Commit | Line | Data |
---|---|---|
45051539 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
3c7b4e6b CM |
2 | /* |
3 | * mm/kmemleak.c | |
4 | * | |
5 | * Copyright (C) 2008 ARM Limited | |
6 | * Written by Catalin Marinas <catalin.marinas@arm.com> | |
7 | * | |
3c7b4e6b | 8 | * For more information on the algorithm and kmemleak usage, please see |
22901c6c | 9 | * Documentation/dev-tools/kmemleak.rst. |
3c7b4e6b CM |
10 | * |
11 | * Notes on locking | |
12 | * ---------------- | |
13 | * | |
14 | * The following locks and mutexes are used by kmemleak: | |
15 | * | |
16 | * - kmemleak_lock (rwlock): protects the object_list modifications and | |
17 | * accesses to the object_tree_root. The object_list is the main list | |
18 | * holding the metadata (struct kmemleak_object) for the allocated memory | |
85d3a316 | 19 | * blocks. The object_tree_root is a red black tree used to look-up |
3c7b4e6b CM |
20 | * metadata based on a pointer to the corresponding memory block. The |
21 | * kmemleak_object structures are added to the object_list and | |
22 | * object_tree_root in the create_object() function called from the | |
23 | * kmemleak_alloc() callback and removed in delete_object() called from the | |
24 | * kmemleak_free() callback | |
25 | * - kmemleak_object.lock (spinlock): protects a kmemleak_object. Accesses to | |
26 | * the metadata (e.g. count) are protected by this lock. Note that some | |
27 | * members of this structure may be protected by other means (atomic or | |
28 | * kmemleak_lock). This lock is also held when scanning the corresponding | |
29 | * memory block to avoid the kernel freeing it via the kmemleak_free() | |
30 | * callback. This is less heavyweight than holding a global lock like | |
31 | * kmemleak_lock during scanning | |
32 | * - scan_mutex (mutex): ensures that only one thread may scan the memory for | |
33 | * unreferenced objects at a time. The gray_list contains the objects which | |
34 | * are already referenced or marked as false positives and need to be | |
35 | * scanned. This list is only modified during a scanning episode when the | |
36 | * scan_mutex is held. At the end of a scan, the gray_list is always empty. | |
37 | * Note that the kmemleak_object.use_count is incremented when an object is | |
4698c1f2 CM |
38 | * added to the gray_list and therefore cannot be freed. This mutex also |
39 | * prevents multiple users of the "kmemleak" debugfs file together with | |
40 | * modifications to the memory scanning parameters including the scan_thread | |
41 | * pointer | |
3c7b4e6b | 42 | * |
93ada579 | 43 | * Locks and mutexes are acquired/nested in the following order: |
9d5a4c73 | 44 | * |
93ada579 CM |
45 | * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING) |
46 | * | |
47 | * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex | |
48 | * regions. | |
9d5a4c73 | 49 | * |
3c7b4e6b CM |
50 | * The kmemleak_object structures have a use_count incremented or decremented |
51 | * using the get_object()/put_object() functions. When the use_count becomes | |
52 | * 0, this count can no longer be incremented and put_object() schedules the | |
53 | * kmemleak_object freeing via an RCU callback. All calls to the get_object() | |
54 | * function must be protected by rcu_read_lock() to avoid accessing a freed | |
55 | * structure. | |
56 | */ | |
57 | ||
ae281064 JP |
58 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
59 | ||
3c7b4e6b CM |
60 | #include <linux/init.h> |
61 | #include <linux/kernel.h> | |
62 | #include <linux/list.h> | |
3f07c014 | 63 | #include <linux/sched/signal.h> |
29930025 | 64 | #include <linux/sched/task.h> |
68db0cf1 | 65 | #include <linux/sched/task_stack.h> |
3c7b4e6b CM |
66 | #include <linux/jiffies.h> |
67 | #include <linux/delay.h> | |
b95f1b31 | 68 | #include <linux/export.h> |
3c7b4e6b | 69 | #include <linux/kthread.h> |
85d3a316 | 70 | #include <linux/rbtree.h> |
3c7b4e6b CM |
71 | #include <linux/fs.h> |
72 | #include <linux/debugfs.h> | |
73 | #include <linux/seq_file.h> | |
74 | #include <linux/cpumask.h> | |
75 | #include <linux/spinlock.h> | |
154221c3 | 76 | #include <linux/module.h> |
3c7b4e6b CM |
77 | #include <linux/mutex.h> |
78 | #include <linux/rcupdate.h> | |
79 | #include <linux/stacktrace.h> | |
80 | #include <linux/cache.h> | |
81 | #include <linux/percpu.h> | |
57c8a661 | 82 | #include <linux/memblock.h> |
9099daed | 83 | #include <linux/pfn.h> |
3c7b4e6b CM |
84 | #include <linux/mmzone.h> |
85 | #include <linux/slab.h> | |
86 | #include <linux/thread_info.h> | |
87 | #include <linux/err.h> | |
88 | #include <linux/uaccess.h> | |
89 | #include <linux/string.h> | |
90 | #include <linux/nodemask.h> | |
91 | #include <linux/mm.h> | |
179a8100 | 92 | #include <linux/workqueue.h> |
04609ccc | 93 | #include <linux/crc32.h> |
3c7b4e6b CM |
94 | |
95 | #include <asm/sections.h> | |
96 | #include <asm/processor.h> | |
60063497 | 97 | #include <linux/atomic.h> |
3c7b4e6b | 98 | |
e79ed2f1 | 99 | #include <linux/kasan.h> |
3c7b4e6b | 100 | #include <linux/kmemleak.h> |
029aeff5 | 101 | #include <linux/memory_hotplug.h> |
3c7b4e6b CM |
102 | |
103 | /* | |
104 | * Kmemleak configuration and common defines. | |
105 | */ | |
106 | #define MAX_TRACE 16 /* stack trace length */ | |
3c7b4e6b | 107 | #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */ |
3c7b4e6b CM |
108 | #define SECS_FIRST_SCAN 60 /* delay before the first scan */ |
109 | #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */ | |
af98603d | 110 | #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */ |
3c7b4e6b CM |
111 | |
112 | #define BYTES_PER_POINTER sizeof(void *) | |
113 | ||
216c04b0 | 114 | /* GFP bitmask for kmemleak internal allocations */ |
20b5c303 | 115 | #define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC)) | \ |
6ae4bd1f | 116 | __GFP_NORETRY | __GFP_NOMEMALLOC | \ |
df9576de | 117 | __GFP_NOWARN) |
216c04b0 | 118 | |
3c7b4e6b CM |
119 | /* scanning area inside a memory block */ |
120 | struct kmemleak_scan_area { | |
121 | struct hlist_node node; | |
c017b4be CM |
122 | unsigned long start; |
123 | size_t size; | |
3c7b4e6b CM |
124 | }; |
125 | ||
a1084c87 LR |
126 | #define KMEMLEAK_GREY 0 |
127 | #define KMEMLEAK_BLACK -1 | |
128 | ||
3c7b4e6b CM |
129 | /* |
130 | * Structure holding the metadata for each allocated memory block. | |
131 | * Modifications to such objects should be made while holding the | |
132 | * object->lock. Insertions or deletions from object_list, gray_list or | |
85d3a316 | 133 | * rb_node are already protected by the corresponding locks or mutex (see |
3c7b4e6b CM |
134 | * the notes on locking above). These objects are reference-counted |
135 | * (use_count) and freed using the RCU mechanism. | |
136 | */ | |
137 | struct kmemleak_object { | |
138 | spinlock_t lock; | |
f66abf09 | 139 | unsigned int flags; /* object status flags */ |
3c7b4e6b CM |
140 | struct list_head object_list; |
141 | struct list_head gray_list; | |
85d3a316 | 142 | struct rb_node rb_node; |
3c7b4e6b CM |
143 | struct rcu_head rcu; /* object_list lockless traversal */ |
144 | /* object usage count; object freed when use_count == 0 */ | |
145 | atomic_t use_count; | |
146 | unsigned long pointer; | |
147 | size_t size; | |
94f4a161 CM |
148 | /* pass surplus references to this pointer */ |
149 | unsigned long excess_ref; | |
3c7b4e6b CM |
150 | /* minimum number of a pointers found before it is considered leak */ |
151 | int min_count; | |
152 | /* the total number of pointers found pointing to this object */ | |
153 | int count; | |
04609ccc CM |
154 | /* checksum for detecting modified objects */ |
155 | u32 checksum; | |
3c7b4e6b CM |
156 | /* memory ranges to be scanned inside an object (empty for all) */ |
157 | struct hlist_head area_list; | |
158 | unsigned long trace[MAX_TRACE]; | |
159 | unsigned int trace_len; | |
160 | unsigned long jiffies; /* creation timestamp */ | |
161 | pid_t pid; /* pid of the current task */ | |
162 | char comm[TASK_COMM_LEN]; /* executable name */ | |
163 | }; | |
164 | ||
165 | /* flag representing the memory block allocation status */ | |
166 | #define OBJECT_ALLOCATED (1 << 0) | |
167 | /* flag set after the first reporting of an unreference object */ | |
168 | #define OBJECT_REPORTED (1 << 1) | |
169 | /* flag set to not scan the object */ | |
170 | #define OBJECT_NO_SCAN (1 << 2) | |
dba82d94 CM |
171 | /* flag set to fully scan the object when scan_area allocation failed */ |
172 | #define OBJECT_FULL_SCAN (1 << 3) | |
3c7b4e6b | 173 | |
154221c3 | 174 | #define HEX_PREFIX " " |
0494e082 SS |
175 | /* number of bytes to print per line; must be 16 or 32 */ |
176 | #define HEX_ROW_SIZE 16 | |
177 | /* number of bytes to print at a time (1, 2, 4, 8) */ | |
178 | #define HEX_GROUP_SIZE 1 | |
179 | /* include ASCII after the hex output */ | |
180 | #define HEX_ASCII 1 | |
181 | /* max number of lines to be printed */ | |
182 | #define HEX_MAX_LINES 2 | |
183 | ||
3c7b4e6b CM |
184 | /* the list of all allocated objects */ |
185 | static LIST_HEAD(object_list); | |
186 | /* the list of gray-colored objects (see color_gray comment below) */ | |
187 | static LIST_HEAD(gray_list); | |
85d3a316 ML |
188 | /* search tree for object boundaries */ |
189 | static struct rb_root object_tree_root = RB_ROOT; | |
190 | /* rw_lock protecting the access to object_list and object_tree_root */ | |
3c7b4e6b CM |
191 | static DEFINE_RWLOCK(kmemleak_lock); |
192 | ||
193 | /* allocation caches for kmemleak internal data */ | |
194 | static struct kmem_cache *object_cache; | |
195 | static struct kmem_cache *scan_area_cache; | |
196 | ||
197 | /* set if tracing memory operations is enabled */ | |
8910ae89 | 198 | static int kmemleak_enabled; |
c5f3b1a5 CM |
199 | /* same as above but only for the kmemleak_free() callback */ |
200 | static int kmemleak_free_enabled; | |
3c7b4e6b | 201 | /* set in the late_initcall if there were no errors */ |
8910ae89 | 202 | static int kmemleak_initialized; |
3c7b4e6b | 203 | /* enables or disables early logging of the memory operations */ |
8910ae89 | 204 | static int kmemleak_early_log = 1; |
5f79020c | 205 | /* set if a kmemleak warning was issued */ |
8910ae89 | 206 | static int kmemleak_warning; |
5f79020c | 207 | /* set if a fatal kmemleak error has occurred */ |
8910ae89 | 208 | static int kmemleak_error; |
3c7b4e6b CM |
209 | |
210 | /* minimum and maximum address that may be valid pointers */ | |
211 | static unsigned long min_addr = ULONG_MAX; | |
212 | static unsigned long max_addr; | |
213 | ||
3c7b4e6b | 214 | static struct task_struct *scan_thread; |
acf4968e | 215 | /* used to avoid reporting of recently allocated objects */ |
3c7b4e6b | 216 | static unsigned long jiffies_min_age; |
acf4968e | 217 | static unsigned long jiffies_last_scan; |
3c7b4e6b CM |
218 | /* delay between automatic memory scannings */ |
219 | static signed long jiffies_scan_wait; | |
