 | * delete it. | |
653 | */ | |
654 | static void delete_object_full(unsigned long ptr) | |
655 | { | |
656 | struct kmemleak_object *object; | |
657 | ||
e781a9ab | 658 | object = find_and_remove_object(ptr, 0); |
53238a60 CM |
659 | if (!object) { |
660 | #ifdef DEBUG | |
661 | kmemleak_warn("Freeing unknown object at 0x%08lx\n", | |
662 | ptr); | |
663 | #endif | |
664 | return; | |
665 | } | |
666 | __delete_object(object); | |
53238a60 CM |
667 | } |
668 | ||
669 | /* | |
670 | * Look up the metadata (struct kmemleak_object) corresponding to ptr and | |
671 | * delete it. If the memory block is partially freed, the function may create | |
672 | * additional metadata for the remaining parts of the block. | |
673 | */ | |
674 | static void delete_object_part(unsigned long ptr, size_t size) | |
675 | { | |
676 | struct kmemleak_object *object; | |
677 | unsigned long start, end; | |
678 | ||
e781a9ab | 679 | object = find_and_remove_object(ptr, 1); |
53238a60 CM |
680 | if (!object) { |
681 | #ifdef DEBUG | |
682 | kmemleak_warn("Partially freeing unknown object at 0x%08lx " | |
683 | "(size %zu)\n", ptr, size); | |
684 | #endif | |
685 | return; | |
686 | } | |
53238a60 CM |
687 | |
688 | /* | |
689 | * Create one or two objects that may result from the memory block | |
690 | * split. Note that partial freeing is only done by free_bootmem() and | |
691 | * this happens before kmemleak_init() is called. The path below is | |
692 | * only executed during early log recording in kmemleak_init(), so | |
693 | * GFP_KERNEL is enough. | |
694 | */ | |
695 | start = object->pointer; | |
696 | end = object->pointer + object->size; | |
697 | if (ptr > start) | |
698 | create_object(start, ptr - start, object->min_count, | |
699 | GFP_KERNEL); | |
700 | if (ptr + size < end) | |
701 | create_object(ptr + size, end - ptr - size, object->min_count, | |
702 | GFP_KERNEL); | |
703 | ||
e781a9ab | 704 | __delete_object(object); |
53238a60 | 705 | } |
a1084c87 LR |
706 | |
707 | static void __paint_it(struct kmemleak_object *object, int color) | |
708 | { | |
709 | object->min_count = color; | |
710 | if (color == KMEMLEAK_BLACK) | |
711 | object->flags |= OBJECT_NO_SCAN; | |
712 | } | |
713 | ||
714 | static void paint_it(struct kmemleak_object *object, int color) | |
3c7b4e6b CM |
715 | { |
716 | unsigned long flags; | |
a1084c87 LR |
717 | |
718 | spin_lock_irqsave(&object->lock, flags); | |
719 | __paint_it(object, color); | |
720 | spin_unlock_irqrestore(&object->lock, flags); | |
721 | } | |
722 | ||
723 | static void paint_ptr(unsigned long ptr, int color) | |
724 | { | |
3c7b4e6b CM |
725 | struct kmemleak_object *object; |
726 | ||
727 | object = find_and_get_object(ptr, 0); | |
728 | if (!object) { | |
a1084c87 LR |
729 | kmemleak_warn("Trying to color unknown object " |
730 | "at 0x%08lx as %s\n", ptr, | |
731 | (color == KMEMLEAK_GREY) ? "Grey" : | |
732 | (color == KMEMLEAK_BLACK) ? "Black" : "Unknown"); | |
3c7b4e6b CM |
733 | return; |
734 | } | |
a1084c87 | 735 | paint_it(object, color); |
3c7b4e6b CM |
736 | put_object(object); |
737 | } | |
738 | ||
a1084c87 | 739 | /* |
145b64b9 | 740 | * Mark an object permanently as gray-colored so that it can no longer be |
a1084c87 LR |
741 | * reported as a leak. This is used in general to mark a false positive. |
742 | */ | |
743 | static void make_gray_object(unsigned long ptr) | |
744 | { | |
745 | paint_ptr(ptr, KMEMLEAK_GREY); | |
746 | } | |
747 | ||
3c7b4e6b CM |
748 | /* |
749 | * Mark the object as black-colored so that it is ignored from scans and | |
750 | * reporting. | |
751 | */ | |
752 | static void make_black_object(unsigned long ptr) | |
753 | { | |
a1084c87 | 754 | paint_ptr(ptr, KMEMLEAK_BLACK); |
3c7b4e6b CM |
755 | } |
756 | ||
757 | /* | |
758 | * Add a scanning area to the object. If at least one such area is added, | |
759 | * kmemleak will only scan these ranges rather than the whole memory block. | |
760 | */ | |
c017b4be | 761 | static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp) |
3c7b4e6b CM |
762 | { |
763 | unsigned long flags; | |
764 | struct kmemleak_object *object; | |
765 | struct kmemleak_scan_area *area; | |
766 | ||
c017b4be | 767 | object = find_and_get_object(ptr, 1); |
3c7b4e6b | 768 | if (!object) { |
ae281064 JP |
769 | kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n", |
770 | ptr); | |
3c7b4e6b CM |
771 | return; |
772 | } | |
773 | ||
6ae4bd1f | 774 | area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp)); |
3c7b4e6b | 775 | if (!area) { |
6ae4bd1f | 776 | pr_warning("Cannot allocate a scan area\n"); |
3c7b4e6b CM |
777 | goto out; |
778 | } | |
779 | ||
780 | spin_lock_irqsave(&object->lock, flags); | |
7f88f88f CM |
781 | if (size == SIZE_MAX) { |
782 | size = object->pointer + object->size - ptr; | |
783 | } else if (ptr + size > object->pointer + object->size) { | |
ae281064 | 784 | kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr); |
3c7b4e6b CM |
785 | dump_object_info(object); |
786 | kmem_cache_free(scan_area_cache, area); | |
787 | goto out_unlock; | |
788 | } | |
789 | ||
790 | INIT_HLIST_NODE(&area->node); | |
c017b4be CM |
791 | area->start = ptr; |
792 | area->size = size; | |
3c7b4e6b CM |
793 | |
794 | hlist_add_head(&area->node, &object->area_list); | |
795 | out_unlock: | |
796 | spin_unlock_irqrestore(&object->lock, flags); | |
797 | out: | |
798 | put_object(object); | |
799 | } | |
800 | ||
801 | /* | |
802 | * Set the OBJECT_NO_SCAN flag for the object corresponding to the give | |
803 | * pointer. Such object will not be scanned by kmemleak but references to it | |
804 | * are searched. | |
805 | */ | |
806 | static void object_no_scan(unsigned long ptr) | |
807 | { | |
808 | unsigned long flags; | |
809 | struct kmemleak_object *object; | |
810 | ||
811 | object = find_and_get_object(ptr, 0); | |
812 | if (!object) { | |
ae281064 | 813 | kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr); |
3c7b4e6b CM |
814 | return; |
815 | } | |
816 | ||
817 | spin_lock_irqsave(&object->lock, flags); | |
818 | object->flags |= OBJECT_NO_SCAN; | |
819 | spin_unlock_irqrestore(&object->lock, flags); | |
820 | put_object(object); | |
821 | } | |
822 | ||
823 | /* | |
824 | * Log an early kmemleak_* call to the early_log buffer. These calls will be | |
