]>
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 | * | |
8c96f1bc | 16 | * - kmemleak_lock (raw_spinlock_t): protects the object_list modifications and |
0c24e061 PW |
17 | * accesses to the object_tree_root (or object_phys_tree_root). The |
18 | * object_list is the main list holding the metadata (struct kmemleak_object) | |
19 | * for the allocated memory blocks. The object_tree_root and object_phys_tree_root | |
20 | * are red black trees used to look-up metadata based on a pointer to the | |
21 | * corresponding memory block. The object_phys_tree_root is for objects | |
22 | * allocated with physical address. The kmemleak_object structures are | |
23 | * added to the object_list and object_tree_root (or object_phys_tree_root) | |
24 | * in the create_object() function called from the kmemleak_alloc() (or | |
25 | * kmemleak_alloc_phys()) callback and removed in delete_object() called from | |
26 | * the kmemleak_free() callback | |
8c96f1bc HZ |
27 | * - kmemleak_object.lock (raw_spinlock_t): protects a kmemleak_object. |
28 | * Accesses to the metadata (e.g. count) are protected by this lock. Note | |
29 | * that some members of this structure may be protected by other means | |
30 | * (atomic or kmemleak_lock). This lock is also held when scanning the | |
31 | * corresponding memory block to avoid the kernel freeing it via the | |
32 | * kmemleak_free() callback. This is less heavyweight than holding a global | |
33 | * lock like kmemleak_lock during scanning. | |
3c7b4e6b CM |
34 | * - scan_mutex (mutex): ensures that only one thread may scan the memory for |
35 | * unreferenced objects at a time. The gray_list contains the objects which | |
36 | * are already referenced or marked as false positives and need to be | |
37 | * scanned. This list is only modified during a scanning episode when the | |
38 | * scan_mutex is held. At the end of a scan, the gray_list is always empty. | |
39 | * Note that the kmemleak_object.use_count is incremented when an object is | |
4698c1f2 CM |
40 | * added to the gray_list and therefore cannot be freed. This mutex also |
41 | * prevents multiple users of the "kmemleak" debugfs file together with | |
42 | * modifications to the memory scanning parameters including the scan_thread | |
43 | * pointer | |
3c7b4e6b | 44 | * |
93ada579 | 45 | * Locks and mutexes are acquired/nested in the following order: |
9d5a4c73 | 46 | * |
93ada579 CM |
47 | * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING) |
48 | * | |
49 | * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex | |
50 | * regions. | |
9d5a4c73 | 51 | * |
3c7b4e6b CM |
52 | * The kmemleak_object structures have a use_count incremented or decremented |
53 | * using the get_object()/put_object() functions. When the use_count becomes | |
54 | * 0, this count can no longer be incremented and put_object() schedules the | |
55 | * kmemleak_object freeing via an RCU callback. All calls to the get_object() | |
56 | * function must be protected by rcu_read_lock() to avoid accessing a freed | |
57 | * structure. | |
58 | */ | |
59 | ||
ae281064 JP |
60 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
61 | ||
3c7b4e6b CM |
62 | #include <linux/init.h> |
63 | #include <linux/kernel.h> | |
64 | #include <linux/list.h> | |
3f07c014 | 65 | #include <linux/sched/signal.h> |
29930025 | 66 | #include <linux/sched/task.h> |
68db0cf1 | 67 | #include <linux/sched/task_stack.h> |
3c7b4e6b CM |
68 | #include <linux/jiffies.h> |
69 | #include <linux/delay.h> | |
b95f1b31 | 70 | #include <linux/export.h> |
3c7b4e6b | 71 | #include <linux/kthread.h> |
85d3a316 | 72 | #include <linux/rbtree.h> |
3c7b4e6b CM |
73 | #include <linux/fs.h> |
74 | #include <linux/debugfs.h> | |
75 | #include <linux/seq_file.h> | |
76 | #include <linux/cpumask.h> | |
77 | #include <linux/spinlock.h> | |
154221c3 | 78 | #include <linux/module.h> |
3c7b4e6b CM |
79 | #include <linux/mutex.h> |
80 | #include <linux/rcupdate.h> | |
81 | #include <linux/stacktrace.h> | |
82 | #include <linux/cache.h> | |
83 | #include <linux/percpu.h> | |
57c8a661 | 84 | #include <linux/memblock.h> |
9099daed | 85 | #include <linux/pfn.h> |
3c7b4e6b CM |
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> |
95511580 | 102 | #include <linux/kfence.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 */ |
79d37050 NA |
118 | #define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC | \ |
119 | __GFP_NOLOCKDEP)) | \ | |
6ae4bd1f | 120 | __GFP_NORETRY | __GFP_NOMEMALLOC | \ |
df9576de | 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 { | |
8c96f1bc | 142 | raw_spinlock_t lock; |
f66abf09 | 143 | unsigned int flags; /* object status flags */ |
3c7b4e6b CM |
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; | |
94f4a161 CM |
152 | /* pass surplus references to this pointer */ |
153 | unsigned long excess_ref; | |
3c7b4e6b CM |
154 | /* minimum number of a pointers found before it is considered leak */ |
155 | int min_count; | |
156 | /* the total number of pointers found pointing to this object */ | |
157 | int count; | |
04609ccc CM |
158 | /* checksum for detecting modified objects */ |
159 | u32 checksum; | |
3c7b4e6b CM |
160 | /* memory ranges to be scanned inside an object (empty for all) */ |
161 | struct hlist_head area_list; | |
162 | unsigned long trace[MAX_TRACE]; | |
163 | unsigned int trace_len; | |
164 | unsigned long jiffies; /* creation timestamp */ | |
165 | pid_t pid; /* pid of the current task */ | |
166 | char comm[TASK_COMM_LEN]; /* executable name */ | |
167 | }; | |
168 | ||
169 | /* flag representing the memory block allocation status */ | |
170 | #define OBJECT_ALLOCATED (1 << 0) | |
171 | /* flag set after the first reporting of an unreference object */ | |
172 | #define OBJECT_REPORTED (1 << 1) | |
173 | /* flag set to not scan the object */ | |
174 | #define OBJECT_NO_SCAN (1 << 2) | |
dba82d94 CM |
175 | /* flag set to fully scan the object when scan_area allocation failed */ |
176 | #define OBJECT_FULL_SCAN (1 << 3) | |
8e0c4ab3 PW |
177 | /* flag set for object allocated with physical address */ |
178 | #define OBJECT_PHYS (1 << 4) | |
3c7b4e6b | 179 | |
154221c3 | 180 | #define HEX_PREFIX " " |
0494e082 SS |
181 | /* number of bytes to print per line; must be 16 or 32 */ |
182 | #define HEX_ROW_SIZE 16 | |
183 | /* number of bytes to print at a time (1, 2, 4, 8) */ | |
184 | #define HEX_GROUP_SIZE 1 | |
185 | /* include ASCII after the hex output */ | |
186 | #define HEX_ASCII 1 | |
187 | /* max number of lines to be printed */ | |
188 | #define HEX_MAX_LINES 2 | |
189 | ||
3c7b4e6b CM |
190 | /* the list of all allocated objects */ |
191 | static LIST_HEAD(object_list); | |
192 | /* the list of gray-colored objects (see color_gray comment below) */ | |
193 | static LIST_HEAD(gray_list); | |
0647398a | 194 | /* memory pool allocation */ |
c5665868 | 195 | static struct kmemleak_object mem_pool[CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE]; |
0647398a CM |
196 | static int mem_pool_free_count = ARRAY_SIZE(mem_pool); |
197 | static LIST_HEAD(mem_pool_free_list); | |
85d3a316 ML |
198 | /* search tree for object boundaries */ |
199 | static struct rb_root object_tree_root = RB_ROOT; | |
0c24e061 PW |
200 | /* search tree for object (with OBJECT_PHYS flag) boundaries */ |
201 | static struct rb_root object_phys_tree_root = RB_ROOT; | |
202 | /* protecting the access to object_list, object_tree_root (or object_phys_tree_root) */ | |
8c96f1bc | 203 | static DEFINE_RAW_SPINLOCK(kmemleak_lock); |
3c7b4e6b CM |
204 | |
205 | /* allocation caches for kmemleak internal data */ | |
206 | static struct kmem_cache *object_cache; | |
207 | static struct kmem_cache *scan_area_cache; | |
208 | ||
209 | /* set if tracing memory operations is enabled */ | |
c5665868 | 210 | static int kmemleak_enabled = 1; |
c5f3b1a5 | 211 | /* same as above but only for the kmemleak_free() callback */ |
c5665868 | 212 | static int kmemleak_free_enabled = 1; |
3c7b4e6b | 213 | /* set in the late_initcall if there were no errors */ |
8910ae89 | 214 | static int kmemleak_initialized; |
5f79020c | 215 | /* set if a kmemleak warning was issued */ |
8910ae89 | 216 | static int kmemleak_warning; |
5f79020c | 217 | /* set if a fatal kmemleak error has occurred */ |
8910ae89 | 218 | static int kmemleak_error; |
3c7b4e6b CM |
219 | |
220 | /* minimum and maximum address that may be valid pointers */ | |
221 | static unsigned long min_addr = ULONG_MAX; | |
222 | static unsigned long max_addr; | |
223 | ||
3c7b4e6b | 224 | static struct task_struct *scan_thread; |
acf4968e | 225 | /* used to avoid reporting of recently allocated objects */ |
3c7b4e6b | 226 | static unsigned long jiffies_min_age; |
acf4968e | 227 | static unsigned long jiffies_last_scan; |
3c7b4e6b | 228 | /* delay between automatic memory scannings */ |
54dd200c | 229 | static unsigned long jiffies_scan_wait; |
3c7b4e6b | 230 | /* enables or disables the task stacks scanning */ |
e0a2a160 | 231 | static int kmemleak_stack_scan = 1; |
4698c1f2 | 232 | /* protects the memory scanning, parameters and debug/kmemleak file access */ |
3c7b4e6b | 233 | static DEFINE_MUTEX(scan_mutex); |
ab0155a2 JB |
234 | /* setting kmemleak=on, will set this var, skipping the disable */ |
235 | static int kmemleak_skip_disable; | |
dc9b3f42 LZ |
236 | /* If there are leaks that can be reported */ |
237 | static bool kmemleak_found_leaks; | |
3c7b4e6b | 238 | |
154221c3 VW |
239 | static bool kmemleak_verbose; |
240 | module_param_named(verbose, kmemleak_verbose, bool, 0600); | |
241 | ||
3c7b4e6b CM |
242 | static void kmemleak_disable(void); |
243 | ||
244 | /* | |
245 | * Print a warning and dump the stack trace. | |
246 | */ | |
5f79020c | 247 | #define kmemleak_warn(x...) do { \ |
598d8091 | 248 | pr_warn(x); \ |
5f79020c | 249 | dump_stack(); \ |
8910ae89 | 250 | kmemleak_warning = 1; \ |
3c7b4e6b CM |
251 | } while (0) |
252 | ||
253 | /* | |
25985edc | 254 | * Macro invoked when a serious kmemleak condition occurred and cannot be |
2030117d | 255 | * recovered from. Kmemleak will be disabled and further allocation/freeing |
3c7b4e6b CM |
256 | * tracing no longer available. |
257 | */ | |
000814f4 | 258 | #define kmemleak_stop(x...) do { \ |
3c7b4e6b CM |
259 | kmemleak_warn(x); \ |
260 | kmemleak_disable(); \ | |
261 | } while (0) | |
262 | ||
154221c3 VW |
263 | #define warn_or_seq_printf(seq, fmt, ...) do { \ |
264 | if (seq) \ | |