220 | /* enables or disables the task stacks scanning */ | |
e0a2a160 | 221 | static int kmemleak_stack_scan = 1; |
4698c1f2 | 222 | /* protects the memory scanning, parameters and debug/kmemleak file access */ |
3c7b4e6b | 223 | static DEFINE_MUTEX(scan_mutex); |
ab0155a2 JB |
224 | /* setting kmemleak=on, will set this var, skipping the disable */ |
225 | static int kmemleak_skip_disable; | |
dc9b3f42 LZ |
226 | /* If there are leaks that can be reported */ |
227 | static bool kmemleak_found_leaks; | |
3c7b4e6b | 228 | |
154221c3 VW |
229 | static bool kmemleak_verbose; |
230 | module_param_named(verbose, kmemleak_verbose, bool, 0600); | |
231 | ||
3c7b4e6b | 232 | /* |
2030117d | 233 | * Early object allocation/freeing logging. Kmemleak is initialized after the |
3c7b4e6b | 234 | * kernel allocator. However, both the kernel allocator and kmemleak may |
2030117d | 235 | * allocate memory blocks which need to be tracked. Kmemleak defines an |
3c7b4e6b CM |
236 | * arbitrary buffer to hold the allocation/freeing information before it is |
237 | * fully initialized. | |
238 | */ | |
239 | ||
240 | /* kmemleak operation type for early logging */ | |
241 | enum { | |
242 | KMEMLEAK_ALLOC, | |
f528f0b8 | 243 | KMEMLEAK_ALLOC_PERCPU, |
3c7b4e6b | 244 | KMEMLEAK_FREE, |
53238a60 | 245 | KMEMLEAK_FREE_PART, |
f528f0b8 | 246 | KMEMLEAK_FREE_PERCPU, |
3c7b4e6b CM |
247 | KMEMLEAK_NOT_LEAK, |
248 | KMEMLEAK_IGNORE, | |
249 | KMEMLEAK_SCAN_AREA, | |
94f4a161 CM |
250 | KMEMLEAK_NO_SCAN, |
251 | KMEMLEAK_SET_EXCESS_REF | |
3c7b4e6b CM |
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 */ | |
f66abf09 | 260 | int min_count; /* minimum reference count */ |
3c7b4e6b | 261 | const void *ptr; /* allocated/freed memory block */ |
94f4a161 CM |
262 | union { |
263 | size_t size; /* memory block size */ | |
264 | unsigned long excess_ref; /* surplus reference passing */ | |
265 | }; | |
fd678967 CM |
266 | unsigned long trace[MAX_TRACE]; /* stack trace */ |
267 | unsigned int trace_len; /* stack trace length */ | |
3c7b4e6b CM |
268 | }; |
269 | ||
270 | /* early logging buffer and current position */ | |
a6186d89 CM |
271 | static struct early_log |
272 | early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata; | |
273 | static int crt_early_log __initdata; | |
3c7b4e6b CM |
274 | |
275 | static void kmemleak_disable(void); | |
276 | ||
277 | /* | |
278 | * Print a warning and dump the stack trace. | |
279 | */ | |
5f79020c | 280 | #define kmemleak_warn(x...) do { \ |
598d8091 | 281 | pr_warn(x); \ |
5f79020c | 282 | dump_stack(); \ |
8910ae89 | 283 | kmemleak_warning = 1; \ |
3c7b4e6b CM |
284 | } while (0) |
285 | ||
286 | /* | |
25985edc | 287 | * Macro invoked when a serious kmemleak condition occurred and cannot be |
2030117d | 288 | * recovered from. Kmemleak will be disabled and further allocation/freeing |
3c7b4e6b CM |
289 | * tracing no longer available. |
290 | */ | |
000814f4 | 291 | #define kmemleak_stop(x...) do { \ |
3c7b4e6b CM |
292 | kmemleak_warn(x); \ |
293 | kmemleak_disable(); \ | |
294 | } while (0) | |
295 | ||
154221c3 VW |
296 | #define warn_or_seq_printf(seq, fmt, ...) do { \ |
297 | if (seq) \ | |
298 | seq_printf(seq, fmt, ##__VA_ARGS__); \ | |
299 | else \ | |
300 | pr_warn(fmt, ##__VA_ARGS__); \ | |
301 | } while (0) | |
302 | ||
303 | static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type, | |
304 | int rowsize, int groupsize, const void *buf, | |
305 | size_t len, bool ascii) | |
306 | { | |
307 | if (seq) | |
308 | seq_hex_dump(seq, HEX_PREFIX, prefix_type, rowsize, groupsize, | |
309 | buf, len, ascii); | |
310 | else | |
311 | print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type, | |
312 | rowsize, groupsize, buf, len, ascii); | |
313 | } | |
314 | ||
0494e082 SS |
315 | /* |
316 | * Printing of the objects hex dump to the seq file. The number of lines to be | |
317 | * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The | |
318 | * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called | |
319 | * with the object->lock held. | |
320 | */ | |
321 | static void hex_dump_object(struct seq_file *seq, | |
322 | struct kmemleak_object *object) | |
323 | { | |
324 | const u8 *ptr = (const u8 *)object->pointer; | |
6fc37c49 | 325 | size_t len; |
0494e082 SS |
326 | |
327 | /* limit the number of lines to HEX_MAX_LINES */ | |
6fc37c49 | 328 | len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE); |
0494e082 | 329 | |
154221c3 | 330 | warn_or_seq_printf(seq, " hex dump (first %zu bytes):\n", len); |
5c335fe0 | 331 | kasan_disable_current(); |
154221c3 VW |
332 | warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE, |
333 | HEX_GROUP_SIZE, ptr, len, HEX_ASCII); | |
5c335fe0 | 334 | kasan_enable_current(); |
0494e082 SS |
335 | } |
336 | ||
3c7b4e6b CM |
337 | /* |
338 | * Object colors, encoded with count and min_count: | |
339 | * - white - orphan object, not enough references to it (count < min_count) | |
340 | * - gray - not orphan, not marked as false positive (min_count == 0) or | |
341 | * sufficient references to it (count >= min_count) | |
342 | * - black - ignore, it doesn't contain references (e.g. text section) | |
343 | * (min_count == -1). No function defined for this color. | |
344 | * Newly created objects don't have any color assigned (object->count == -1) | |
345 | * before the next memory scan when they become white. | |
346 | */ | |
4a558dd6 | 347 | static bool color_white(const struct kmemleak_object *object) |
3c7b4e6b | 348 | { |
a1084c87 LR |
349 | return object->count != KMEMLEAK_BLACK && |
350 | object->count < object->min_count; | |
3c7b4e6b CM |
351 | } |
352 | ||
4a558dd6 | 353 | static bool color_gray(const struct kmemleak_object *object) |
3c7b4e6b | 354 | { |
a1084c87 LR |
355 | return object->min_count != KMEMLEAK_BLACK && |
356 | object->count >= object->min_count; | |
3c7b4e6b CM |
357 | } |
358 | ||
3c7b4e6b CM |
359 | /* |
360 | * Objects are considered unreferenced only if their color is white, they have | |
361 | * not be deleted and have a minimum age to avoid false positives caused by | |
362 | * pointers temporarily stored in CPU registers. | |
363 | */ | |
4a558dd6 | 364 | static bool unreferenced_object(struct kmemleak_object *object) |
3c7b4e6b | 365 | { |
04609ccc | 366 | return (color_white(object) && object->flags & OBJECT_ALLOCATED) && |
acf4968e CM |
367 | time_before_eq(object->jiffies + jiffies_min_age, |
368 | jiffies_last_scan); | |
3c7b4e6b CM |
369 | } |
370 | ||
371 | /* | |
bab4a34a CM |
372 | * Printing of the unreferenced objects information to the seq file. The |
373 | * print_unreferenced function must be called with the object->lock held. | |
3c7b4e6b | 374 | */ |
3c7b4e6b CM |
375 | static void print_unreferenced(struct seq_file *seq, |
376 | struct kmemleak_object *object) | |
377 | { | |
378 | int i; | |
fefdd336 | 379 | unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies); |
3c7b4e6b | 380 | |
154221c3 | 381 | warn_or_seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n", |
bab4a34a | 382 | object->pointer, object->size); |
154221c3 | 383 | warn_or_seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n", |
fefdd336 CM |
384 | object->comm, object->pid, object->jiffies, |
385 | msecs_age / 1000, msecs_age % 1000); | |
0494e082 | 386 | hex_dump_object(seq, object); |
154221c3 | 387 | warn_or_seq_printf(seq, " backtrace:\n"); |
3c7b4e6b CM |
388 | |
389 | for (i = 0; i < object->trace_len; i++) { | |
390 | void *ptr = (void *)object->trace[i]; | |
154221c3 | 391 | warn_or_seq_printf(seq, " [<%p>] %pS\n", ptr, ptr); |
3c7b4e6b CM |
392 | } |
393 | } | |
394 | ||
395 | /* | |
396 | * Print the kmemleak_object information. This function is used mainly for | |
397 | * debugging special cases when kmemleak operations. It must be called with | |
398 | * the object->lock held. | |
399 | */ | |
400 | static void dump_object_info(struct kmemleak_object *object) | |
401 | { | |
ae281064 | 402 | pr_notice("Object 0x%08lx (size %zu):\n", |
85d3a316 | 403 | object->pointer, object->size); |
3c7b4e6b CM |
404 | pr_notice(" comm \"%s\", pid %d, jiffies %lu\n", |
405 | object->comm, object->pid, object->jiffies); | |
406 | pr_notice(" min_count = %d\n", object->min_count); | |
407 | pr_notice(" count = %d\n", object->count); | |
f66abf09 | 408 | pr_notice(" flags = 0x%x\n", object->flags); |
aae0ad7a | 409 | pr_notice(" checksum = %u\n", object->checksum); |
3c7b4e6b | 410 | pr_notice(" backtrace:\n"); |
07984aad | 411 | stack_trace_print(object->trace, object->trace_len, 4); |
3c7b4e6b CM |
412 | } |
413 | ||
414 | /* | |
85d3a316 | 415 | * Look-up a memory block metadata (kmemleak_object) in the object search |
3c7b4e6b CM |
416 | * tree based on a pointer value. If alias is 0, only values pointing to the |
417 | * beginning of the memory block are allowed. The kmemleak_lock must be held | |
418 | * when calling this function. | |
419 | */ | |
420 | static struct kmemleak_object *lookup_object(unsigned long ptr, int alias) | |
421 | { | |
85d3a316 ML |
422 | struct rb_node *rb = object_tree_root.rb_node; |
423 | ||
424 | while (rb) { | |
425 | struct kmemleak_object *object = | |
426 | rb_entry(rb, struct kmemleak_object, rb_node); | |
427 | if (ptr < object->pointer) | |
428 | rb = object->rb_node.rb_left; | |
429 | else if (object->pointer + object->size <= ptr) | |
430 | rb = object->rb_node.rb_right; | |
431 | else if (object->pointer == ptr || alias) | |
432 | return object; | |
433 | else { | |
5f79020c CM |
434 | kmemleak_warn("Found object by alias at 0x%08lx\n", |
435 | ptr); | |
a7686a45 | 436 | dump_object_info(object); |
85d3a316 | 437 | break; |
3c7b4e6b | 438 | } |
85d3a316 ML |
439 | } |
440 | return NULL; | |
3c7b4e6b CM |
441 | } |
442 | ||
443 | /* | |
444 | * Increment the object use_count. Return 1 if successful or 0 otherwise. Note | |
445 | * that once an object's use_count reached 0, the RCU freeing was already | |
446 | * registered and the object should no longer be used. This function must be | |
447 | * called under the protection of rcu_read_lock(). | |
448 | */ | |
449 | static int get_object(struct kmemleak_object *object) | |
450 | { | |
451 | return atomic_inc_not_zero(&object->use_count); | |
452 | } | |
453 | ||
454 | /* | |
455 | * RCU callback to free a kmemleak_object. | |
456 | */ | |
457 | static void free_object_rcu(struct rcu_head *rcu) | |
458 | { | |
b67bfe0d | 459 | struct hlist_node *tmp; |
3c7b4e6b CM |
460 | struct kmemleak_scan_area *area; |
461 | struct kmemleak_object *object = | |
462 | container_of(rcu, struct kmemleak_object, rcu); | |
463 | ||
464 | /* | |
465 | * Once use_count is 0 (guaranteed by put_object), there is no other | |
466 | * code accessing this object, hence no need for locking. | |
467 | */ | |
b67bfe0d SL |
468 | hlist_for_each_entry_safe(area, tmp, &object->area_list, node) { |