825 | * processed later once kmemleak is fully initialized. | |
826 | */ | |
a6186d89 | 827 | static void __init log_early(int op_type, const void *ptr, size_t size, |
c017b4be | 828 | int min_count) |
3c7b4e6b CM |
829 | { |
830 | unsigned long flags; | |
831 | struct early_log *log; | |
832 | ||
8910ae89 | 833 | if (kmemleak_error) { |
b6693005 CM |
834 | /* kmemleak stopped recording, just count the requests */ |
835 | crt_early_log++; | |
836 | return; | |
837 | } | |
838 | ||
3c7b4e6b | 839 | if (crt_early_log >= ARRAY_SIZE(early_log)) { |
a9d9058a | 840 | kmemleak_disable(); |
3c7b4e6b CM |
841 | return; |
842 | } | |
843 | ||
844 | /* | |
845 | * There is no need for locking since the kernel is still in UP mode | |
846 | * at this stage. Disabling the IRQs is enough. | |
847 | */ | |
848 | local_irq_save(flags); | |
849 | log = &early_log[crt_early_log]; | |
850 | log->op_type = op_type; | |
851 | log->ptr = ptr; | |
852 | log->size = size; | |
853 | log->min_count = min_count; | |
5f79020c | 854 | log->trace_len = __save_stack_trace(log->trace); |
3c7b4e6b CM |
855 | crt_early_log++; |
856 | local_irq_restore(flags); | |
857 | } | |
858 | ||
fd678967 CM |
859 | /* |
860 | * Log an early allocated block and populate the stack trace. | |
861 | */ | |
862 | static void early_alloc(struct early_log *log) | |
863 | { | |
864 | struct kmemleak_object *object; | |
865 | unsigned long flags; | |
866 | int i; | |
867 | ||
8910ae89 | 868 | if (!kmemleak_enabled || !log->ptr || IS_ERR(log->ptr)) |
fd678967 CM |
869 | return; |
870 | ||
871 | /* | |
872 | * RCU locking needed to ensure object is not freed via put_object(). | |
873 | */ | |
874 | rcu_read_lock(); | |
875 | object = create_object((unsigned long)log->ptr, log->size, | |
c1bcd6b3 | 876 | log->min_count, GFP_ATOMIC); |
0d5d1aad CM |
877 | if (!object) |
878 | goto out; | |
fd678967 CM |
879 | spin_lock_irqsave(&object->lock, flags); |
880 | for (i = 0; i < log->trace_len; i++) | |
881 | object->trace[i] = log->trace[i]; | |
882 | object->trace_len = log->trace_len; | |
883 | spin_unlock_irqrestore(&object->lock, flags); | |
0d5d1aad | 884 | out: |
fd678967 CM |
885 | rcu_read_unlock(); |
886 | } | |
887 | ||
f528f0b8 CM |
888 | /* |
889 | * Log an early allocated block and populate the stack trace. | |
890 | */ | |
891 | static void early_alloc_percpu(struct early_log *log) | |
892 | { | |
893 | unsigned int cpu; | |
894 | const void __percpu *ptr = log->ptr; | |
895 | ||
896 | for_each_possible_cpu(cpu) { | |
897 | log->ptr = per_cpu_ptr(ptr, cpu); | |
898 | early_alloc(log); | |
899 | } | |
900 | } | |
901 | ||
a2b6bf63 CM |
902 | /** |
903 | * kmemleak_alloc - register a newly allocated object | |
904 | * @ptr: pointer to beginning of the object | |
905 | * @size: size of the object | |
906 | * @min_count: minimum number of references to this object. If during memory | |
907 | * scanning a number of references less than @min_count is found, | |
908 | * the object is reported as a memory leak. If @min_count is 0, | |
909 | * the object is never reported as a leak. If @min_count is -1, | |
910 | * the object is ignored (not scanned and not reported as a leak) | |
911 | * @gfp: kmalloc() flags used for kmemleak internal memory allocations | |
912 | * | |
913 | * This function is called from the kernel allocators when a new object | |
914 | * (memory block) is allocated (kmem_cache_alloc, kmalloc, vmalloc etc.). | |
3c7b4e6b | 915 | */ |
a6186d89 CM |
916 | void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count, |
917 | gfp_t gfp) | |
3c7b4e6b CM |
918 | { |
919 | pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count); | |
920 | ||
8910ae89 | 921 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
3c7b4e6b | 922 | create_object((unsigned long)ptr, size, min_count, gfp); |
8910ae89 | 923 | else if (kmemleak_early_log) |
c017b4be | 924 | log_early(KMEMLEAK_ALLOC, ptr, size, min_count); |
3c7b4e6b CM |
925 | } |
926 | EXPORT_SYMBOL_GPL(kmemleak_alloc); | |
927 | ||
f528f0b8 CM |
928 | /** |
929 | * kmemleak_alloc_percpu - register a newly allocated __percpu object | |
930 | * @ptr: __percpu pointer to beginning of the object | |
931 | * @size: size of the object | |
932 | * | |
933 | * This function is called from the kernel percpu allocator when a new object | |
934 | * (memory block) is allocated (alloc_percpu). It assumes GFP_KERNEL | |
935 | * allocation. | |
936 | */ | |
937 | void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size) | |
938 | { | |
939 | unsigned int cpu; | |
940 | ||
941 | pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size); | |
942 | ||
943 | /* | |
944 | * Percpu allocations are only scanned and not reported as leaks | |
945 | * (min_count is set to 0). | |
946 | */ | |
8910ae89 | 947 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
f528f0b8 CM |
948 | for_each_possible_cpu(cpu) |
949 | create_object((unsigned long)per_cpu_ptr(ptr, cpu), | |
950 | size, 0, GFP_KERNEL); | |
8910ae89 | 951 | else if (kmemleak_early_log) |
f528f0b8 CM |
952 | log_early(KMEMLEAK_ALLOC_PERCPU, ptr, size, 0); |
953 | } | |
954 | EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu); | |
955 | ||
a2b6bf63 CM |
956 | /** |
957 | * kmemleak_free - unregister a previously registered object | |
958 | * @ptr: pointer to beginning of the object | |
959 | * | |
960 | * This function is called from the kernel allocators when an object (memory | |
961 | * block) is freed (kmem_cache_free, kfree, vfree etc.). | |
3c7b4e6b | 962 | */ |
a6186d89 | 963 | void __ref kmemleak_free(const void *ptr) |
3c7b4e6b CM |
964 | { |
965 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
966 | ||
c5f3b1a5 | 967 | if (kmemleak_free_enabled && ptr && !IS_ERR(ptr)) |
53238a60 | 968 | delete_object_full((unsigned long)ptr); |
8910ae89 | 969 | else if (kmemleak_early_log) |
c017b4be | 970 | log_early(KMEMLEAK_FREE, ptr, 0, 0); |
3c7b4e6b CM |
971 | } |
972 | EXPORT_SYMBOL_GPL(kmemleak_free); | |
973 | ||
a2b6bf63 CM |
974 | /** |
975 | * kmemleak_free_part - partially unregister a previously registered object | |
976 | * @ptr: pointer to the beginning or inside the object. This also | |
977 | * represents the start of the range to be freed | |