265 | seq_printf(seq, fmt, ##__VA_ARGS__); \ | |
266 | else \ | |
267 | pr_warn(fmt, ##__VA_ARGS__); \ | |
268 | } while (0) | |
269 | ||
270 | static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type, | |
271 | int rowsize, int groupsize, const void *buf, | |
272 | size_t len, bool ascii) | |
273 | { | |
274 | if (seq) | |
275 | seq_hex_dump(seq, HEX_PREFIX, prefix_type, rowsize, groupsize, | |
276 | buf, len, ascii); | |
277 | else | |
278 | print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type, | |
279 | rowsize, groupsize, buf, len, ascii); | |
280 | } | |
281 | ||
0494e082 SS |
282 | /* |
283 | * Printing of the objects hex dump to the seq file. The number of lines to be | |
284 | * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The | |
285 | * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called | |
286 | * with the object->lock held. | |
287 | */ | |
288 | static void hex_dump_object(struct seq_file *seq, | |
289 | struct kmemleak_object *object) | |
290 | { | |
291 | const u8 *ptr = (const u8 *)object->pointer; | |
6fc37c49 | 292 | size_t len; |
0494e082 | 293 | |
0c24e061 PW |
294 | if (WARN_ON_ONCE(object->flags & OBJECT_PHYS)) |
295 | return; | |
296 | ||
0494e082 | 297 | /* limit the number of lines to HEX_MAX_LINES */ |
6fc37c49 | 298 | len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE); |
0494e082 | 299 | |
154221c3 | 300 | warn_or_seq_printf(seq, " hex dump (first %zu bytes):\n", len); |
5c335fe0 | 301 | kasan_disable_current(); |
154221c3 | 302 | warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE, |
6c7a00b8 | 303 | HEX_GROUP_SIZE, kasan_reset_tag((void *)ptr), len, HEX_ASCII); |
5c335fe0 | 304 | kasan_enable_current(); |
0494e082 SS |
305 | } |
306 | ||
3c7b4e6b CM |
307 | /* |
308 | * Object colors, encoded with count and min_count: | |
309 | * - white - orphan object, not enough references to it (count < min_count) | |
310 | * - gray - not orphan, not marked as false positive (min_count == 0) or | |
311 | * sufficient references to it (count >= min_count) | |
312 | * - black - ignore, it doesn't contain references (e.g. text section) | |
313 | * (min_count == -1). No function defined for this color. | |
314 | * Newly created objects don't have any color assigned (object->count == -1) | |
315 | * before the next memory scan when they become white. | |
316 | */ | |
4a558dd6 | 317 | static bool color_white(const struct kmemleak_object *object) |
3c7b4e6b | 318 | { |
a1084c87 LR |
319 | return object->count != KMEMLEAK_BLACK && |
320 | object->count < object->min_count; | |
3c7b4e6b CM |
321 | } |
322 | ||
4a558dd6 | 323 | static bool color_gray(const struct kmemleak_object *object) |
3c7b4e6b | 324 | { |
a1084c87 LR |
325 | return object->min_count != KMEMLEAK_BLACK && |
326 | object->count >= object->min_count; | |
3c7b4e6b CM |
327 | } |
328 | ||
3c7b4e6b CM |
329 | /* |
330 | * Objects are considered unreferenced only if their color is white, they have | |
331 | * not be deleted and have a minimum age to avoid false positives caused by | |
332 | * pointers temporarily stored in CPU registers. | |
333 | */ | |
4a558dd6 | 334 | static bool unreferenced_object(struct kmemleak_object *object) |
3c7b4e6b | 335 | { |
04609ccc | 336 | return (color_white(object) && object->flags & OBJECT_ALLOCATED) && |
acf4968e CM |
337 | time_before_eq(object->jiffies + jiffies_min_age, |
338 | jiffies_last_scan); | |
3c7b4e6b CM |
339 | } |
340 | ||
341 | /* | |
bab4a34a CM |
342 | * Printing of the unreferenced objects information to the seq file. The |
343 | * print_unreferenced function must be called with the object->lock held. | |
3c7b4e6b | 344 | */ |
3c7b4e6b CM |
345 | static void print_unreferenced(struct seq_file *seq, |
346 | struct kmemleak_object *object) | |
347 | { | |
348 | int i; | |
fefdd336 | 349 | unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies); |
3c7b4e6b | 350 | |
154221c3 | 351 | warn_or_seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n", |
bab4a34a | 352 | object->pointer, object->size); |
154221c3 | 353 | warn_or_seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n", |
fefdd336 CM |
354 | object->comm, object->pid, object->jiffies, |
355 | msecs_age / 1000, msecs_age % 1000); | |
0494e082 | 356 | hex_dump_object(seq, object); |
154221c3 | 357 | warn_or_seq_printf(seq, " backtrace:\n"); |
3c7b4e6b CM |
358 | |
359 | for (i = 0; i < object->trace_len; i++) { | |
360 | void *ptr = (void *)object->trace[i]; | |
154221c3 | 361 | warn_or_seq_printf(seq, " [<%p>] %pS\n", ptr, ptr); |
3c7b4e6b CM |
362 | } |
363 | } | |
364 | ||
365 | /* | |
366 | * Print the kmemleak_object information. This function is used mainly for | |
367 | * debugging special cases when kmemleak operations. It must be called with | |
368 | * the object->lock held. | |
369 | */ | |
370 | static void dump_object_info(struct kmemleak_object *object) | |
371 | { | |
ae281064 | 372 | pr_notice("Object 0x%08lx (size %zu):\n", |
85d3a316 | 373 | object->pointer, object->size); |
3c7b4e6b CM |
374 | pr_notice(" comm \"%s\", pid %d, jiffies %lu\n", |
375 | object->comm, object->pid, object->jiffies); | |
376 | pr_notice(" min_count = %d\n", object->min_count); | |
377 | pr_notice(" count = %d\n", object->count); | |
f66abf09 | 378 | pr_notice(" flags = 0x%x\n", object->flags); |
aae0ad7a | 379 | pr_notice(" checksum = %u\n", object->checksum); |
3c7b4e6b | 380 | pr_notice(" backtrace:\n"); |
07984aad | 381 | stack_trace_print(object->trace, object->trace_len, 4); |
3c7b4e6b CM |
382 | } |
383 | ||
384 | /* | |
85d3a316 | 385 | * Look-up a memory block metadata (kmemleak_object) in the object search |
3c7b4e6b CM |
386 | * tree based on a pointer value. If alias is 0, only values pointing to the |
387 | * beginning of the memory block are allowed. The kmemleak_lock must be held | |
388 | * when calling this function. | |
389 | */ | |
0c24e061 PW |
390 | static struct kmemleak_object *__lookup_object(unsigned long ptr, int alias, |
391 | bool is_phys) | |
3c7b4e6b | 392 | { |
0c24e061 PW |
393 | struct rb_node *rb = is_phys ? object_phys_tree_root.rb_node : |
394 | object_tree_root.rb_node; | |
ad1a3e15 | 395 | unsigned long untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr); |
85d3a316 ML |
396 | |
397 | while (rb) { | |
ad1a3e15 KYL |
398 | struct kmemleak_object *object; |
399 | unsigned long untagged_objp; | |
400 | ||
401 | object = rb_entry(rb, struct kmemleak_object, rb_node); | |
402 | untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer); | |
403 | ||
404 | if (untagged_ptr < untagged_objp) | |
85d3a316 | 405 | rb = object->rb_node.rb_left; |
ad1a3e15 | 406 | else if (untagged_objp + object->size <= untagged_ptr) |
85d3a316 | 407 | rb = object->rb_node.rb_right; |
ad1a3e15 | 408 | else if (untagged_objp == untagged_ptr || alias) |
85d3a316 ML |
409 | return object; |
410 | else { | |
5f79020c CM |
411 | kmemleak_warn("Found object by alias at 0x%08lx\n", |
412 | ptr); | |
a7686a45 | 413 | dump_object_info(object); |
85d3a316 | 414 | break; |
3c7b4e6b | 415 | } |
85d3a316 ML |
416 | } |
417 | return NULL; | |
3c7b4e6b CM |
418 | } |
419 | ||
0c24e061 PW |
420 | /* Look-up a kmemleak object which allocated with virtual address. */ |
421 | static struct kmemleak_object *lookup_object(unsigned long ptr, int alias) | |
422 | { | |
423 | return __lookup_object(ptr, alias, false); | |
424 | } | |
425 | ||
3c7b4e6b CM |
426 | /* |
427 | * Increment the object use_count. Return 1 if successful or 0 otherwise. Note | |
428 | * that once an object's use_count reached 0, the RCU freeing was already | |
429 | * registered and the object should no longer be used. This function must be | |
430 | * called under the protection of rcu_read_lock(). | |
431 | */ | |
432 | static int get_object(struct kmemleak_object *object) | |
433 | { | |
434 | return atomic_inc_not_zero(&object->use_count); | |
435 | } | |
436 | ||
0647398a CM |
437 | /* |
438 | * Memory pool allocation and freeing. kmemleak_lock must not be held. | |
439 | */ | |
440 | static struct kmemleak_object *mem_pool_alloc(gfp_t gfp) | |
441 | { | |
442 | unsigned long flags; | |
443 | struct kmemleak_object *object; | |
444 | ||
445 | /* try the slab allocator first */ | |
c5665868 CM |
446 | if (object_cache) { |
447 | object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp)); | |
448 | if (object) | |
449 | return object; | |
450 | } | |
0647398a CM |
451 | |
452 | /* slab allocation failed, try the memory pool */ | |
8c96f1bc | 453 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
0647398a CM |
454 | object = list_first_entry_or_null(&mem_pool_free_list, |
455 | typeof(*object), object_list); | |
456 | if (object) | |
457 | list_del(&object->object_list); | |
458 | else if (mem_pool_free_count) | |
459 | object = &mem_pool[--mem_pool_free_count]; | |
c5665868 CM |
460 | else |
461 | pr_warn_once("Memory pool empty, consider increasing CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE\n"); | |
8c96f1bc | 462 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
0647398a CM |
463 | |
464 | return object; | |
465 | } | |
466 | ||
467 | /* | |
468 | * Return the object to either the slab allocator or the memory pool. | |
469 | */ | |
470 | static void mem_pool_free(struct kmemleak_object *object) | |
471 | { | |
472 | unsigned long flags; | |
473 | ||
474 | if (object < mem_pool || object >= mem_pool + ARRAY_SIZE(mem_pool)) { | |
475 | kmem_cache_free(object_cache, object); | |
476 | return; | |
477 | } | |
478 | ||
479 | /* add the object to the memory pool free list */ | |
8c96f1bc | 480 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
0647398a | 481 | list_add(&object->object_list, &mem_pool_free_list); |
8c96f1bc | 482 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
0647398a CM |
483 | } |
484 | ||
3c7b4e6b CM |
485 | /* |
486 | * RCU callback to free a kmemleak_object. | |
487 | */ | |
488 | static void free_object_rcu(struct rcu_head *rcu) | |
489 | { | |
b67bfe0d | 490 | struct hlist_node *tmp; |
3c7b4e6b CM |
491 | struct kmemleak_scan_area *area; |
492 | struct kmemleak_object *object = | |
493 | container_of(rcu, struct kmemleak_object, rcu); | |
494 | ||
495 | /* | |
496 | * Once use_count is 0 (guaranteed by put_object), there is no other | |
497 | * code accessing this object, hence no need for locking. | |
498 | */ | |
b67bfe0d SL |
499 | hlist_for_each_entry_safe(area, tmp, &object->area_list, node) { |