469 | hlist_del(&area->node); | |
3c7b4e6b CM |
470 | kmem_cache_free(scan_area_cache, area); |
471 | } | |
472 | kmem_cache_free(object_cache, object); | |
473 | } | |
474 | ||
475 | /* | |
476 | * Decrement the object use_count. Once the count is 0, free the object using | |
477 | * an RCU callback. Since put_object() may be called via the kmemleak_free() -> | |
478 | * delete_object() path, the delayed RCU freeing ensures that there is no | |
479 | * recursive call to the kernel allocator. Lock-less RCU object_list traversal | |
480 | * is also possible. | |
481 | */ | |
482 | static void put_object(struct kmemleak_object *object) | |
483 | { | |
484 | if (!atomic_dec_and_test(&object->use_count)) | |
485 | return; | |
486 | ||
487 | /* should only get here after delete_object was called */ | |
488 | WARN_ON(object->flags & OBJECT_ALLOCATED); | |
489 | ||
490 | call_rcu(&object->rcu, free_object_rcu); | |
491 | } | |
492 | ||
493 | /* | |
85d3a316 | 494 | * Look up an object in the object search tree and increase its use_count. |
3c7b4e6b CM |
495 | */ |
496 | static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias) | |
497 | { | |
498 | unsigned long flags; | |
9fbed254 | 499 | struct kmemleak_object *object; |
3c7b4e6b CM |
500 | |
501 | rcu_read_lock(); | |
502 | read_lock_irqsave(&kmemleak_lock, flags); | |
93ada579 | 503 | object = lookup_object(ptr, alias); |
3c7b4e6b CM |
504 | read_unlock_irqrestore(&kmemleak_lock, flags); |
505 | ||
506 | /* check whether the object is still available */ | |
507 | if (object && !get_object(object)) | |
508 | object = NULL; | |
509 | rcu_read_unlock(); | |
510 | ||
511 | return object; | |
512 | } | |
513 | ||
e781a9ab CM |
514 | /* |
515 | * Look up an object in the object search tree and remove it from both | |
516 | * object_tree_root and object_list. The returned object's use_count should be | |
517 | * at least 1, as initially set by create_object(). | |
518 | */ | |
519 | static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias) | |
520 | { | |
521 | unsigned long flags; | |
522 | struct kmemleak_object *object; | |
523 | ||
524 | write_lock_irqsave(&kmemleak_lock, flags); | |
525 | object = lookup_object(ptr, alias); | |
526 | if (object) { | |
527 | rb_erase(&object->rb_node, &object_tree_root); | |
528 | list_del_rcu(&object->object_list); | |
529 | } | |
530 | write_unlock_irqrestore(&kmemleak_lock, flags); | |
531 | ||
532 | return object; | |
533 | } | |
534 | ||
fd678967 CM |
535 | /* |
536 | * Save stack trace to the given array of MAX_TRACE size. | |
537 | */ | |
538 | static int __save_stack_trace(unsigned long *trace) | |
539 | { | |
07984aad | 540 | return stack_trace_save(trace, MAX_TRACE, 2); |
fd678967 CM |
541 | } |
542 | ||
3c7b4e6b CM |
543 | /* |
544 | * Create the metadata (struct kmemleak_object) corresponding to an allocated | |
545 | * memory block and add it to the object_list and object_tree_root. | |
546 | */ | |
fd678967 CM |
547 | static struct kmemleak_object *create_object(unsigned long ptr, size_t size, |
548 | int min_count, gfp_t gfp) | |
3c7b4e6b CM |
549 | { |
550 | unsigned long flags; | |
85d3a316 ML |
551 | struct kmemleak_object *object, *parent; |
552 | struct rb_node **link, *rb_parent; | |
a2f77575 | 553 | unsigned long untagged_ptr; |
3c7b4e6b | 554 | |
6ae4bd1f | 555 | object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp)); |
3c7b4e6b | 556 | if (!object) { |
598d8091 | 557 | pr_warn("Cannot allocate a kmemleak_object structure\n"); |
6ae4bd1f | 558 | kmemleak_disable(); |
fd678967 | 559 | return NULL; |
3c7b4e6b CM |
560 | } |
561 | ||
562 | INIT_LIST_HEAD(&object->object_list); | |
563 | INIT_LIST_HEAD(&object->gray_list); | |
564 | INIT_HLIST_HEAD(&object->area_list); | |
565 | spin_lock_init(&object->lock); | |
566 | atomic_set(&object->use_count, 1); | |
04609ccc | 567 | object->flags = OBJECT_ALLOCATED; |
3c7b4e6b CM |
568 | object->pointer = ptr; |
569 | object->size = size; | |
94f4a161 | 570 | object->excess_ref = 0; |
3c7b4e6b | 571 | object->min_count = min_count; |
04609ccc | 572 | object->count = 0; /* white color initially */ |
3c7b4e6b | 573 | object->jiffies = jiffies; |
04609ccc | 574 | object->checksum = 0; |
3c7b4e6b CM |
575 | |
576 | /* task information */ | |
577 | if (in_irq()) { | |
578 | object->pid = 0; | |
579 | strncpy(object->comm, "hardirq", sizeof(object->comm)); | |
6ef90569 | 580 | } else if (in_serving_softirq()) { |
3c7b4e6b CM |
581 | object->pid = 0; |
582 | strncpy(object->comm, "softirq", sizeof(object->comm)); | |
583 | } else { | |
584 | object->pid = current->pid; | |
585 | /* | |
586 | * There is a small chance of a race with set_task_comm(), | |
587 | * however using get_task_comm() here may cause locking | |
588 | * dependency issues with current->alloc_lock. In the worst | |
589 | * case, the command line is not correct. | |
590 | */ | |
591 | strncpy(object->comm, current->comm, sizeof(object->comm)); | |
592 | } | |
593 | ||
594 | /* kernel backtrace */ | |
fd678967 | 595 | object->trace_len = __save_stack_trace(object->trace); |
3c7b4e6b | 596 | |
3c7b4e6b | 597 | write_lock_irqsave(&kmemleak_lock, flags); |
0580a181 | 598 | |
a2f77575 AK |
599 | untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr); |
600 | min_addr = min(min_addr, untagged_ptr); | |
601 | max_addr = max(max_addr, untagged_ptr + size); | |
85d3a316 ML |
602 | link = &object_tree_root.rb_node; |
603 | rb_parent = NULL; | |
604 | while (*link) { | |
605 | rb_parent = *link; | |
606 | parent = rb_entry(rb_parent, struct kmemleak_object, rb_node); | |
607 | if (ptr + size <= parent->pointer) | |
608 | link = &parent->rb_node.rb_left; | |
609 | else if (parent->pointer + parent->size <= ptr) | |
610 | link = &parent->rb_node.rb_right; | |
611 | else { | |
756a025f | 612 | kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n", |
85d3a316 | 613 | ptr); |
9d5a4c73 CM |
614 | /* |
615 | * No need for parent->lock here since "parent" cannot | |
616 | * be freed while the kmemleak_lock is held. | |
617 | */ | |
618 | dump_object_info(parent); | |
85d3a316 | 619 | kmem_cache_free(object_cache, object); |
9d5a4c73 | 620 | object = NULL; |
85d3a316 ML |
621 | goto out; |
622 | } | |
3c7b4e6b | 623 | } |
85d3a316 ML |
624 | rb_link_node(&object->rb_node, rb_parent, link); |
625 | rb_insert_color(&object->rb_node, &object_tree_root); | |
626 | ||
3c7b4e6b CM |
627 | list_add_tail_rcu(&object->object_list, &object_list); |
628 | out: | |
629 | write_unlock_irqrestore(&kmemleak_lock, flags); | |
fd678967 | 630 | return object; |
3c7b4e6b CM |
631 | } |
632 | ||
633 | /* | |
e781a9ab | 634 | * Mark the object as not allocated and schedule RCU freeing via put_object(). |
3c7b4e6b | 635 | */ |
53238a60 | 636 | static void __delete_object(struct kmemleak_object *object) |
3c7b4e6b CM |
637 | { |
638 | unsigned long flags; | |
3c7b4e6b | 639 | |
3c7b4e6b | 640 | WARN_ON(!(object->flags & OBJECT_ALLOCATED)); |
e781a9ab | 641 | WARN_ON(atomic_read(&object->use_count) < 1); |
3c7b4e6b CM |
642 | |
643 | /* | |
644 | * Locking here also ensures that the corresponding memory block | |
645 | * cannot be freed when it is being scanned. | |
646 | */ | |
647 | spin_lock_irqsave(&object->lock, flags); | |
3c7b4e6b CM |
648 | object->flags &= ~OBJECT_ALLOCATED; |
649 | spin_unlock_irqrestore(&object->lock, flags); | |
650 | put_object(object); | |
651 | } | |
652 | ||
53238a60 CM |
653 | /* |
654 | * Look up the metadata (struct kmemleak_object) corresponding to ptr and | |
655 | * delete it. | |
656 | */ | |
657 | static void delete_object_full(unsigned long ptr) | |
658 | { | |
659 | struct kmemleak_object *object; | |
660 | ||
e781a9ab | 661 | object = find_and_remove_object(ptr, 0); |
53238a60 CM |
662 | if (!object) { |
663 | #ifdef DEBUG | |
664 | kmemleak_warn("Freeing unknown object at 0x%08lx\n", | |
665 | ptr); | |
666 | #endif | |
667 | return; | |
668 | } | |
669 | __delete_object(object); | |
53238a60 CM |
670 | } |
671 | ||
672 | /* | |
673 | * Look up the metadata (struct kmemleak_object) corresponding to ptr and | |
674 | * delete it. If the memory block is partially freed, the function may create | |
675 | * additional metadata for the remaining parts of the block. | |
676 | */ | |
677 | static void delete_object_part(unsigned long ptr, size_t size) | |
678 | { | |
679 | struct kmemleak_object *object; | |
680 | unsigned long start, end; | |
681 | ||
e781a9ab | 682 | object = find_and_remove_object(ptr, 1); |
53238a60 CM |
683 | if (!object) { |
684 | #ifdef DEBUG | |
756a025f JP |
685 | kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n", |
686 | ptr, size); | |
53238a60 CM |
687 | #endif |
688 | return; | |
689 | } | |
53238a60 CM |
690 | |
691 | /* | |
692 | * Create one or two objects that may result from the memory block | |
693 | * split. Note that partial freeing is only done by free_bootmem() and | |
694 | * this happens before kmemleak_init() is called. The path below is | |
695 | * only executed during early log recording in kmemleak_init(), so | |
696 | * GFP_KERNEL is enough. | |
697 | */ | |
698 | start = object->pointer; | |
699 | end = object->pointer + object->size; | |
700 | if (ptr > start) | |
701 | create_object(start, ptr - start, object->min_count, | |
702 | GFP_KERNEL); | |
703 | if (ptr + size < end) | |
704 | create_object(ptr + size, end - ptr - size, object->min_count, | |
705 | GFP_KERNEL); | |
706 | ||
e781a9ab | 707 | __delete_object(object); |
53238a60 | 708 | } |
a1084c87 LR |
709 | |
710 | static void __paint_it(struct kmemleak_object *object, int color) | |
711 | { | |
712 | object->min_count = color; | |
713 | if (color == KMEMLEAK_BLACK) | |
714 | object->flags |= OBJECT_NO_SCAN; | |
715 | } | |
716 | ||
717 | static void paint_it(struct kmemleak_object *object, int color) | |
3c7b4e6b CM |
718 | { |
719 | unsigned long flags; | |
a1084c87 LR |
720 | |
721 | spin_lock_irqsave(&object->lock, flags); | |
722 | __paint_it(object, color); | |
723 | spin_unlock_irqrestore(&object->lock, flags); | |
724 | } | |
725 | ||
726 | static void paint_ptr(unsigned long ptr, int color) | |
727 | { | |
3c7b4e6b CM |
728 | struct kmemleak_object *object; |
729 | ||
730 | object = find_and_get_object(ptr, 0); | |
731 | if (!object) { | |
756a025f JP |
732 | kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n", |
733 | ptr, | |
a1084c87 LR |
734 | (color == KMEMLEAK_GREY) ? "Grey" : |
735 | (color == KMEMLEAK_BLACK) ? "Black" : "Unknown"); | |
3c7b4e6b CM |
736 | return; |
737 | } | |
a1084c87 | 738 | paint_it(object, color); |
3c7b4e6b CM |
739 | put_object(object); |
740 | } | |
741 | ||
a1084c87 | 742 | /* |
145b64b9 | 743 | * Mark an object permanently as gray-colored so that it can no longer be |
a1084c87 LR |
744 | * reported as a leak. This is used in general to mark a false positive. |
745 | */ | |
746 | static void make_gray_object(unsigned long ptr) | |