978 | * @size: size to be unregistered | |
979 | * | |
980 | * This function is called when only a part of a memory block is freed | |
981 | * (usually from the bootmem allocator). | |
53238a60 | 982 | */ |
a6186d89 | 983 | void __ref kmemleak_free_part(const void *ptr, size_t size) |
53238a60 CM |
984 | { |
985 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
986 | ||
8910ae89 | 987 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
53238a60 | 988 | delete_object_part((unsigned long)ptr, size); |
8910ae89 | 989 | else if (kmemleak_early_log) |
c017b4be | 990 | log_early(KMEMLEAK_FREE_PART, ptr, size, 0); |
53238a60 CM |
991 | } |
992 | EXPORT_SYMBOL_GPL(kmemleak_free_part); | |
993 | ||
f528f0b8 CM |
994 | /** |
995 | * kmemleak_free_percpu - unregister a previously registered __percpu object | |
996 | * @ptr: __percpu pointer to beginning of the object | |
997 | * | |
998 | * This function is called from the kernel percpu allocator when an object | |
999 | * (memory block) is freed (free_percpu). | |
1000 | */ | |
1001 | void __ref kmemleak_free_percpu(const void __percpu *ptr) | |
1002 | { | |
1003 | unsigned int cpu; | |
1004 | ||
1005 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1006 | ||
c5f3b1a5 | 1007 | if (kmemleak_free_enabled && ptr && !IS_ERR(ptr)) |
f528f0b8 CM |
1008 | for_each_possible_cpu(cpu) |
1009 | delete_object_full((unsigned long)per_cpu_ptr(ptr, | |
1010 | cpu)); | |
8910ae89 | 1011 | else if (kmemleak_early_log) |
f528f0b8 CM |
1012 | log_early(KMEMLEAK_FREE_PERCPU, ptr, 0, 0); |
1013 | } | |
1014 | EXPORT_SYMBOL_GPL(kmemleak_free_percpu); | |
1015 | ||
ffe2c748 CM |
1016 | /** |
1017 | * kmemleak_update_trace - update object allocation stack trace | |
1018 | * @ptr: pointer to beginning of the object | |
1019 | * | |
1020 | * Override the object allocation stack trace for cases where the actual | |
1021 | * allocation place is not always useful. | |
1022 | */ | |
1023 | void __ref kmemleak_update_trace(const void *ptr) | |
1024 | { | |
1025 | struct kmemleak_object *object; | |
1026 | unsigned long flags; | |
1027 | ||
1028 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1029 | ||
1030 | if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr)) | |
1031 | return; | |
1032 | ||
1033 | object = find_and_get_object((unsigned long)ptr, 1); | |
1034 | if (!object) { | |
1035 | #ifdef DEBUG | |
1036 | kmemleak_warn("Updating stack trace for unknown object at %p\n", | |
1037 | ptr); | |
1038 | #endif | |
1039 | return; | |
1040 | } | |
1041 | ||
1042 | spin_lock_irqsave(&object->lock, flags); | |
1043 | object->trace_len = __save_stack_trace(object->trace); | |
1044 | spin_unlock_irqrestore(&object->lock, flags); | |
1045 | ||
1046 | put_object(object); | |
1047 | } | |
1048 | EXPORT_SYMBOL(kmemleak_update_trace); | |
1049 | ||
a2b6bf63 CM |
1050 | /** |
1051 | * kmemleak_not_leak - mark an allocated object as false positive | |
1052 | * @ptr: pointer to beginning of the object | |
1053 | * | |
1054 | * Calling this function on an object will cause the memory block to no longer | |
1055 | * be reported as leak and always be scanned. | |
3c7b4e6b | 1056 | */ |
a6186d89 | 1057 | void __ref kmemleak_not_leak(const void *ptr) |
3c7b4e6b CM |
1058 | { |
1059 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1060 | ||
8910ae89 | 1061 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
3c7b4e6b | 1062 | make_gray_object((unsigned long)ptr); |
8910ae89 | 1063 | else if (kmemleak_early_log) |
c017b4be | 1064 | log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0); |
3c7b4e6b CM |
1065 | } |
1066 | EXPORT_SYMBOL(kmemleak_not_leak); | |
1067 | ||
a2b6bf63 CM |
1068 | /** |
1069 | * kmemleak_ignore - ignore an allocated object | |
1070 | * @ptr: pointer to beginning of the object | |
1071 | * | |
1072 | * Calling this function on an object will cause the memory block to be | |
1073 | * ignored (not scanned and not reported as a leak). This is usually done when | |
1074 | * it is known that the corresponding block is not a leak and does not contain | |
1075 | * any references to other allocated memory blocks. | |
3c7b4e6b | 1076 | */ |
a6186d89 | 1077 | void __ref kmemleak_ignore(const void *ptr) |
3c7b4e6b CM |
1078 | { |
1079 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1080 | ||
8910ae89 | 1081 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
3c7b4e6b | 1082 | make_black_object((unsigned long)ptr); |
8910ae89 | 1083 | else if (kmemleak_early_log) |
c017b4be | 1084 | log_early(KMEMLEAK_IGNORE, ptr, 0, 0); |
3c7b4e6b CM |
1085 | } |
1086 | EXPORT_SYMBOL(kmemleak_ignore); | |
1087 | ||
a2b6bf63 CM |
1088 | /** |
1089 | * kmemleak_scan_area - limit the range to be scanned in an allocated object | |
1090 | * @ptr: pointer to beginning or inside the object. This also | |
1091 | * represents the start of the scan area | |
1092 | * @size: size of the scan area | |
1093 | * @gfp: kmalloc() flags used for kmemleak internal memory allocations | |
1094 | * | |
1095 | * This function is used when it is known that only certain parts of an object | |
1096 | * contain references to other objects. Kmemleak will only scan these areas | |
1097 | * reducing the number false negatives. | |
3c7b4e6b | 1098 | */ |
c017b4be | 1099 | void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp) |
3c7b4e6b CM |
1100 | { |
1101 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1102 | ||
8910ae89 | 1103 | if (kmemleak_enabled && ptr && size && !IS_ERR(ptr)) |
c017b4be | 1104 | add_scan_area((unsigned long)ptr, size, gfp); |
8910ae89 | 1105 | else if (kmemleak_early_log) |
c017b4be | 1106 | log_early(KMEMLEAK_SCAN_AREA, ptr, size, 0); |
3c7b4e6b CM |
1107 | } |
1108 | EXPORT_SYMBOL(kmemleak_scan_area); | |
1109 | ||
a2b6bf63 CM |
1110 | /** |
1111 | * kmemleak_no_scan - do not scan an allocated object | |
1112 | * @ptr: pointer to beginning of the object | |
1113 | * | |
1114 | * This function notifies kmemleak not to scan the given memory block. Useful | |
1115 | * in situations where it is known that the given object does not contain any | |
1116 | * references to other objects. Kmemleak will not scan such objects reducing | |
1117 | * the number of false negatives. | |
3c7b4e6b | 1118 | */ |