500 | hlist_del(&area->node); | |
3c7b4e6b CM |
501 | kmem_cache_free(scan_area_cache, area); |
502 | } | |
0647398a | 503 | mem_pool_free(object); |
3c7b4e6b CM |
504 | } |
505 | ||
506 | /* | |
507 | * Decrement the object use_count. Once the count is 0, free the object using | |
508 | * an RCU callback. Since put_object() may be called via the kmemleak_free() -> | |
509 | * delete_object() path, the delayed RCU freeing ensures that there is no | |
510 | * recursive call to the kernel allocator. Lock-less RCU object_list traversal | |
511 | * is also possible. | |
512 | */ | |
513 | static void put_object(struct kmemleak_object *object) | |
514 | { | |
515 | if (!atomic_dec_and_test(&object->use_count)) | |
516 | return; | |
517 | ||
518 | /* should only get here after delete_object was called */ | |
519 | WARN_ON(object->flags & OBJECT_ALLOCATED); | |
520 | ||
c5665868 CM |
521 | /* |
522 | * It may be too early for the RCU callbacks, however, there is no | |
523 | * concurrent object_list traversal when !object_cache and all objects | |
524 | * came from the memory pool. Free the object directly. | |
525 | */ | |
526 | if (object_cache) | |
527 | call_rcu(&object->rcu, free_object_rcu); | |
528 | else | |
529 | free_object_rcu(&object->rcu); | |
3c7b4e6b CM |
530 | } |
531 | ||
532 | /* | |
85d3a316 | 533 | * Look up an object in the object search tree and increase its use_count. |
3c7b4e6b | 534 | */ |
0c24e061 PW |
535 | static struct kmemleak_object *__find_and_get_object(unsigned long ptr, int alias, |
536 | bool is_phys) | |
3c7b4e6b CM |
537 | { |
538 | unsigned long flags; | |
9fbed254 | 539 | struct kmemleak_object *object; |
3c7b4e6b CM |
540 | |
541 | rcu_read_lock(); | |
8c96f1bc | 542 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
0c24e061 | 543 | object = __lookup_object(ptr, alias, is_phys); |
8c96f1bc | 544 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
3c7b4e6b CM |
545 | |
546 | /* check whether the object is still available */ | |
547 | if (object && !get_object(object)) | |
548 | object = NULL; | |
549 | rcu_read_unlock(); | |
550 | ||
551 | return object; | |
552 | } | |
553 | ||
0c24e061 PW |
554 | /* Look up and get an object which allocated with virtual address. */ |
555 | static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias) | |
556 | { | |
557 | return __find_and_get_object(ptr, alias, false); | |
558 | } | |
559 | ||
2abd839a | 560 | /* |
0c24e061 PW |
561 | * Remove an object from the object_tree_root (or object_phys_tree_root) |
562 | * and object_list. Must be called with the kmemleak_lock held _if_ kmemleak | |
563 | * is still enabled. | |
2abd839a CM |
564 | */ |
565 | static void __remove_object(struct kmemleak_object *object) | |
566 | { | |
0c24e061 PW |
567 | rb_erase(&object->rb_node, object->flags & OBJECT_PHYS ? |
568 | &object_phys_tree_root : | |
569 | &object_tree_root); | |
2abd839a CM |
570 | list_del_rcu(&object->object_list); |
571 | } | |
572 | ||
e781a9ab CM |
573 | /* |
574 | * Look up an object in the object search tree and remove it from both | |
0c24e061 PW |
575 | * object_tree_root (or object_phys_tree_root) and object_list. The |
576 | * returned object's use_count should be at least 1, as initially set | |
577 | * by create_object(). | |
e781a9ab | 578 | */ |
0c24e061 PW |
579 | static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias, |
580 | bool is_phys) | |
e781a9ab CM |
581 | { |
582 | unsigned long flags; | |
583 | struct kmemleak_object *object; | |
584 | ||
8c96f1bc | 585 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
0c24e061 | 586 | object = __lookup_object(ptr, alias, is_phys); |
2abd839a CM |
587 | if (object) |
588 | __remove_object(object); | |
8c96f1bc | 589 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
e781a9ab CM |
590 | |
591 | return object; | |
592 | } | |
593 | ||
fd678967 CM |
594 | /* |
595 | * Save stack trace to the given array of MAX_TRACE size. | |
596 | */ | |
597 | static int __save_stack_trace(unsigned long *trace) | |
598 | { | |
07984aad | 599 | return stack_trace_save(trace, MAX_TRACE, 2); |
fd678967 CM |
600 | } |
601 | ||
3c7b4e6b CM |
602 | /* |
603 | * Create the metadata (struct kmemleak_object) corresponding to an allocated | |
0c24e061 PW |
604 | * memory block and add it to the object_list and object_tree_root (or |
605 | * object_phys_tree_root). | |
3c7b4e6b | 606 | */ |
8e0c4ab3 PW |
607 | static struct kmemleak_object *__create_object(unsigned long ptr, size_t size, |
608 | int min_count, gfp_t gfp, | |
609 | bool is_phys) | |
3c7b4e6b CM |
610 | { |
611 | unsigned long flags; | |
85d3a316 ML |
612 | struct kmemleak_object *object, *parent; |
613 | struct rb_node **link, *rb_parent; | |
a2f77575 | 614 | unsigned long untagged_ptr; |
ad1a3e15 | 615 | unsigned long untagged_objp; |
3c7b4e6b | 616 | |
0647398a | 617 | object = mem_pool_alloc(gfp); |
3c7b4e6b | 618 | if (!object) { |
598d8091 | 619 | pr_warn("Cannot allocate a kmemleak_object structure\n"); |
6ae4bd1f | 620 | kmemleak_disable(); |
fd678967 | 621 | return NULL; |
3c7b4e6b CM |
622 | } |
623 | ||
624 | INIT_LIST_HEAD(&object->object_list); | |
625 | INIT_LIST_HEAD(&object->gray_list); | |
626 | INIT_HLIST_HEAD(&object->area_list); | |
8c96f1bc | 627 | raw_spin_lock_init(&object->lock); |
3c7b4e6b | 628 | atomic_set(&object->use_count, 1); |
8e0c4ab3 | 629 | object->flags = OBJECT_ALLOCATED | (is_phys ? OBJECT_PHYS : 0); |
3c7b4e6b | 630 | object->pointer = ptr; |
95511580 | 631 | object->size = kfence_ksize((void *)ptr) ?: size; |
94f4a161 | 632 | object->excess_ref = 0; |
3c7b4e6b | 633 | object->min_count = min_count; |
04609ccc | 634 | object->count = 0; /* white color initially */ |
3c7b4e6b | 635 | object->jiffies = jiffies; |
04609ccc | 636 | object->checksum = 0; |
3c7b4e6b CM |
637 | |
638 | /* task information */ | |
ea0eafea | 639 | if (in_hardirq()) { |
3c7b4e6b CM |
640 | object->pid = 0; |
641 | strncpy(object->comm, "hardirq", sizeof(object->comm)); | |
6ef90569 | 642 | } else if (in_serving_softirq()) { |
3c7b4e6b CM |
643 | object->pid = 0; |
644 | strncpy(object->comm, "softirq", sizeof(object->comm)); | |
645 | } else { | |
646 | object->pid = current->pid; | |
647 | /* | |
648 | * There is a small chance of a race with set_task_comm(), | |
649 | * however using get_task_comm() here may cause locking | |
650 | * dependency issues with current->alloc_lock. In the worst | |
651 | * case, the command line is not correct. | |
652 | */ | |
653 | strncpy(object->comm, current->comm, sizeof(object->comm)); | |
654 | } | |
655 | ||
656 | /* kernel backtrace */ | |
fd678967 | 657 | object->trace_len = __save_stack_trace(object->trace); |
3c7b4e6b | 658 | |
8c96f1bc | 659 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
0580a181 | 660 | |
a2f77575 | 661 | untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr); |
0c24e061 PW |
662 | /* |
663 | * Only update min_addr and max_addr with object | |
664 | * storing virtual address. | |
665 | */ | |
666 | if (!is_phys) { | |
667 | min_addr = min(min_addr, untagged_ptr); | |
668 | max_addr = max(max_addr, untagged_ptr + size); | |
669 | } | |
670 | link = is_phys ? &object_phys_tree_root.rb_node : | |
671 | &object_tree_root.rb_node; | |
85d3a316 ML |
672 | rb_parent = NULL; |
673 | while (*link) { | |
674 | rb_parent = *link; | |
675 | parent = rb_entry(rb_parent, struct kmemleak_object, rb_node); | |
ad1a3e15 KYL |
676 | untagged_objp = (unsigned long)kasan_reset_tag((void *)parent->pointer); |
677 | if (untagged_ptr + size <= untagged_objp) | |
85d3a316 | 678 | link = &parent->rb_node.rb_left; |
ad1a3e15 | 679 | else if (untagged_objp + parent->size <= untagged_ptr) |
85d3a316 ML |
680 | link = &parent->rb_node.rb_right; |
681 | else { | |
756a025f | 682 | kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n", |
85d3a316 | 683 | ptr); |
9d5a4c73 CM |
684 | /* |
685 | * No need for parent->lock here since "parent" cannot | |
686 | * be freed while the kmemleak_lock is held. | |
687 | */ | |
688 | dump_object_info(parent); | |
85d3a316 | 689 | kmem_cache_free(object_cache, object); |
9d5a4c73 | 690 | object = NULL; |
85d3a316 ML |
691 | goto out; |
692 | } | |
3c7b4e6b | 693 | } |
85d3a316 | 694 | rb_link_node(&object->rb_node, rb_parent, link); |
0c24e061 PW |
695 | rb_insert_color(&object->rb_node, is_phys ? &object_phys_tree_root : |
696 | &object_tree_root); | |
85d3a316 | 697 | |
3c7b4e6b CM |
698 | list_add_tail_rcu(&object->object_list, &object_list); |
699 | out: | |
8c96f1bc | 700 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
fd678967 | 701 | return object; |
3c7b4e6b CM |
702 | } |
703 | ||
8e0c4ab3 PW |
704 | /* Create kmemleak object which allocated with virtual address. */ |
705 | static struct kmemleak_object *create_object(unsigned long ptr, size_t size, | |
706 | int min_count, gfp_t gfp) | |
707 | { | |
708 | return __create_object(ptr, size, min_count, gfp, false); | |
709 | } | |
710 | ||
711 | /* Create kmemleak object which allocated with physical address. */ | |
712 | static struct kmemleak_object *create_object_phys(unsigned long ptr, size_t size, | |
713 | int min_count, gfp_t gfp) | |
714 | { | |
715 | return __create_object(ptr, size, min_count, gfp, true); | |
716 | } | |
717 | ||
3c7b4e6b | 718 | /* |
e781a9ab | 719 | * Mark the object as not allocated and schedule RCU freeing via put_object(). |
3c7b4e6b | 720 | */ |
53238a60 | 721 | static void __delete_object(struct kmemleak_object *object) |
3c7b4e6b CM |
722 | { |
723 | unsigned long flags; | |
3c7b4e6b | 724 | |
3c7b4e6b | 725 | WARN_ON(!(object->flags & OBJECT_ALLOCATED)); |
e781a9ab | 726 | WARN_ON(atomic_read(&object->use_count) < 1); |
3c7b4e6b CM |
727 | |
728 | /* | |
729 | * Locking here also ensures that the corresponding memory block | |
730 | * cannot be freed when it is being scanned. | |
731 | */ | |
8c96f1bc | 732 | raw_spin_lock_irqsave(&object->lock, flags); |
3c7b4e6b | 733 | object->flags &= ~OBJECT_ALLOCATED; |
8c96f1bc | 734 | raw_spin_unlock_irqrestore(&object->lock, flags); |
3c7b4e6b CM |
735 | put_object(object); |
736 | } | |
737 | ||
53238a60 CM |
738 | /* |
739 | * Look up the metadata (struct kmemleak_object) corresponding to ptr and | |
740 | * delete it. | |
741 | */ | |
742 | static void delete_object_full(unsigned long ptr) | |
743 | { | |
744 | struct kmemleak_object *object; | |
745 | ||
0c24e061 | 746 | object = find_and_remove_object(ptr, 0, false); |