747 | { | |
748 | paint_ptr(ptr, KMEMLEAK_GREY); | |
749 | } | |
750 | ||
3c7b4e6b CM |
751 | /* |
752 | * Mark the object as black-colored so that it is ignored from scans and | |
753 | * reporting. | |
754 | */ | |
755 | static void make_black_object(unsigned long ptr) | |
756 | { | |
a1084c87 | 757 | paint_ptr(ptr, KMEMLEAK_BLACK); |
3c7b4e6b CM |
758 | } |
759 | ||
760 | /* | |
761 | * Add a scanning area to the object. If at least one such area is added, | |
762 | * kmemleak will only scan these ranges rather than the whole memory block. | |
763 | */ | |
c017b4be | 764 | static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp) |
3c7b4e6b CM |
765 | { |
766 | unsigned long flags; | |
767 | struct kmemleak_object *object; | |
768 | struct kmemleak_scan_area *area; | |
769 | ||
c017b4be | 770 | object = find_and_get_object(ptr, 1); |
3c7b4e6b | 771 | if (!object) { |
ae281064 JP |
772 | kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n", |
773 | ptr); | |
3c7b4e6b CM |
774 | return; |
775 | } | |
776 | ||
6ae4bd1f | 777 | area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp)); |
3c7b4e6b CM |
778 | |
779 | spin_lock_irqsave(&object->lock, flags); | |
dba82d94 CM |
780 | if (!area) { |
781 | pr_warn_once("Cannot allocate a scan area, scanning the full object\n"); | |
782 | /* mark the object for full scan to avoid false positives */ | |
783 | object->flags |= OBJECT_FULL_SCAN; | |
784 | goto out_unlock; | |
785 | } | |
7f88f88f CM |
786 | if (size == SIZE_MAX) { |
787 | size = object->pointer + object->size - ptr; | |
788 | } else if (ptr + size > object->pointer + object->size) { | |
ae281064 | 789 | kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr); |
3c7b4e6b CM |
790 | dump_object_info(object); |
791 | kmem_cache_free(scan_area_cache, area); | |
792 | goto out_unlock; | |
793 | } | |
794 | ||
795 | INIT_HLIST_NODE(&area->node); | |
c017b4be CM |
796 | area->start = ptr; |
797 | area->size = size; | |
3c7b4e6b CM |
798 | |
799 | hlist_add_head(&area->node, &object->area_list); | |
800 | out_unlock: | |
801 | spin_unlock_irqrestore(&object->lock, flags); | |
3c7b4e6b CM |
802 | put_object(object); |
803 | } | |
804 | ||
94f4a161 CM |
805 | /* |
806 | * Any surplus references (object already gray) to 'ptr' are passed to | |
807 | * 'excess_ref'. This is used in the vmalloc() case where a pointer to | |
808 | * vm_struct may be used as an alternative reference to the vmalloc'ed object | |
809 | * (see free_thread_stack()). | |
810 | */ | |
811 | static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref) | |
812 | { | |
813 | unsigned long flags; | |
814 | struct kmemleak_object *object; | |
815 | ||
816 | object = find_and_get_object(ptr, 0); | |
817 | if (!object) { | |
818 | kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n", | |
819 | ptr); | |
820 | return; | |
821 | } | |
822 | ||
823 | spin_lock_irqsave(&object->lock, flags); | |
824 | object->excess_ref = excess_ref; | |
825 | spin_unlock_irqrestore(&object->lock, flags); | |
826 | put_object(object); | |
827 | } | |
828 | ||
3c7b4e6b CM |
829 | /* |
830 | * Set the OBJECT_NO_SCAN flag for the object corresponding to the give | |
831 | * pointer. Such object will not be scanned by kmemleak but references to it | |
832 | * are searched. | |
833 | */ | |
834 | static void object_no_scan(unsigned long ptr) | |
835 | { | |
836 | unsigned long flags; | |
837 | struct kmemleak_object *object; | |
838 | ||
839 | object = find_and_get_object(ptr, 0); | |
840 | if (!object) { | |
ae281064 | 841 | kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr); |
3c7b4e6b CM |
842 | return; |
843 | } | |
844 | ||
845 | spin_lock_irqsave(&object->lock, flags); | |
846 | object->flags |= OBJECT_NO_SCAN; | |
847 | spin_unlock_irqrestore(&object->lock, flags); | |
848 | put_object(object); | |
849 | } | |
850 | ||
851 | /* | |
852 | * Log an early kmemleak_* call to the early_log buffer. These calls will be | |
853 | * processed later once kmemleak is fully initialized. | |
854 | */ | |
a6186d89 | 855 | static void __init log_early(int op_type, const void *ptr, size_t size, |
c017b4be | 856 | int min_count) |
3c7b4e6b CM |
857 | { |
858 | unsigned long flags; | |
859 | struct early_log *log; | |
860 | ||
8910ae89 | 861 | if (kmemleak_error) { |
b6693005 CM |
862 | /* kmemleak stopped recording, just count the requests */ |
863 | crt_early_log++; | |
864 | return; | |
865 | } | |
866 | ||
3c7b4e6b | 867 | if (crt_early_log >= ARRAY_SIZE(early_log)) { |
21cd3a60 | 868 | crt_early_log++; |
a9d9058a | 869 | kmemleak_disable(); |
3c7b4e6b CM |
870 | return; |
871 | } | |
872 | ||
873 | /* | |
874 | * There is no need for locking since the kernel is still in UP mode | |
875 | * at this stage. Disabling the IRQs is enough. | |
876 | */ | |
877 | local_irq_save(flags); | |
878 | log = &early_log[crt_early_log]; | |
879 | log->op_type = op_type; | |
880 | log->ptr = ptr; | |
881 | log->size = size; | |
882 | log->min_count = min_count; | |
5f79020c | 883 | log->trace_len = __save_stack_trace(log->trace); |
3c7b4e6b CM |
884 | crt_early_log++; |
885 | local_irq_restore(flags); | |
886 | } | |
887 | ||
fd678967 CM |
888 | /* |
889 | * Log an early allocated block and populate the stack trace. | |
890 | */ | |
891 | static void early_alloc(struct early_log *log) | |
892 | { | |
893 | struct kmemleak_object *object; | |
894 | unsigned long flags; | |
895 | int i; | |
896 | ||
8910ae89 | 897 | if (!kmemleak_enabled || !log->ptr || IS_ERR(log->ptr)) |
fd678967 CM |
898 | return; |
899 | ||
900 | /* | |
901 | * RCU locking needed to ensure object is not freed via put_object(). | |
902 | */ | |
903 | rcu_read_lock(); | |
904 | object = create_object((unsigned long)log->ptr, log->size, | |
c1bcd6b3 | 905 | log->min_count, GFP_ATOMIC); |
0d5d1aad CM |
906 | if (!object) |
907 | goto out; | |
fd678967 CM |
908 | spin_lock_irqsave(&object->lock, flags); |
909 | for (i = 0; i < log->trace_len; i++) | |
910 | object->trace[i] = log->trace[i]; | |
911 | object->trace_len = log->trace_len; | |
912 | spin_unlock_irqrestore(&object->lock, flags); | |
0d5d1aad | 913 | out: |
fd678967 CM |
914 | rcu_read_unlock(); |
915 | } | |
916 | ||
f528f0b8 CM |
917 | /* |
918 | * Log an early allocated block and populate the stack trace. | |
919 | */ | |
920 | static void early_alloc_percpu(struct early_log *log) | |
921 | { | |
922 | unsigned int cpu; | |
923 | const void __percpu *ptr = log->ptr; | |
924 | ||
925 | for_each_possible_cpu(cpu) { | |
926 | log->ptr = per_cpu_ptr(ptr, cpu); | |
927 | early_alloc(log); | |
928 | } | |
929 | } | |
930 | ||
a2b6bf63 CM |
931 | /** |
932 | * kmemleak_alloc - register a newly allocated object | |
933 | * @ptr: pointer to beginning of the object | |
934 | * @size: size of the object | |
935 | * @min_count: minimum number of references to this object. If during memory | |
936 | * scanning a number of references less than @min_count is found, | |
937 | * the object is reported as a memory leak. If @min_count is 0, | |
938 | * the object is never reported as a leak. If @min_count is -1, | |
939 | * the object is ignored (not scanned and not reported as a leak) | |
940 | * @gfp: kmalloc() flags used for kmemleak internal memory allocations | |
941 | * | |
942 | * This function is called from the kernel allocators when a new object | |
94f4a161 | 943 | * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.). |
3c7b4e6b | 944 | */ |
a6186d89 CM |
945 | void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count, |
946 | gfp_t gfp) | |
3c7b4e6b CM |
947 | { |
948 | pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count); | |
949 | ||
8910ae89 | 950 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
3c7b4e6b | 951 | create_object((unsigned long)ptr, size, min_count, gfp); |
8910ae89 | 952 | else if (kmemleak_early_log) |
c017b4be | 953 | log_early(KMEMLEAK_ALLOC, ptr, size, min_count); |
3c7b4e6b CM |
954 | } |
955 | EXPORT_SYMBOL_GPL(kmemleak_alloc); | |
956 | ||
f528f0b8 CM |
957 | /** |
958 | * kmemleak_alloc_percpu - register a newly allocated __percpu object | |
959 | * @ptr: __percpu pointer to beginning of the object | |
960 | * @size: size of the object | |
8a8c35fa | 961 | * @gfp: flags used for kmemleak internal memory allocations |
f528f0b8 CM |
962 | * |
963 | * This function is called from the kernel percpu allocator when a new object | |
8a8c35fa | 964 | * (memory block) is allocated (alloc_percpu). |
f528f0b8 | 965 | */ |
8a8c35fa LF |
966 | void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size, |
967 | gfp_t gfp) | |
f528f0b8 CM |
968 | { |
969 | unsigned int cpu; | |
970 | ||
971 | pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size); | |
972 | ||
973 | /* | |
974 | * Percpu allocations are only scanned and not reported as leaks | |
975 | * (min_count is set to 0). | |
976 | */ | |
8910ae89 | 977 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
f528f0b8 CM |
978 | for_each_possible_cpu(cpu) |
979 | create_object((unsigned long)per_cpu_ptr(ptr, cpu), | |
8a8c35fa | 980 | size, 0, gfp); |
8910ae89 | 981 | else if (kmemleak_early_log) |
f528f0b8 CM |
982 | log_early(KMEMLEAK_ALLOC_PERCPU, ptr, size, 0); |
983 | } | |
984 | EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu); | |
985 | ||
94f4a161 CM |
986 | /** |
987 | * kmemleak_vmalloc - register a newly vmalloc'ed object | |
988 | * @area: pointer to vm_struct | |
989 | * @size: size of the object | |
990 | * @gfp: __vmalloc() flags used for kmemleak internal memory allocations | |
991 | * | |
992 | * This function is called from the vmalloc() kernel allocator when a new | |
993 | * object (memory block) is allocated. | |
994 | */ | |
995 | void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp) | |
996 | { | |
997 | pr_debug("%s(0x%p, %zu)\n", __func__, area, size); | |
998 | ||
999 | /* | |
1000 | * A min_count = 2 is needed because vm_struct contains a reference to | |
1001 | * the virtual address of the vmalloc'ed block. | |
1002 | */ | |
1003 | if (kmemleak_enabled) { | |
1004 | create_object((unsigned long)area->addr, size, 2, gfp); | |
1005 | object_set_excess_ref((unsigned long)area, | |
1006 | (unsigned long)area->addr); | |
1007 | } else if (kmemleak_early_log) { | |
1008 | log_early(KMEMLEAK_ALLOC, area->addr, size, 2); | |
1009 | /* reusing early_log.size for storing area->addr */ | |
1010 | log_early(KMEMLEAK_SET_EXCESS_REF, | |
1011 | area, (unsigned long)area->addr, 0); | |
1012 | } | |
1013 | } | |
1014 | EXPORT_SYMBOL_GPL(kmemleak_vmalloc); | |
1015 | ||
a2b6bf63 CM |
1016 | /** |
1017 | * kmemleak_free - unregister a previously registered object | |
1018 | * @ptr: pointer to beginning of the object | |
1019 | * | |
1020 | * This function is called from the kernel allocators when an object (memory | |
1021 | * block) is freed (kmem_cache_free, kfree, vfree etc.). | |
