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