a6186d89 | 1119 | void __ref kmemleak_no_scan(const void *ptr) |
3c7b4e6b CM |
1120 | { |
1121 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1122 | ||
8910ae89 | 1123 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
3c7b4e6b | 1124 | object_no_scan((unsigned long)ptr); |
8910ae89 | 1125 | else if (kmemleak_early_log) |
c017b4be | 1126 | log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0); |
3c7b4e6b CM |
1127 | } |
1128 | EXPORT_SYMBOL(kmemleak_no_scan); | |
1129 | ||
04609ccc CM |
1130 | /* |
1131 | * Update an object's checksum and return true if it was modified. | |
1132 | */ | |
1133 | static bool update_checksum(struct kmemleak_object *object) | |
1134 | { | |
1135 | u32 old_csum = object->checksum; | |
1136 | ||
1137 | if (!kmemcheck_is_obj_initialized(object->pointer, object->size)) | |
1138 | return false; | |
1139 | ||
e79ed2f1 | 1140 | kasan_disable_current(); |
04609ccc | 1141 | object->checksum = crc32(0, (void *)object->pointer, object->size); |
e79ed2f1 AR |
1142 | kasan_enable_current(); |
1143 | ||
04609ccc CM |
1144 | return object->checksum != old_csum; |
1145 | } | |
1146 | ||
3c7b4e6b CM |
1147 | /* |
1148 | * Memory scanning is a long process and it needs to be interruptable. This | |
25985edc | 1149 | * function checks whether such interrupt condition occurred. |
3c7b4e6b CM |
1150 | */ |
1151 | static int scan_should_stop(void) | |
1152 | { | |
8910ae89 | 1153 | if (!kmemleak_enabled) |
3c7b4e6b CM |
1154 | return 1; |
1155 | ||
1156 | /* | |
1157 | * This function may be called from either process or kthread context, | |
1158 | * hence the need to check for both stop conditions. | |
1159 | */ | |
1160 | if (current->mm) | |
1161 | return signal_pending(current); | |
1162 | else | |
1163 | return kthread_should_stop(); | |
1164 | ||
1165 | return 0; | |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * Scan a memory block (exclusive range) for valid pointers and add those | |
1170 | * found to the gray list. | |
1171 | */ | |
1172 | static void scan_block(void *_start, void *_end, | |
4b8a9674 | 1173 | struct kmemleak_object *scanned, int allow_resched) |
3c7b4e6b CM |
1174 | { |
1175 | unsigned long *ptr; | |
1176 | unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER); | |
1177 | unsigned long *end = _end - (BYTES_PER_POINTER - 1); | |
1178 | ||
1179 | for (ptr = start; ptr < end; ptr++) { | |
3c7b4e6b | 1180 | struct kmemleak_object *object; |
8e019366 PE |
1181 | unsigned long flags; |
1182 | unsigned long pointer; | |
3c7b4e6b | 1183 | |
4b8a9674 CM |
1184 | if (allow_resched) |
1185 | cond_resched(); | |
3c7b4e6b CM |
1186 | if (scan_should_stop()) |
1187 | break; | |
1188 | ||
8e019366 PE |
1189 | /* don't scan uninitialized memory */ |
1190 | if (!kmemcheck_is_obj_initialized((unsigned long)ptr, | |
1191 | BYTES_PER_POINTER)) | |
1192 | continue; | |
1193 | ||
e79ed2f1 | 1194 | kasan_disable_current(); |
8e019366 | 1195 | pointer = *ptr; |
e79ed2f1 | 1196 | kasan_enable_current(); |
8e019366 | 1197 | |
3c7b4e6b CM |
1198 | object = find_and_get_object(pointer, 1); |
1199 | if (!object) | |
1200 | continue; | |
1201 | if (object == scanned) { | |
1202 | /* self referenced, ignore */ | |
1203 | put_object(object); | |
1204 | continue; | |
1205 | } | |
1206 | ||
1207 | /* | |
1208 | * Avoid the lockdep recursive warning on object->lock being | |
1209 | * previously acquired in scan_object(). These locks are | |
1210 | * enclosed by scan_mutex. | |
1211 | */ | |
1212 | spin_lock_irqsave_nested(&object->lock, flags, | |
1213 | SINGLE_DEPTH_NESTING); | |
1214 | if (!color_white(object)) { | |
1215 | /* non-orphan, ignored or new */ | |
1216 | spin_unlock_irqrestore(&object->lock, flags); | |
1217 | put_object(object); | |
1218 | continue; | |
1219 | } | |
1220 | ||
1221 | /* | |
1222 | * Increase the object's reference count (number of pointers | |
1223 | * to the memory block). If this count reaches the required | |
1224 | * minimum, the object's color will become gray and it will be | |
1225 | * added to the gray_list. | |
1226 | */ | |
1227 | object->count++; | |
0587da40 | 1228 | if (color_gray(object)) { |
3c7b4e6b | 1229 | list_add_tail(&object->gray_list, &gray_list); |
0587da40 CM |
1230 | spin_unlock_irqrestore(&object->lock, flags); |
1231 | continue; | |
1232 | } | |
1233 | ||
3c7b4e6b | 1234 | spin_unlock_irqrestore(&object->lock, flags); |
0587da40 | 1235 | put_object(object); |
3c7b4e6b CM |
1236 | } |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * Scan a memory block corresponding to a kmemleak_object. A condition is | |
1241 | * that object->use_count >= 1. | |
1242 | */ | |
1243 | static void scan_object(struct kmemleak_object *object) | |
1244 | { | |
1245 | struct kmemleak_scan_area *area; | |
3c7b4e6b CM |
1246 | unsigned long flags; |
1247 | ||
1248 | /* | |
21ae2956 UKK |
1249 | * Once the object->lock is acquired, the corresponding memory block |
1250 | * cannot be freed (the same lock is acquired in delete_object). | |
3c7b4e6b CM |
1251 | */ |
1252 | spin_lock_irqsave(&object->lock, flags); | |
1253 | if (object->flags & OBJECT_NO_SCAN) | |
1254 | goto out; | |
1255 | if (!(object->flags & OBJECT_ALLOCATED)) | |
1256 | /* already freed object */ | |
1257 | goto out; | |
af98603d CM |
1258 | if (hlist_empty(&object->area_list)) { |
1259 | void *start = (void *)object->pointer; | |
1260 | void *end = (void *)(object->pointer + object->size); | |
1261 | ||
1262 | while (start < end && (object->flags & OBJECT_ALLOCATED) && | |
1263 | !(object->flags & OBJECT_NO_SCAN)) { | |
1264 | scan_block(start, min(start + MAX_SCAN_SIZE, end), | |
1265 | object, 0); | |
1266 | start += MAX_SCAN_SIZE; | |
1267 | ||
1268 | spin_unlock_irqrestore(&object->lock, flags); | |
1269 | cond_resched(); | |
1270 | spin_lock_irqsave(&object->lock, flags); | |
1271 | } | |
1272 | } else | |
b67bfe0d | 1273 | hlist_for_each_entry(area, &object->area_list, node) |
c017b4be CM |
1274 | scan_block((void *)area->start, |
1275 | (void *)(area->start + area->size), | |
1276 | object, 0); | |
3c7b4e6b CM |
1277 | out: |
1278 | spin_unlock_irqrestore(&object->lock, flags); | |
1279 | } | |
1280 | ||
04609ccc CM |
1281 | /* |
1282 | * Scan the objects already referenced (gray objects). More objects will be | |