3c7b4e6b | 1022 | */ |
a6186d89 | 1023 | void __ref kmemleak_free(const void *ptr) |
3c7b4e6b CM |
1024 | { |
1025 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1026 | ||
c5f3b1a5 | 1027 | if (kmemleak_free_enabled && ptr && !IS_ERR(ptr)) |
53238a60 | 1028 | delete_object_full((unsigned long)ptr); |
8910ae89 | 1029 | else if (kmemleak_early_log) |
c017b4be | 1030 | log_early(KMEMLEAK_FREE, ptr, 0, 0); |
3c7b4e6b CM |
1031 | } |
1032 | EXPORT_SYMBOL_GPL(kmemleak_free); | |
1033 | ||
a2b6bf63 CM |
1034 | /** |
1035 | * kmemleak_free_part - partially unregister a previously registered object | |
1036 | * @ptr: pointer to the beginning or inside the object. This also | |
1037 | * represents the start of the range to be freed | |
1038 | * @size: size to be unregistered | |
1039 | * | |
1040 | * This function is called when only a part of a memory block is freed | |
1041 | * (usually from the bootmem allocator). | |
53238a60 | 1042 | */ |
a6186d89 | 1043 | void __ref kmemleak_free_part(const void *ptr, size_t size) |
53238a60 CM |
1044 | { |
1045 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1046 | ||
8910ae89 | 1047 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
53238a60 | 1048 | delete_object_part((unsigned long)ptr, size); |
8910ae89 | 1049 | else if (kmemleak_early_log) |
c017b4be | 1050 | log_early(KMEMLEAK_FREE_PART, ptr, size, 0); |
53238a60 CM |
1051 | } |
1052 | EXPORT_SYMBOL_GPL(kmemleak_free_part); | |
1053 | ||
f528f0b8 CM |
1054 | /** |
1055 | * kmemleak_free_percpu - unregister a previously registered __percpu object | |
1056 | * @ptr: __percpu pointer to beginning of the object | |
1057 | * | |
1058 | * This function is called from the kernel percpu allocator when an object | |
1059 | * (memory block) is freed (free_percpu). | |
1060 | */ | |
1061 | void __ref kmemleak_free_percpu(const void __percpu *ptr) | |
1062 | { | |
1063 | unsigned int cpu; | |
1064 | ||
1065 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1066 | ||
c5f3b1a5 | 1067 | if (kmemleak_free_enabled && ptr && !IS_ERR(ptr)) |
f528f0b8 CM |
1068 | for_each_possible_cpu(cpu) |
1069 | delete_object_full((unsigned long)per_cpu_ptr(ptr, | |
1070 | cpu)); | |
8910ae89 | 1071 | else if (kmemleak_early_log) |
f528f0b8 CM |
1072 | log_early(KMEMLEAK_FREE_PERCPU, ptr, 0, 0); |
1073 | } | |
1074 | EXPORT_SYMBOL_GPL(kmemleak_free_percpu); | |
1075 | ||
ffe2c748 CM |
1076 | /** |
1077 | * kmemleak_update_trace - update object allocation stack trace | |
1078 | * @ptr: pointer to beginning of the object | |
1079 | * | |
1080 | * Override the object allocation stack trace for cases where the actual | |
1081 | * allocation place is not always useful. | |
1082 | */ | |
1083 | void __ref kmemleak_update_trace(const void *ptr) | |
1084 | { | |
1085 | struct kmemleak_object *object; | |
1086 | unsigned long flags; | |
1087 | ||
1088 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1089 | ||
1090 | if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr)) | |
1091 | return; | |
1092 | ||
1093 | object = find_and_get_object((unsigned long)ptr, 1); | |
1094 | if (!object) { | |
1095 | #ifdef DEBUG | |
1096 | kmemleak_warn("Updating stack trace for unknown object at %p\n", | |
1097 | ptr); | |
1098 | #endif | |
1099 | return; | |
1100 | } | |
1101 | ||
1102 | spin_lock_irqsave(&object->lock, flags); | |
1103 | object->trace_len = __save_stack_trace(object->trace); | |
1104 | spin_unlock_irqrestore(&object->lock, flags); | |
1105 | ||
1106 | put_object(object); | |
1107 | } | |
1108 | EXPORT_SYMBOL(kmemleak_update_trace); | |
1109 | ||
a2b6bf63 CM |
1110 | /** |
1111 | * kmemleak_not_leak - mark an allocated object as false positive | |
1112 | * @ptr: pointer to beginning of the object | |
1113 | * | |
1114 | * Calling this function on an object will cause the memory block to no longer | |
1115 | * be reported as leak and always be scanned. | |
3c7b4e6b | 1116 | */ |
a6186d89 | 1117 | void __ref kmemleak_not_leak(const void *ptr) |
3c7b4e6b CM |
1118 | { |
1119 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1120 | ||
8910ae89 | 1121 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
3c7b4e6b | 1122 | make_gray_object((unsigned long)ptr); |
8910ae89 | 1123 | else if (kmemleak_early_log) |
c017b4be | 1124 | log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0); |
3c7b4e6b CM |
1125 | } |
1126 | EXPORT_SYMBOL(kmemleak_not_leak); | |
1127 | ||
a2b6bf63 CM |
1128 | /** |
1129 | * kmemleak_ignore - ignore an allocated object | |
1130 | * @ptr: pointer to beginning of the object | |
1131 | * | |
1132 | * Calling this function on an object will cause the memory block to be | |
1133 | * ignored (not scanned and not reported as a leak). This is usually done when | |
1134 | * it is known that the corresponding block is not a leak and does not contain | |
1135 | * any references to other allocated memory blocks. | |
3c7b4e6b | 1136 | */ |
a6186d89 | 1137 | void __ref kmemleak_ignore(const void *ptr) |
3c7b4e6b CM |
1138 | { |
1139 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1140 | ||
8910ae89 | 1141 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
3c7b4e6b | 1142 | make_black_object((unsigned long)ptr); |
8910ae89 | 1143 | else if (kmemleak_early_log) |
c017b4be | 1144 | log_early(KMEMLEAK_IGNORE, ptr, 0, 0); |
3c7b4e6b CM |
1145 | } |
1146 | EXPORT_SYMBOL(kmemleak_ignore); | |
1147 | ||
a2b6bf63 CM |
1148 | /** |
1149 | * kmemleak_scan_area - limit the range to be scanned in an allocated object | |
1150 | * @ptr: pointer to beginning or inside the object. This also | |
1151 | * represents the start of the scan area | |
1152 | * @size: size of the scan area | |
1153 | * @gfp: kmalloc() flags used for kmemleak internal memory allocations | |
1154 | * | |
1155 | * This function is used when it is known that only certain parts of an object | |
1156 | * contain references to other objects. Kmemleak will only scan these areas | |
1157 | * reducing the number false negatives. | |
3c7b4e6b | 1158 | */ |
c017b4be | 1159 | void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp) |
3c7b4e6b CM |
1160 | { |
1161 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1162 | ||
8910ae89 | 1163 | if (kmemleak_enabled && ptr && size && !IS_ERR(ptr)) |
c017b4be | 1164 | add_scan_area((unsigned long)ptr, size, gfp); |
8910ae89 | 1165 | else if (kmemleak_early_log) |
c017b4be | 1166 | log_early(KMEMLEAK_SCAN_AREA, ptr, size, 0); |
3c7b4e6b CM |
1167 | } |
1168 | EXPORT_SYMBOL(kmemleak_scan_area); | |
1169 | ||
a2b6bf63 CM |
1170 | /** |
1171 | * kmemleak_no_scan - do not scan an allocated object | |
1172 | * @ptr: pointer to beginning of the object | |
1173 | * | |
1174 | * This function notifies kmemleak not to scan the given memory block. Useful | |
1175 | * in situations where it is known that the given object does not contain any | |
1176 | * references to other objects. Kmemleak will not scan such objects reducing | |
1177 | * the number of false negatives. | |
3c7b4e6b | 1178 | */ |
a6186d89 | 1179 | void __ref kmemleak_no_scan(const void *ptr) |
3c7b4e6b CM |
1180 | { |
1181 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1182 | ||
8910ae89 | 1183 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
3c7b4e6b | 1184 | object_no_scan((unsigned long)ptr); |
8910ae89 | 1185 | else if (kmemleak_early_log) |
c017b4be | 1186 | log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0); |
3c7b4e6b CM |
1187 | } |
1188 | EXPORT_SYMBOL(kmemleak_no_scan); | |
1189 | ||
9099daed CM |
1190 | /** |
1191 | * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical | |
1192 | * address argument | |
e8b098fc MR |
1193 | * @phys: physical address of the object |
1194 | * @size: size of the object | |
1195 | * @min_count: minimum number of references to this object. | |
1196 | * See kmemleak_alloc() | |
1197 | * @gfp: kmalloc() flags used for kmemleak internal memory allocations | |
9099daed CM |
1198 | */ |
1199 | void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, int min_count, | |
1200 | gfp_t gfp) | |
1201 | { | |
1202 | if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn) | |
1203 | kmemleak_alloc(__va(phys), size, min_count, gfp); | |
1204 | } | |
1205 | EXPORT_SYMBOL(kmemleak_alloc_phys); | |
1206 | ||
1207 | /** | |
1208 | * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a | |
1209 | * physical address argument | |
e8b098fc MR |
1210 | * @phys: physical address if the beginning or inside an object. This |
1211 | * also represents the start of the range to be freed | |
1212 | * @size: size to be unregistered | |
9099daed CM |
1213 | */ |
1214 | void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size) | |
1215 | { | |
1216 | if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn) | |
1217 | kmemleak_free_part(__va(phys), size); | |
1218 | } | |
1219 | EXPORT_SYMBOL(kmemleak_free_part_phys); | |
1220 | ||
1221 | /** | |
1222 | * kmemleak_not_leak_phys - similar to kmemleak_not_leak but taking a physical | |
1223 | * address argument | |
e8b098fc | 1224 | * @phys: physical address of the object |
9099daed CM |
1225 | */ |
1226 | void __ref kmemleak_not_leak_phys(phys_addr_t phys) | |
1227 | { | |
1228 | if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn) | |
1229 | kmemleak_not_leak(__va(phys)); | |
1230 | } | |
1231 | EXPORT_SYMBOL(kmemleak_not_leak_phys); | |
1232 | ||
1233 | /** | |
1234 | * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical | |
1235 | * address argument | |
e8b098fc | 1236 | * @phys: physical address of the object |
9099daed CM |
1237 | */ |
1238 | void __ref kmemleak_ignore_phys(phys_addr_t phys) | |
1239 | { | |
1240 | if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn) | |
1241 | kmemleak_ignore(__va(phys)); | |
1242 | } | |
1243 | EXPORT_SYMBOL(kmemleak_ignore_phys); | |
1244 | ||
04609ccc CM |
1245 | /* |
1246 | * Update an object's checksum and return true if it was modified. | |
1247 | */ | |
1248 | static bool update_checksum(struct kmemleak_object *object) | |
1249 | { | |
1250 | u32 old_csum = object->checksum; | |
1251 | ||
e79ed2f1 | 1252 | kasan_disable_current(); |
04609ccc | 1253 | object->checksum = crc32(0, (void *)object->pointer, object->size); |
e79ed2f1 AR |
1254 | kasan_enable_current(); |
1255 | ||
04609ccc CM |
1256 | return object->checksum != old_csum; |
1257 | } | |
1258 | ||
04f70d13 CM |
1259 | /* |
1260 | * Update an object's references. object->lock must be held by the caller. | |
1261 | */ | |
1262 | static void update_refs(struct kmemleak_object *object) | |
1263 | { | |
1264 | if (!color_white(object)) { | |
1265 | /* non-orphan, ignored or new */ | |
1266 | return; | |
1267 | } | |
1268 | ||
1269 | /* | |
1270 | * Increase the object's reference count (number of pointers to the | |
1271 | * memory block). If this count reaches the required minimum, the | |
1272 | * object's color will become gray and it will be added to the | |