53238a60 CM |
747 | if (!object) { |
748 | #ifdef DEBUG | |
749 | kmemleak_warn("Freeing unknown object at 0x%08lx\n", | |
750 | ptr); | |
751 | #endif | |
752 | return; | |
753 | } | |
754 | __delete_object(object); | |
53238a60 CM |
755 | } |
756 | ||
757 | /* | |
758 | * Look up the metadata (struct kmemleak_object) corresponding to ptr and | |
759 | * delete it. If the memory block is partially freed, the function may create | |
760 | * additional metadata for the remaining parts of the block. | |
761 | */ | |
0c24e061 | 762 | static void delete_object_part(unsigned long ptr, size_t size, bool is_phys) |
53238a60 CM |
763 | { |
764 | struct kmemleak_object *object; | |
765 | unsigned long start, end; | |
766 | ||
0c24e061 | 767 | object = find_and_remove_object(ptr, 1, is_phys); |
53238a60 CM |
768 | if (!object) { |
769 | #ifdef DEBUG | |
756a025f JP |
770 | kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n", |
771 | ptr, size); | |
53238a60 CM |
772 | #endif |
773 | return; | |
774 | } | |
53238a60 CM |
775 | |
776 | /* | |
777 | * Create one or two objects that may result from the memory block | |
778 | * split. Note that partial freeing is only done by free_bootmem() and | |
c5665868 | 779 | * this happens before kmemleak_init() is called. |
53238a60 CM |
780 | */ |
781 | start = object->pointer; | |
782 | end = object->pointer + object->size; | |
783 | if (ptr > start) | |
8e0c4ab3 | 784 | __create_object(start, ptr - start, object->min_count, |
0c24e061 | 785 | GFP_KERNEL, is_phys); |
53238a60 | 786 | if (ptr + size < end) |
8e0c4ab3 | 787 | __create_object(ptr + size, end - ptr - size, object->min_count, |
0c24e061 | 788 | GFP_KERNEL, is_phys); |
53238a60 | 789 | |
e781a9ab | 790 | __delete_object(object); |
53238a60 | 791 | } |
a1084c87 LR |
792 | |
793 | static void __paint_it(struct kmemleak_object *object, int color) | |
794 | { | |
795 | object->min_count = color; | |
796 | if (color == KMEMLEAK_BLACK) | |
797 | object->flags |= OBJECT_NO_SCAN; | |
798 | } | |
799 | ||
800 | static void paint_it(struct kmemleak_object *object, int color) | |
3c7b4e6b CM |
801 | { |
802 | unsigned long flags; | |
a1084c87 | 803 | |
8c96f1bc | 804 | raw_spin_lock_irqsave(&object->lock, flags); |
a1084c87 | 805 | __paint_it(object, color); |
8c96f1bc | 806 | raw_spin_unlock_irqrestore(&object->lock, flags); |
a1084c87 LR |
807 | } |
808 | ||
0c24e061 | 809 | static void paint_ptr(unsigned long ptr, int color, bool is_phys) |
a1084c87 | 810 | { |
3c7b4e6b CM |
811 | struct kmemleak_object *object; |
812 | ||
0c24e061 | 813 | object = __find_and_get_object(ptr, 0, is_phys); |
3c7b4e6b | 814 | if (!object) { |
756a025f JP |
815 | kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n", |
816 | ptr, | |
a1084c87 LR |
817 | (color == KMEMLEAK_GREY) ? "Grey" : |
818 | (color == KMEMLEAK_BLACK) ? "Black" : "Unknown"); | |
3c7b4e6b CM |
819 | return; |
820 | } | |
a1084c87 | 821 | paint_it(object, color); |
3c7b4e6b CM |
822 | put_object(object); |
823 | } | |
824 | ||
a1084c87 | 825 | /* |
145b64b9 | 826 | * Mark an object permanently as gray-colored so that it can no longer be |
a1084c87 LR |
827 | * reported as a leak. This is used in general to mark a false positive. |
828 | */ | |
829 | static void make_gray_object(unsigned long ptr) | |
830 | { | |
0c24e061 | 831 | paint_ptr(ptr, KMEMLEAK_GREY, false); |
a1084c87 LR |
832 | } |
833 | ||
3c7b4e6b CM |
834 | /* |
835 | * Mark the object as black-colored so that it is ignored from scans and | |
836 | * reporting. | |
837 | */ | |
0c24e061 | 838 | static void make_black_object(unsigned long ptr, bool is_phys) |
3c7b4e6b | 839 | { |
0c24e061 | 840 | paint_ptr(ptr, KMEMLEAK_BLACK, is_phys); |
3c7b4e6b CM |
841 | } |
842 | ||
843 | /* | |
844 | * Add a scanning area to the object. If at least one such area is added, | |
845 | * kmemleak will only scan these ranges rather than the whole memory block. | |
846 | */ | |
c017b4be | 847 | static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp) |
3c7b4e6b CM |
848 | { |
849 | unsigned long flags; | |
850 | struct kmemleak_object *object; | |
c5665868 | 851 | struct kmemleak_scan_area *area = NULL; |
bfc8089f KYL |
852 | unsigned long untagged_ptr; |
853 | unsigned long untagged_objp; | |
3c7b4e6b | 854 | |
c017b4be | 855 | object = find_and_get_object(ptr, 1); |
3c7b4e6b | 856 | if (!object) { |
ae281064 JP |
857 | kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n", |
858 | ptr); | |
3c7b4e6b CM |
859 | return; |
860 | } | |
861 | ||
bfc8089f KYL |
862 | untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr); |
863 | untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer); | |
864 | ||
c5665868 CM |
865 | if (scan_area_cache) |
866 | area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp)); | |
3c7b4e6b | 867 | |
8c96f1bc | 868 | raw_spin_lock_irqsave(&object->lock, flags); |
dba82d94 CM |
869 | if (!area) { |
870 | pr_warn_once("Cannot allocate a scan area, scanning the full object\n"); | |
871 | /* mark the object for full scan to avoid false positives */ | |
872 | object->flags |= OBJECT_FULL_SCAN; | |
873 | goto out_unlock; | |
874 | } | |
7f88f88f | 875 | if (size == SIZE_MAX) { |
bfc8089f KYL |
876 | size = untagged_objp + object->size - untagged_ptr; |
877 | } else if (untagged_ptr + size > untagged_objp + object->size) { | |
ae281064 | 878 | kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr); |
3c7b4e6b CM |
879 | dump_object_info(object); |
880 | kmem_cache_free(scan_area_cache, area); | |
881 | goto out_unlock; | |
882 | } | |
883 | ||
884 | INIT_HLIST_NODE(&area->node); | |
c017b4be CM |
885 | area->start = ptr; |
886 | area->size = size; | |
3c7b4e6b CM |
887 | |
888 | hlist_add_head(&area->node, &object->area_list); | |
889 | out_unlock: | |
8c96f1bc | 890 | raw_spin_unlock_irqrestore(&object->lock, flags); |
3c7b4e6b CM |
891 | put_object(object); |
892 | } | |
893 | ||
94f4a161 CM |
894 | /* |
895 | * Any surplus references (object already gray) to 'ptr' are passed to | |
896 | * 'excess_ref'. This is used in the vmalloc() case where a pointer to | |
897 | * vm_struct may be used as an alternative reference to the vmalloc'ed object | |
898 | * (see free_thread_stack()). | |
899 | */ | |
900 | static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref) | |
901 | { | |
902 | unsigned long flags; | |
903 | struct kmemleak_object *object; | |
904 | ||
905 | object = find_and_get_object(ptr, 0); | |
906 | if (!object) { | |
907 | kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n", | |
908 | ptr); | |
909 | return; | |
910 | } | |
911 | ||
8c96f1bc | 912 | raw_spin_lock_irqsave(&object->lock, flags); |
94f4a161 | 913 | object->excess_ref = excess_ref; |
8c96f1bc | 914 | raw_spin_unlock_irqrestore(&object->lock, flags); |
94f4a161 CM |
915 | put_object(object); |
916 | } | |
917 | ||
3c7b4e6b CM |
918 | /* |
919 | * Set the OBJECT_NO_SCAN flag for the object corresponding to the give | |
920 | * pointer. Such object will not be scanned by kmemleak but references to it | |
921 | * are searched. | |
922 | */ | |
923 | static void object_no_scan(unsigned long ptr) | |
924 | { | |
925 | unsigned long flags; | |
926 | struct kmemleak_object *object; | |
927 | ||
928 | object = find_and_get_object(ptr, 0); | |
929 | if (!object) { | |
ae281064 | 930 | kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr); |
3c7b4e6b CM |
931 | return; |
932 | } | |
933 | ||
8c96f1bc | 934 | raw_spin_lock_irqsave(&object->lock, flags); |
3c7b4e6b | 935 | object->flags |= OBJECT_NO_SCAN; |
8c96f1bc | 936 | raw_spin_unlock_irqrestore(&object->lock, flags); |
3c7b4e6b CM |
937 | put_object(object); |
938 | } | |
939 | ||
a2b6bf63 CM |
940 | /** |
941 | * kmemleak_alloc - register a newly allocated object | |
942 | * @ptr: pointer to beginning of the object | |
943 | * @size: size of the object | |
944 | * @min_count: minimum number of references to this object. If during memory | |
945 | * scanning a number of references less than @min_count is found, | |
946 | * the object is reported as a memory leak. If @min_count is 0, | |
947 | * the object is never reported as a leak. If @min_count is -1, | |
948 | * the object is ignored (not scanned and not reported as a leak) | |
949 | * @gfp: kmalloc() flags used for kmemleak internal memory allocations | |
950 | * | |
951 | * This function is called from the kernel allocators when a new object | |
94f4a161 | 952 | * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.). |
3c7b4e6b | 953 | */ |
a6186d89 CM |
954 | void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count, |
955 | gfp_t gfp) | |
3c7b4e6b CM |
956 | { |
957 | pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count); | |
958 | ||
8910ae89 | 959 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
3c7b4e6b | 960 | create_object((unsigned long)ptr, size, min_count, gfp); |
3c7b4e6b CM |
961 | } |
962 | EXPORT_SYMBOL_GPL(kmemleak_alloc); | |
963 | ||
f528f0b8 CM |
964 | /** |
965 | * kmemleak_alloc_percpu - register a newly allocated __percpu object | |
966 | * @ptr: __percpu pointer to beginning of the object | |
967 | * @size: size of the object | |
8a8c35fa | 968 | * @gfp: flags used for kmemleak internal memory allocations |
f528f0b8 CM |
969 | * |
970 | * This function is called from the kernel percpu allocator when a new object | |
8a8c35fa | 971 | * (memory block) is allocated (alloc_percpu). |
f528f0b8 | 972 | */ |
8a8c35fa LF |
973 | void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size, |
974 | gfp_t gfp) | |
f528f0b8 CM |
975 | { |
976 | unsigned int cpu; | |
977 | ||
978 | pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size); | |
979 | ||
980 | /* | |
981 | * Percpu allocations are only scanned and not reported as leaks | |
982 | * (min_count is set to 0). | |
983 | */ | |
8910ae89 | 984 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
f528f0b8 CM |
985 | for_each_possible_cpu(cpu) |
986 | create_object((unsigned long)per_cpu_ptr(ptr, cpu), | |
8a8c35fa | 987 | size, 0, gfp); |
f528f0b8 CM |
988 | } |
989 | EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu); | |
990 | ||
94f4a161 CM |
991 | /** |
992 | * kmemleak_vmalloc - register a newly vmalloc'ed object | |
993 | * @area: pointer to vm_struct | |
994 | * @size: size of the object | |
995 | * @gfp: __vmalloc() flags used for kmemleak internal memory allocations | |
996 | * | |