1283 | * referenced and, if there are no memory leaks, all the objects are scanned. | |
1284 | */ | |
1285 | static void scan_gray_list(void) | |
1286 | { | |
1287 | struct kmemleak_object *object, *tmp; | |
1288 | ||
1289 | /* | |
1290 | * The list traversal is safe for both tail additions and removals | |
1291 | * from inside the loop. The kmemleak objects cannot be freed from | |
1292 | * outside the loop because their use_count was incremented. | |
1293 | */ | |
1294 | object = list_entry(gray_list.next, typeof(*object), gray_list); | |
1295 | while (&object->gray_list != &gray_list) { | |
1296 | cond_resched(); | |
1297 | ||
1298 | /* may add new objects to the list */ | |
1299 | if (!scan_should_stop()) | |
1300 | scan_object(object); | |
1301 | ||
1302 | tmp = list_entry(object->gray_list.next, typeof(*object), | |
1303 | gray_list); | |
1304 | ||
1305 | /* remove the object from the list and release it */ | |
1306 | list_del(&object->gray_list); | |
1307 | put_object(object); | |
1308 | ||
1309 | object = tmp; | |
1310 | } | |
1311 | WARN_ON(!list_empty(&gray_list)); | |
1312 | } | |
1313 | ||
3c7b4e6b CM |
1314 | /* |
1315 | * Scan data sections and all the referenced memory blocks allocated via the | |
1316 | * kernel's standard allocators. This function must be called with the | |
1317 | * scan_mutex held. | |
1318 | */ | |
1319 | static void kmemleak_scan(void) | |
1320 | { | |
1321 | unsigned long flags; | |
04609ccc | 1322 | struct kmemleak_object *object; |
3c7b4e6b | 1323 | int i; |
4698c1f2 | 1324 | int new_leaks = 0; |
3c7b4e6b | 1325 | |
acf4968e CM |
1326 | jiffies_last_scan = jiffies; |
1327 | ||
3c7b4e6b CM |
1328 | /* prepare the kmemleak_object's */ |
1329 | rcu_read_lock(); | |
1330 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1331 | spin_lock_irqsave(&object->lock, flags); | |
1332 | #ifdef DEBUG | |
1333 | /* | |
1334 | * With a few exceptions there should be a maximum of | |
1335 | * 1 reference to any object at this point. | |
1336 | */ | |
1337 | if (atomic_read(&object->use_count) > 1) { | |
ae281064 | 1338 | pr_debug("object->use_count = %d\n", |
3c7b4e6b CM |
1339 | atomic_read(&object->use_count)); |
1340 | dump_object_info(object); | |
1341 | } | |
1342 | #endif | |
1343 | /* reset the reference count (whiten the object) */ | |
1344 | object->count = 0; | |
1345 | if (color_gray(object) && get_object(object)) | |
1346 | list_add_tail(&object->gray_list, &gray_list); | |
1347 | ||
1348 | spin_unlock_irqrestore(&object->lock, flags); | |
1349 | } | |
1350 | rcu_read_unlock(); | |
1351 | ||
1352 | /* data/bss scanning */ | |
4b8a9674 CM |
1353 | scan_block(_sdata, _edata, NULL, 1); |
1354 | scan_block(__bss_start, __bss_stop, NULL, 1); | |
3c7b4e6b CM |
1355 | |
1356 | #ifdef CONFIG_SMP | |
1357 | /* per-cpu sections scanning */ | |
1358 | for_each_possible_cpu(i) | |
1359 | scan_block(__per_cpu_start + per_cpu_offset(i), | |
4b8a9674 | 1360 | __per_cpu_end + per_cpu_offset(i), NULL, 1); |
3c7b4e6b CM |
1361 | #endif |
1362 | ||
1363 | /* | |
029aeff5 | 1364 | * Struct page scanning for each node. |
3c7b4e6b | 1365 | */ |
bfc8c901 | 1366 | get_online_mems(); |
3c7b4e6b | 1367 | for_each_online_node(i) { |
108bcc96 CS |
1368 | unsigned long start_pfn = node_start_pfn(i); |
1369 | unsigned long end_pfn = node_end_pfn(i); | |
3c7b4e6b CM |
1370 | unsigned long pfn; |
1371 | ||
1372 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
1373 | struct page *page; | |
1374 | ||
1375 | if (!pfn_valid(pfn)) | |
1376 | continue; | |
1377 | page = pfn_to_page(pfn); | |
1378 | /* only scan if page is in use */ | |
1379 | if (page_count(page) == 0) | |
1380 | continue; | |
4b8a9674 | 1381 | scan_block(page, page + 1, NULL, 1); |
3c7b4e6b CM |
1382 | } |
1383 | } | |
bfc8c901 | 1384 | put_online_mems(); |
3c7b4e6b CM |
1385 | |
1386 | /* | |
43ed5d6e | 1387 | * Scanning the task stacks (may introduce false negatives). |
3c7b4e6b CM |
1388 | */ |
1389 | if (kmemleak_stack_scan) { | |
43ed5d6e CM |
1390 | struct task_struct *p, *g; |
1391 | ||
3c7b4e6b | 1392 | read_lock(&tasklist_lock); |
43ed5d6e CM |
1393 | do_each_thread(g, p) { |
1394 | scan_block(task_stack_page(p), task_stack_page(p) + | |
1395 | THREAD_SIZE, NULL, 0); | |
1396 | } while_each_thread(g, p); | |
3c7b4e6b CM |
1397 | read_unlock(&tasklist_lock); |
1398 | } | |
1399 | ||
1400 | /* | |
1401 | * Scan the objects already referenced from the sections scanned | |
04609ccc | 1402 | * above. |
3c7b4e6b | 1403 | */ |
04609ccc | 1404 | scan_gray_list(); |
2587362e CM |
1405 | |
1406 | /* | |
04609ccc CM |
1407 | * Check for new or unreferenced objects modified since the previous |
1408 | * scan and color them gray until the next scan. | |
2587362e CM |
1409 | */ |
1410 | rcu_read_lock(); | |
1411 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1412 | spin_lock_irqsave(&object->lock, flags); | |
04609ccc CM |
1413 | if (color_white(object) && (object->flags & OBJECT_ALLOCATED) |
1414 | && update_checksum(object) && get_object(object)) { | |
1415 | /* color it gray temporarily */ | |
1416 | object->count = object->min_count; | |
2587362e CM |
1417 | list_add_tail(&object->gray_list, &gray_list); |
1418 | } | |
1419 | spin_unlock_irqrestore(&object->lock, flags); | |
1420 | } | |
1421 | rcu_read_unlock(); | |
1422 | ||
04609ccc CM |
1423 | /* |
1424 | * Re-scan the gray list for modified unreferenced objects. | |
1425 | */ | |
1426 | scan_gray_list(); | |
4698c1f2 | 1427 | |
17bb9e0d | 1428 | /* |
04609ccc | 1429 | * If scanning was stopped do not report any new unreferenced objects. |
17bb9e0d | 1430 | */ |
04609ccc | 1431 | if (scan_should_stop()) |
17bb9e0d CM |
1432 | return; |
1433 | ||
4698c1f2 CM |
1434 | /* |
1435 | * Scanning result reporting. | |
1436 | */ | |
1437 | rcu_read_lock(); | |
1438 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1439 | spin_lock_irqsave(&object->lock, flags); | |
1440 | if (unreferenced_object(object) && | |
1441 | !(object->flags & OBJECT_REPORTED)) { | |
1442 | object->flags |= OBJECT_REPORTED; | |
1443 | new_leaks++; | |
1444 | } | |
1445 | spin_unlock_irqrestore(&object->lock, flags); | |
1446 | } | |
1447 | rcu_read_unlock(); | |
1448 | ||
dc9b3f42 LZ |
1449 | if (new_leaks) { |
1450 | kmemleak_found_leaks = true; | |