1273 | * gray_list. | |
1274 | */ | |
1275 | object->count++; | |
1276 | if (color_gray(object)) { | |
1277 | /* put_object() called when removing from gray_list */ | |
1278 | WARN_ON(!get_object(object)); | |
1279 | list_add_tail(&object->gray_list, &gray_list); | |
1280 | } | |
1281 | } | |
1282 | ||
3c7b4e6b CM |
1283 | /* |
1284 | * Memory scanning is a long process and it needs to be interruptable. This | |
25985edc | 1285 | * function checks whether such interrupt condition occurred. |
3c7b4e6b CM |
1286 | */ |
1287 | static int scan_should_stop(void) | |
1288 | { | |
8910ae89 | 1289 | if (!kmemleak_enabled) |
3c7b4e6b CM |
1290 | return 1; |
1291 | ||
1292 | /* | |
1293 | * This function may be called from either process or kthread context, | |
1294 | * hence the need to check for both stop conditions. | |
1295 | */ | |
1296 | if (current->mm) | |
1297 | return signal_pending(current); | |
1298 | else | |
1299 | return kthread_should_stop(); | |
1300 | ||
1301 | return 0; | |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * Scan a memory block (exclusive range) for valid pointers and add those | |
1306 | * found to the gray list. | |
1307 | */ | |
1308 | static void scan_block(void *_start, void *_end, | |
93ada579 | 1309 | struct kmemleak_object *scanned) |
3c7b4e6b CM |
1310 | { |
1311 | unsigned long *ptr; | |
1312 | unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER); | |
1313 | unsigned long *end = _end - (BYTES_PER_POINTER - 1); | |
93ada579 | 1314 | unsigned long flags; |
a2f77575 | 1315 | unsigned long untagged_ptr; |
3c7b4e6b | 1316 | |
93ada579 | 1317 | read_lock_irqsave(&kmemleak_lock, flags); |
3c7b4e6b | 1318 | for (ptr = start; ptr < end; ptr++) { |
3c7b4e6b | 1319 | struct kmemleak_object *object; |
8e019366 | 1320 | unsigned long pointer; |
94f4a161 | 1321 | unsigned long excess_ref; |
3c7b4e6b CM |
1322 | |
1323 | if (scan_should_stop()) | |
1324 | break; | |
1325 | ||
e79ed2f1 | 1326 | kasan_disable_current(); |
8e019366 | 1327 | pointer = *ptr; |
e79ed2f1 | 1328 | kasan_enable_current(); |
8e019366 | 1329 | |
a2f77575 AK |
1330 | untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer); |
1331 | if (untagged_ptr < min_addr || untagged_ptr >= max_addr) | |
93ada579 CM |
1332 | continue; |
1333 | ||
1334 | /* | |
1335 | * No need for get_object() here since we hold kmemleak_lock. | |
1336 | * object->use_count cannot be dropped to 0 while the object | |
1337 | * is still present in object_tree_root and object_list | |
1338 | * (with updates protected by kmemleak_lock). | |
1339 | */ | |
1340 | object = lookup_object(pointer, 1); | |
3c7b4e6b CM |
1341 | if (!object) |
1342 | continue; | |
93ada579 | 1343 | if (object == scanned) |
3c7b4e6b | 1344 | /* self referenced, ignore */ |
3c7b4e6b | 1345 | continue; |
3c7b4e6b CM |
1346 | |
1347 | /* | |
1348 | * Avoid the lockdep recursive warning on object->lock being | |
1349 | * previously acquired in scan_object(). These locks are | |
1350 | * enclosed by scan_mutex. | |
1351 | */ | |
93ada579 | 1352 | spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING); |
94f4a161 CM |
1353 | /* only pass surplus references (object already gray) */ |
1354 | if (color_gray(object)) { | |
1355 | excess_ref = object->excess_ref; | |
1356 | /* no need for update_refs() if object already gray */ | |
1357 | } else { | |
1358 | excess_ref = 0; | |
1359 | update_refs(object); | |
1360 | } | |
93ada579 | 1361 | spin_unlock(&object->lock); |
94f4a161 CM |
1362 | |
1363 | if (excess_ref) { | |
1364 | object = lookup_object(excess_ref, 0); | |
1365 | if (!object) | |
1366 | continue; | |
1367 | if (object == scanned) | |
1368 | /* circular reference, ignore */ | |
1369 | continue; | |
1370 | spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING); | |
1371 | update_refs(object); | |
1372 | spin_unlock(&object->lock); | |
1373 | } | |
93ada579 CM |
1374 | } |
1375 | read_unlock_irqrestore(&kmemleak_lock, flags); | |
1376 | } | |
0587da40 | 1377 | |
93ada579 CM |
1378 | /* |
1379 | * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency. | |
1380 | */ | |
dce5b0bd | 1381 | #ifdef CONFIG_SMP |
93ada579 CM |
1382 | static void scan_large_block(void *start, void *end) |
1383 | { | |
1384 | void *next; | |
1385 | ||
1386 | while (start < end) { | |
1387 | next = min(start + MAX_SCAN_SIZE, end); | |
1388 | scan_block(start, next, NULL); | |
1389 | start = next; | |
1390 | cond_resched(); | |
3c7b4e6b CM |
1391 | } |
1392 | } | |
dce5b0bd | 1393 | #endif |
3c7b4e6b CM |
1394 | |
1395 | /* | |
1396 | * Scan a memory block corresponding to a kmemleak_object. A condition is | |
1397 | * that object->use_count >= 1. | |
1398 | */ | |
1399 | static void scan_object(struct kmemleak_object *object) | |
1400 | { | |
1401 | struct kmemleak_scan_area *area; | |
3c7b4e6b CM |
1402 | unsigned long flags; |
1403 | ||
1404 | /* | |
21ae2956 UKK |
1405 | * Once the object->lock is acquired, the corresponding memory block |
1406 | * cannot be freed (the same lock is acquired in delete_object). | |
3c7b4e6b CM |
1407 | */ |
1408 | spin_lock_irqsave(&object->lock, flags); | |
1409 | if (object->flags & OBJECT_NO_SCAN) | |
1410 | goto out; | |
1411 | if (!(object->flags & OBJECT_ALLOCATED)) | |
1412 | /* already freed object */ | |
1413 | goto out; | |
dba82d94 CM |
1414 | if (hlist_empty(&object->area_list) || |
1415 | object->flags & OBJECT_FULL_SCAN) { | |
af98603d CM |
1416 | void *start = (void *)object->pointer; |
1417 | void *end = (void *)(object->pointer + object->size); | |
93ada579 CM |
1418 | void *next; |
1419 | ||
1420 | do { | |
1421 | next = min(start + MAX_SCAN_SIZE, end); | |
1422 | scan_block(start, next, object); | |
af98603d | 1423 | |
93ada579 CM |
1424 | start = next; |
1425 | if (start >= end) | |
1426 | break; | |
af98603d CM |
1427 | |
1428 | spin_unlock_irqrestore(&object->lock, flags); | |
1429 | cond_resched(); | |
1430 | spin_lock_irqsave(&object->lock, flags); | |
93ada579 | 1431 | } while (object->flags & OBJECT_ALLOCATED); |
af98603d | 1432 | } else |
b67bfe0d | 1433 | hlist_for_each_entry(area, &object->area_list, node) |
c017b4be CM |
1434 | scan_block((void *)area->start, |
1435 | (void *)(area->start + area->size), | |
93ada579 | 1436 | object); |
3c7b4e6b CM |
1437 | out: |
1438 | spin_unlock_irqrestore(&object->lock, flags); | |
1439 | } | |
1440 | ||
04609ccc CM |
1441 | /* |
1442 | * Scan the objects already referenced (gray objects). More objects will be | |
1443 | * referenced and, if there are no memory leaks, all the objects are scanned. | |
1444 | */ | |
1445 | static void scan_gray_list(void) | |
1446 | { | |
1447 | struct kmemleak_object *object, *tmp; | |
1448 | ||
1449 | /* | |
1450 | * The list traversal is safe for both tail additions and removals | |
1451 | * from inside the loop. The kmemleak objects cannot be freed from | |
1452 | * outside the loop because their use_count was incremented. | |
1453 | */ | |
1454 | object = list_entry(gray_list.next, typeof(*object), gray_list); | |
1455 | while (&object->gray_list != &gray_list) { | |
1456 | cond_resched(); | |
1457 | ||
1458 | /* may add new objects to the list */ | |
1459 | if (!scan_should_stop()) | |
1460 | scan_object(object); | |
1461 | ||
1462 | tmp = list_entry(object->gray_list.next, typeof(*object), | |
1463 | gray_list); | |
1464 | ||
1465 | /* remove the object from the list and release it */ | |
1466 | list_del(&object->gray_list); | |
1467 | put_object(object); | |
1468 | ||
1469 | object = tmp; | |
1470 | } | |
1471 | WARN_ON(!list_empty(&gray_list)); | |
1472 | } | |
1473 | ||
3c7b4e6b CM |
1474 | /* |
1475 | * Scan data sections and all the referenced memory blocks allocated via the | |
1476 | * kernel's standard allocators. This function must be called with the | |
1477 | * scan_mutex held. | |
1478 | */ | |
1479 | static void kmemleak_scan(void) | |
1480 | { | |
1481 | unsigned long flags; | |
04609ccc | 1482 | struct kmemleak_object *object; |
3c7b4e6b | 1483 | int i; |
4698c1f2 | 1484 | int new_leaks = 0; |
3c7b4e6b | 1485 | |
acf4968e CM |
1486 | jiffies_last_scan = jiffies; |
1487 | ||
3c7b4e6b CM |
1488 | /* prepare the kmemleak_object's */ |
1489 | rcu_read_lock(); | |
1490 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1491 | spin_lock_irqsave(&object->lock, flags); | |
1492 | #ifdef DEBUG | |
1493 | /* | |
1494 | * With a few exceptions there should be a maximum of | |
1495 | * 1 reference to any object at this point. | |
1496 | */ | |
1497 | if (atomic_read(&object->use_count) > 1) { | |
ae281064 | 1498 | pr_debug("object->use_count = %d\n", |
3c7b4e6b CM |
1499 | atomic_read(&object->use_count)); |
1500 | dump_object_info(object); | |
1501 | } | |
1502 | #endif | |
1503 | /* reset the reference count (whiten the object) */ | |
1504 | object->count = 0; | |
1505 | if (color_gray(object) && get_object(object)) | |
1506 | list_add_tail(&object->gray_list, &gray_list); | |
1507 | ||
1508 | spin_unlock_irqrestore(&object->lock, flags); | |
1509 | } | |
1510 | rcu_read_unlock(); | |
1511 | ||
3c7b4e6b CM |
1512 | #ifdef CONFIG_SMP |
1513 | /* per-cpu sections scanning */ | |
1514 | for_each_possible_cpu(i) | |
93ada579 CM |
1515 | scan_large_block(__per_cpu_start + per_cpu_offset(i), |
1516 | __per_cpu_end + per_cpu_offset(i)); | |
3c7b4e6b CM |
1517 | #endif |
1518 | ||
1519 | /* | |
029aeff5 | 1520 | * Struct page scanning for each node. |
3c7b4e6b | 1521 | */ |
bfc8c901 | 1522 | get_online_mems(); |
3c7b4e6b | 1523 | for_each_online_node(i) { |
108bcc96 CS |
1524 | unsigned long start_pfn = node_start_pfn(i); |
1525 | unsigned long end_pfn = node_end_pfn(i); | |
3c7b4e6b CM |
1526 | unsigned long pfn; |
1527 | ||
1528 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
9f1eb38e | 1529 | struct page *page = pfn_to_online_page(pfn); |
3c7b4e6b | 1530 | |
9f1eb38e OS |
1531 | if (!page) |
1532 | continue; | |
1533 | ||
1534 | /* only scan pages belonging to this node */ | |
1535 | if (page_to_nid(page) != i) | |
3c7b4e6b | 1536 | continue; |
3c7b4e6b CM |
1537 | /* only scan if page is in use */ |
1538 | if (page_count(page) == 0) | |
1539 | continue; | |
93ada579 | 1540 | scan_block(page, page + 1, NULL); |
13ab183d | 1541 | if (!(pfn & 63)) |
bde5f6bc | 1542 | cond_resched(); |
3c7b4e6b CM |
1543 | } |
1544 | } | |
bfc8c901 | 1545 | put_online_mems(); |
3c7b4e6b CM |
1546 | |
1547 | /* | |
43ed5d6e | 1548 | * Scanning the task stacks (may introduce false negatives). |
3c7b4e6b CM |
1549 | */ |
1550 | if (kmemleak_stack_scan) { | |
43ed5d6e CM |
1551 | struct task_struct *p, *g; |
1552 | ||
3c7b4e6b | 1553 | read_lock(&tasklist_lock); |
43ed5d6e | 1554 | do_each_thread(g, p) { |
37df49f4 CM |
1555 | void *stack = try_get_task_stack(p); |
1556 | if (stack) { | |
1557 | scan_block(stack, stack + THREAD_SIZE, NULL); | |
1558 | put_task_stack(p); | |
1559 | } | |
43ed5d6e | 1560 | } while_each_thread(g, p); |