997 | * This function is called from the vmalloc() kernel allocator when a new | |
998 | * object (memory block) is allocated. | |
999 | */ | |
1000 | void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp) | |
1001 | { | |
1002 | pr_debug("%s(0x%p, %zu)\n", __func__, area, size); | |
1003 | ||
1004 | /* | |
1005 | * A min_count = 2 is needed because vm_struct contains a reference to | |
1006 | * the virtual address of the vmalloc'ed block. | |
1007 | */ | |
1008 | if (kmemleak_enabled) { | |
1009 | create_object((unsigned long)area->addr, size, 2, gfp); | |
1010 | object_set_excess_ref((unsigned long)area, | |
1011 | (unsigned long)area->addr); | |
94f4a161 CM |
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); |
3c7b4e6b CM |
1029 | } |
1030 | EXPORT_SYMBOL_GPL(kmemleak_free); | |
1031 | ||
a2b6bf63 CM |
1032 | /** |
1033 | * kmemleak_free_part - partially unregister a previously registered object | |
1034 | * @ptr: pointer to the beginning or inside the object. This also | |
1035 | * represents the start of the range to be freed | |
1036 | * @size: size to be unregistered | |
1037 | * | |
1038 | * This function is called when only a part of a memory block is freed | |
1039 | * (usually from the bootmem allocator). | |
53238a60 | 1040 | */ |
a6186d89 | 1041 | void __ref kmemleak_free_part(const void *ptr, size_t size) |
53238a60 CM |
1042 | { |
1043 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1044 | ||
8910ae89 | 1045 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
0c24e061 | 1046 | delete_object_part((unsigned long)ptr, size, false); |
53238a60 CM |
1047 | } |
1048 | EXPORT_SYMBOL_GPL(kmemleak_free_part); | |
1049 | ||
f528f0b8 CM |
1050 | /** |
1051 | * kmemleak_free_percpu - unregister a previously registered __percpu object | |
1052 | * @ptr: __percpu pointer to beginning of the object | |
1053 | * | |
1054 | * This function is called from the kernel percpu allocator when an object | |
1055 | * (memory block) is freed (free_percpu). | |
1056 | */ | |
1057 | void __ref kmemleak_free_percpu(const void __percpu *ptr) | |
1058 | { | |
1059 | unsigned int cpu; | |
1060 | ||
1061 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1062 | ||
c5f3b1a5 | 1063 | if (kmemleak_free_enabled && ptr && !IS_ERR(ptr)) |
f528f0b8 CM |
1064 | for_each_possible_cpu(cpu) |
1065 | delete_object_full((unsigned long)per_cpu_ptr(ptr, | |
1066 | cpu)); | |
f528f0b8 CM |
1067 | } |
1068 | EXPORT_SYMBOL_GPL(kmemleak_free_percpu); | |
1069 | ||
ffe2c748 CM |
1070 | /** |
1071 | * kmemleak_update_trace - update object allocation stack trace | |
1072 | * @ptr: pointer to beginning of the object | |
1073 | * | |
1074 | * Override the object allocation stack trace for cases where the actual | |
1075 | * allocation place is not always useful. | |
1076 | */ | |
1077 | void __ref kmemleak_update_trace(const void *ptr) | |
1078 | { | |
1079 | struct kmemleak_object *object; | |
1080 | unsigned long flags; | |
1081 | ||
1082 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1083 | ||
1084 | if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr)) | |
1085 | return; | |
1086 | ||
1087 | object = find_and_get_object((unsigned long)ptr, 1); | |
1088 | if (!object) { | |
1089 | #ifdef DEBUG | |
1090 | kmemleak_warn("Updating stack trace for unknown object at %p\n", | |
1091 | ptr); | |
1092 | #endif | |
1093 | return; | |
1094 | } | |
1095 | ||
8c96f1bc | 1096 | raw_spin_lock_irqsave(&object->lock, flags); |
ffe2c748 | 1097 | object->trace_len = __save_stack_trace(object->trace); |
8c96f1bc | 1098 | raw_spin_unlock_irqrestore(&object->lock, flags); |
ffe2c748 CM |
1099 | |
1100 | put_object(object); | |
1101 | } | |
1102 | EXPORT_SYMBOL(kmemleak_update_trace); | |
1103 | ||
a2b6bf63 CM |
1104 | /** |
1105 | * kmemleak_not_leak - mark an allocated object as false positive | |
1106 | * @ptr: pointer to beginning of the object | |
1107 | * | |
1108 | * Calling this function on an object will cause the memory block to no longer | |
1109 | * be reported as leak and always be scanned. | |
3c7b4e6b | 1110 | */ |
a6186d89 | 1111 | void __ref kmemleak_not_leak(const void *ptr) |
3c7b4e6b CM |
1112 | { |
1113 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1114 | ||
8910ae89 | 1115 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
3c7b4e6b | 1116 | make_gray_object((unsigned long)ptr); |
3c7b4e6b CM |
1117 | } |
1118 | EXPORT_SYMBOL(kmemleak_not_leak); | |
1119 | ||
a2b6bf63 CM |
1120 | /** |
1121 | * kmemleak_ignore - ignore an allocated object | |
1122 | * @ptr: pointer to beginning of the object | |
1123 | * | |
1124 | * Calling this function on an object will cause the memory block to be | |
1125 | * ignored (not scanned and not reported as a leak). This is usually done when | |
1126 | * it is known that the corresponding block is not a leak and does not contain | |
1127 | * any references to other allocated memory blocks. | |
3c7b4e6b | 1128 | */ |
a6186d89 | 1129 | void __ref kmemleak_ignore(const void *ptr) |
3c7b4e6b CM |
1130 | { |
1131 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1132 | ||
8910ae89 | 1133 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
0c24e061 | 1134 | make_black_object((unsigned long)ptr, false); |
3c7b4e6b CM |
1135 | } |
1136 | EXPORT_SYMBOL(kmemleak_ignore); | |
1137 | ||
a2b6bf63 CM |
1138 | /** |
1139 | * kmemleak_scan_area - limit the range to be scanned in an allocated object | |
1140 | * @ptr: pointer to beginning or inside the object. This also | |
1141 | * represents the start of the scan area | |
1142 | * @size: size of the scan area | |
1143 | * @gfp: kmalloc() flags used for kmemleak internal memory allocations | |
1144 | * | |
1145 | * This function is used when it is known that only certain parts of an object | |
1146 | * contain references to other objects. Kmemleak will only scan these areas | |
1147 | * reducing the number false negatives. | |
3c7b4e6b | 1148 | */ |
c017b4be | 1149 | void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp) |
3c7b4e6b CM |
1150 | { |
1151 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1152 | ||
8910ae89 | 1153 | if (kmemleak_enabled && ptr && size && !IS_ERR(ptr)) |
c017b4be | 1154 | add_scan_area((unsigned long)ptr, size, gfp); |
3c7b4e6b CM |
1155 | } |
1156 | EXPORT_SYMBOL(kmemleak_scan_area); | |
1157 | ||
a2b6bf63 CM |
1158 | /** |
1159 | * kmemleak_no_scan - do not scan an allocated object | |
1160 | * @ptr: pointer to beginning of the object | |
1161 | * | |
1162 | * This function notifies kmemleak not to scan the given memory block. Useful | |
1163 | * in situations where it is known that the given object does not contain any | |
1164 | * references to other objects. Kmemleak will not scan such objects reducing | |
1165 | * the number of false negatives. | |
3c7b4e6b | 1166 | */ |
a6186d89 | 1167 | void __ref kmemleak_no_scan(const void *ptr) |
3c7b4e6b CM |
1168 | { |
1169 | pr_debug("%s(0x%p)\n", __func__, ptr); | |
1170 | ||
8910ae89 | 1171 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
3c7b4e6b | 1172 | object_no_scan((unsigned long)ptr); |
3c7b4e6b CM |
1173 | } |
1174 | EXPORT_SYMBOL(kmemleak_no_scan); | |
1175 | ||
9099daed CM |
1176 | /** |
1177 | * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical | |
1178 | * address argument | |
e8b098fc MR |
1179 | * @phys: physical address of the object |
1180 | * @size: size of the object | |
e8b098fc | 1181 | * @gfp: kmalloc() flags used for kmemleak internal memory allocations |
9099daed | 1182 | */ |
c200d900 | 1183 | void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, gfp_t gfp) |
9099daed | 1184 | { |
8e0c4ab3 PW |
1185 | pr_debug("%s(0x%pa, %zu)\n", __func__, &phys, size); |
1186 | ||
23c2d497 | 1187 | if (PHYS_PFN(phys) >= min_low_pfn && PHYS_PFN(phys) < max_low_pfn) |
8e0c4ab3 PW |
1188 | /* |
1189 | * Create object with OBJECT_PHYS flag and | |
1190 | * assume min_count 0. | |
1191 | */ | |
0c24e061 | 1192 | create_object_phys((unsigned long)phys, size, 0, gfp); |
9099daed CM |
1193 | } |
1194 | EXPORT_SYMBOL(kmemleak_alloc_phys); | |
1195 | ||
1196 | /** | |
1197 | * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a | |
1198 | * physical address argument | |
e8b098fc MR |
1199 | * @phys: physical address if the beginning or inside an object. This |
1200 | * also represents the start of the range to be freed | |
1201 | * @size: size to be unregistered | |
9099daed CM |
1202 | */ |
1203 | void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size) | |
1204 | { | |
0c24e061 PW |
1205 | pr_debug("%s(0x%pa)\n", __func__, &phys); |
1206 | ||
23c2d497 | 1207 | if (PHYS_PFN(phys) >= min_low_pfn && PHYS_PFN(phys) < max_low_pfn) |
0c24e061 | 1208 | delete_object_part((unsigned long)phys, size, true); |
9099daed CM |
1209 | } |
1210 | EXPORT_SYMBOL(kmemleak_free_part_phys); | |
1211 | ||
9099daed CM |
1212 | /** |
1213 | * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical | |
1214 | * address argument | |
e8b098fc | 1215 | * @phys: physical address of the object |
9099daed CM |
1216 | */ |
1217 | void __ref kmemleak_ignore_phys(phys_addr_t phys) | |
1218 | { | |
0c24e061 PW |
1219 | pr_debug("%s(0x%pa)\n", __func__, &phys); |
1220 | ||
23c2d497 | 1221 | if (PHYS_PFN(phys) >= min_low_pfn && PHYS_PFN(phys) < max_low_pfn) |
0c24e061 | 1222 | make_black_object((unsigned long)phys, true); |
9099daed CM |
1223 | } |
1224 | EXPORT_SYMBOL(kmemleak_ignore_phys); | |
1225 | ||
04609ccc CM |
1226 | /* |
1227 | * Update an object's checksum and return true if it was modified. | |
1228 | */ | |
1229 | static bool update_checksum(struct kmemleak_object *object) | |
1230 | { | |
1231 | u32 old_csum = object->checksum; | |
1232 | ||
0c24e061 PW |
1233 | if (WARN_ON_ONCE(object->flags & OBJECT_PHYS)) |
1234 | return false; | |
1235 | ||
e79ed2f1 | 1236 | kasan_disable_current(); |
69d0b54d | 1237 | kcsan_disable_current(); |
6c7a00b8 | 1238 | object->checksum = crc32(0, kasan_reset_tag((void *)object->pointer), object->size); |
e79ed2f1 | 1239 | kasan_enable_current(); |
69d0b54d | 1240 | kcsan_enable_current(); |
e79ed2f1 | 1241 | |
04609ccc CM |
1242 | return object->checksum != old_csum; |
1243 | } | |
1244 | ||
04f70d13 CM |
1245 | /* |
1246 | * Update an object's references. object->lock must be held by the caller. | |
1247 | */ | |
1248 | static void update_refs(struct kmemleak_object *object) | |
1249 | { | |
1250 | if (!color_white(object)) { | |
1251 | /* non-orphan, ignored or new */ | |