1451 | ||
4698c1f2 CM |
1452 | pr_info("%d new suspected memory leaks (see " |
1453 | "/sys/kernel/debug/kmemleak)\n", new_leaks); | |
dc9b3f42 | 1454 | } |
4698c1f2 | 1455 | |
3c7b4e6b CM |
1456 | } |
1457 | ||
1458 | /* | |
1459 | * Thread function performing automatic memory scanning. Unreferenced objects | |
1460 | * at the end of a memory scan are reported but only the first time. | |
1461 | */ | |
1462 | static int kmemleak_scan_thread(void *arg) | |
1463 | { | |
1464 | static int first_run = 1; | |
1465 | ||
ae281064 | 1466 | pr_info("Automatic memory scanning thread started\n"); |
bf2a76b3 | 1467 | set_user_nice(current, 10); |
3c7b4e6b CM |
1468 | |
1469 | /* | |
1470 | * Wait before the first scan to allow the system to fully initialize. | |
1471 | */ | |
1472 | if (first_run) { | |
1473 | first_run = 0; | |
1474 | ssleep(SECS_FIRST_SCAN); | |
1475 | } | |
1476 | ||
1477 | while (!kthread_should_stop()) { | |
3c7b4e6b CM |
1478 | signed long timeout = jiffies_scan_wait; |
1479 | ||
1480 | mutex_lock(&scan_mutex); | |
3c7b4e6b | 1481 | kmemleak_scan(); |
3c7b4e6b | 1482 | mutex_unlock(&scan_mutex); |
4698c1f2 | 1483 | |
3c7b4e6b CM |
1484 | /* wait before the next scan */ |
1485 | while (timeout && !kthread_should_stop()) | |
1486 | timeout = schedule_timeout_interruptible(timeout); | |
1487 | } | |
1488 | ||
ae281064 | 1489 | pr_info("Automatic memory scanning thread ended\n"); |
3c7b4e6b CM |
1490 | |
1491 | return 0; | |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * Start the automatic memory scanning thread. This function must be called | |
4698c1f2 | 1496 | * with the scan_mutex held. |
3c7b4e6b | 1497 | */ |
7eb0d5e5 | 1498 | static void start_scan_thread(void) |
3c7b4e6b CM |
1499 | { |
1500 | if (scan_thread) | |
1501 | return; | |
1502 | scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak"); | |
1503 | if (IS_ERR(scan_thread)) { | |
ae281064 | 1504 | pr_warning("Failed to create the scan thread\n"); |
3c7b4e6b CM |
1505 | scan_thread = NULL; |
1506 | } | |
1507 | } | |
1508 | ||
1509 | /* | |
1510 | * Stop the automatic memory scanning thread. This function must be called | |
4698c1f2 | 1511 | * with the scan_mutex held. |
3c7b4e6b | 1512 | */ |
7eb0d5e5 | 1513 | static void stop_scan_thread(void) |
3c7b4e6b CM |
1514 | { |
1515 | if (scan_thread) { | |
1516 | kthread_stop(scan_thread); | |
1517 | scan_thread = NULL; | |
1518 | } | |
1519 | } | |
1520 | ||
1521 | /* | |
1522 | * Iterate over the object_list and return the first valid object at or after | |
1523 | * the required position with its use_count incremented. The function triggers | |
1524 | * a memory scanning when the pos argument points to the first position. | |
1525 | */ | |
1526 | static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos) | |
1527 | { | |
1528 | struct kmemleak_object *object; | |
1529 | loff_t n = *pos; | |
b87324d0 CM |
1530 | int err; |
1531 | ||
1532 | err = mutex_lock_interruptible(&scan_mutex); | |
1533 | if (err < 0) | |
1534 | return ERR_PTR(err); | |
3c7b4e6b | 1535 | |
3c7b4e6b CM |
1536 | rcu_read_lock(); |
1537 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1538 | if (n-- > 0) | |
1539 | continue; | |
1540 | if (get_object(object)) | |
1541 | goto out; | |
1542 | } | |
1543 | object = NULL; | |
1544 | out: | |
3c7b4e6b CM |
1545 | return object; |
1546 | } | |
1547 | ||
1548 | /* | |
1549 | * Return the next object in the object_list. The function decrements the | |
1550 | * use_count of the previous object and increases that of the next one. | |
1551 | */ | |
1552 | static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
1553 | { | |
1554 | struct kmemleak_object *prev_obj = v; | |
1555 | struct kmemleak_object *next_obj = NULL; | |
58fac095 | 1556 | struct kmemleak_object *obj = prev_obj; |
3c7b4e6b CM |
1557 | |
1558 | ++(*pos); | |
3c7b4e6b | 1559 | |
58fac095 | 1560 | list_for_each_entry_continue_rcu(obj, &object_list, object_list) { |
52c3ce4e CM |
1561 | if (get_object(obj)) { |
1562 | next_obj = obj; | |
3c7b4e6b | 1563 | break; |
52c3ce4e | 1564 | } |
3c7b4e6b | 1565 | } |
288c857d | 1566 | |
3c7b4e6b CM |
1567 | put_object(prev_obj); |
1568 | return next_obj; | |
1569 | } | |
1570 | ||
1571 | /* | |
1572 | * Decrement the use_count of the last object required, if any. | |
1573 | */ | |
1574 | static void kmemleak_seq_stop(struct seq_file *seq, void *v) | |
1575 | { | |
b87324d0 CM |
1576 | if (!IS_ERR(v)) { |
1577 | /* | |
1578 | * kmemleak_seq_start may return ERR_PTR if the scan_mutex | |
1579 | * waiting was interrupted, so only release it if !IS_ERR. | |
1580 | */ | |
f5886c7f | 1581 | rcu_read_unlock(); |
b87324d0 CM |
1582 | mutex_unlock(&scan_mutex); |
1583 | if (v) | |
1584 | put_object(v); | |
1585 | } | |
3c7b4e6b CM |
1586 | } |
1587 | ||
1588 | /* | |
1589 | * Print the information for an unreferenced object to the seq file. | |
1590 | */ | |
1591 | static int kmemleak_seq_show(struct seq_file *seq, void *v) | |
1592 | { | |
1593 | struct kmemleak_object *object = v; | |
1594 | unsigned long flags; | |
1595 | ||
1596 | spin_lock_irqsave(&object->lock, flags); | |
288c857d | 1597 | if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object)) |
17bb9e0d | 1598 | print_unreferenced(seq, object); |
3c7b4e6b CM |
1599 | spin_unlock_irqrestore(&object->lock, flags); |
1600 | return 0; | |
1601 | } | |
1602 | ||
1603 | static const struct seq_operations kmemleak_seq_ops = { | |
1604 | .start = kmemleak_seq_start, | |
1605 | .next = kmemleak_seq_next, | |
1606 | .stop = kmemleak_seq_stop, | |
1607 | .show = kmemleak_seq_show, | |
1608 | }; | |
1609 | ||
1610 | static int kmemleak_open(struct inode *inode, struct file *file) | |
1611 | { | |
b87324d0 | 1612 | return seq_open(file, &kmemleak_seq_ops); |
3c7b4e6b CM |
1613 | } |
1614 | ||
189d84ed CM |
1615 | static int dump_str_object_info(const char *str) |
1616 | { | |
1617 | unsigned long flags; | |
1618 | struct kmemleak_object *object; | |
1619 | unsigned long addr; | |
1620 | ||
dc053733 AP |
1621 | if (kstrtoul(str, 0, &addr)) |
1622 | return -EINVAL; | |
189d84ed CM |
1623 | object = find_and_get_object(addr, 0); |
1624 | if (!object) { | |
1625 | pr_info("Unknown object at 0x%08lx\n", addr); | |
1626 | return -EINVAL; | |
1627 | } | |