3c7b4e6b CM |
1561 | read_unlock(&tasklist_lock); |
1562 | } | |
1563 | ||
1564 | /* | |
1565 | * Scan the objects already referenced from the sections scanned | |
04609ccc | 1566 | * above. |
3c7b4e6b | 1567 | */ |
04609ccc | 1568 | scan_gray_list(); |
2587362e CM |
1569 | |
1570 | /* | |
04609ccc CM |
1571 | * Check for new or unreferenced objects modified since the previous |
1572 | * scan and color them gray until the next scan. | |
2587362e CM |
1573 | */ |
1574 | rcu_read_lock(); | |
1575 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1576 | spin_lock_irqsave(&object->lock, flags); | |
04609ccc CM |
1577 | if (color_white(object) && (object->flags & OBJECT_ALLOCATED) |
1578 | && update_checksum(object) && get_object(object)) { | |
1579 | /* color it gray temporarily */ | |
1580 | object->count = object->min_count; | |
2587362e CM |
1581 | list_add_tail(&object->gray_list, &gray_list); |
1582 | } | |
1583 | spin_unlock_irqrestore(&object->lock, flags); | |
1584 | } | |
1585 | rcu_read_unlock(); | |
1586 | ||
04609ccc CM |
1587 | /* |
1588 | * Re-scan the gray list for modified unreferenced objects. | |
1589 | */ | |
1590 | scan_gray_list(); | |
4698c1f2 | 1591 | |
17bb9e0d | 1592 | /* |
04609ccc | 1593 | * If scanning was stopped do not report any new unreferenced objects. |
17bb9e0d | 1594 | */ |
04609ccc | 1595 | if (scan_should_stop()) |
17bb9e0d CM |
1596 | return; |
1597 | ||
4698c1f2 CM |
1598 | /* |
1599 | * Scanning result reporting. | |
1600 | */ | |
1601 | rcu_read_lock(); | |
1602 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1603 | spin_lock_irqsave(&object->lock, flags); | |
1604 | if (unreferenced_object(object) && | |
1605 | !(object->flags & OBJECT_REPORTED)) { | |
1606 | object->flags |= OBJECT_REPORTED; | |
154221c3 VW |
1607 | |
1608 | if (kmemleak_verbose) | |
1609 | print_unreferenced(NULL, object); | |
1610 | ||
4698c1f2 CM |
1611 | new_leaks++; |
1612 | } | |
1613 | spin_unlock_irqrestore(&object->lock, flags); | |
1614 | } | |
1615 | rcu_read_unlock(); | |
1616 | ||
dc9b3f42 LZ |
1617 | if (new_leaks) { |
1618 | kmemleak_found_leaks = true; | |
1619 | ||
756a025f JP |
1620 | pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n", |
1621 | new_leaks); | |
dc9b3f42 | 1622 | } |
4698c1f2 | 1623 | |
3c7b4e6b CM |
1624 | } |
1625 | ||
1626 | /* | |
1627 | * Thread function performing automatic memory scanning. Unreferenced objects | |
1628 | * at the end of a memory scan are reported but only the first time. | |
1629 | */ | |
1630 | static int kmemleak_scan_thread(void *arg) | |
1631 | { | |
d53ce042 | 1632 | static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN); |
3c7b4e6b | 1633 | |
ae281064 | 1634 | pr_info("Automatic memory scanning thread started\n"); |
bf2a76b3 | 1635 | set_user_nice(current, 10); |
3c7b4e6b CM |
1636 | |
1637 | /* | |
1638 | * Wait before the first scan to allow the system to fully initialize. | |
1639 | */ | |
1640 | if (first_run) { | |
98c42d94 | 1641 | signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000); |
3c7b4e6b | 1642 | first_run = 0; |
98c42d94 VN |
1643 | while (timeout && !kthread_should_stop()) |
1644 | timeout = schedule_timeout_interruptible(timeout); | |
3c7b4e6b CM |
1645 | } |
1646 | ||
1647 | while (!kthread_should_stop()) { | |
3c7b4e6b CM |
1648 | signed long timeout = jiffies_scan_wait; |
1649 | ||
1650 | mutex_lock(&scan_mutex); | |
3c7b4e6b | 1651 | kmemleak_scan(); |
3c7b4e6b | 1652 | mutex_unlock(&scan_mutex); |
4698c1f2 | 1653 | |
3c7b4e6b CM |
1654 | /* wait before the next scan */ |
1655 | while (timeout && !kthread_should_stop()) | |
1656 | timeout = schedule_timeout_interruptible(timeout); | |
1657 | } | |
1658 | ||
ae281064 | 1659 | pr_info("Automatic memory scanning thread ended\n"); |
3c7b4e6b CM |
1660 | |
1661 | return 0; | |
1662 | } | |
1663 | ||
1664 | /* | |
1665 | * Start the automatic memory scanning thread. This function must be called | |
4698c1f2 | 1666 | * with the scan_mutex held. |
3c7b4e6b | 1667 | */ |
7eb0d5e5 | 1668 | static void start_scan_thread(void) |
3c7b4e6b CM |
1669 | { |
1670 | if (scan_thread) | |
1671 | return; | |
1672 | scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak"); | |
1673 | if (IS_ERR(scan_thread)) { | |
598d8091 | 1674 | pr_warn("Failed to create the scan thread\n"); |
3c7b4e6b CM |
1675 | scan_thread = NULL; |
1676 | } | |
1677 | } | |
1678 | ||
1679 | /* | |
914b6dff | 1680 | * Stop the automatic memory scanning thread. |
3c7b4e6b | 1681 | */ |
7eb0d5e5 | 1682 | static void stop_scan_thread(void) |
3c7b4e6b CM |
1683 | { |
1684 | if (scan_thread) { | |
1685 | kthread_stop(scan_thread); | |
1686 | scan_thread = NULL; | |
1687 | } | |
1688 | } | |
1689 | ||
1690 | /* | |
1691 | * Iterate over the object_list and return the first valid object at or after | |
1692 | * the required position with its use_count incremented. The function triggers | |
1693 | * a memory scanning when the pos argument points to the first position. | |
1694 | */ | |
1695 | static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos) | |
1696 | { | |
1697 | struct kmemleak_object *object; | |
1698 | loff_t n = *pos; | |
b87324d0 CM |
1699 | int err; |
1700 | ||
1701 | err = mutex_lock_interruptible(&scan_mutex); | |
1702 | if (err < 0) | |
1703 | return ERR_PTR(err); | |
3c7b4e6b | 1704 | |
3c7b4e6b CM |
1705 | rcu_read_lock(); |
1706 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1707 | if (n-- > 0) | |
1708 | continue; | |
1709 | if (get_object(object)) | |
1710 | goto out; | |
1711 | } | |
1712 | object = NULL; | |
1713 | out: | |
3c7b4e6b CM |
1714 | return object; |
1715 | } | |
1716 | ||
1717 | /* | |
1718 | * Return the next object in the object_list. The function decrements the | |
1719 | * use_count of the previous object and increases that of the next one. | |
1720 | */ | |
1721 | static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
1722 | { | |
1723 | struct kmemleak_object *prev_obj = v; | |
1724 | struct kmemleak_object *next_obj = NULL; | |
58fac095 | 1725 | struct kmemleak_object *obj = prev_obj; |
3c7b4e6b CM |
1726 | |
1727 | ++(*pos); | |
3c7b4e6b | 1728 | |
58fac095 | 1729 | list_for_each_entry_continue_rcu(obj, &object_list, object_list) { |
52c3ce4e CM |
1730 | if (get_object(obj)) { |
1731 | next_obj = obj; | |
3c7b4e6b | 1732 | break; |
52c3ce4e | 1733 | } |
3c7b4e6b | 1734 | } |
288c857d | 1735 | |
3c7b4e6b CM |
1736 | put_object(prev_obj); |
1737 | return next_obj; | |
1738 | } | |
1739 | ||
1740 | /* | |
1741 | * Decrement the use_count of the last object required, if any. | |
1742 | */ | |
1743 | static void kmemleak_seq_stop(struct seq_file *seq, void *v) | |
1744 | { | |
b87324d0 CM |
1745 | if (!IS_ERR(v)) { |
1746 | /* | |
1747 | * kmemleak_seq_start may return ERR_PTR if the scan_mutex | |
1748 | * waiting was interrupted, so only release it if !IS_ERR. | |
1749 | */ | |
f5886c7f | 1750 | rcu_read_unlock(); |
b87324d0 CM |
1751 | mutex_unlock(&scan_mutex); |
1752 | if (v) | |
1753 | put_object(v); | |
1754 | } | |
3c7b4e6b CM |
1755 | } |
1756 | ||
1757 | /* | |
1758 | * Print the information for an unreferenced object to the seq file. | |
1759 | */ | |
1760 | static int kmemleak_seq_show(struct seq_file *seq, void *v) | |
1761 | { | |
1762 | struct kmemleak_object *object = v; | |
1763 | unsigned long flags; | |
1764 | ||
1765 | spin_lock_irqsave(&object->lock, flags); | |
288c857d | 1766 | if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object)) |
17bb9e0d | 1767 | print_unreferenced(seq, object); |
3c7b4e6b CM |
1768 | spin_unlock_irqrestore(&object->lock, flags); |
1769 | return 0; | |
1770 | } | |
1771 | ||
1772 | static const struct seq_operations kmemleak_seq_ops = { | |
1773 | .start = kmemleak_seq_start, | |
1774 | .next = kmemleak_seq_next, | |
1775 | .stop = kmemleak_seq_stop, | |
1776 | .show = kmemleak_seq_show, | |
1777 | }; | |
1778 | ||
1779 | static int kmemleak_open(struct inode *inode, struct file *file) | |
1780 | { | |
b87324d0 | 1781 | return seq_open(file, &kmemleak_seq_ops); |
3c7b4e6b CM |
1782 | } |
1783 | ||
189d84ed CM |
1784 | static int dump_str_object_info(const char *str) |
1785 | { | |
1786 | unsigned long flags; | |
1787 | struct kmemleak_object *object; | |
1788 | unsigned long addr; | |
1789 | ||
dc053733 AP |
1790 | if (kstrtoul(str, 0, &addr)) |
1791 | return -EINVAL; | |
189d84ed CM |
1792 | object = find_and_get_object(addr, 0); |
1793 | if (!object) { | |
1794 | pr_info("Unknown object at 0x%08lx\n", addr); | |
1795 | return -EINVAL; | |
1796 | } | |
1797 | ||
1798 | spin_lock_irqsave(&object->lock, flags); | |
1799 | dump_object_info(object); | |
1800 | spin_unlock_irqrestore(&object->lock, flags); | |
1801 | ||
1802 | put_object(object); | |
1803 | return 0; | |
1804 | } | |
1805 | ||
30b37101 LR |
1806 | /* |
1807 | * We use grey instead of black to ensure we can do future scans on the same | |
1808 | * objects. If we did not do future scans these black objects could | |
1809 | * potentially contain references to newly allocated objects in the future and | |
1810 | * we'd end up with false positives. | |
1811 | */ | |
1812 | static void kmemleak_clear(void) | |
1813 | { | |
1814 | struct kmemleak_object *object; | |
1815 | unsigned long flags; | |
1816 | ||
1817 | rcu_read_lock(); | |
1818 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1819 | spin_lock_irqsave(&object->lock, flags); | |
1820 | if ((object->flags & OBJECT_REPORTED) && | |
1821 | unreferenced_object(object)) | |
a1084c87 | 1822 | __paint_it(object, KMEMLEAK_GREY); |
30b37101 LR |
1823 | spin_unlock_irqrestore(&object->lock, flags); |
1824 | } | |
1825 | rcu_read_unlock(); | |
dc9b3f42 LZ |
1826 | |
1827 | kmemleak_found_leaks = false; | |
30b37101 LR |
1828 | } |
1829 | ||
c89da70c LZ |
1830 | static void __kmemleak_do_cleanup(void); |
1831 | ||
3c7b4e6b CM |
1832 | /* |
1833 | * File write operation to configure kmemleak at run-time. The following | |
1834 | * commands can be written to the /sys/kernel/debug/kmemleak file: | |
1835 | * off - disable kmemleak (irreversible) | |
1836 | * stack=on - enable the task stacks scanning | |
1837 | * stack=off - disable the tasks stacks scanning | |
1838 | * scan=on - start the automatic memory scanning thread | |
1839 | * scan=off - stop the automatic memory scanning thread | |
1840 | * scan=... - set the automatic memory scanning period in seconds (0 to | |
1841 | * disable it) | |
4698c1f2 | 1842 | * scan - trigger a memory scan |
30b37101 | 1843 | * clear - mark all current reported unreferenced kmemleak objects as |
c89da70c LZ |
1844 | * grey to ignore printing them, or free all kmemleak objects |
1845 | * if kmemleak has been disabled. | |
189d84ed | 1846 | * dump=... - dump information about the object found at the given address |
3c7b4e6b CM |
1847 | */ |
1848 | static ssize_t kmemleak_write(struct file *file, const char __user *user_buf, | |