1252 | return; | |
1253 | } | |
1254 | ||
1255 | /* | |
1256 | * Increase the object's reference count (number of pointers to the | |
1257 | * memory block). If this count reaches the required minimum, the | |
1258 | * object's color will become gray and it will be added to the | |
1259 | * gray_list. | |
1260 | */ | |
1261 | object->count++; | |
1262 | if (color_gray(object)) { | |
1263 | /* put_object() called when removing from gray_list */ | |
1264 | WARN_ON(!get_object(object)); | |
1265 | list_add_tail(&object->gray_list, &gray_list); | |
1266 | } | |
1267 | } | |
1268 | ||
3c7b4e6b | 1269 | /* |
0b5121ef | 1270 | * Memory scanning is a long process and it needs to be interruptible. This |
25985edc | 1271 | * function checks whether such interrupt condition occurred. |
3c7b4e6b CM |
1272 | */ |
1273 | static int scan_should_stop(void) | |
1274 | { | |
8910ae89 | 1275 | if (!kmemleak_enabled) |
3c7b4e6b CM |
1276 | return 1; |
1277 | ||
1278 | /* | |
1279 | * This function may be called from either process or kthread context, | |
1280 | * hence the need to check for both stop conditions. | |
1281 | */ | |
1282 | if (current->mm) | |
1283 | return signal_pending(current); | |
1284 | else | |
1285 | return kthread_should_stop(); | |
1286 | ||
1287 | return 0; | |
1288 | } | |
1289 | ||
1290 | /* | |
1291 | * Scan a memory block (exclusive range) for valid pointers and add those | |
1292 | * found to the gray list. | |
1293 | */ | |
1294 | static void scan_block(void *_start, void *_end, | |
93ada579 | 1295 | struct kmemleak_object *scanned) |
3c7b4e6b CM |
1296 | { |
1297 | unsigned long *ptr; | |
1298 | unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER); | |
1299 | unsigned long *end = _end - (BYTES_PER_POINTER - 1); | |
93ada579 | 1300 | unsigned long flags; |
a2f77575 | 1301 | unsigned long untagged_ptr; |
3c7b4e6b | 1302 | |
8c96f1bc | 1303 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
3c7b4e6b | 1304 | for (ptr = start; ptr < end; ptr++) { |
3c7b4e6b | 1305 | struct kmemleak_object *object; |
8e019366 | 1306 | unsigned long pointer; |
94f4a161 | 1307 | unsigned long excess_ref; |
3c7b4e6b CM |
1308 | |
1309 | if (scan_should_stop()) | |
1310 | break; | |
1311 | ||
e79ed2f1 | 1312 | kasan_disable_current(); |
6c7a00b8 | 1313 | pointer = *(unsigned long *)kasan_reset_tag((void *)ptr); |
e79ed2f1 | 1314 | kasan_enable_current(); |
8e019366 | 1315 | |
a2f77575 AK |
1316 | untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer); |
1317 | if (untagged_ptr < min_addr || untagged_ptr >= max_addr) | |
93ada579 CM |
1318 | continue; |
1319 | ||
1320 | /* | |
1321 | * No need for get_object() here since we hold kmemleak_lock. | |
1322 | * object->use_count cannot be dropped to 0 while the object | |
1323 | * is still present in object_tree_root and object_list | |
1324 | * (with updates protected by kmemleak_lock). | |
1325 | */ | |
1326 | object = lookup_object(pointer, 1); | |
3c7b4e6b CM |
1327 | if (!object) |
1328 | continue; | |
93ada579 | 1329 | if (object == scanned) |
3c7b4e6b | 1330 | /* self referenced, ignore */ |
3c7b4e6b | 1331 | continue; |
3c7b4e6b CM |
1332 | |
1333 | /* | |
1334 | * Avoid the lockdep recursive warning on object->lock being | |
1335 | * previously acquired in scan_object(). These locks are | |
1336 | * enclosed by scan_mutex. | |
1337 | */ | |
8c96f1bc | 1338 | raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING); |
94f4a161 CM |
1339 | /* only pass surplus references (object already gray) */ |
1340 | if (color_gray(object)) { | |
1341 | excess_ref = object->excess_ref; | |
1342 | /* no need for update_refs() if object already gray */ | |
1343 | } else { | |
1344 | excess_ref = 0; | |
1345 | update_refs(object); | |
1346 | } | |
8c96f1bc | 1347 | raw_spin_unlock(&object->lock); |
94f4a161 CM |
1348 | |
1349 | if (excess_ref) { | |
1350 | object = lookup_object(excess_ref, 0); | |
1351 | if (!object) | |
1352 | continue; | |
1353 | if (object == scanned) | |
1354 | /* circular reference, ignore */ | |
1355 | continue; | |
8c96f1bc | 1356 | raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING); |
94f4a161 | 1357 | update_refs(object); |
8c96f1bc | 1358 | raw_spin_unlock(&object->lock); |
94f4a161 | 1359 | } |
93ada579 | 1360 | } |
8c96f1bc | 1361 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
93ada579 | 1362 | } |
0587da40 | 1363 | |
93ada579 CM |
1364 | /* |
1365 | * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency. | |
1366 | */ | |
dce5b0bd | 1367 | #ifdef CONFIG_SMP |
93ada579 CM |
1368 | static void scan_large_block(void *start, void *end) |
1369 | { | |
1370 | void *next; | |
1371 | ||
1372 | while (start < end) { | |
1373 | next = min(start + MAX_SCAN_SIZE, end); | |
1374 | scan_block(start, next, NULL); | |
1375 | start = next; | |
1376 | cond_resched(); | |
3c7b4e6b CM |
1377 | } |
1378 | } | |
dce5b0bd | 1379 | #endif |
3c7b4e6b CM |
1380 | |
1381 | /* | |
1382 | * Scan a memory block corresponding to a kmemleak_object. A condition is | |
1383 | * that object->use_count >= 1. | |
1384 | */ | |
1385 | static void scan_object(struct kmemleak_object *object) | |
1386 | { | |
1387 | struct kmemleak_scan_area *area; | |
3c7b4e6b | 1388 | unsigned long flags; |
0c24e061 | 1389 | void *obj_ptr; |
3c7b4e6b CM |
1390 | |
1391 | /* | |
21ae2956 UKK |
1392 | * Once the object->lock is acquired, the corresponding memory block |
1393 | * cannot be freed (the same lock is acquired in delete_object). | |
3c7b4e6b | 1394 | */ |
8c96f1bc | 1395 | raw_spin_lock_irqsave(&object->lock, flags); |
3c7b4e6b CM |
1396 | if (object->flags & OBJECT_NO_SCAN) |
1397 | goto out; | |
1398 | if (!(object->flags & OBJECT_ALLOCATED)) | |
1399 | /* already freed object */ | |
1400 | goto out; | |
0c24e061 PW |
1401 | |
1402 | obj_ptr = object->flags & OBJECT_PHYS ? | |
1403 | __va((phys_addr_t)object->pointer) : | |
1404 | (void *)object->pointer; | |
1405 | ||
dba82d94 CM |
1406 | if (hlist_empty(&object->area_list) || |
1407 | object->flags & OBJECT_FULL_SCAN) { | |
0c24e061 PW |
1408 | void *start = obj_ptr; |
1409 | void *end = obj_ptr + object->size; | |
93ada579 CM |
1410 | void *next; |
1411 | ||
1412 | do { | |
1413 | next = min(start + MAX_SCAN_SIZE, end); | |
1414 | scan_block(start, next, object); | |
af98603d | 1415 | |
93ada579 CM |
1416 | start = next; |
1417 | if (start >= end) | |
1418 | break; | |
af98603d | 1419 | |
8c96f1bc | 1420 | raw_spin_unlock_irqrestore(&object->lock, flags); |
af98603d | 1421 | cond_resched(); |
8c96f1bc | 1422 | raw_spin_lock_irqsave(&object->lock, flags); |
93ada579 | 1423 | } while (object->flags & OBJECT_ALLOCATED); |
af98603d | 1424 | } else |
b67bfe0d | 1425 | hlist_for_each_entry(area, &object->area_list, node) |
c017b4be CM |
1426 | scan_block((void *)area->start, |
1427 | (void *)(area->start + area->size), | |
93ada579 | 1428 | object); |
3c7b4e6b | 1429 | out: |
8c96f1bc | 1430 | raw_spin_unlock_irqrestore(&object->lock, flags); |
3c7b4e6b CM |
1431 | } |
1432 | ||
04609ccc CM |
1433 | /* |
1434 | * Scan the objects already referenced (gray objects). More objects will be | |
1435 | * referenced and, if there are no memory leaks, all the objects are scanned. | |
1436 | */ | |
1437 | static void scan_gray_list(void) | |
1438 | { | |
1439 | struct kmemleak_object *object, *tmp; | |
1440 | ||
1441 | /* | |
1442 | * The list traversal is safe for both tail additions and removals | |
1443 | * from inside the loop. The kmemleak objects cannot be freed from | |
1444 | * outside the loop because their use_count was incremented. | |
1445 | */ | |
1446 | object = list_entry(gray_list.next, typeof(*object), gray_list); | |
1447 | while (&object->gray_list != &gray_list) { | |
1448 | cond_resched(); | |
1449 | ||
1450 | /* may add new objects to the list */ | |
1451 | if (!scan_should_stop()) | |
1452 | scan_object(object); | |
1453 | ||
1454 | tmp = list_entry(object->gray_list.next, typeof(*object), | |
1455 | gray_list); | |
1456 | ||
1457 | /* remove the object from the list and release it */ | |
1458 | list_del(&object->gray_list); | |
1459 | put_object(object); | |
1460 | ||
1461 | object = tmp; | |
1462 | } | |
1463 | WARN_ON(!list_empty(&gray_list)); | |
1464 | } | |
1465 | ||
3c7b4e6b CM |
1466 | /* |
1467 | * Scan data sections and all the referenced memory blocks allocated via the | |
1468 | * kernel's standard allocators. This function must be called with the | |
1469 | * scan_mutex held. | |
1470 | */ | |
1471 | static void kmemleak_scan(void) | |
1472 | { | |
1473 | unsigned long flags; | |
04609ccc | 1474 | struct kmemleak_object *object; |
c10a0f87 LY |
1475 | struct zone *zone; |
1476 | int __maybe_unused i; | |
4698c1f2 | 1477 | int new_leaks = 0; |
3c7b4e6b | 1478 | |
acf4968e CM |
1479 | jiffies_last_scan = jiffies; |
1480 | ||
3c7b4e6b CM |
1481 | /* prepare the kmemleak_object's */ |
1482 | rcu_read_lock(); | |
1483 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
8c96f1bc | 1484 | raw_spin_lock_irqsave(&object->lock, flags); |
3c7b4e6b CM |
1485 | #ifdef DEBUG |
1486 | /* | |
1487 | * With a few exceptions there should be a maximum of | |
1488 | * 1 reference to any object at this point. | |
1489 | */ | |
1490 | if (atomic_read(&object->use_count) > 1) { | |
ae281064 | 1491 | pr_debug("object->use_count = %d\n", |
3c7b4e6b CM |
1492 | atomic_read(&object->use_count)); |
1493 | dump_object_info(object); | |
1494 | } | |
1495 | #endif | |
1496 | /* reset the reference count (whiten the object) */ | |
1497 | object->count = 0; | |
1498 | if (color_gray(object) && get_object(object)) | |
1499 | list_add_tail(&object->gray_list, &gray_list); | |
1500 | ||
8c96f1bc | 1501 | raw_spin_unlock_irqrestore(&object->lock, flags); |
3c7b4e6b CM |
1502 | } |
1503 | rcu_read_unlock(); | |
1504 | ||
3c7b4e6b CM |
1505 | #ifdef CONFIG_SMP |
1506 | /* per-cpu sections scanning */ | |
1507 | for_each_possible_cpu(i) | |
93ada579 CM |
1508 | scan_large_block(__per_cpu_start + per_cpu_offset(i), |
1509 | __per_cpu_end + per_cpu_offset(i)); | |
3c7b4e6b CM |
1510 | #endif |
1511 | ||
1512 | /* | |
029aeff5 | 1513 | * Struct page scanning for each node. |
3c7b4e6b | 1514 | */ |
bfc8c901 | 1515 | get_online_mems(); |
c10a0f87 LY |
1516 | for_each_populated_zone(zone) { |
1517 | unsigned long start_pfn = zone->zone_start_pfn; | |
1518 | unsigned long end_pfn = zone_end_pfn(zone); | |
3c7b4e6b CM |
1519 | unsigned long pfn; |
1520 | ||