1628 | ||
1629 | spin_lock_irqsave(&object->lock, flags); | |
1630 | dump_object_info(object); | |
1631 | spin_unlock_irqrestore(&object->lock, flags); | |
1632 | ||
1633 | put_object(object); | |
1634 | return 0; | |
1635 | } | |
1636 | ||
30b37101 LR |
1637 | /* |
1638 | * We use grey instead of black to ensure we can do future scans on the same | |
1639 | * objects. If we did not do future scans these black objects could | |
1640 | * potentially contain references to newly allocated objects in the future and | |
1641 | * we'd end up with false positives. | |
1642 | */ | |
1643 | static void kmemleak_clear(void) | |
1644 | { | |
1645 | struct kmemleak_object *object; | |
1646 | unsigned long flags; | |
1647 | ||
1648 | rcu_read_lock(); | |
1649 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1650 | spin_lock_irqsave(&object->lock, flags); | |
1651 | if ((object->flags & OBJECT_REPORTED) && | |
1652 | unreferenced_object(object)) | |
a1084c87 | 1653 | __paint_it(object, KMEMLEAK_GREY); |
30b37101 LR |
1654 | spin_unlock_irqrestore(&object->lock, flags); |
1655 | } | |
1656 | rcu_read_unlock(); | |
dc9b3f42 LZ |
1657 | |
1658 | kmemleak_found_leaks = false; | |
30b37101 LR |
1659 | } |
1660 | ||
c89da70c LZ |
1661 | static void __kmemleak_do_cleanup(void); |
1662 | ||
3c7b4e6b CM |
1663 | /* |
1664 | * File write operation to configure kmemleak at run-time. The following | |
1665 | * commands can be written to the /sys/kernel/debug/kmemleak file: | |
1666 | * off - disable kmemleak (irreversible) | |
1667 | * stack=on - enable the task stacks scanning | |
1668 | * stack=off - disable the tasks stacks scanning | |
1669 | * scan=on - start the automatic memory scanning thread | |
1670 | * scan=off - stop the automatic memory scanning thread | |
1671 | * scan=... - set the automatic memory scanning period in seconds (0 to | |
1672 | * disable it) | |
4698c1f2 | 1673 | * scan - trigger a memory scan |
30b37101 | 1674 | * clear - mark all current reported unreferenced kmemleak objects as |
c89da70c LZ |
1675 | * grey to ignore printing them, or free all kmemleak objects |
1676 | * if kmemleak has been disabled. | |
189d84ed | 1677 | * dump=... - dump information about the object found at the given address |
3c7b4e6b CM |
1678 | */ |
1679 | static ssize_t kmemleak_write(struct file *file, const char __user *user_buf, | |
1680 | size_t size, loff_t *ppos) | |
1681 | { | |
1682 | char buf[64]; | |
1683 | int buf_size; | |
b87324d0 | 1684 | int ret; |
3c7b4e6b CM |
1685 | |
1686 | buf_size = min(size, (sizeof(buf) - 1)); | |
1687 | if (strncpy_from_user(buf, user_buf, buf_size) < 0) | |
1688 | return -EFAULT; | |
1689 | buf[buf_size] = 0; | |
1690 | ||
b87324d0 CM |
1691 | ret = mutex_lock_interruptible(&scan_mutex); |
1692 | if (ret < 0) | |
1693 | return ret; | |
1694 | ||
c89da70c | 1695 | if (strncmp(buf, "clear", 5) == 0) { |
8910ae89 | 1696 | if (kmemleak_enabled) |
c89da70c LZ |
1697 | kmemleak_clear(); |
1698 | else | |
1699 | __kmemleak_do_cleanup(); | |
1700 | goto out; | |
1701 | } | |
1702 | ||
8910ae89 | 1703 | if (!kmemleak_enabled) { |
c89da70c LZ |
1704 | ret = -EBUSY; |
1705 | goto out; | |
1706 | } | |
1707 | ||
3c7b4e6b CM |
1708 | if (strncmp(buf, "off", 3) == 0) |
1709 | kmemleak_disable(); | |
1710 | else if (strncmp(buf, "stack=on", 8) == 0) | |
1711 | kmemleak_stack_scan = 1; | |
1712 | else if (strncmp(buf, "stack=off", 9) == 0) | |
1713 | kmemleak_stack_scan = 0; | |
1714 | else if (strncmp(buf, "scan=on", 7) == 0) | |
1715 | start_scan_thread(); | |
1716 | else if (strncmp(buf, "scan=off", 8) == 0) | |
1717 | stop_scan_thread(); | |
1718 | else if (strncmp(buf, "scan=", 5) == 0) { | |
1719 | unsigned long secs; | |
3c7b4e6b | 1720 | |
3dbb95f7 | 1721 | ret = kstrtoul(buf + 5, 0, &secs); |
b87324d0 CM |
1722 | if (ret < 0) |
1723 | goto out; | |
3c7b4e6b CM |
1724 | stop_scan_thread(); |
1725 | if (secs) { | |
1726 | jiffies_scan_wait = msecs_to_jiffies(secs * 1000); | |
1727 | start_scan_thread(); | |
1728 | } | |
4698c1f2 CM |
1729 | } else if (strncmp(buf, "scan", 4) == 0) |
1730 | kmemleak_scan(); | |
189d84ed CM |
1731 | else if (strncmp(buf, "dump=", 5) == 0) |
1732 | ret = dump_str_object_info(buf + 5); | |
4698c1f2 | 1733 | else |
b87324d0 CM |
1734 | ret = -EINVAL; |
1735 | ||
1736 | out: | |
1737 | mutex_unlock(&scan_mutex); | |
1738 | if (ret < 0) | |
1739 | return ret; | |
3c7b4e6b CM |
1740 | |
1741 | /* ignore the rest of the buffer, only one command at a time */ | |
1742 | *ppos += size; | |
1743 | return size; | |
1744 | } | |
1745 | ||
1746 | static const struct file_operations kmemleak_fops = { | |
1747 | .owner = THIS_MODULE, | |
1748 | .open = kmemleak_open, | |
1749 | .read = seq_read, | |
1750 | .write = kmemleak_write, | |
1751 | .llseek = seq_lseek, | |
5f3bf19a | 1752 | .release = seq_release, |
3c7b4e6b CM |
1753 | }; |
1754 | ||
c89da70c LZ |
1755 | static void __kmemleak_do_cleanup(void) |
1756 | { | |
1757 | struct kmemleak_object *object; | |
1758 | ||
1759 | rcu_read_lock(); | |
1760 | list_for_each_entry_rcu(object, &object_list, object_list) | |
1761 | delete_object_full(object->pointer); | |
1762 | rcu_read_unlock(); | |
1763 | } | |
1764 | ||
3c7b4e6b | 1765 | /* |
74341703 CM |
1766 | * Stop the memory scanning thread and free the kmemleak internal objects if |
1767 | * no previous scan thread (otherwise, kmemleak may still have some useful | |
1768 | * information on memory leaks). | |
3c7b4e6b | 1769 | */ |
179a8100 | 1770 | static void kmemleak_do_cleanup(struct work_struct *work) |
3c7b4e6b | 1771 | { |
3c7b4e6b | 1772 | stop_scan_thread(); |
3c7b4e6b | 1773 | |
c5f3b1a5 CM |
1774 | /* |
1775 | * Once the scan thread has stopped, it is safe to no longer track | |
1776 | * object freeing. Ordering of the scan thread stopping and the memory | |
1777 | * accesses below is guaranteed by the kthread_stop() function. | |
1778 | */ | |
1779 | kmemleak_free_enabled = 0; | |
1780 | ||
c89da70c LZ |
1781 | if (!kmemleak_found_leaks) |
1782 | __kmemleak_do_cleanup(); | |
1783 | else | |
1784 | pr_info("Kmemleak disabled without freeing internal data. " | |
1785 | "Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\"\n"); | |
3c7b4e6b CM |
1786 | } |
1787 | ||