1849 | size_t size, loff_t *ppos) | |
1850 | { | |
1851 | char buf[64]; | |
1852 | int buf_size; | |
b87324d0 | 1853 | int ret; |
3c7b4e6b CM |
1854 | |
1855 | buf_size = min(size, (sizeof(buf) - 1)); | |
1856 | if (strncpy_from_user(buf, user_buf, buf_size) < 0) | |
1857 | return -EFAULT; | |
1858 | buf[buf_size] = 0; | |
1859 | ||
b87324d0 CM |
1860 | ret = mutex_lock_interruptible(&scan_mutex); |
1861 | if (ret < 0) | |
1862 | return ret; | |
1863 | ||
c89da70c | 1864 | if (strncmp(buf, "clear", 5) == 0) { |
8910ae89 | 1865 | if (kmemleak_enabled) |
c89da70c LZ |
1866 | kmemleak_clear(); |
1867 | else | |
1868 | __kmemleak_do_cleanup(); | |
1869 | goto out; | |
1870 | } | |
1871 | ||
8910ae89 | 1872 | if (!kmemleak_enabled) { |
4e4dfce2 | 1873 | ret = -EPERM; |
c89da70c LZ |
1874 | goto out; |
1875 | } | |
1876 | ||
3c7b4e6b CM |
1877 | if (strncmp(buf, "off", 3) == 0) |
1878 | kmemleak_disable(); | |
1879 | else if (strncmp(buf, "stack=on", 8) == 0) | |
1880 | kmemleak_stack_scan = 1; | |
1881 | else if (strncmp(buf, "stack=off", 9) == 0) | |
1882 | kmemleak_stack_scan = 0; | |
1883 | else if (strncmp(buf, "scan=on", 7) == 0) | |
1884 | start_scan_thread(); | |
1885 | else if (strncmp(buf, "scan=off", 8) == 0) | |
1886 | stop_scan_thread(); | |
1887 | else if (strncmp(buf, "scan=", 5) == 0) { | |
1888 | unsigned long secs; | |
3c7b4e6b | 1889 | |
3dbb95f7 | 1890 | ret = kstrtoul(buf + 5, 0, &secs); |
b87324d0 CM |
1891 | if (ret < 0) |
1892 | goto out; | |
3c7b4e6b CM |
1893 | stop_scan_thread(); |
1894 | if (secs) { | |
1895 | jiffies_scan_wait = msecs_to_jiffies(secs * 1000); | |
1896 | start_scan_thread(); | |
1897 | } | |
4698c1f2 CM |
1898 | } else if (strncmp(buf, "scan", 4) == 0) |
1899 | kmemleak_scan(); | |
189d84ed CM |
1900 | else if (strncmp(buf, "dump=", 5) == 0) |
1901 | ret = dump_str_object_info(buf + 5); | |
4698c1f2 | 1902 | else |
b87324d0 CM |
1903 | ret = -EINVAL; |
1904 | ||
1905 | out: | |
1906 | mutex_unlock(&scan_mutex); | |
1907 | if (ret < 0) | |
1908 | return ret; | |
3c7b4e6b CM |
1909 | |
1910 | /* ignore the rest of the buffer, only one command at a time */ | |
1911 | *ppos += size; | |
1912 | return size; | |
1913 | } | |
1914 | ||
1915 | static const struct file_operations kmemleak_fops = { | |
1916 | .owner = THIS_MODULE, | |
1917 | .open = kmemleak_open, | |
1918 | .read = seq_read, | |
1919 | .write = kmemleak_write, | |
1920 | .llseek = seq_lseek, | |
5f3bf19a | 1921 | .release = seq_release, |
3c7b4e6b CM |
1922 | }; |
1923 | ||
c89da70c LZ |
1924 | static void __kmemleak_do_cleanup(void) |
1925 | { | |
1926 | struct kmemleak_object *object; | |
1927 | ||
1928 | rcu_read_lock(); | |
1929 | list_for_each_entry_rcu(object, &object_list, object_list) | |
1930 | delete_object_full(object->pointer); | |
1931 | rcu_read_unlock(); | |
1932 | } | |
1933 | ||
3c7b4e6b | 1934 | /* |
74341703 CM |
1935 | * Stop the memory scanning thread and free the kmemleak internal objects if |
1936 | * no previous scan thread (otherwise, kmemleak may still have some useful | |
1937 | * information on memory leaks). | |
3c7b4e6b | 1938 | */ |
179a8100 | 1939 | static void kmemleak_do_cleanup(struct work_struct *work) |
3c7b4e6b | 1940 | { |
3c7b4e6b | 1941 | stop_scan_thread(); |
3c7b4e6b | 1942 | |
914b6dff | 1943 | mutex_lock(&scan_mutex); |
c5f3b1a5 | 1944 | /* |
914b6dff VM |
1945 | * Once it is made sure that kmemleak_scan has stopped, it is safe to no |
1946 | * longer track object freeing. Ordering of the scan thread stopping and | |
1947 | * the memory accesses below is guaranteed by the kthread_stop() | |
1948 | * function. | |
c5f3b1a5 CM |
1949 | */ |
1950 | kmemleak_free_enabled = 0; | |
914b6dff | 1951 | mutex_unlock(&scan_mutex); |
c5f3b1a5 | 1952 | |
c89da70c LZ |
1953 | if (!kmemleak_found_leaks) |
1954 | __kmemleak_do_cleanup(); | |
1955 | else | |
756a025f | 1956 | pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n"); |
3c7b4e6b CM |
1957 | } |
1958 | ||
179a8100 | 1959 | static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup); |
3c7b4e6b CM |
1960 | |
1961 | /* | |
1962 | * Disable kmemleak. No memory allocation/freeing will be traced once this | |
1963 | * function is called. Disabling kmemleak is an irreversible operation. | |
1964 | */ | |
1965 | static void kmemleak_disable(void) | |
1966 | { | |
1967 | /* atomically check whether it was already invoked */ | |
8910ae89 | 1968 | if (cmpxchg(&kmemleak_error, 0, 1)) |
3c7b4e6b CM |
1969 | return; |
1970 | ||
1971 | /* stop any memory operation tracing */ | |
8910ae89 | 1972 | kmemleak_enabled = 0; |
fcf3a5b6 | 1973 | kmemleak_early_log = 0; |
3c7b4e6b CM |
1974 | |
1975 | /* check whether it is too early for a kernel thread */ | |
8910ae89 | 1976 | if (kmemleak_initialized) |
179a8100 | 1977 | schedule_work(&cleanup_work); |
c5f3b1a5 CM |
1978 | else |
1979 | kmemleak_free_enabled = 0; | |
3c7b4e6b CM |
1980 | |
1981 | pr_info("Kernel memory leak detector disabled\n"); | |
1982 | } | |
1983 | ||
1984 | /* | |
1985 | * Allow boot-time kmemleak disabling (enabled by default). | |
1986 | */ | |
8bd30c10 | 1987 | static int __init kmemleak_boot_config(char *str) |
3c7b4e6b CM |
1988 | { |
1989 | if (!str) | |
1990 | return -EINVAL; | |
1991 | if (strcmp(str, "off") == 0) | |
1992 | kmemleak_disable(); | |
ab0155a2 JB |
1993 | else if (strcmp(str, "on") == 0) |
1994 | kmemleak_skip_disable = 1; | |
1995 | else | |
3c7b4e6b CM |
1996 | return -EINVAL; |
1997 | return 0; | |
1998 | } | |
1999 | early_param("kmemleak", kmemleak_boot_config); | |
2000 | ||
5f79020c CM |
2001 | static void __init print_log_trace(struct early_log *log) |
2002 | { | |
5f79020c | 2003 | pr_notice("Early log backtrace:\n"); |
07984aad | 2004 | stack_trace_print(log->trace, log->trace_len, 2); |
5f79020c CM |
2005 | } |
2006 | ||
3c7b4e6b | 2007 | /* |
2030117d | 2008 | * Kmemleak initialization. |
3c7b4e6b CM |
2009 | */ |
2010 | void __init kmemleak_init(void) | |
2011 | { | |
2012 | int i; | |
2013 | unsigned long flags; | |
2014 | ||
ab0155a2 JB |
2015 | #ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF |
2016 | if (!kmemleak_skip_disable) { | |
2017 | kmemleak_disable(); | |
2018 | return; | |
2019 | } | |
2020 | #endif | |
2021 | ||
3c7b4e6b CM |
2022 | jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE); |
2023 | jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000); | |
2024 | ||
2025 | object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE); | |
2026 | scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE); | |
3c7b4e6b | 2027 | |
21cd3a60 | 2028 | if (crt_early_log > ARRAY_SIZE(early_log)) |
598d8091 JP |
2029 | pr_warn("Early log buffer exceeded (%d), please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n", |
2030 | crt_early_log); | |
b6693005 | 2031 | |
3c7b4e6b CM |
2032 | /* the kernel is still in UP mode, so disabling the IRQs is enough */ |
2033 | local_irq_save(flags); | |
3551a928 | 2034 | kmemleak_early_log = 0; |
8910ae89 | 2035 | if (kmemleak_error) { |
b6693005 CM |
2036 | local_irq_restore(flags); |
2037 | return; | |
c5f3b1a5 | 2038 | } else { |
8910ae89 | 2039 | kmemleak_enabled = 1; |
c5f3b1a5 CM |
2040 | kmemleak_free_enabled = 1; |
2041 | } | |
3c7b4e6b CM |
2042 | local_irq_restore(flags); |
2043 | ||
298a32b1 CM |
2044 | /* register the data/bss sections */ |
2045 | create_object((unsigned long)_sdata, _edata - _sdata, | |
2046 | KMEMLEAK_GREY, GFP_ATOMIC); | |
2047 | create_object((unsigned long)__bss_start, __bss_stop - __bss_start, | |
2048 | KMEMLEAK_GREY, GFP_ATOMIC); | |
2049 | /* only register .data..ro_after_init if not within .data */ | |
2050 | if (__start_ro_after_init < _sdata || __end_ro_after_init > _edata) | |
2051 | create_object((unsigned long)__start_ro_after_init, | |
2052 | __end_ro_after_init - __start_ro_after_init, | |
2053 | KMEMLEAK_GREY, GFP_ATOMIC); | |
2054 | ||
3c7b4e6b CM |
2055 | /* |
2056 | * This is the point where tracking allocations is safe. Automatic | |
2057 | * scanning is started during the late initcall. Add the early logged | |
2058 | * callbacks to the kmemleak infrastructure. | |
2059 | */ | |
2060 | for (i = 0; i < crt_early_log; i++) { | |
2061 | struct early_log *log = &early_log[i]; | |
2062 | ||
2063 | switch (log->op_type) { | |
2064 | case KMEMLEAK_ALLOC: | |
fd678967 | 2065 | early_alloc(log); |
3c7b4e6b | 2066 | break; |
f528f0b8 CM |
2067 | case KMEMLEAK_ALLOC_PERCPU: |
2068 | early_alloc_percpu(log); | |
2069 | break; | |
3c7b4e6b CM |
2070 | case KMEMLEAK_FREE: |
2071 | kmemleak_free(log->ptr); | |
2072 | break; | |
53238a60 CM |
2073 | case KMEMLEAK_FREE_PART: |
2074 | kmemleak_free_part(log->ptr, log->size); | |
2075 | break; | |
f528f0b8 CM |
2076 | case KMEMLEAK_FREE_PERCPU: |
2077 | kmemleak_free_percpu(log->ptr); | |
2078 | break; | |
3c7b4e6b CM |
2079 | case KMEMLEAK_NOT_LEAK: |
2080 | kmemleak_not_leak(log->ptr); | |
2081 | break; | |
2082 | case KMEMLEAK_IGNORE: | |
2083 | kmemleak_ignore(log->ptr); | |
2084 | break; | |
2085 | case KMEMLEAK_SCAN_AREA: | |
c017b4be | 2086 | kmemleak_scan_area(log->ptr, log->size, GFP_KERNEL); |
3c7b4e6b CM |
2087 | break; |
2088 | case KMEMLEAK_NO_SCAN: | |
2089 | kmemleak_no_scan(log->ptr); | |
2090 | break; | |
94f4a161 CM |
2091 | case KMEMLEAK_SET_EXCESS_REF: |
2092 | object_set_excess_ref((unsigned long)log->ptr, | |
2093 | log->excess_ref); | |
2094 | break; | |
3c7b4e6b | 2095 | default: |
5f79020c CM |
2096 | kmemleak_warn("Unknown early log operation: %d\n", |
2097 | log->op_type); | |
2098 | } | |
2099 | ||
8910ae89 | 2100 | if (kmemleak_warning) { |
5f79020c | 2101 | print_log_trace(log); |
8910ae89 | 2102 | kmemleak_warning = 0; |
3c7b4e6b CM |
2103 | } |
2104 | } | |
2105 | } | |
2106 | ||
2107 | /* | |
2108 | * Late initialization function. | |
2109 | */ | |
2110 | static int __init kmemleak_late_init(void) | |
2111 | { | |
8910ae89 | 2112 | kmemleak_initialized = 1; |
3c7b4e6b | 2113 | |
282401df | 2114 | debugfs_create_file("kmemleak", 0644, NULL, NULL, &kmemleak_fops); |
b353756b | 2115 | |
8910ae89 | 2116 | if (kmemleak_error) { |
3c7b4e6b | 2117 | /* |
25985edc | 2118 | * Some error occurred and kmemleak was disabled. There is a |
3c7b4e6b CM |
2119 | * small chance that kmemleak_disable() was called immediately |
2120 | * after setting kmemleak_initialized and we may end up with | |
2121 | * two clean-up threads but serialized by scan_mutex. | |
2122 | */ | |
179a8100 | 2123 | schedule_work(&cleanup_work); |
3c7b4e6b CM |
2124 | return -ENOMEM; |
2125 | } | |
2126 | ||
d53ce042 SK |
2127 | if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) { |
2128 | mutex_lock(&scan_mutex); | |
2129 | start_scan_thread(); | |
2130 | mutex_unlock(&scan_mutex); | |
2131 | } | |
3c7b4e6b CM |
2132 | |
2133 | pr_info("Kernel memory leak detector initialized\n"); | |
2134 | ||
2135 | return 0; | |
2136 | } | |
2137 | late_initcall(kmemleak_late_init); |