1521 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
9f1eb38e | 1522 | struct page *page = pfn_to_online_page(pfn); |
3c7b4e6b | 1523 | |
9f1eb38e OS |
1524 | if (!page) |
1525 | continue; | |
1526 | ||
c10a0f87 LY |
1527 | /* only scan pages belonging to this zone */ |
1528 | if (page_zone(page) != zone) | |
3c7b4e6b | 1529 | continue; |
3c7b4e6b CM |
1530 | /* only scan if page is in use */ |
1531 | if (page_count(page) == 0) | |
1532 | continue; | |
93ada579 | 1533 | scan_block(page, page + 1, NULL); |
13ab183d | 1534 | if (!(pfn & 63)) |
bde5f6bc | 1535 | cond_resched(); |
3c7b4e6b CM |
1536 | } |
1537 | } | |
bfc8c901 | 1538 | put_online_mems(); |
3c7b4e6b CM |
1539 | |
1540 | /* | |
43ed5d6e | 1541 | * Scanning the task stacks (may introduce false negatives). |
3c7b4e6b CM |
1542 | */ |
1543 | if (kmemleak_stack_scan) { | |
43ed5d6e CM |
1544 | struct task_struct *p, *g; |
1545 | ||
c4b28963 DB |
1546 | rcu_read_lock(); |
1547 | for_each_process_thread(g, p) { | |
37df49f4 CM |
1548 | void *stack = try_get_task_stack(p); |
1549 | if (stack) { | |
1550 | scan_block(stack, stack + THREAD_SIZE, NULL); | |
1551 | put_task_stack(p); | |
1552 | } | |
c4b28963 DB |
1553 | } |
1554 | rcu_read_unlock(); | |
3c7b4e6b CM |
1555 | } |
1556 | ||
1557 | /* | |
1558 | * Scan the objects already referenced from the sections scanned | |
04609ccc | 1559 | * above. |
3c7b4e6b | 1560 | */ |
04609ccc | 1561 | scan_gray_list(); |
2587362e CM |
1562 | |
1563 | /* | |
04609ccc CM |
1564 | * Check for new or unreferenced objects modified since the previous |
1565 | * scan and color them gray until the next scan. | |
2587362e CM |
1566 | */ |
1567 | rcu_read_lock(); | |
1568 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
8c96f1bc | 1569 | raw_spin_lock_irqsave(&object->lock, flags); |
04609ccc CM |
1570 | if (color_white(object) && (object->flags & OBJECT_ALLOCATED) |
1571 | && update_checksum(object) && get_object(object)) { | |
1572 | /* color it gray temporarily */ | |
1573 | object->count = object->min_count; | |
2587362e CM |
1574 | list_add_tail(&object->gray_list, &gray_list); |
1575 | } | |
8c96f1bc | 1576 | raw_spin_unlock_irqrestore(&object->lock, flags); |
2587362e CM |
1577 | } |
1578 | rcu_read_unlock(); | |
1579 | ||
04609ccc CM |
1580 | /* |
1581 | * Re-scan the gray list for modified unreferenced objects. | |
1582 | */ | |
1583 | scan_gray_list(); | |
4698c1f2 | 1584 | |
17bb9e0d | 1585 | /* |
04609ccc | 1586 | * If scanning was stopped do not report any new unreferenced objects. |
17bb9e0d | 1587 | */ |
04609ccc | 1588 | if (scan_should_stop()) |
17bb9e0d CM |
1589 | return; |
1590 | ||
4698c1f2 CM |
1591 | /* |
1592 | * Scanning result reporting. | |
1593 | */ | |
1594 | rcu_read_lock(); | |
1595 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
8c96f1bc | 1596 | raw_spin_lock_irqsave(&object->lock, flags); |
4698c1f2 CM |
1597 | if (unreferenced_object(object) && |
1598 | !(object->flags & OBJECT_REPORTED)) { | |
1599 | object->flags |= OBJECT_REPORTED; | |
154221c3 VW |
1600 | |
1601 | if (kmemleak_verbose) | |
1602 | print_unreferenced(NULL, object); | |
1603 | ||
4698c1f2 CM |
1604 | new_leaks++; |
1605 | } | |
8c96f1bc | 1606 | raw_spin_unlock_irqrestore(&object->lock, flags); |
4698c1f2 CM |
1607 | } |
1608 | rcu_read_unlock(); | |
1609 | ||
dc9b3f42 LZ |
1610 | if (new_leaks) { |
1611 | kmemleak_found_leaks = true; | |
1612 | ||
756a025f JP |
1613 | pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n", |
1614 | new_leaks); | |
dc9b3f42 | 1615 | } |
4698c1f2 | 1616 | |
3c7b4e6b CM |
1617 | } |
1618 | ||
1619 | /* | |
1620 | * Thread function performing automatic memory scanning. Unreferenced objects | |
1621 | * at the end of a memory scan are reported but only the first time. | |
1622 | */ | |
1623 | static int kmemleak_scan_thread(void *arg) | |
1624 | { | |
d53ce042 | 1625 | static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN); |
3c7b4e6b | 1626 | |
ae281064 | 1627 | pr_info("Automatic memory scanning thread started\n"); |
bf2a76b3 | 1628 | set_user_nice(current, 10); |
3c7b4e6b CM |
1629 | |
1630 | /* | |
1631 | * Wait before the first scan to allow the system to fully initialize. | |
1632 | */ | |
1633 | if (first_run) { | |
98c42d94 | 1634 | signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000); |
3c7b4e6b | 1635 | first_run = 0; |
98c42d94 VN |
1636 | while (timeout && !kthread_should_stop()) |
1637 | timeout = schedule_timeout_interruptible(timeout); | |
3c7b4e6b CM |
1638 | } |
1639 | ||
1640 | while (!kthread_should_stop()) { | |
54dd200c | 1641 | signed long timeout = READ_ONCE(jiffies_scan_wait); |
3c7b4e6b CM |
1642 | |
1643 | mutex_lock(&scan_mutex); | |
3c7b4e6b | 1644 | kmemleak_scan(); |
3c7b4e6b | 1645 | mutex_unlock(&scan_mutex); |
4698c1f2 | 1646 | |
3c7b4e6b CM |
1647 | /* wait before the next scan */ |
1648 | while (timeout && !kthread_should_stop()) | |
1649 | timeout = schedule_timeout_interruptible(timeout); | |
1650 | } | |
1651 | ||
ae281064 | 1652 | pr_info("Automatic memory scanning thread ended\n"); |
3c7b4e6b CM |
1653 | |
1654 | return 0; | |
1655 | } | |
1656 | ||
1657 | /* | |
1658 | * Start the automatic memory scanning thread. This function must be called | |
4698c1f2 | 1659 | * with the scan_mutex held. |
3c7b4e6b | 1660 | */ |
7eb0d5e5 | 1661 | static void start_scan_thread(void) |
3c7b4e6b CM |
1662 | { |
1663 | if (scan_thread) | |
1664 | return; | |
1665 | scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak"); | |
1666 | if (IS_ERR(scan_thread)) { | |
598d8091 | 1667 | pr_warn("Failed to create the scan thread\n"); |
3c7b4e6b CM |
1668 | scan_thread = NULL; |
1669 | } | |
1670 | } | |
1671 | ||
1672 | /* | |
914b6dff | 1673 | * Stop the automatic memory scanning thread. |
3c7b4e6b | 1674 | */ |
7eb0d5e5 | 1675 | static void stop_scan_thread(void) |
3c7b4e6b CM |
1676 | { |
1677 | if (scan_thread) { | |
1678 | kthread_stop(scan_thread); | |
1679 | scan_thread = NULL; | |
1680 | } | |
1681 | } | |
1682 | ||
1683 | /* | |
1684 | * Iterate over the object_list and return the first valid object at or after | |
1685 | * the required position with its use_count incremented. The function triggers | |
1686 | * a memory scanning when the pos argument points to the first position. | |
1687 | */ | |
1688 | static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos) | |
1689 | { | |
1690 | struct kmemleak_object *object; | |
1691 | loff_t n = *pos; | |
b87324d0 CM |
1692 | int err; |
1693 | ||
1694 | err = mutex_lock_interruptible(&scan_mutex); | |
1695 | if (err < 0) | |
1696 | return ERR_PTR(err); | |
3c7b4e6b | 1697 | |
3c7b4e6b CM |
1698 | rcu_read_lock(); |
1699 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
1700 | if (n-- > 0) | |
1701 | continue; | |
1702 | if (get_object(object)) | |
1703 | goto out; | |
1704 | } | |
1705 | object = NULL; | |
1706 | out: | |
3c7b4e6b CM |
1707 | return object; |
1708 | } | |
1709 | ||
1710 | /* | |
1711 | * Return the next object in the object_list. The function decrements the | |
1712 | * use_count of the previous object and increases that of the next one. | |
1713 | */ | |
1714 | static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
1715 | { | |
1716 | struct kmemleak_object *prev_obj = v; | |
1717 | struct kmemleak_object *next_obj = NULL; | |
58fac095 | 1718 | struct kmemleak_object *obj = prev_obj; |
3c7b4e6b CM |
1719 | |
1720 | ++(*pos); | |
3c7b4e6b | 1721 | |
58fac095 | 1722 | list_for_each_entry_continue_rcu(obj, &object_list, object_list) { |
52c3ce4e CM |
1723 | if (get_object(obj)) { |
1724 | next_obj = obj; | |
3c7b4e6b | 1725 | break; |
52c3ce4e | 1726 | } |
3c7b4e6b | 1727 | } |
288c857d | 1728 | |
3c7b4e6b CM |
1729 | put_object(prev_obj); |
1730 | return next_obj; | |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * Decrement the use_count of the last object required, if any. | |
1735 | */ | |
1736 | static void kmemleak_seq_stop(struct seq_file *seq, void *v) | |
1737 | { | |
b87324d0 CM |
1738 | if (!IS_ERR(v)) { |
1739 | /* | |
1740 | * kmemleak_seq_start may return ERR_PTR if the scan_mutex | |
1741 | * waiting was interrupted, so only release it if !IS_ERR. | |
1742 | */ | |
f5886c7f | 1743 | rcu_read_unlock(); |
b87324d0 CM |
1744 | mutex_unlock(&scan_mutex); |
1745 | if (v) | |
1746 | put_object(v); | |
1747 | } | |
3c7b4e6b CM |
1748 | } |
1749 | ||
1750 | /* | |
1751 | * Print the information for an unreferenced object to the seq file. | |
1752 | */ | |
1753 | static int kmemleak_seq_show(struct seq_file *seq, void *v) | |
1754 | { | |
1755 | struct kmemleak_object *object = v; | |
1756 | unsigned long flags; | |
1757 | ||
8c96f1bc | 1758 | raw_spin_lock_irqsave(&object->lock, flags); |
288c857d | 1759 | if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object)) |
17bb9e0d | 1760 | print_unreferenced(seq, object); |
8c96f1bc | 1761 | raw_spin_unlock_irqrestore(&object->lock, flags); |
3c7b4e6b CM |
1762 | return 0; |
1763 | } | |
1764 | ||
1765 | static const struct seq_operations kmemleak_seq_ops = { | |
1766 | .start = kmemleak_seq_start, | |
1767 | .next = kmemleak_seq_next, | |
1768 | .stop = kmemleak_seq_stop, | |
1769 | .show = kmemleak_seq_show, | |
1770 | }; | |
1771 | ||
1772 | static int kmemleak_open(struct inode *inode, struct file *file) | |
1773 | { | |
b87324d0 | 1774 | return seq_open(file, &kmemleak_seq_ops); |
3c7b4e6b CM |
1775 | } |
1776 | ||
189d84ed CM |
1777 | static int dump_str_object_info(const char *str) |
1778 | { | |
1779 | unsigned long flags; | |
1780 | struct kmemleak_object *object; | |
1781 | unsigned long addr; | |
1782 | ||
dc053733 AP |
1783 | if (kstrtoul(str, 0, &addr)) |
1784 | return -EINVAL; | |
189d84ed CM |
1785 | object = find_and_get_object(addr, 0); |
1786 | if (!object) { | |
1787 | pr_info("Unknown object at 0x%08lx\n", addr); | |
1788 | return -EINVAL; | |
1789 | } | |
1790 | ||
8c96f1bc | 1791 | raw_spin_lock_irqsave(&object->lock, flags); |
189d84ed | 1792 | dump_object_info(object); |
8c96f1bc | 1793 | raw_spin_unlock_irqrestore(&object->lock, flags); |
189d84ed CM |
1794 | |
1795 | put_object(object); | |
1796 | return 0; | |
1797 | } | |
1798 | ||
30b37101 LR |
1799 | /* |
1800 | * We use grey instead of black to ensure we can do future scans on the same | |
1801 | * objects. If we did not do future scans these black objects could | |
1802 | * potentially contain references to newly allocated objects in the future and | |