179a8100 | 1788 | static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup); |
3c7b4e6b CM |
1789 | |
1790 | /* | |
1791 | * Disable kmemleak. No memory allocation/freeing will be traced once this | |
1792 | * function is called. Disabling kmemleak is an irreversible operation. | |
1793 | */ | |
1794 | static void kmemleak_disable(void) | |
1795 | { | |
1796 | /* atomically check whether it was already invoked */ | |
8910ae89 | 1797 | if (cmpxchg(&kmemleak_error, 0, 1)) |
3c7b4e6b CM |
1798 | return; |
1799 | ||
1800 | /* stop any memory operation tracing */ | |
8910ae89 | 1801 | kmemleak_enabled = 0; |
3c7b4e6b CM |
1802 | |
1803 | /* check whether it is too early for a kernel thread */ | |
8910ae89 | 1804 | if (kmemleak_initialized) |
179a8100 | 1805 | schedule_work(&cleanup_work); |
c5f3b1a5 CM |
1806 | else |
1807 | kmemleak_free_enabled = 0; | |
3c7b4e6b CM |
1808 | |
1809 | pr_info("Kernel memory leak detector disabled\n"); | |
1810 | } | |
1811 | ||
1812 | /* | |
1813 | * Allow boot-time kmemleak disabling (enabled by default). | |
1814 | */ | |
1815 | static int kmemleak_boot_config(char *str) | |
1816 | { | |
1817 | if (!str) | |
1818 | return -EINVAL; | |
1819 | if (strcmp(str, "off") == 0) | |
1820 | kmemleak_disable(); | |
ab0155a2 JB |
1821 | else if (strcmp(str, "on") == 0) |
1822 | kmemleak_skip_disable = 1; | |
1823 | else | |
3c7b4e6b CM |
1824 | return -EINVAL; |
1825 | return 0; | |
1826 | } | |
1827 | early_param("kmemleak", kmemleak_boot_config); | |
1828 | ||
5f79020c CM |
1829 | static void __init print_log_trace(struct early_log *log) |
1830 | { | |
1831 | struct stack_trace trace; | |
1832 | ||
1833 | trace.nr_entries = log->trace_len; | |
1834 | trace.entries = log->trace; | |
1835 | ||
1836 | pr_notice("Early log backtrace:\n"); | |
1837 | print_stack_trace(&trace, 2); | |
1838 | } | |
1839 | ||
3c7b4e6b | 1840 | /* |
2030117d | 1841 | * Kmemleak initialization. |
3c7b4e6b CM |
1842 | */ |
1843 | void __init kmemleak_init(void) | |
1844 | { | |
1845 | int i; | |
1846 | unsigned long flags; | |
1847 | ||
ab0155a2 JB |
1848 | #ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF |
1849 | if (!kmemleak_skip_disable) { | |
3551a928 | 1850 | kmemleak_early_log = 0; |
ab0155a2 JB |
1851 | kmemleak_disable(); |
1852 | return; | |
1853 | } | |
1854 | #endif | |
1855 | ||
3c7b4e6b CM |
1856 | jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE); |
1857 | jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000); | |
1858 | ||
1859 | object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE); | |
1860 | scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE); | |
3c7b4e6b | 1861 | |
b6693005 CM |
1862 | if (crt_early_log >= ARRAY_SIZE(early_log)) |
1863 | pr_warning("Early log buffer exceeded (%d), please increase " | |
1864 | "DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n", crt_early_log); | |
1865 | ||
3c7b4e6b CM |
1866 | /* the kernel is still in UP mode, so disabling the IRQs is enough */ |
1867 | local_irq_save(flags); | |
3551a928 | 1868 | kmemleak_early_log = 0; |
8910ae89 | 1869 | if (kmemleak_error) { |
b6693005 CM |
1870 | local_irq_restore(flags); |
1871 | return; | |
c5f3b1a5 | 1872 | } else { |
8910ae89 | 1873 | kmemleak_enabled = 1; |
c5f3b1a5 CM |
1874 | kmemleak_free_enabled = 1; |
1875 | } | |
3c7b4e6b CM |
1876 | local_irq_restore(flags); |
1877 | ||
1878 | /* | |
1879 | * This is the point where tracking allocations is safe. Automatic | |
1880 | * scanning is started during the late initcall. Add the early logged | |
1881 | * callbacks to the kmemleak infrastructure. | |
1882 | */ | |
1883 | for (i = 0; i < crt_early_log; i++) { | |
1884 | struct early_log *log = &early_log[i]; | |
1885 | ||
1886 | switch (log->op_type) { | |
1887 | case KMEMLEAK_ALLOC: | |
fd678967 | 1888 | early_alloc(log); |
3c7b4e6b | 1889 | break; |
f528f0b8 CM |
1890 | case KMEMLEAK_ALLOC_PERCPU: |
1891 | early_alloc_percpu(log); | |
1892 | break; | |
3c7b4e6b CM |
1893 | case KMEMLEAK_FREE: |
1894 | kmemleak_free(log->ptr); | |
1895 | break; | |
53238a60 CM |
1896 | case KMEMLEAK_FREE_PART: |
1897 | kmemleak_free_part(log->ptr, log->size); | |
1898 | break; | |
f528f0b8 CM |
1899 | case KMEMLEAK_FREE_PERCPU: |
1900 | kmemleak_free_percpu(log->ptr); | |
1901 | break; | |
3c7b4e6b CM |
1902 | case KMEMLEAK_NOT_LEAK: |
1903 | kmemleak_not_leak(log->ptr); | |
1904 | break; | |
1905 | case KMEMLEAK_IGNORE: | |
1906 | kmemleak_ignore(log->ptr); | |
1907 | break; | |
1908 | case KMEMLEAK_SCAN_AREA: | |
c017b4be | 1909 | kmemleak_scan_area(log->ptr, log->size, GFP_KERNEL); |
3c7b4e6b CM |
1910 | break; |
1911 | case KMEMLEAK_NO_SCAN: | |
1912 | kmemleak_no_scan(log->ptr); | |
1913 | break; | |
1914 | default: | |
5f79020c CM |
1915 | kmemleak_warn("Unknown early log operation: %d\n", |
1916 | log->op_type); | |
1917 | } | |
1918 | ||
8910ae89 | 1919 | if (kmemleak_warning) { |
5f79020c | 1920 | print_log_trace(log); |
8910ae89 | 1921 | kmemleak_warning = 0; |
3c7b4e6b CM |
1922 | } |
1923 | } | |
1924 | } | |
1925 | ||
1926 | /* | |
1927 | * Late initialization function. | |
1928 | */ | |
1929 | static int __init kmemleak_late_init(void) | |
1930 | { | |
1931 | struct dentry *dentry; | |
1932 | ||
8910ae89 | 1933 | kmemleak_initialized = 1; |
3c7b4e6b | 1934 | |
8910ae89 | 1935 | if (kmemleak_error) { |
3c7b4e6b | 1936 | /* |
25985edc | 1937 | * Some error occurred and kmemleak was disabled. There is a |
3c7b4e6b CM |
1938 | * small chance that kmemleak_disable() was called immediately |
1939 | * after setting kmemleak_initialized and we may end up with | |
1940 | * two clean-up threads but serialized by scan_mutex. | |
1941 | */ | |
179a8100 | 1942 | schedule_work(&cleanup_work); |
3c7b4e6b CM |
1943 | return -ENOMEM; |
1944 | } | |
1945 | ||
1946 | dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL, | |
1947 | &kmemleak_fops); | |
1948 | if (!dentry) | |
ae281064 | 1949 | pr_warning("Failed to create the debugfs kmemleak file\n"); |
4698c1f2 | 1950 | mutex_lock(&scan_mutex); |
3c7b4e6b | 1951 | start_scan_thread(); |
4698c1f2 | 1952 | mutex_unlock(&scan_mutex); |
3c7b4e6b CM |
1953 | |
1954 | pr_info("Kernel memory leak detector initialized\n"); | |
1955 | ||
1956 | return 0; | |
1957 | } | |
1958 | late_initcall(kmemleak_late_init); |