1803 | * we'd end up with false positives. | |
1804 | */ | |
1805 | static void kmemleak_clear(void) | |
1806 | { | |
1807 | struct kmemleak_object *object; | |
1808 | unsigned long flags; | |
1809 | ||
1810 | rcu_read_lock(); | |
1811 | list_for_each_entry_rcu(object, &object_list, object_list) { | |
8c96f1bc | 1812 | raw_spin_lock_irqsave(&object->lock, flags); |
30b37101 LR |
1813 | if ((object->flags & OBJECT_REPORTED) && |
1814 | unreferenced_object(object)) | |
a1084c87 | 1815 | __paint_it(object, KMEMLEAK_GREY); |
8c96f1bc | 1816 | raw_spin_unlock_irqrestore(&object->lock, flags); |
30b37101 LR |
1817 | } |
1818 | rcu_read_unlock(); | |
dc9b3f42 LZ |
1819 | |
1820 | kmemleak_found_leaks = false; | |
30b37101 LR |
1821 | } |
1822 | ||
c89da70c LZ |
1823 | static void __kmemleak_do_cleanup(void); |
1824 | ||
3c7b4e6b CM |
1825 | /* |
1826 | * File write operation to configure kmemleak at run-time. The following | |
1827 | * commands can be written to the /sys/kernel/debug/kmemleak file: | |
1828 | * off - disable kmemleak (irreversible) | |
1829 | * stack=on - enable the task stacks scanning | |
1830 | * stack=off - disable the tasks stacks scanning | |
1831 | * scan=on - start the automatic memory scanning thread | |
1832 | * scan=off - stop the automatic memory scanning thread | |
1833 | * scan=... - set the automatic memory scanning period in seconds (0 to | |
1834 | * disable it) | |
4698c1f2 | 1835 | * scan - trigger a memory scan |
30b37101 | 1836 | * clear - mark all current reported unreferenced kmemleak objects as |
c89da70c LZ |
1837 | * grey to ignore printing them, or free all kmemleak objects |
1838 | * if kmemleak has been disabled. | |
189d84ed | 1839 | * dump=... - dump information about the object found at the given address |
3c7b4e6b CM |
1840 | */ |
1841 | static ssize_t kmemleak_write(struct file *file, const char __user *user_buf, | |
1842 | size_t size, loff_t *ppos) | |
1843 | { | |
1844 | char buf[64]; | |
1845 | int buf_size; | |
b87324d0 | 1846 | int ret; |
3c7b4e6b CM |
1847 | |
1848 | buf_size = min(size, (sizeof(buf) - 1)); | |
1849 | if (strncpy_from_user(buf, user_buf, buf_size) < 0) | |
1850 | return -EFAULT; | |
1851 | buf[buf_size] = 0; | |
1852 | ||
b87324d0 CM |
1853 | ret = mutex_lock_interruptible(&scan_mutex); |
1854 | if (ret < 0) | |
1855 | return ret; | |
1856 | ||
c89da70c | 1857 | if (strncmp(buf, "clear", 5) == 0) { |
8910ae89 | 1858 | if (kmemleak_enabled) |
c89da70c LZ |
1859 | kmemleak_clear(); |
1860 | else | |
1861 | __kmemleak_do_cleanup(); | |
1862 | goto out; | |
1863 | } | |
1864 | ||
8910ae89 | 1865 | if (!kmemleak_enabled) { |
4e4dfce2 | 1866 | ret = -EPERM; |
c89da70c LZ |
1867 | goto out; |
1868 | } | |
1869 | ||
3c7b4e6b CM |
1870 | if (strncmp(buf, "off", 3) == 0) |
1871 | kmemleak_disable(); | |
1872 | else if (strncmp(buf, "stack=on", 8) == 0) | |
1873 | kmemleak_stack_scan = 1; | |
1874 | else if (strncmp(buf, "stack=off", 9) == 0) | |
1875 | kmemleak_stack_scan = 0; | |
1876 | else if (strncmp(buf, "scan=on", 7) == 0) | |
1877 | start_scan_thread(); | |
1878 | else if (strncmp(buf, "scan=off", 8) == 0) | |
1879 | stop_scan_thread(); | |
1880 | else if (strncmp(buf, "scan=", 5) == 0) { | |
54dd200c YX |
1881 | unsigned secs; |
1882 | unsigned long msecs; | |
3c7b4e6b | 1883 | |
54dd200c | 1884 | ret = kstrtouint(buf + 5, 0, &secs); |
b87324d0 CM |
1885 | if (ret < 0) |
1886 | goto out; | |
54dd200c YX |
1887 | |
1888 | msecs = secs * MSEC_PER_SEC; | |
1889 | if (msecs > UINT_MAX) | |
1890 | msecs = UINT_MAX; | |
1891 | ||
3c7b4e6b | 1892 | stop_scan_thread(); |
54dd200c YX |
1893 | if (msecs) { |
1894 | WRITE_ONCE(jiffies_scan_wait, msecs_to_jiffies(msecs)); | |
3c7b4e6b CM |
1895 | start_scan_thread(); |
1896 | } | |
4698c1f2 CM |
1897 | } else if (strncmp(buf, "scan", 4) == 0) |
1898 | kmemleak_scan(); | |
189d84ed CM |
1899 | else if (strncmp(buf, "dump=", 5) == 0) |
1900 | ret = dump_str_object_info(buf + 5); | |
4698c1f2 | 1901 | else |
b87324d0 CM |
1902 | ret = -EINVAL; |
1903 | ||
1904 | out: | |
1905 | mutex_unlock(&scan_mutex); | |
1906 | if (ret < 0) | |
1907 | return ret; | |
3c7b4e6b CM |
1908 | |
1909 | /* ignore the rest of the buffer, only one command at a time */ | |
1910 | *ppos += size; | |
1911 | return size; | |
1912 | } | |
1913 | ||
1914 | static const struct file_operations kmemleak_fops = { | |
1915 | .owner = THIS_MODULE, | |
1916 | .open = kmemleak_open, | |
1917 | .read = seq_read, | |
1918 | .write = kmemleak_write, | |
1919 | .llseek = seq_lseek, | |
5f3bf19a | 1920 | .release = seq_release, |
3c7b4e6b CM |
1921 | }; |
1922 | ||
c89da70c LZ |
1923 | static void __kmemleak_do_cleanup(void) |
1924 | { | |
2abd839a | 1925 | struct kmemleak_object *object, *tmp; |
c89da70c | 1926 | |
2abd839a CM |
1927 | /* |
1928 | * Kmemleak has already been disabled, no need for RCU list traversal | |
1929 | * or kmemleak_lock held. | |
1930 | */ | |
1931 | list_for_each_entry_safe(object, tmp, &object_list, object_list) { | |
1932 | __remove_object(object); | |
1933 | __delete_object(object); | |
1934 | } | |
c89da70c LZ |
1935 | } |
1936 | ||
3c7b4e6b | 1937 | /* |
74341703 CM |
1938 | * Stop the memory scanning thread and free the kmemleak internal objects if |
1939 | * no previous scan thread (otherwise, kmemleak may still have some useful | |
1940 | * information on memory leaks). | |
3c7b4e6b | 1941 | */ |
179a8100 | 1942 | static void kmemleak_do_cleanup(struct work_struct *work) |
3c7b4e6b | 1943 | { |
3c7b4e6b | 1944 | stop_scan_thread(); |
3c7b4e6b | 1945 | |
914b6dff | 1946 | mutex_lock(&scan_mutex); |
c5f3b1a5 | 1947 | /* |
914b6dff VM |
1948 | * Once it is made sure that kmemleak_scan has stopped, it is safe to no |
1949 | * longer track object freeing. Ordering of the scan thread stopping and | |
1950 | * the memory accesses below is guaranteed by the kthread_stop() | |
1951 | * function. | |
c5f3b1a5 CM |
1952 | */ |
1953 | kmemleak_free_enabled = 0; | |
914b6dff | 1954 | mutex_unlock(&scan_mutex); |
c5f3b1a5 | 1955 | |
c89da70c LZ |
1956 | if (!kmemleak_found_leaks) |
1957 | __kmemleak_do_cleanup(); | |
1958 | else | |
756a025f | 1959 | pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n"); |
3c7b4e6b CM |
1960 | } |
1961 | ||
179a8100 | 1962 | static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup); |
3c7b4e6b CM |
1963 | |
1964 | /* | |
1965 | * Disable kmemleak. No memory allocation/freeing will be traced once this | |
1966 | * function is called. Disabling kmemleak is an irreversible operation. | |
1967 | */ | |
1968 | static void kmemleak_disable(void) | |
1969 | { | |
1970 | /* atomically check whether it was already invoked */ | |
8910ae89 | 1971 | if (cmpxchg(&kmemleak_error, 0, 1)) |
3c7b4e6b CM |
1972 | return; |
1973 | ||
1974 | /* stop any memory operation tracing */ | |
8910ae89 | 1975 | kmemleak_enabled = 0; |
3c7b4e6b CM |
1976 | |
1977 | /* check whether it is too early for a kernel thread */ | |
8910ae89 | 1978 | if (kmemleak_initialized) |
179a8100 | 1979 | schedule_work(&cleanup_work); |
c5f3b1a5 CM |
1980 | else |
1981 | kmemleak_free_enabled = 0; | |
3c7b4e6b CM |
1982 | |
1983 | pr_info("Kernel memory leak detector disabled\n"); | |
1984 | } | |
1985 | ||
1986 | /* | |
1987 | * Allow boot-time kmemleak disabling (enabled by default). | |
1988 | */ | |
8bd30c10 | 1989 | static int __init kmemleak_boot_config(char *str) |
3c7b4e6b CM |
1990 | { |
1991 | if (!str) | |
1992 | return -EINVAL; | |
1993 | if (strcmp(str, "off") == 0) | |
1994 | kmemleak_disable(); | |
ab0155a2 JB |
1995 | else if (strcmp(str, "on") == 0) |
1996 | kmemleak_skip_disable = 1; | |
1997 | else | |
3c7b4e6b CM |
1998 | return -EINVAL; |
1999 | return 0; | |
2000 | } | |
2001 | early_param("kmemleak", kmemleak_boot_config); | |
2002 | ||
2003 | /* | |
2030117d | 2004 | * Kmemleak initialization. |
3c7b4e6b CM |
2005 | */ |
2006 | void __init kmemleak_init(void) | |
2007 | { | |
ab0155a2 JB |
2008 | #ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF |
2009 | if (!kmemleak_skip_disable) { | |
2010 | kmemleak_disable(); | |
2011 | return; | |
2012 | } | |
2013 | #endif | |
2014 | ||
c5665868 CM |
2015 | if (kmemleak_error) |
2016 | return; | |
2017 | ||
3c7b4e6b CM |
2018 | jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE); |
2019 | jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000); | |
2020 | ||
2021 | object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE); | |
2022 | scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE); | |
3c7b4e6b | 2023 | |
298a32b1 CM |
2024 | /* register the data/bss sections */ |
2025 | create_object((unsigned long)_sdata, _edata - _sdata, | |
2026 | KMEMLEAK_GREY, GFP_ATOMIC); | |
2027 | create_object((unsigned long)__bss_start, __bss_stop - __bss_start, | |
2028 | KMEMLEAK_GREY, GFP_ATOMIC); | |
2029 | /* only register .data..ro_after_init if not within .data */ | |
b0d14fc4 | 2030 | if (&__start_ro_after_init < &_sdata || &__end_ro_after_init > &_edata) |
298a32b1 CM |
2031 | create_object((unsigned long)__start_ro_after_init, |
2032 | __end_ro_after_init - __start_ro_after_init, | |
2033 | KMEMLEAK_GREY, GFP_ATOMIC); | |
3c7b4e6b CM |
2034 | } |
2035 | ||
2036 | /* | |
2037 | * Late initialization function. | |
2038 | */ | |
2039 | static int __init kmemleak_late_init(void) | |
2040 | { | |
8910ae89 | 2041 | kmemleak_initialized = 1; |
3c7b4e6b | 2042 | |
282401df | 2043 | debugfs_create_file("kmemleak", 0644, NULL, NULL, &kmemleak_fops); |
b353756b | 2044 | |
8910ae89 | 2045 | if (kmemleak_error) { |
3c7b4e6b | 2046 | /* |
25985edc | 2047 | * Some error occurred and kmemleak was disabled. There is a |
3c7b4e6b CM |
2048 | * small chance that kmemleak_disable() was called immediately |
2049 | * after setting kmemleak_initialized and we may end up with | |
2050 | * two clean-up threads but serialized by scan_mutex. | |
2051 | */ | |
179a8100 | 2052 | schedule_work(&cleanup_work); |
3c7b4e6b CM |
2053 | return -ENOMEM; |
2054 | } | |
2055 | ||
d53ce042 SK |
2056 | if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) { |
2057 | mutex_lock(&scan_mutex); | |
2058 | start_scan_thread(); | |
2059 | mutex_unlock(&scan_mutex); | |
2060 | } | |
3c7b4e6b | 2061 | |
0e965a6b QC |
2062 | pr_info("Kernel memory leak detector initialized (mem pool available: %d)\n", |
2063 | mem_pool_free_count); | |
3c7b4e6b CM |
2064 | |
2065 | return 0; | |
2066 | } | |
2067 | late_initcall(kmemleak_late_init); |